Home

César Emanuel Brandão Gomes Monitorização Fina do - Serv-CPS

1. Appliance Database Figure 4 7 Preliminary Implementation of the NILM system Single Appliance Mains Appliance ON OFF ON OFF Oven 13 13 17 20 Refrigerator 173 179 165 128 Table 4 2 Comparison of detection in single appliance files to detection in aggregated load mains file Analysing table 4 2 one can see that the number of ON OFF events detected are close but not equal These errors arise from the low sampling frequencies on the power flows and the big number of appliances on the mains file That two characteristics result in overlapping events that are seen as noise by the system causing the miss of detections of some events Figure 4 8 represents identification of appliances in the mains file by the JAVA routine the power flow of total load and the power flows of individual loads In figure 4 8alit is represented the identification of a refrigerator in the total load The refrigerator is identified well even when there are other appliance events happening along the function of the refrigerator but it is also seen a false identification caused by a signature very close to that of the refrigerator In figure 4 8b it is represented the identification of an oven The same observations made for 48 Refrigerator Indentification 2000 Mains Power Refrigerator Power 1500 Refrigerator Steps States 1000 w o a 500 0 6 82 6 84 6 86 6 88 6 9 6 9
2. 2 4 2 Ema Ice o P t 2 5 where Phorm is the power that would be spent by a linear load with admittance Y t if the voltage supply is stable at its nominal voltage Vrorm 230 Vrms in Portugal Another model to best fit the normalization should arise by the non linearity of some appliances 18 but this analysis is not relevant for the project For better understanding of the concepts here stated it is presented in figure 2 3 a signa ture space in real and reactive power of the steady state signatures of some appliances Harmonic frequency signatures Assuming that the loads are linear it is expected that if the supplied voltage waveform is sinusoidal the response current should also be si nusoidal But as one may observe this is false for many appliances being them non linear in this aspect This non linearity may express itself as the presence of some power in some harmonics of the base frequency It is possible to consider that inductive motors may have tri angular current waveform containing power in third fifth and other low order odd harmonics 50Hz in Portugal SIGNATURE SPACE ACTON HOUSE 1 708 688 Dehumidifier 5 a E see ater Pump split f 490 d x u O 302 a W gt 289 Ice Maker 5 Bar E 108 raptor x O Light ent Fan o O Rum 188 a 258 588 758 1000 1258 REAL POWER lt W Figure 2 3 Steady State complex power signature space of some household a
3. Residual Energy 1515 048 5358 200 i 0 000421 0 001488 i Table 5 11 Energy Values by integration of Power Flows 82 The values of Measured Energy and Estimated Energy are very close and consequently the Residual Energy is low In terms of absolute ratio the Residual Energy is 0 143 of the Measured Energy This value is the error of the system This result is very good as a measure of the completeness and accuracy of the developed system 5 6 Summary This chapter addressed the evaluation of the system s correctness and accuracy It started by presenting the test bed that would allow the system s test and its calibration Tests were performed on some individual components of the system the Step Detector the Appliance Classifier and the Indirect Power Sensing WSN which results have shown their correct im plementation To finalize the system was tested as a whole and results were presented in order to provide indicators of completeness and correct function It has been verified that the system detects the appliances with great success and that the mean error on the energy by integration of power was 0 143 83 84 Chapter 6 Conclusions and Future Work 6 1 Conclusions A system capable of monitoring a household electrical installation was successfully im plemented The system disambiguates and identifies individual appliance loads within the aggregated load recurring to NIALM techniques The proble
4. ADE7953 Single Phase Multifunction Metering IC with Neutral Current Measurement Rev 0 Analog Devices Inc 2011 P Bartolomeu and V Baptista uMRFs a tiny wireless node 0 1 Micro I O servi os de electr nica Lda Oct 2011 CARACTERIZACAO DA PROCURA DE ENERGIA ELECTRICA EM 2011 ERSE Entidade Reguladora dos Servicos Energ ticos Dec 2010 ChipKIT Max32 Reference Manual Digilent 2011 CR8100 series Split Core Current Transformer CR Magnetics Inc Current Sense Resistors Application Note TTelectronics Current to Current Transformers MGC 1000 Series Magnelab Inc Current Transformers amp Sensors Series PREMO 2008 Feb 2012 Digi Key Corporation Online Available http www digikey com Energy Monitoring Pictail Plus Daughter Board User s Guide Microchip Technology Inc 2011 Evaluation Board User Guide UG 194 Evaluation Board for the ADE7953 Single Phase Energy Metering IC Rev 0 Analog Devices Inc 2011 Frequently Asked Questions FAQs Analog Devices Energy ADE Products Application Note AN 639 Analog Devices 2009 FT232R USB UART IC Datasheet Future Technology Devices International Ltd 2010 C Gomes T Cant o and R Ribeiro Monitoring System Report for a project devel oped in the course of Redes de Comunicac o em Ambiente Industrial Communication Networks in Industrial environment 2011 2012 Feb 2012 G Hart Nonintrusive appliance load monitoring Pr
5. Graphs Tab The graphs tab shows real time traces of RMS Voltage and RMS Current as can be seen in figure It is also shown real time traces of the components of power As it can be seen in figure 4 19 both the measured total power blue trace and the estimated power red trace are present All the traces are updated at 50 Hz 59 Vrms Irms 238 117 Stored Data ir zus ss 2312 Vrms 38 2278 77 211 92 V p 2 i 5 2211 E 55 Irms 217 6 44 8 041 A 2142 33 2108 A Active Power 22 4 ws 141 Measured Estimated 204 a 2007 W 2017 W are 1430 1560 1690 1820 950 2080 2210 2340 1430 1560 1690 1820 asa 2080 2210 2340 Reactive Power TimeC20ms TimeC20ms Measured Estimated Active Power Reactive Power VAR 047 VAR 052 W 2970 2700 Event p 8 280 E 51 Lamp1 OFF 2000 2 xl g A 1890 3 Latest Events E Y 3 1620 o be Wed 25 July 21 59 33 Heater1 ON 1960 E 2 Wed 25 July 21 59 35 Heater1 OFF 1080 Wed 25 July 21 59 37 Heater2 ON gt 34 7 Wed 25 July 21 59 40 Heater1 ON st T Wed 25 July 21 59 44 Fan heater ON so 68 Wed 25 July 21 59 46 Fan_heater OFF 270 25 Wed 25 July 21 59 49 Lamp1 ON ol 1430 1560 1690 1820 1950 2080 2210 2340 1430 1560 1690 1820 1950 2080 2210 2340 Wed 25 July 21 59 50 Lamp1 OFF TimeC20m8 Timec20m9 Figure 4 17
6. 4 3 4 4 8 Graphical User Interface The Graphical User Interface GUI is designed to provide to the user all the system s data of interest being it real time and historical The GUI is connected to the database fetching the the following data about the electric circuit in e RMS Voltage e RMS Current e Normalized Complex Power in active and reactive components It also fetches the following the computed real time data e Expected Power in the active and reactive components e Residual Power in active and reactive components The GUI should also fetch the historical and the computed statistics data but that isn t implemented and should be considered for short term future work Apart from that the GUI listens for the events of the appliances changing states as well as it fetches information about the appliance database A representation of the data transactions of the GUI is represented in figure 58 VRMS Circuit ARMS gq AA Ame Power RT DATA GUI Estimated P Beulen Residual Power Appliances Name Stat Signatures Appliance Database Appliance Changing Stat Figure 4 16 Graphical User Interface entity Figure represents the default aspect of the GUI The GUI is divided in two tabs The Graphs tab is oriented to display the data about the electric installation and plotting the graphs of interest The Apps tab displays the data about the appliances in the circuit
7. A survey of the current sensing technologies is presented in appendix from where tableB 2 is a summary The Rogowski Coil Sensor is the most promising and accurate sensor but the availability of this technology for the general market is scarce So the selected solution is the Current Transformer which presents the best compromise between safety and support as it s the one of the most used technologies in energy metering 18 Shunt Current Hall Effect Rogowski Transformer Sensor Coil Cost Low Medium High Bandwidth Low Medium High High Isolation Low High High High Linearity Good Good Medium Very Good High Current Measurements Low Good Good Very Good Power Dissipation Medium Low Low Low Temperature Drift Medium Low Medium Very Low Saturation No Yes Yes No Table 3 2 Comparative of the various solutions for current measurment Equipment Selection Having chosen the type of sensor it was acquired the Current Transformer CR8349 1500 from CR Magnetics Inc shown in figure This sensor is a general purpose vertical pcb mount current transformer with small footprint It s main characteristics are e maximum current to be linearly sensed 1 75 A e maximum saturation output voltage Vmax 12 8 Vms e frequency range 20 1K Hz 3 2 2 ADE7953 Theory of Operation Figure 3 2 CR8349 Current Transformer 3 2 3 Data Acquisition System Deployment Figure represents a block
8. Figure 4 10 Data Acquisition Logical Entity The data acquisition entity depicted in figure 4 10 is responsible of communicating with the data acquisition system already described Its basic function is to read and write registers from the ADE7953 The data in the registers doesn t have a direct physical meaning thus it is considered to be raw data So this part of the system is also accountable for converting raw data into meaningful physical values i e Volts Amperes Watts Watt Hour etc The data acquisition layer obtains repeatedly at a frequency of 50 Hz the following values e Vrms RMS Voltage e Irms RMS Current e P Complex Power it is composed by two values active power and reactive power The values of Vrms and P are used to compute a sample of normalized power at the same frequency as of that of the sampling of those values Alternatively to the sampling of the values of power consumption from the data acquisition system there is also the possibility of providing the system with data recorded previously by the system by the data logging entity All the obtained and computed values are timestamped and pushed into the database 4 4 2 Step Detection The Step Detection entity depicted in figure 4 11 takes the samples of normalized power from the logical database and sequences the flow in steady periods and changing periods Then identifies step change events in the power flow that are the difference of the average
9. Heater1 ON Wed 25 July 21 59 35 Heater1 OFF Wed 25 July 21 59 37 Heater2 ON Wed 25 July 21 59 40 Heater1 ON Wed 25 July 21 59 44 Fan_heater ON Wed 25 July 21 59 46 Fan_heater OFF Wed 25 July 21 59 49 Lamp1 ON Wed 25 July 21 59 50 Lamp1 OFF Figure 4 20 Informations about the electrical installation Apps Tab The Apps Tab was designed to provide visual information about the appliances Not only the current state but some historical data and other calculations should be displayed But in the current implementation it only shows the appliances in the system and their detected states Figure 4 21 shows the Apps Tab in operation As one may see there are three types of states for appliances ON and OFF are self explanatory The Undetermined state is the default for the appliances at start up Since the appliances states are detected at their changes if no state change is detected one should not infer the appliance s state although it may be possible for special cases such as no power being drawn which means all appliances are OFF their state is left Undetermined 61 One feature that was projected to have but is missing is that when clicked on any of the appliances historical statistics and data on appliances power consumption such as the times of operation and the energy cost of its usage was to be shown This feature should be considered for future short term implementation Graphs Apps
10. Tolerance 5 State OFF y Database Update DATABASE Update iance Database Appliance State Appliance Database Appliance Event Signal Figure 4 12 Appliance Classification Process 4 4 4 Data Logging Despite the system being designed for real time monitoring not all data expected to be produced is real time data Historical data for statistics and other calculations are of the highest importance There is also interest in improving the system by the development of more sophisticated algorithms To satisfy these requirements it was developed a Data Logging module that records the data that may be of interest to further experiences and processing Figure represents 53 the data logging entity The entities recorded in the database are e Data Acquisition Vrms Irms and Normalized Power Step Detection steady states and steps e Appliance Classification events associated with every appliance individually in separate files and all appliances together in a single file Estimated Power the power resulting from the sum of power being spent by all the appliances detected to be ON e Indirect Power Sensing WSN all the detected changes of state of the monitored appliances Indirect Power a Sensing WSN lt x Las co Data a Acquisition i VRMS Sensors Events x IRMS O Power ronan a States Steps Appliance Appliance Detections Events Classification Fi
11. UART Protocol Taking advantage of the software developed to test the UART interface the protocol here developed is very similar to the one in ADE7953 s datasheet 4 but without the time and baudrate constraints Communication is byte oriented and each field in following figures represent a byte The protocol is constituted by 3 types of messages frames Write This message represented in figure F 1 is sent to gateway to issue a write in the ADE7953 SFD and EFD are Start of Frame Delimiter and End of Frame Delimiter and are used for frame consistency check and synchronization Next there isa WRITE command to signalize that the PC is issuing a write Next there are the 2 bytes of the register address MSB first After the address comes the data which the number of bytes is dependant of the register to write The data is sent LSB first SFD WRITE ADDR_MSB ADDR_LSB DATA_LSB gt DATA_MSB EFD Figure F 1 UART Write Message Read This message represented in figure is sent to gateway to issue a read of a register of ADE7953 Again SFD and EFD are used for the same as before Following comes a byte of command that signalizes the gateway that the PC is issuing a read of the register which address follows as in previous message After this message there will be a reply message with the data from the register addressed SFD READ ADDR_MSB ADDR_LSB EFD Figure F 2 UART Rea
12. Photojoy 20 luz which falls around living room lighting 25 The tuning isn t very good but that can easily be arranged by changing R to a more suitable value It is here suggested that changing R to a smaller value in order to be able to raise the thresholds as they may be too low for normal use Taking into account that the sensors would be placed near the lamps the light intensity is much greater than the assumed A final assembled version of the developed circuit device can be seen in figure D 3 112 Figure D 3 Assembled Sensor Circuitry 113 114 Appendix E Current and Magnetic Field The Ampere s Law that states The line integral of the magnetic field intensity H about any closed path is equal to the current enclosed within that path 26 That can be stated mathematically as f HedL I 4 E 1 which for a infinite round conductor as in figure can derive the relationship of the equation E 2 between the current and the magnetic flux B where B uH and u is the magnetic permeability of the material Concentric Magnetic Conducting Wire Field Vectors H Conductor lt A P Dann 7 u u vvy Hal y di Current Eu DE e b Magnetic Fields around a current car tions rying conductor Figure E 1 Current carrying conductor Ber E 2 Orr r is the radius of the equi magnetic flux circumference to the center of the conductor 115 116 Appendix F
13. 0 2r r cos 9 r amp r sen Y 27 r cos 0 r amp r sen 0 pi a The first term of the equation corresponds to the component of one of the conductors and it is additive The second term corresponds to the component of the other conductor and it is subtractive because the current flow is in the inverse direction and so the field is symmetric to the first Magnetic Flux vs Angle 90 180 a b Figure 3 16 Magnetic Flux around a two Conductor Cable la Cut section of a generic two conductor cable fb Magnetic Flux vs Angle From the plot one can infer that the function maximum is obtained int 6 0 180 with the value Ip Ip T 2r r r 2r r ri Bmax and is minimum at 9 90 270 with the value Bmin 0 T since at that points the contribution of the magnetic flux of each conductor is equal in norm but with inverse directions and so they cancel themselves If a hall effect sensor is to be attached to a power chord of an appliance it should be pos sible to extract information about the current being carried in that conductor This method 33 can be used to detect and measure the current flowing in the chord but the uncertainty in the cable placement and the possible sensor position drift makes it only viable to detect if the appliance is ON or OFF In conclusion if one can find a sensor with sufficient sensitivity to sense magnetic fluxes around a conducto
14. 4 All the registers are addressed with a 2 byte value The most significant byte of the address contains an indication of the register s length being the registers length equal to register length address s MSB 1 x 8 bits 3 1 The registers lengths and the corresponding address format are represented in table 3 3 where X means don t care 21 Register Length Address 8 bit OxOXX 16 bit Ox1XX 24 bit 0x2XX 32 bit 0x3XX Table 3 3 Register s adressing scheme The communication interfaces are three e 4 pin SPI interface e 2 pin bidirectional I C interface e 2 pin UART interface 4 Despite being carried out v a different interfaces the steps necessary to access the registers are generically the same consisting in read or write one register at a time in an ask reply fashion Figure 3 5 shows a minimal representation of a data transaction Command Figure 3 5 Minimal data transaction 3 3 2 ADE IO INTERFACE BOARD Z To develop the system it is necessary to implement communication between the acquisi tion system and a PC The ADE7958 Evaluation Board already comes with an interface to communicate with a PC via USB emulating a Virtual COMM That interface is done with a independent board the ADE IO INTERFACE BOARD Z that connects to interface de picted in figure 3 3 This PC interface was developed to communicate directly to proprietary software in LabVIEW an
15. 500 1000 1500 Active Power W Figure 5 4 Appliance Classification in Q plane 77 Reactive Power VAR Reactive Power VAR Reactive Power VAR Heaterl ON ape Pied 770 780 790 800 81 820 830 840 Active Power W Heater2 ON T T er T 7 a u 20 a u 4 a NI 20 if te et 40h N P 80 A 4 1160 1180 1200 1220 1240 1260 Active Power W 7 T 7 BE it 4 O amp AK A T NZ 35 40 o 50 55 Active Power W Reactive Power VAR 8 o w m 8 5 a Heaterl OFF Te B 0 4 Active Power W Heater2 OFF T r r sol Fi 4 60 F 4 2 T 40 F E 5 E a rro o tee 2 E o E E a gt Na 20 a 7 4 1260 1240 20 1200 1180 1160 Active Power W Fan OFF T r r 7 14 a inc O 4 ae S 12 4 10 4 E 6 x sf o woe 5h en Ez E E 55 DD von v 45 Active Power W Figure 5 5 Appliances Individual Interval of identification a 78 Fan Heater ON Fan Heater OF a e T T T T T T ern T gt E 7 BE oOo A Rs Reactive Power VAR 2 E Du gt f Reactive Power VAR a E N a 0 Ns von gt en E T T T NL L 504 Pn E E cd ob ee p Fi L 1 AA rl 1 1 1 1 1 1 li
16. Feature Summary It was carried out a market survey on meter equipment The complete results are reported in appendix A while table 3 1 presents a brief summary and comparison of the solutions found Analysing the data in table is possible to conclude that the viable equipments are Analog Devices ADE7953 EVM Evaluation Board and Microchip s MCP3909 dsPIC33F 17 Analog Devices Microchip s Maxim 78M6613 Energy Monitorin System renin EVM Pe EVMs ei delas E J Vrms yes yes yes yes yes Irms yes yes yes yes yes M asurments Active Power yes yes yes yes yes Reactive Power yes yes yes yes yes Power Factor yes yes yes yes no Line Frequency yes yes Not Specified yes yes Resolution 24 bits 16 bits 22 bits 16 bit 16 bit A D Performance Measurments 2nd to the 31st Sampling harmonic 16hz for Availability 1 23 kHz measurments Not Specified 1hz 50 hz Communications Communications USB SPI 12C UART USB UART USB UART MODBUS RS485 USB UART Current Sensin Provided E None Une ar Ante Shunt Resistor None Shunt Resitor Current Sensing Current Transformer Shunt Current Transformer Current Sensing Resistor di dt Shunt Resistor Available sensor Current TransformerCurrent Transformer Shunt Resitor Table 3 1 Systems main features 3 Phase Meter Reference Design The reasons for this statement are they both are full functional systems removing all the designing and prototyping work the
17. From figure 8lit can be seen that Estimated Power follows 3000 2000 1000 Active Power W Measured Power vs Estimated Power T 500 1000 1500 Time seconds 2000 2500 3000 Measured Power Estimated Power Residual Power Reactive Power VAR L 500 1000 1500 Time seconds Figure 5 8 Comparison of Measured Power with Estimated Power the Measured Power very closely The Residual Power is most of the time close to zero The peaks are due to the delay on the detection A characterization of the error of the system can be quantified by the norm of the Residual Power compared with the Measured Power Value Mean V A Standard Deviation V A Norm of Measured Power 1306 44 834 61 Norm of Estimated Power 1305 96 836 84 Norm of Residual Power 20 91 96 15 Table 5 10 Comparison of Norms of Power Values From table it can be seen that the Measured Power and the Estimated Power values are very close and consequently the Residual Power value is low Knowing the values of power it is straightforward to get the values of energy by integra tion Table shows the energy of the different power flows Value Energy J Energy kWh Measured Energy 3883990 521 63343 642 i 1 078886 0 017595 i Estimated Energy 3882473 341 68701 747 i 1 078465 0 019084 i
18. Here it will be discussed three classes of appliances models e ON OFF e Finite State Machine FSM e Continuously Variable ON OFF model The ON OFF model is the simplest model assuming the form of a Boolean Switch Function allowing an appliance to be either ON or OFF at a given time t This model suits almost all appliances in a household environment with great accuracy FSM model The Finite State Machine Model FSM model is suitable to appliances with more than two different ON states 1 e states of function which have different power consumption as it allows for an arbitrary number of states and state transitions Some representations of FSM models for common appliances are shown in figure The circles denote the states identified by name and operating power The arcs denote the allowed state transitions and are labelled with the correspondent signature change in power that occurs in that state transition For simplicity power is only represented with scalar real power but it is here advised that following in this document power will be identified with a vector of real and imaginary power i e active and reactive power As one can see the ON OFF model is a particular case of the FSM model one with only two states what makes this model more robust and ideal for modelling appliances An important notice to this model is that the signatures labelling the arcs cannot be chosen arbitrarily As one can see in figure 2 2 a and b th
19. and home environments One way of promoting this reduction is giving energy awareness to electricity consumers giving them feedback on how and where the energy is being spent Some studies state that direct feedback on energy consumption can lead to savings between 10 to 15 in the electrical bill 20 To provide energy awareness there s the need of a fine grained energy monitor capable of monitoring the majority of appliances within a household environment individually There are two main ways of accomplishing this The first consists in deploying one sensor at each appliance to be monitored and measure directly the power being spent by it recording the desired data This technique presents the drawbacks of being very expensive and difficult to install as it requires intrusion and modification of the house s electrical circuit Further it can be very difficult to monitor appliances that don t allow direct physical access such as embedded ovens The second is a concept introduced in 1992 by George W Hart that is Non Intrusive Appliance Load Monitoring a technique for monitoring the electrical consumption of individual appliances of a household facility without the necessity of installing individual sensors for each appliance By sophisticated analysis of current and voltage of the total load ideally at the main panel circuit breaker it is estimated the number and nature of the individual loads and their individual energy consumption as w
20. batteries 5 The nodes provide direct digital input output to the micro controller via connection pins which will be used to interface to external sensors sensing the environment 5 3 6 Deployment of Indirect Power Sensing The indirect sensing method is based on the fact that appliances may emit measurable physical signals when consuming energy 21 To indicate if some appliances are ON OFF according to the appliance model followed in the project it is necessary to implement sensory circuitry which output is directly correlated with the appliance s power state Since there is no need to estimate the power consumption from this the circuits output was chosen to be digital indicating only an ON or an OFF state The circuits developed were based on light sensors and magnetic sensors Operation and implementation details will follow 3 6 1 Ambient Light Sensing Circuit Lamps emit light when consuming power Sensing the light intensity near a lamp is a way to discriminate its internal state There exist such sensors called ambient light sensors Ambient light sensors are photo detectors designed to perceive the power emitted in the visible spectrum These sensors shall be ideal for the end here demanded as they can sense the most important output of a lamp which should be the easiest to detect There exist essentially two types of light sensors Light Dependant Resistors CdS Cells Commonly known as LDRs these sensors vary thei
21. delay in bins of 50ms From the figure it can be concluded that the most common delay in Lamp detections is in the interval of 200 250 ms while the most common delay for Fan heater detections is in the interval of 100 150 ms From the results in figure 5 7Jand from table 5 8 and from the fact that the WSN detection 79 Delay on Lamp ON detection Delay on Lamp OFF detection T T T ocurences ocurences o ro E po 350 300 250 200 150 100 50 0 50 100 150 200 250 300 350 350 300 250 200 150 100 50 0 50 100 150 200 250 300 350 Delay ms elay ms Jitter on Fan heater ON detection Jitter on Fan heater ON detection 12 T T T 12 Scere errr rc comer ter en 10 ety eet RA ter L a T BE Pcs AAN ocurences o T ocurences o T 350 300 250 200 150 100 50 0 50 100 150 200 250 300 350 350 300 250 200 150 100 50 O 50 100 150 200 250 300 350 Jitter ms Jitter ms Figure 5 7 Histograms of the delay between the detection by the WSN and by the load disambiguation algorithm ON OFF Appliance Min ms Mean ms Mar ms Min ms Mean ms Maz ms Lamp 220 97 265 240 131 300 Fan heater 325 1 160 240 57 180 Table 5 8 Analysis of the delay in the WSN detection only failed once in the realized tests it can be inferred that the Indirect Power Sensing WSN is reliable in detection of the desired events and has a performance that is comparable to t
22. eek end die oes Seok nas 100 ea eS bes eee ND 101 B 5 Hal effect principle 101 B 6 Hall effect principle magnetic field present ooo 102 B 7 Simplified linear current sensor lO cc 102 B 8 Hall effect linear current sensor Al o ooo 103 B 9 Rogowski Coil current sensor simple representation 00 103 iv B 10 Examples of Rogowski Coil current sensors 2 0 104 C 1 Simple model of a Current Transformer 2 000004 107 C 2 Current Sensor Circuit 2 A 108 C 3 Current Sensor Board aaa ee 109 D 1 Light Sensor basic Circuit 2222 22 a 111 D 2 Light Sensor Circuit Schematic 2 o e e 112 D 3 Assembled Sensor Circuitry 2 a 113 E 1 Current carrying conductor soo oe ee 115 F 1 UART Write Message 2 Co oo ee 117 F 2 UART Read Message 2 a 117 F 3 Gateway Fata Message 22222 aa ee 118 G 1 Hall Effect Sensor Conditioning Circuit o o e 119 G 2 Schematic of the Indirect Current Sensor Circuit 121 G 3 Fully Assembled Version of Indirect Current Sensor Circuit 122 vi List of Tables 3 1 Systems main features 2 nn 18 3 2 Comparative of the various solutions for current measurment 19 3 3 Register s adressing scheme 2 22 2 Comm nn 22 4 1 Single Appliance Characterization of some appliances in REDD 45 4 2 Comparison of detection in single appliance files to de
23. fetches the correspondent appliance from the Appliance Database It checks if the announced change is valid If so the appliance entry in the Appliance Database is updated and a notification is fired in the Logical Database signalizing an appliance state change event 57 4 4 7 Statistics and Calculations Having power consumption and times of state changes for every appliance available it is possible to compute some very interesting statistics Those may be Energy Spent Cost by appliance variation of power consumption by time of day weekday weekend duration of time each state is visited etc None of this statistics are currently computed but with the data provided by the logging system it could be achieved with minimum effort Nonetheless there are some calculations made by the system The system computes the estimated power that is the sum of the characteristic power of the appliances that are on the ON state at each moment Recalling from boolean switch process a t of equation 2 1 the estimated power Pestimated is N Pestimatea t gt Prot gt 4 2 i 1 where P is the power consumed by the ith appliance when in ON state Another calculation made is the residual power Presidual Which is the difference between the total measured power and the estimated power which is analogous to e t in equation 2 2 and can be interpreted as a measure of completeness and accuracy of the system 18 Presidual Protal Pestimated
24. field passes through the Hall element a Lorentz force is exhorted and a perpendicular Hall voltage proportional to the magnetic flux is generated 19 figure B 6 Vr amp IxB B 4 As one may see a Hall effect current sensor doesn t measure the current directly mea suring the magnetic flux generated by it instead So it s common to see Hall effect sensors coupled with flux concentrators in order to have better measurements In figure B 7 there is such a configuration in which the sensor is inserted in the air gap Characterizing the Hall effect current sensor it is a simple sensor that provides galvanic isolation making it safe when measuring high currents It is possible to measure AC and DC currents with it over a wide range of values ranging from few milliamperes to hundreds of 101 CORE com WITH N TURNS Figure B 7 Simplified linear current sensor amperes The frequency response is said to be outstanding There is also virtually no danger of damaging the magnetic circuit by high currents One may have special care since these devices can have large temperature drift if not compensated In figure B 8 is presented an example of a Hall effect linear current sensor 102 Figure B 8 Hall effect linear current sensor 41 B 4 Rogowski Coil Sensor A Rogowski Coil Sensor is an air cored coil with mutual inductance with the conductor carrying the primary current Its theory of operation is ba
25. interface 26 limitations the overhead introduced consumes valuable bus time that may jeopardize the data synchronization and that in other models could be used to send real data Furthermore issuing single access to registers introduces delays that are inconvenient A more suitable implementation would be such that in case of read of registers could issue repetitive reads of a list of register cycling through them In case of the write of registers it could be also be improved by being able to write to several registers in a single access Note that writes aren t critical since they are used almost only during the configuration phase of the device and in sporadic maintenance Such a model would allow for the data to be read by the PC at almost the maximum of the speed of the UART by eliminating the overheads and the round trip times assuming that the Gateway Device communication is much faster than the PC Gateway communication which is usually true since the gateway is implemented to a single purpose being very efficient in one task while the PC is designed to perform a high number of tasks simultaneously turning it mediocre in each task An example of an alternative method for high performance access is represented in fig ure 3 11 A repetitive read is issued by a sending special command of cycle read and a list of registers to read the gateway than cycles the list repetitively by issuing reads in the ADE7953 and when it has a new ve
26. on table 5 7 Appliance Tolerance Heaterl 5 Heater2 5 Fan 25 Fan Heater 5 Lamp 5 Table 5 7 Appliance classification tolerances In figure 5 4 is represented the classification of the appliances in the complex power plane The ellipses limit the intervals where the appliances are detected while the dots represent the scatter of the identification of the appliances Figures 5 5 and 5 6 show a scatter of detections of the appliances within their tolerances 76 Big loads such as the Radiator Heater Heater1 and Heater2 and the Fan Heater show small scatters being well defined within their intervals The Fan heater seems to have a larger scatter which is mainly due to the effects of the motor Small loads such as the Lamps and the Fan have greater scatters since they are more prone to the inherent errors of the system and load variation The Fan seems to have the greater scatter This may be caused by being the smallest load detected and also because of the effects of the slow transient of its motor Moreover the effect of switching the appliance ON in cold start or while the motor is still spinning seems to have a very relevant effect 100 80 4 Fan Heater OFF Heater2 ON ob Heater1 ON a Lamp OFF Fan ON P N J Fan OFF 4 ol y Lamp ON 4 Heater1 OFF 3 y Reactive Power VAR 44 s0 Heater2 OFF Fan Heater ON so 100 l 1500 1000 500 0
27. outputs the correct information about the Lamp s state Magnetic Sensor Nodes Calibration The sensor placement requires it to be very close to the supplying cable Figure 5 3 shows an example of correct placement of the magnetic sensor in which the sensor is in direct touch with the cable and strapped with a nylon clamp Figure 5 3 Placement of magnetic sensor Calibration Method 1 2 Tune potentiometer on board to approximately mid position Attach sensor to the cable of the appliance to be monitored and supply it Turn the appliance ON Install oscilloscope probe on the output of the second amplifier Rotate sensor around the cable and find the position where the signal has maximum amplitude Fix sensor position Check signal s amplitude tune potentiometer until the amplifier outputs a signal with amplitude close to 3V in order to put the system in saturation to guarantee correct operation Check sensor s digital output and verify if it is stable and if reflects appliance s state If not 73 e Some tuning in the potentiometer may be needed 9 Turn the appliance OFF 10 Check sensor s digital output and verify if it is stable and if reflects appliance s state If not e Some tuning in the potentiometer may be needed This method is to be repeated until the results are satisfactory 5 3 Appliance Characterization Before presenting the results and tests to the syst
28. power of consecutive steady periods intercalated with changing periods The steps are identified by its characteristic value of Power and its time stamp of occurrence The steps are then pushed to the database No further discussion on this aspect is needed since this subject has already been thoroughly discussed 51 Normalized Power Flow ATABASE Steps Queue Step1 Step2 Step3 Step4 Figure 4 11 Step Detection entity 4 4 3 Appliance Classification Although appliance classification has been already discussed in concept and implemen tation the implementation that will be here presented is different from the one previously exposed Before presenting the actual implementation of the algorithm it is important to present some concepts on the database that are necessary for the appliance disambiguation Appliance Database The system here developed only identifies appliances known a priori This means that the all appliances passive of being detected must be first identified and catalogued with their characterization Therefore it was implemented in the database an appliance database in which all the appliances of interest must be described in order to provide sufficient data for appliance detection and tracking of operating state An entry of the appliance database has the following parameters e Name Name of the appliance for identification e ON Signature Complex power s
29. que me transmitiram Agrade o ainda Micro I O Servi os de Electr nica LDA pela ced ncia dos uMRFs sem os quais n o poderia ter feito a rede de sensores sem fios para medi o indirecta de pot ncia Palavras chave Resumo Monitoriza o N o Intrusiva de Cargas Redes de Sensores sem Fios Medi o Indirecta de Pot ncia Hoje em dia assistimos a uma grande press o e esfor o no sentido de esti mular a conserva o de energia seja devido a limita es econ micas pre ocupa es ambientais ou normas regulat rias No topo do consumo de energia est a energia sob a forma de electricidade Para se conseguir incentivar a utiliza o eficiente da energia necess rio fornecer consciencializa o energ tica Se os consumidores de elecricidade tiverem acesso a informa o detalhada sobre como quando e que apare lhos est o a consumir energia estes poder o tomar boas e bem suportadas decis es e anuir mudar os seus h bitos de modo a reduzir o consumo de energia e por fim gastar menos dinheiro O projecto aqui apresentado desenvolve as bases para um aparelho que fornece informa o detalhada sobre o consumo de energia el ctrica para uma casa informa o a ser fornecida dever ser separada individualmente para virtualmente cada aparelho singular presente A t cnica escolhida para desenvolver o aparelho foi Monitoriza o N o In trusiva de Cargas de Aparelhos A partir de um nico ponto de medida o s
30. sensors make the installation costs lower than the deployment of a network of single power measurement equipment turning it cost effective and advantageous 1 3 Document Outline The document outline is as follows Chapter 2 Presents a theoretical and technical exposition of the Non Intrusive Ap pliance Load Monitoring branch of research analysing its deployment advantages and limitations Addresses the problem of disambiguating electrical identical loads present ing the concept of Indirect Power Sensing Chapter 3 Presents a High level description of the system s architecture and discusses its hardware implementation concerning development material selection and deploy ment It is described the implementation of the Data aquisition system that monitors the parameters of an electrical installation the communication interface between the Data acquisition system and the computational system and the hardware for the deploy ment of the WSN for Indirect Power Sensing Chapter 4 Presents a discussion of the system s algorithms and software logical com ponents namely the Load Disambiguation and Appliance Classification algorithms in the Fine Grained Electricity Monitor and its integration with the Indirect Power sensing WSN The system s output and user interface are also discussed Chapter 5 Presents the methods and tests that enabled the validation of the system The description of the test bed and the calibration of the
31. the PC at a rate of 48000bps Being the UART configured at 115200bps the actual data rate is gt x 115200 92160 bps again the configuration is more than sufficient for the specified requirements 3 4 3 Communication Model The communication model implemented in this gateway is coherent with the memory model implemented in the 4DE7953 The accesses are done to single registers and issued via the UART to the gateway A write to a register is issued by sending a packet to the gateway with a write command the address and the data to write The gateway then is responsible to communicate with the device and forward the write transaction as depicted in figure 3 9 A read is issued by sending a read command with the register s address to the gateway Then the gateway fetches the data in the register from the device and replies it via UART A read transaction is depicted in figure 25 Gateway ADE7953 Write Address Data Address Write Data UART SPI Figure 3 9 Gateway Write PC Gateway ADE7953 Read Address Address Read Data Data UART SPI Figure 3 10 Gateway Read This communication model was chosen by its simplicity and it is adequate for testing and configuration purposes However it has some limitations namely it is not suitable for high throughput systems that need synchronized data at very high sampling rate Considering that the PC Gateway is much more slower than the Gateway device because of structural
32. the measured load is 500 W at a time t the best estimate is 0 1 1 0 as it solves to e t 0 in equation If the measured aggregated load at time t At increases to 501 W the best estimate would then be 1 0 0 1 resulting again in e t 0 But it implies that every appliance switched state in a short interval At However normal household environment a change of state of all appliances is very unlikely and improbable From this it comes the suggestion of a criterion for describing the changes in the load model Switch Continuity Principle In a small time interval we expect only a small number of appliances to change state in a typical load 18 As it will be shown in the following chapters this principle will be the basis of the system s NIALM implementation This principle introduces a very important notion that is in a small enough interval it is expected that the number of appliances changing state to be usually zero and sometimes one and very rarely more than one From this section it is important to retain the importance of the Power measurement As it will be a representation of the circuit s load it is also of the highest value the Switch Continuity Principle as it will be the concept which we will build up the project 2 3 Appliance Models Having a brief view on how the appliances combine and how they may behave it is important to know how they can be modelled taking into account their function and behaviour
33. 2 6 94 6 96 Time seconds 10 a Oven Indentification 3000 Mains Power Oven Power 2000 0 1000 2000 5 35 5 352 5 354 5 356 5 358 5 36 5 362 5 364 5 366 5 368 5 37 Time seconds x 10 b Oven and Refrigerator Indentification Mains Power Oven Power Refrigerator Power Oven Steps States 5 35 5 352 5 354 5 356 5 358 5 36 5 362 5 364 5 366 5 368 5 37 Time seconds x10 c Figure 4 8 Appliance Identification in mains aggregated load la Refrigerator Detection Oven Detection lc Refrigerator and Oven Detection refrigerator apply in this case although here it is shown a miss of an identification caused by the events that are very close due to low sampling frequency In figure 4 8c it is represented the simultaneous detection of refrigerator and oven From the figures here presented and the 49 data in table 4 2 it s established the correctness of the suggested algorithm It is important now to develop the infrastructures and interfaces to apply the algorithm to real data acquired by the data acquisition system and to develop a full functioning Monitoring system 4 4 Fine Grained Electricity Monitor With the proposed algorithm verified it is now opportune to develop the Fine Grained Electricity Monitor The system developed can be decomposed in various logical entities which are represented in figure 4 9 Indirect Power Sensing WSN L Data Loggin Statistics and Ste
34. 7 4 Repeat step 1 and check current accuracy The calibration method is to be repeated for any of the channels until one gets the desired accuracy Checking Voltage Channel Accuracy Let Vineasured be the mean value of RMS voltage at the data acquisition s voltage input measured in 10 seconds intervals Let Vactual be the mean value of samples of the VRMS register taken at 10Hz during the same interval Knowing that between the Voltage input of the data acquisition system and the ADE7953 s voltage ADC channel is a 1 1000 voltage divider it is easy to compute the voltage at the ADC s input disregarding any error in the voltage divider which should be very small It is possible to compute the value that was expected in the VRMS register namely Vexpected that is V2 Vexpected Vmeasured X 500 x 9032007 5 4 Thus the gain error is E Vexpected eTTOT gain 1 5 5 Vactual The Fluke 287 True RMS multimeter s accuracy is accuracy of reading number of least significant digits 5 6 For the range of the measurements used which was 500V AC for a signal at 50H z is accuracy 3 of reading 25 x 0 01V 5 7 Table 5 1 shows the results of the assessment of the gain error comparing it to the accuracy of the reference error gain represents the ratio in percentage of the gain error of the ADC to Vactual and accuracy represents the accuracy of reference multimeter regarding Vineasured It may be obs
35. ARGAIN O O CLKIN CLKOUT COMPUTATIONAL BLOCK FOR TOTAL REACTIVE POWER CONFIGURATION AND CONTROL CFIDEN POWER FACTOR SAG Figure A 1 ADE7953 functional block diagram 89 A 1 1 ADE7953 Single Phase Multifunction Metering IC with Neutral Current Measurement The ADE7953 functional block diagram presented in figure A 1 It has one voltage chan nel and two current channels and measures line voltage and current and calculates active reactive apparent energy over programmable periods plus is capable of waveform sampling in over a 1 23 kHz bandwidth in the measurements e Vims Trs e Active Power e Reactive Power e Power Factor The IC measures some power line figures with voltage sag detection and period measure ment no load detection and zero crossing detection is also provided A variety of current sensors are supported such as are Rogowski Coil Sensors current transformers or low value shunt resistors due to its PGAs and Integrators in the current channel path All ADE7953 features can be accessed via a three different communication interfaces e 4 pin SPI interface e 2 pin bidirectional I2C interface e 2 pin UART interface A 1 2 ADE7953 Evaluation Board The ADE7953 evaluation kit includes two boards that together allow the performance of the ADE7953 single phase energy mete
36. Departamento de Universidade de Aveiro Electr nica Telecomunica es e Inform tica 2012 C sar Emanuel Monitoriza o Fina do Consumo de Energia El ctrica Brand o Gomes Fine Grained Monitoring of Electric Energy Consumption Departamento de Universidade de Aveiro Electr nica Telecomunica es e Inform tica C sar Emanuel Brand o Gomes 2012 Monitoriza o Fina do Consumo de Energia El ctrica Fine Grained Monitoring of Electric Energy Consumption Disserta o apresentada Universidade de Aveiro para cumprimento dos requisitos necess rios obten o do grau de Mestre em Engenharia Electr nica e Telecomunica es realizada sob a orienta o cient fica do Professor Doutor Paulo Bacelar Reis Pedreiras Professor Auxiliar do Departamento de Electr nica Telecomunica es e Inform tica da Universidade de Aveiro e do Professor Doutor Alexandre Manuel Moutela Nunes da Mota Professor As sociado do Departamento de Electr nica Telecomunica es e Inform tica da Universidade de Aveiro Trabalho financiado por Fundos Nacionais atrav s da FCT Funda o para a Ci ncia e a Tecnologia no mbito do projecto Serv CPS PTDC EAA AUT 122362 2010 Trabalho financiado pelo programa Europeu FEDER atrav s do Programa Operacional Factores de Competitividade COMPETE e pelo Governo Portugu s atrav s da FCT Fundac o para a Ci ncia e a Tecnologia no mbito do projecto HaR
37. Graphical User Interface default appearance Vrms Irms 111 2345 oo 2312 es 2 E E gt E 1430 1560 1690 1820 1950 2080 2210 2340 1430 1560 1690 1820 1950 2080 2210 2340 TimeC20ms TimeC20ms Figure 4 18 Graphics with traces of RMS Voltage and RMS Current Reactive Power VAR 2970 86 2700 6 2430 E S 51 2160 7 2 34 A 1890 2 e 1620 E 5 1350 1080 3 gt 810 540 2704 85 0 1430 1560 1690 1820 1950 2080 2210 2340 1430 1560 1690 1820 1960 2080 2210 2340 TimeC20ms Time 20ms Figure 4 19 Estimated Power 60 Graphics with traces of the components of complex power of Total Power and In parallel with the graphs it is also displayed some information in text form about the electric installation and about the events that are happening in the appliances As shown in figure it is displayed the values of e Vrms RMS Voltage e Irms RMS Current e Complex power components in the measured total power and in estimated power There is also a field called Events that is updated at the detection of an appliance event and flashes so it can get the user s attention Under that field there is also a scrolling list of the last 10 events that have happened appended to their time stamp Vrms 211 92 V Irms 8 041A Active Power Measured Estimated 2007 W 2017 W Reactive Power Measured Estimated 047 VAR 052 W Event Lamp1 OFF Latest Events Wed 25 July 21 59 33
38. LSBs must be made to get the Voltage Meter Constant The linear regression result was a voltage meter constant of VimeterConstant 3 887799914 x 10 Vams LSB 5 12 with a correlation factor R 0 9995 5 13 the offset was b 0 6726 VRMS 5 14 that represents an error of erroroffset 0 3 which is very small and lower than the reference meter s accuracy For being so small and for system s simplicity this error was not taken into account Table compares the value of RMS voltage at the utility s electrical installation mea sured by the reference multimeter Vrms with the measurements by the data acquisition system VrMsactual The error of the data acquisition system in relation with the reference measurements is lower than the provided accuracy and is mainly due to the discarding of offset on the calibration Method of calibration for Current Meter Constant A list of ordered pairs IRMS LS Bs Current Input A Currents List needs to be ob tained by the following steps 1 Set System s load 70 Vrms V VRM Sactual v ETTOT Mconstant accuracy 7 211 05 210 42 0 30 0 42 213 93 213 19 0 35 0 42 218 70 218 06 0 29 0 41 219 13 218 41 0 33 0 41 221 93 221 19 0 34 0 41 224 60 224 25 0 16 0 41 225 36 224 62 0 33 0 41 227 96 227 28 0 30 0 41 228 24 227 43 0 35 0 41 Table 5 3 Results of the voltage meter constant fitting 2 Get IRMS register from ADE7953 d
39. Signatures Non intrusive Signatures are those that can be measured passively by sensing the ag gregated load The previously mentioned power steps qualify as non intrusive signatures This non intrusiveness is such that there s no need to observe individual loads or to modify the loads in any way Within the non intrusive signatures we can also classify the observed signatures according to its nature The steady state signatures have information about the appliances state of operation that it is is continuously present in the aggregated load The transient signatures have information about appliance s state changes which is only present for brief moments at the times of transition Steady State Signatures Steady State Signatures arise from the difference of steady state properties between appli ance s states of operation and can be measured in different parameters that will be discussed further ahead Besides they have some interesting properties that are worth taking in to ac count They are continuously present in the load indicating the appliance s operating state which turns it easier to assess consistency of appliance s detections They are the signatures that satisfy ZLSC taking a very important place in detecting appliance s modelled with FSM models They are additive when two or more happen coincidentally Fundamental frequency signatures The measurements available at the utility s fun damental frequency are p
40. TES FCOMP 01 0124 FEDER 007220 or 1 FCT COMPETE i EEE Funda o para a Ci ncia e a Tecn ologia o j ri the jury presidente president vogais examiners committee Professor Doutor Jos Alberto Gouveia Fonseca Professor Associado da Universidade de Aveiro Professor Doutor Paulo Jos Lopes Machado Portugal Professor Auxiliar da Faculdade de Engenharia da Universidade do Porto Professor Doutor Paulo Bacelar Reis Pedreiras Professor Auxiliar da Universidade de Aveiro Professor Doutor Alexandre Manuel Moutela Nunes da Mota Professor Associado da Universidade de Aveiro agradecimentos acknowledgements Gostaria de agradecer ao Professor Doutor Paulo Pedreiras a sugest o deste tema de dissertac o o qual desde o inicio me interessou por ser de elevada praticabilidade e liga o ao mundo real Agrade o em conjunto aos Professor Doutor Paulo Pedreiras e ao Professor Doutor Alexandre Mota pela excelente orienta o que me proporcionaram Agrade o a todos os que me acompanharam no meu percurso acad mico nos meus altos e baixos Agrade o a quem me acompanhou nas minhas in meras discuss es e me ajudou a estimular o meu racioc nio e a questionar o mundo que me rodeia Gostaria tamb m de agradecer aos meus pais por me proporcionarem a oportunidade de realizar este curso numa rea que me cativa e num local onde me sinto confort vel Agrade o tambem pela excelente educa o que me proporcionaram e pelos valores
41. Turn OFFs Turn ONs Turn OF Fs Heaterl 9 9 10 10 Heater2 5 5 5 5 Fan 8 8 5 5 Lampl 8 7 9 8 Lamp2 8 8 12 12 Fan heater 13 12 13 12 Table 5 9 Detection of appliances in a random sequence of events The two level Heater Radiator has shown to be the most easy and straightforward load to detect being detected correctly all the times in the test The extra detections denoted in the table were due to human error during the execution of the sequence of events The Fan is not one of the best appliances to detect via NIALM because it is very small and any noise and variations on the load have great impact in its correct detection Thus it was noted during the tests that every time that the Fan was switched ON after the Fan heater was switched ON the system would fail to detect the fan It is believed that this behaviour is due to the slow transients of the fan heater and it is affecting the detection of the Fan preventing its step from being detected The Lamps were noted to be detected well by the WSN during the tests The extra detections noted in the table were due to some cross talk between lamps and also between the environment due to non optimal conditions The Fan heater was detected correctly by the WSN 81 From the Measured Power Estimated Power and the Residual Power one can verify the accuracy of the system by characterizing the quality of the estimation Figure 5 8 shows plots of the above referred power flows
42. V regulator that can provide up to 500 mA 7 This platform meets all the connectivity and communications requirements It exceeds the SPI specifications of the ADE7953 it has USB connectivity in order to communicate with a PC that is good because that s an interface that the majority of PCs have and the UART has high bitrates Plus it is capable of supplying power to the communications interface of the energy monitoring board without any other connection than USB 3 4 2 Gateway Deployment Figure 3 7 presents a diagram of the gateway and its interface with the rest of the system 3 3V PIC32MX795F512L Regulator USB to UART MARTRI SPI E UART Translator o FTDI232R MASTER Processing Unit Analog and DSP Frontend chipKit Max32 Figure 3 7 SPI gateway implementation ADE7953 s SPI interface ADE7953 s SPI interface is a fully functional SPI device that uses all four communication pins It operates in slave mode which means that a clock must be provided by the master This clock can be at maximum 5M Hz As said before registers are accessed one at a time The access to a register is encapsulated in a SPI packet Which is always initiated by the master consisting in two initial bytes for the register s address Then it is transmitted a command byte to state if the access will be a read or a write Finally the data is sent from ADE7958 to the master in case of a read or written to ADE7953 by the master i
43. Y Lamp2 Undetermined Figure 4 21 Apps Tab of Graphical User Interface 4 5 Summary In this chapter it was presented the algorithms and structures for a Fine Grained Electricity Monitoring system The algorithms were tested and validated This led to a preliminary implementation in software that was also tested and validated The algorithm s preliminary implementation in software was the basis for the development of the system s implementation The implementation consists in e Data Acquisition e Appliance Classification e Indirect Power Sensing WSN e Statistics and Calculations e Data Logging e Graphical User Interface To the Indirect Power Sensing WSN it was seen that it is self configurable and that it integrates with the PC like a data device Thus it notifies the system about the changes of state in individual appliances 62 In terms of inputs outputs the system gets RMS Voltage RMS Current and Complex Power from the Data Acquisition System and gets individual states of some appliances from the Indirect Power Sensing WSN It produces as output appliance disambiguation calcula tions and statistics on historical data and logging of all data of interest Finally it was presented the Graphical User Interface G UI the ultimate output for user s feedback It shows real time data on the electric installation Vrms Irms Power and the calculated Expected Power On the appliances side the GUI shows the appliances p
44. a A BET ail 1 990 1000 1010 1020 1030 1040 1050 1060 1070 1080 1080 1070 1060 1050 1040 1030 1020 1010 1000 990 Active Power W Active Power W Lamp ON Lamp OFF wen a 10 we fie Reactive Power VAR Reactive Power VAR A o o j N 7 A 1 1 A aaa 4 1 eT ii 4 1 102 104 106 108 110 112 108 106 104 102 Active Power W Active Power W Figure 5 6 Appliances Individual Interval of identification b 5 4 3 Indirect Power Sensing WSN The WSN is monitoring two lamps by light sensors and the fan heater with a hall sensor When put to function it was observed that the WSN detected almost all performed events It was not possible to characterize the detection accuracy due to lack of time for thorough tests 5 4 4 Simultaneous detection of appliances The system was configured to try to identify the loads simultaneously by the WSN and by the Load Disambiguation Algorithm It was found that the WSN only failed to detect the lamps events 1 time out of 54 events that were introduced in the test More the WSN hasn t failed to detect the Fan heater in a test of 60 events From the data obtained it was possible to compute the relative delay between the the detection of the events on the load disambiguation algorithm and the detection on the WSN In the following the detection by the load disambiguation algorithm is being used as the time reference Figure 5 7 shows histograms of the
45. a turn ON is performed briefly after a turn OFF while the motor is still spinning where it can be observed that the detected step deviation is almost as small as those of the resistive loads From the above discussion it follows immediately that a certain amount of time must elapse between the occurrence a state change and the identification of the appliance that experienced that change This latency is defined as the detection time It was verified that the step detector also performs well when there are various sources of steps in the system and even when the steady state of the system is not null The conclusions are the same of those of when the system is tested with single appliances The Analysis of the deviation from steady state signatures will allow the existence of a starting point in defining the tolerances for appliance detection in the Appliance Classifier 5 4 2 Appliance Classifier The appliances were catalogued in the Appliance Classifier with the steady state signatures from table 5 5Jand with tolerances with the immediate unit upper bound in percentage of the maximum deviation as described in table It was found that some of the appliances were not being detected correctly in the total load For this reason the tolerances were tuned in order that all the appliances were identified and correctly classified taking into account that none of the intervals corresponding to the signatures would overlap The tolerances were tuned to those
46. an see the accurate fitting of the values from the data acquisition system with the error errormconstant always smaller than the measurements accuracy accuracy This means the system meets the accuracy of the reference and that discarding the offset in the scaling of the values is not significant 71 Irus A RM Sactual A errormeonstant accuracy 0 188 0 182 3 42 29 82 3 230 3 299 2 14 514 7 920 7 922 0 04 4 78 Table 5 4 Results of the current meter constant fitting 5 2 2 Calibration of Sensors in Indirect Power Sensing WSN Because the nodes in the Indirect Power Sensing WSN perform measurements it is also necessary to guarantee that they produce the correct outputs to the designated inputs To this matter it is necessary to calibrate them in site to check its proper function Light Sensor Nodes Calibration The sensor should be placed near the lamp to be monitored An example of a correct placement is shown in figure 5 2 Figure 5 2 Light Sensor Placement Calibration Method 1 Place sensor node 2 Turn the Lamp ON 3 Check if sensor s output reflects Lamp s state If not e Tune potentiometer on board until the sensor s output is correct 4 Turn the Lamp OFF 5 Check if sensor s output reflects Lamp s state If not 72 e Tune potentiometer on board until the sensor s output is correct This method is to be repeated until the sensor node
47. appendix G The output of the sensor to our interest is a sinusoidal wave at 50H z with few millivolts of amplitude 34 SS49 gt Z gt Po Band Pass Filter High Pass Amplifier hmitt Tri i p Rectifier and Filter Schmitt trigger Figure 3 18 Block Diagram for indirect current sensing The sensor s output passes then through a band pass filter to remove other frequencies than that of interest High frequency noise may appear due to radio signals The signal has very low amplitude so it passes through an amplifier with high pass characteristic in order to amplify the desired signal and reject any type of DC Being the signal at an amplitude in the order of the units of Volt it is necessary to convert it to a DC signal in order to further convert it to the digital signal The signal is passed then through a Rectifier and Filter that generates a DC voltage corresponding to the maximum voltage of the Steady State sinusoidal wave In correct function the output here is a high dc voltage when there s current flowing in the cable and null voltage otherwise The signal passes then through a Schmitt Trigger that converts the signal into a digital value indicating the presence or not of current It functions as a comparator with hysteresis A fully assembled version of the circuit can be seen in figure 3 19 Figure 3 19 Fully Assembled Version of Indirect Current Sensor Circuit 3 7 Summary This chapter presented the architectur
48. ate the problems of the fluctuation of utility s voltage as stated in subsection 2 5 3 Step Detection Taking normalized power as an input this block implements an algorithm that detects step like changes and finds their times and sizes For this the power flow is decomposed in steady and changing periods A steady period is defined as a sequence with a minimum number of samples in which they do not vary more than predetermined tolerance in both of the power components All the other periods in between are said to be changing periods Having the steady periods been defined the steady state power of that period is the average of the In Portugal the nominal voltage is commonly 230 V for homes with BTN Baixa Tens o Normal 11 samples in that period in order to minimise the noise Lets call these values the steady states Having the steady states defined the steps are calculated by the differentiation of consecutive steady states with this value comes too a time stamp that allows the identification of the time of the step The output of this block is a sequence of time stamped step change complex power values An illustration of a possible instance of the algorithm is figure 2 5 where it is represented the power and a measure of how much the power is changing that is used to sequence the power flow in steady and changing periods There we may also see the detected steps represented in the center of the steady periods This may not be t
49. ational Engine for calculations and a programmable 80515 at 5MIPS for memory access SFR control and communications It has digital I O ports for interfacing with other devices with its functional block depicted in figure A 5 With appropriate firmware the IC is capable of the following measurements e Vims e I ms apparent power e real power reactive power phase angle e power factor e line frequency The communications are done via UART with a maximum baud rate of 38 4 kbit s 2 A 3 2 78M6613 Evaluation Boards Maxim provides two evaluation boards to assess 78M6613 performance 78M6613 EVM Evaluation Board figure A 6al is asystem designed to perform mea surements up to two single AC loads The current sensing is done by current resistor shunt and the communication is made via USB emulating a serial port The measure ment capabilities are the same as the single IC 1 78M6613SP EVM Split Phase Evaluation Board figure A 6b is a system designed to perform measurements to a single electrical outlet The system provides two inde pendent phases measurement since in the US each outlet has two 120V ms phases and one neutral The current sensing is done through current transformers and the com munication is made via USB emulating a serial port The measurement capabilities are the same as the single IC 3 93 AO Al A3 XIN XOUT CKTEST TEST TX osc RTCLK 32KHz 32KHz 4 9152MHz CK 2
50. ators weren t powerful enough But with today s digital technologies it is possible to build very powerful digital integrators easily with an outstanding accuracy 23 Following up nowadays it can be observed a certain effort of the manufacturers of silicon solutions for energy metering of supporting Rogowski Coil Sensors by developing energy metering ICs with already built in digital integrators for these sensors Some examples of Rogowski Coil Sensors may be found in figure a Rigid Rogowski Coil 11 b Flexible Rogowski Coil 11 c Flexible Rogowski Coil 31 Figure B 10 Examples of Rogowski Coil current sensors 104 B 5 Prices Survey In the development of a technology project one must not forget about the total system cost so it is important to choose the most cost effective solution bringing a balance through price and functionality So a minor survey on current sensors was conducted which lead to the following findings Current Shunts The current shunts are widely available in general components market For the project s range international standard IEC 61036 for energy metering states that the power consumption per channel cannot exceed 2W Regarding that to allow headroom considering 3W low resistance current shunts it was found that it is possible to find a good set of products for a price range from 0 5 to 2 US dollars 12 Current Transformer The current transformer are also available for the general c
51. ble appliance is one which has an infinite number of states and is suitable for few appliances such as light dimmers sewing machines or variable speed drills which have a continuous range of power levels and so do not generate step change signatures not fitting in ON OFF or FSM models 18 Again this model wont be used in this project not only due to its complexity but also the existence of appliances that are described by this model is small and though not very significant in household environments 2 4 Signatures Knowing what are the models of appliances suitable to be found in the load it is needed a consistent way of identifying the appliances in the aggregated load The previous sections used power steps solemnly as signatures but as it will be explained later in this chapter that may not be the most effective way of identifying appliances in the load nor the only one We start then by the definition of signature Signature signature is a measurable parameter of the aggregated load that gives infor mation about the nature or operating state of an individual appliance in the load 18 Now that the reader understands what a signature is it is important to refer that this concept is the fundamental basis for NIALM as signatures are the pieces of data that allow for load disaggregation The signatures can be subdivided in Non intrusive Signatures and Intrusive Signatures which will be now discussed 2 4 1 Non intrusive
52. cation In a preliminary implementation this structure has the parameters e Name The name of the appliance 46 e ON Signature Complex power Steady State Signature of the ON state e OFF Signature Complex power Steady State Signature of the OFF state e State Flag indicating if the appliance is ON or OFF There is also a database where the data about all the appliances is stored in order to be further accessed The classification starts with the computing of the minimum euclidean distance between a given step to every signature of all the appliances in the database with the goal of identifying the closest signature and consequently the appliance and the event an ON or an OFF that is candidate to have occurred The identification ends by the verification if the step is contained within a predetermined tolerance If so the step is classified with the event and the appliance The state data of the appliance in the database is also updated 4 3 3 Process data flow Having the methods and structures been defined it can now explained the processing flow of the data The power samples are pushed into the window Next the steady power discriminator takes the standard deviation and the mean from the window and discriminates the steady states from the non steady states After the discrimination the step identification is made taking into account the consecutive steady states The steps are then taken into appliance classification
53. ck with an fixed overhead of 24bits per register access while I C can operate at a maximum clock of 400kHz with 1 bit per clock with an overhead of 16bits plus 1 start and 1 stop bit plus lack bit per byte written in the bus Besides all that SPI can be considered simpler than FC 3 4 SPI PC Gateway Implementation As normal PCs don t provide external SPI interface To fulfil the communications im plementation it is demanded to implement a gateway to provide the needed communication interface Some requirements are to be taken into account The communication has to have high bitrates and should be compatible i e should be in a connectivity interface present in the majority of computers 3 4 1 Chipkit Max32 Figure 3 6 chipKit Max32 development board Again for fast integration it was chosen to develop the gateway with an integrated so lution in this case a microcontroller development board The platform chosen was chipKIT Max32 by Digilent figure B 6 a microcontroller board based on Microchip PIC32MX795F512L For the project s interest this platform has the following features e PIC32 high performance 32 bit micro controller at 80 MHz with a 5 stage pipelined MIPS core 35 23 e Hardware SPI driver performing with maximum of 40 MHz clock 34 e Hardware UART driver with performance up to 20M bps 34 e FTDI FT232R UART to serial UART that may perform at a maximum of 3 Mbaud at 7 or 8 bit symbols 7 16 e 3 3
54. d Message Data Message This message represented in figure F 1 is sent from gateway to PC via UART figure F 3 and represents the data in the register from the read immediately issued 117 before Its length is variable with the size of the register and is sent LSB first There is a lack of consistency check in this message that can make the protocol unreliable in case of transmission errors But its simplicity turns the coding in the PC easier this is acceptable since the error probability is very low DATA_LSB DATA MSB Figure F 3 Gateway Fata Message The protocol itself may lack some error detection mechanisms But that could be easily implemented with a checksum at the end of every message 118 Appendix G Indirect Current Sensor It is not intuitive what values to expect at the hall sensor s output so lets take by hy pothesis a load of Pjoad 1kW connected to the power outlet at V 230 Vrms with a generic two round conductor cable with the dimensions d 2mm for the conductors and de 5 mm for the whole conductor Assuming that the sensor is directly put in contact with the cable at O 0 from equa tion 3 2 and from the sensor s sensibility 1 4mV Gauss the sensor output will be O Gauss Vout 4 64 MVrms The sensor output is a sinusoidal wave because the current is sinusoidal generating also a sinusoidal magnetic flux The SS49E linear hall sensor s output is a sinusoidal voltage with
55. d allows to test the system with all the communication technolo gies allowed in the ADE7953 I C SPI UART 14 But the protocol it uses to communicate with the LabVIEW program is not available nor the program s API making it impossible to develop custom software around it It is then necessary to deploy another way of communicating with the acquisition system 3 3 3 Communicating Directly UART Due to its simplicity it was first tested the UART implementation directly connecting the PC to the ADE7953 through an USB UART converter that displayed in the PC as virtual COMM Though it was possible to establish communication it didn t provide enough performance it was limited to a 4800bps and the connection wasn t reliable the culprit is it s 22 timing scheme that was not easy to guarantee by general purpose operative systems which have scheduling granularity in the order of the of tens of milliseconds while the protocol has constraints in the order of milliseconds 4 Posteriorly to this test it was found one configuration on the virtual comm driver that could have produced this behaviour but another solution with much better performance was already developed SPI Since the UART interface did not meet the requirements communicating via another interface is imperative The choice was SPI because it has the capability of being faster in data transactions than JC It can operate at a maximum clock of 5M Hz with one bit per clo
56. d in figure where it is represented the tran simpedance circuit attached to a Schmitt trigger to output digital levels regarding the appli ance s state ON or OFF The design and analysis of the light sensor circuit is presented in appendix D The sensor is designed to be placed near a lamp where the light intensity is very high A final assembled version of the developed circuit device can be seen in figure b Figure 3 15 Light Sensor al Light Sensor Circuit Schematic Ib Assembled Sensor Cir cuitry 3 6 2 Magnetic Hall Effect Sensing Proposition one could measure detect current flowing through a current carrying con ductor by measuring the magnetic field around it with a magnetic hall effect sensor Hall effect sensors are semiconductor based devices that when subjected to a magnetic field output a voltage proportional to the field strength 19 32 Taking into account that this project aims to detect household appliances non intrusively the place where one could place the sensor is attached to the electric cable of the appliance which generally is a two wire cable as represented in figure From the Ampere s Law which an analysis is presented in appendix E it was derived the dependence of the magnetic flux at a distance r of the center of a fixed position two conductor cable The magnetic flux varies with the angle 0 as expressed in equation 3 2 A normalized plot of such equation is shown in figure B
57. dBus The current sensing is done by Current Transformers 44 A 5 Microchip s PIC18F87J72 Single Phase Energy Meter Energy Monitoring PICtail Plus Daughter Board Microchip s Energy Monitoring PICtail Plus Daughter Board is a low cost power monitor based on the PIC18F87J72 single chip as the energy meter IC figure A 8 Being a low cost design the current sensors are shunt resistors The system relies on the microcontroller s 16 bit sigma delta ADCs and in its computational in order to do the measurements Vims rms e active power e reactive power 96 MICROCHIP no Figure A 8 Microchip s PIC18F87J72 Energy Monitoring PICtail 83 e active energy e reactive energy e line frequency The measurements are available at every line period corresponding to a 50 Hz sampling frequency v a USB emulating a serial port 33 97 98 Appendix B Current Sensing Technologies Survey In the current days the most used technologies in current measurement are low resistance current shunts current transformers Rogowski coils and Hall effect based current sensors In this section it is presented a survey of the main characteristics of each solution in order to choose the most adequate B 1 Low Resistance Current Shunts Current shunts are the most cheap and simple solution to measure current They can be accurately modelled by a series connection of a low resistance shunt and a parasitic inductance with the load w
58. del in order to support multi state appliances as there are already various examples of such appliances in any home That solu tion would pass through implementation of an appliance model based on finite state machines as suggested in Nonintrusive appliance load monitoring 18 As the appliance pool rises there may be the need of deployment of other techniques for appliance identification so it s proposed augmenting the appliance detection with transient analysis as documented in Power signature analysis 24 harmonic signatures as suggested in Nonintrusive appliance load monitoring 18 and probabilistic and stochastic analysis of appliance s usage as in Circuit Level Load Monitoring for Household Energy Management 28 To account with the augmented parameters for appliance detection and identification there could be implemented Artificial Neural Networks with the previously mentioned data as in puts that through sophisticated functions and self learning could output the states of the appliances in the system as documented in Real Time Recognition and Profiling of Appliances through a Single Blectricity Sensor 38 One aspect of the current system that needs to be improved is the use of the computed data to produce valuable statistics on energy consumption such as detailed electrical bill discriminating the appliance s cost intervals of consumption variance of consumption on a time basis or temperature basis etc That may be one of the ea
59. des galvanic isolation making it safe for measuring high currents This solution is very precise over a wide range of frequencies up to hundreds of kilohertz but it suffers of the problem of not being able of measure nor admit DC due to saturation of the magnetic core Also current transformers can introduce phase shift and special care should be taken to compensate this error Considering power one can say that it consumes little power compared with the current shunt as it introduces nearly a null load to the system 100 In terms of price it has a higher cost than resistance shunts but it can be cheap if made small Some examples of current transformers are presented in figure CR3110 3000 CR3111 3000 a Split Core Current Trans f b Solid Core Current Trans ormer 8 mer Figure B 4 Examples of Current Transformers B 3 Hall Effect Current Sensor Hall Effect Current Sensors are high performance current sensors based on the Hall effect The theory behind Hall effect states that when a current carrying conductor is placed into a magnetic field a voltage will be generated perpendicular both to the current and the field 19 So the elementary sensing system based on hall effect is based on a thin semiconductor sheet Hall element which has a current passing through it figure B 5 x A V 0 gt Figure B 5 Hall effect principle 19 AO Il When a perpendicular magnetic
60. diagram of the ADE7953 Evaluation Board in terms of supplies and main inputs outputs I O Power Supplies As one may see in figure 3 3 the board has two separate power connections The power in the board is in two parts one for the hot part where the analogue front end is and one for the cold part dedicated to communications The reason for this isolation is the protection VDD_COMM VDD_AFE ADE7953 Figure 3 3 ADE7953 evaluation board I O of user and user equipment connected to the communications interface since that in the hot part the ground may be coupled to high voltage signals due to the direct connection of the system s ground to one of the voltage s inputs Both power inputs are rated to have an independent 3 3Vpc 10 voltage supply 4 The one connected to the hot part was based on a 3 3V regulator a UA78M33 from Texas Instruments connected to the lab bench supply The other one was deployed in conjunction with the communication by a micro controller development board connected to a PC that will be further discussed further ahead Voltage Sensing The voltage sensing of the board can be directly driven by the utility s voltage signal The Vp input shall be driven with the Phase conductor while the Vy shall be driven with the Neutral one This is done by connecting the conductors of the configurations power strip in parallel as depicted in figure to the board Current Sensing As mentio
61. e results of the appliance characterization Sig is the mean of the steps detected and STD is their standard deviation Num events is the number of steps detected for that state Turn ON Turn OFF Appliance Sig W STD W Num Events Sig W STD W Num Events Oven 1640 24 46 1639 24 46 Refrigerator 186 13 662 192 42 642 Dishwasher 545 350 107 452 291 126 Lighting 235 ol 28 241 46 27 Microwave 1524 44 241 1514 100 242 Bathroom GFI 1607 21 43 1607 21 43 Washer Dryer 2706 32 107 2706 32 107 Table 4 1 Single Appliance Characterization of some appliances in REDD Analysing the results one can verify that the algorithm is indeed capable of discriminating step like changes in well behaved power flows Thus the standard deviation of the steps is very low in the majority of the appliances detected The OFF signatures are very close to the symmetric of ON apart from a small error thus satisfying Zero Loop Sum Constraint And the number of detections of ON states is coherent with the number of detections for OFF states satisfying in this way the Uniqueness Constraint described in section One of the appliances the dishwasher stands out having big Standard Deviation and signatures very different in absolute and even not coherent in the number of detection The reason for this is that the appliance is not well described by the ON OFF model for which the algorithm was des
62. e solution and the hardware for the project The data acquisition system is based on ADE7953 Evalutation Board a system that mon itors an electricity facility returning at high sampling rates various characteristics of the electric circuit 35 The communication between PC and data acquisition system is intermediated by a gate way that communicates with the PC via virtual COMM and to ADE7958 via SPI The indirect sensing WSN is deployed by uMRF s nodes using PIC 8 bit micro controllers for processing and will communicate wirelessly via radio using IEEE 802 15 4 protocol Indirect Power Sensing is deployed by circuits attached to uMRF s The sensors are be based on photo transistor for light sensing and hall effect sensors for current sensing The sensing is deployed in such a way that the circuits output digital values indicating ON or OFF states 36 Chapter 4 Software Implementation 4 1 Load Disambiguation and Appliance Classification Before presenting the software solution for the Fine Grained Electricity Monitor it is im portant to acknowledge the underlying methods that will be in use Following the NIALM analysis presented in Chapter 2 it will be here described the main algorithm for load disam biguation and appliance classification depicted in figure 4 1 1 Data Aquisition Measure Power and Voltage RMS Data 2 Normalize y nom Perm 7 P V Normalized complex power 3 Step Detection Step C
63. e transition from ON to OFF is the negative of the transition form OFF to ON and furthermore the sum of the signatures of the cycling of the states in is zero Generalizing the following constraint can be stated Figure 2 2 Finite State Appliance Models a generic 1000 W two state appliance no frost refrigerator three way Lamp a clothes dryer e three speed fan Based on figure 4 of Zero Loop Sum Constraint ZLSC The sum of the power changes in any cycle of state transitions is zero 18 Anyone with background in electric circuits will see that this constraint is analogous to Kirchoff s voltage law being the discrete space analogue to the constraint that the curl of conservative vector field is zero Another valuable constraint that may be used in FSM models in order to simplify the build of models for the appliances is Uniqueness Constraint UC Distinct states in a FSM appliance model have distinct op erating power levels 18 Being the analysis of FSM important to the subject of NIALM in order to correctly build methods and algorithms for learning the appliances in the circuit it wont be that important in this project because it s complexity wont bring much performance improvements For this reason the project will be focused in the ON OFF model Continuously Variable At last comes the Continuously Variable model which arises from a generalization of the FSM model in such a way that a continuously varia
64. ect Power Sensing of individual appliances in order to augment the data available for load disambiguation These WSN has too some requirements It has to cause the minimum intrusion to customer s space has to be quick and easy to install and preferably cheap This implies that the wireless sensor nodes need to be small battery operated and have to have low power consumption for high autonomy 3 5 1 uMRESs The WSN nodes used were the the uMRFs provided by Micro I O Servi os de Elec tr nica figure 3 14 They are developed for embedded solutions with wireless applications and they satisfy the needed requirements The nodes are compact and have small profile they are battery operated by a rechargeable Li Ion battery The processing capabilities are provided by a 8 bit low power micro controller from microchip PIC18F26K20 with nanowatt XLP The wireless capabilities are provided by IEEE 802 15 4 compliant radio transceiver MRF24J40MA which protocol is designed for low power applications They provide USB 30 Li loN Batt connector Accelerometer Temperature sensor Reset button HM IEEE802 15 4 Debug button transceiver ICSP connector JB Batt Charge Mini USB control ia connector Batt Charge E jumper m MCU Debug RS232 USB LEDs converter Figure 3 14 uMREs wireless nodef5 connectivity by a USB UART converter The USB connection can also be used to provide power to the node and to recharge
65. ed by a time stamp and a steady state signature in complex power Here it is tested the ability of the system to detect steps of the different appliances individually For each appliance a test was performed by turning ON the system and introducing a sequence of 21 consecutive switches with a period of approximately 4 seconds between switches to let the power stabilize The step detector parameters used were window__size 10 step threshold 25 W and STD_Threshold 5W In the results of the test it was analysed the deviation of the detections from the steady state signature as well as the number of detections Table 5 6 presents the results of these tests For each appliance tested Nr Detections is the number of steps detected Mean Deviation is the mean of the deviations from the steps detected to the steady state signature with Value being the norm of the Mean Deviation and Ratio the ratio between the mean deviation and the appliance s signature Max Deviation represents the maximum deviation detected Value and Ratio values have the same meanings as stated before Mean Deviation Max Deviation Appliance Nr Detections Value W Ratio ValuelW Ratio Heaterl 21 3 09 0 38 4 41 0 54 Heater2 21 3 78 0 31 6 02 0 49 Fan 21 6 95 17 48 9 35 23 49 Fan Heater 21 28 26 2 74 32 52 3 15 Lamp 21 0 46 0 43 1 39 1 30 Table 5 6 Results of Single Appliance tests on step detector From tab
66. ell as other relevant statistics The hardware cost of this implementation is comparatively very low turning the complexity to the software 18 1 2 Proposition It is hereby proposed the development of a fine grained energy monitor capable of mon itoring the majority of appliances within a household environment based on Non Intrusive Appliance Load Monitoring The system ought to provide direct feedback in real time by showing the total power being spent as well as what appliances are turned ON and historical feedback by doing with electricity what is done with the telephone bill e discriminating in detail how the energy was spent what appliances were in function and when and how much energy they were spending and its cost Although a Non Intrusive Appliance Load Monitoring system may identify dif ferent appliances within the total load there is a clear limitation in distinguishing identical loads such as lamps or other appliances that may be equal in different rooms of a house To address this limitation it is also proposed the deployment of a Wireless Sensor Net work that monitors environmental and physical parameters e g light temperature vibration magnetic field that allow the identification of appliances that cannot be disam biguated with The data gathered by the WSN is used to augment the data for the INIALM monitor to distinguish similar equal appliances The simplicity of the network and the inexpensiveness of the
67. em it will be presented the appliance characterization This is done here to justify the values of power detected in the system while testing its accuracy and validity The appliances were characterized by steady state and by turn ON and turn OFF tran sients in normalized power consumption recorded by the acquisition system at a sampling frequency of 50Hz for approximately 10 seconds The steady state signature was found by taking one sample of the power consumption of the appliance after approximately 10 seconds of being turned ON on a cold start assuming that all the transients are over after that period The rise time is the time that the appliance s power consumption takes to rise from 1 to 99 of the steady state signature after being turned ON The fall time is the time that the appliance s power consumption takes to fall from 99 to 1 of the steady state signature after being turned OFF The appliances available in the test bed are e Two Level Radiator Heater this appliance is categorized in two appliances Heater1 and Heater2 since the system is incapable of treating multi state appliances e Two 100W Lamps These lamps were specifically introduced for having 2 identical appliances in the system in order to exploit the problem of identification of similar loads with the Indirect Power Sensing WSN Only one will be characterized it is assumed that the other to be equal e Desk Fan e Fan Heater the characterization of the a
68. equal to Gs0H 500 and gain at DC equal to Gde 1 The need for high gain comes from the low amplitude of the signal at sensor s output of units of millivolt while comparing to the desired output is a digital value of 3V The need for no gain at dc comes from that as we are dealing with such high gains any undesired offset could saturate the amplifiers 1 Order Low Pass Filter This stage implements a low pass filter with corner fre quency at fe 50 Hz to filter any high frequency noise amplified by the previous stage High Pass Amplifier with Variable Gain This stage implements an Amplifier with High Pass characteristic The gain at 50 Hz equal varies from 1 to 100 and the gain at DC is always equal to Gac 1 The need for variable gain comes from the ignorance a priori about the appliance to monitor and its power The gain is tunable via a potentiometer in order to manually adjust to environment s characteristics The need for no gain at dc comes from the same reason as the previous amplifier Rectifier and Filter This stage rectifies and filters the amplified signal corresponding to the current generating a DC voltage corresponding to the maximum voltage of the sinusoidal signal In correct function the output here is a high dc voltage when there s current flowing in the cable and null voltage otherwise Schmitt Trigger This stage aims to convert the dc voltage supplied by the previous stage to a digital value indicating t
69. erved that the gain error is always smaller than the reference s accuracy which means that the calibration may exceed the reference s performance therefore it is not possible to improve the calibration with this reference Vmeasurea V Vexpected LS B Vactual LSB eTTOTgain accuracy 211 05 5391562 5412301 0 38 0 42 213 93 5465136 5483521 0 34 0 42 218 70 5586992 5617941 0 39 0 41 219 13 5597977 5617941 0 36 0 41 221 93 5669507 5689227 0 35 0 41 224 60 5737716 5768089 0 53 0 41 225 36 5757131 5777579 0 35 0 41 227 96 9823551 5845952 0 38 0 41 228 24 5830704 5849892 0 33 0 41 Table 5 1 Assertion of gain error following the calibration of voltage channel 68 Checking Current Channel Accuracy Let IUmeasurea be the mean value of RMS voltage at the ADE7953 s current ADC channel input measured in 10 seconds intervals Let actual be the mean value of samples of the IRMS register taken at 10Hz during the same interval Following the same principles as those on the ADC s voltage channel Tezpectea is the expected value at the IRMS register for IUmeasured and is computed as v2 Texpected LUmeasured x 500 x 10 3 x 9032007 gt 5 8 and so the gain error is I eTTOT gain 1 zenp cted 5 9 Tactual The range of measurement used in the multimeter was 50mV AC with a signal of 50Hz The accuracy is accuracy 0 3 of reading 25 x 0 001mV 5 10 Table 5 2 sh
70. essly They don t need wirings for communications neither for power supply as they can be made battery operated and if well designed they can have big autonomy 13 2 8 Summary In this chapter it was given an introduction and an analysis of the concepts underlying the Non Intrusive Appliance Load Monitoring identifying its main requirements and its main results A NIALM system is able to decode all the information from only one point of measure ment at the total load without the need for any more sensors at any appliance There are various appliance models being the most important the ON OFF and the FSM mod els although the majority of the appliances are accurately described by the first In order to detect the appliances the algorithm looks for signatures being the steady state the most straightforward easy to understand and deploy and the most informative due to its perma nent existence in the load Within the steady state signatures the normalized power is one of the best because it provides independence from voltage fluctuations It was shown that it can be relatively easy to deploy a NIALM system based on step detection on normalized power and that such a system would permit the production of valuable information about power consumption being able to disaggregate the energy in components for each appliance One of the major limitations of this technologies is the difficulty in distinguishing electrically similar loads To address t
71. few millivolt of amplitude standing on a approximate V 2 5 V e value when sensing current near a current carrying cable To transform this waveform onto the desired output one has to pass the signal through various filtering and amplification stages The objective of indirect sensing is detecting the power state of the appliance With the hall effect sensor it is possible to do so by detecting the current flowing in the cable The aim is then to develop a circuit which output is a digital value indicating the presence or not of current in a cable The circuit will be attached to uMRES so the digital levels of the circuit must be 0 V and 3 3V The solution developed contemplates the stages represented in the figure G 1 1st Order Band Pass High Pass Amplifier 1st Order Low Pass High Pass Amplifier Schmitt Trigger fc 50Hz G 50Hz 500 fc 50Hz with Variable Gain Rectifier and Filter G 50Hz 0 5 G DC 1 G 50Hz 0 5 1 lt G 50Hz lt 100 G DC 1 Figure G 1 Hall Effect Sensor Conditioning Circuit 1 Order Band Pass Filter This stage implements a band pass filter with central frequency fe 50 Hz It eliminates the DC of the sensors output and filters high frequency noise due to electromagnetic interferences It passes the only component of interest that is the 50 Hz sinusoidal wave corresponding to the current 119 High Pass Amplifier This stage implements an Amplifier with High Pass character istic with the gain at 50 Hz
72. gure 4 13 Data logging entity 4 4 5 Indirect Power Sensing WSN The Indirect Power Sensing WSN is a Wireless Sensor Network designed to monitor individual appliances and provide its internal power state to the Monitor It is demanded to be easy to configure and have low power consumption The network was developed with the following principles in mind Star Topology There is one master node responsible for network maintenance and coor dination Self Configuration The network configures itself when deployed favouring easiness of installation 54 Event Driven Communication The network tends to communicate only when there is new information on the nodes using minimum energy when not in function An in depth analysis of the WSN development and operation is documented in the report of the project Monitoring System developed in the course of Redes de Comunica o em Ambiente Industrial Communication Networks in Industrial Environment Network Topology The network is deployed in a logical star topology with a master node in the center connected to a PC and the slaves attached to the appliances to monitor Communication only occurs between master and slaves Network Self Configuration The network configures it self at start up The slave requests the master for registering in the network and awaits for reply The master receives the request acknowledges the presence of the slave in the network and sends a reply to t
73. hanges 4 Appliance Classification Figure 4 1 Load Disambiguation Algorithm Representation 37 11 13 15 17 19 21 23 25 27 29 31 33 35 37 4 1 1 Initial considerations To develop the algorithm it is necessary to choose the signature type that will be used to detect the appliances as well as their model The signature type used is the steady state with nnormalized complex power which is tolerant to variations in voltage supply The model in use will be the ON OFF model where each appliance has unique steady state ON and OFF signatures and where the OFF is the symmetric of the ON signature For identification purposes each appliance has a tolerance within its signature which results in the signatures being characterized by circles in the P Q plane in Phase Quadrature same as Real Imaginary for each state The circles must not overlap for any appliance signature 4 1 2 Pseudo code Implementation In this subsection it is described the Load Disambiguation Algorithm in Pseudo code to complement the block diagram of figure 4 1 Data Acquisition and Normalizationx at sampling frequency Hz Vrms get RMS Voltage Power get Complex Power Norm Power Nominal Voltage V 2 Power New Norm Power Value Norm Power Step Detection Steady State Detection x last_window_steady false at sampling frequency Hz power window push Norm_ Power
74. he appliance detection in the Non Intrusive Load technique It can be concluded that the the detecting of appliances via WSN does not compromise the accuracy and correctness of the system and can be used as true detector for single appliances without the requirement of correlating information about the power of the electrical installation 80 5 5 Complete System Test and Validation To verify the correctness and accuracy of the developed system it was simulated the function of a home environment with all the appliances switching ON and OFF independently from one another The system was set up to detect the Fan and the Two level radiator by the load disam biguation algorithm and the 2 Lamps and the Fan heater individually by the Indirect Power Sensing WSN A sequence of 100 switch events was performed by switching ON and OFF the appliances To avoid introducing dependence upon switching events were performed following a random sequence of numbers from random org 37 website The periodicity of the switches was 30 seconds to guarantee that the power is steady thus it is expected that all the transients are over and to be physically possible by manual operation The following data was collected e Measured Power e Expected Power e Residual Power e Events detected for each Appliance Table 5 9 presents the events performed and those which were detected by the system Appliance Performed Detected Turn ONs
75. he ideal output code Thus ADE7953 provides gain error calibration for each of its ADC channels The gain adjustment allowed is in the range of 0 5 1 5 The gain adjustment is set in appropriate registers and is regulated by equation 5 1 Ost ix om pue input 02400000 with 0x200000 lt GAIN lt 0x600000 The gain error calibration for voltage and current channels can be performed using the values of VRMS and IRMS from the registers of ADE7953 and with the measurements of RMS voltage at ADC s channels input Knowing that the ADC has a full scale input of 500mV and that the expected value in the registers of VRMS and IRMS at full scale input 500 mVems is 90320074 4 the calibration equation becomes v2 Value Expected AIN 0x4 2 nun Value Actual en which can be modified in order to be applied iteratively to Value Expected AIN GAIN _ s a Value Actual 2 with the order of the iteration The first iteration assumes GAIN _ 0x400000 The calibration method for any of the ADC channels is 1 Get VRMS or IRMS registers from ADE7953 during an interval of 10 seconds at a sampling frequency of 10 Hz Compute the mean value to get Value Actual Parallelly measure the mean value of Vrms at ADC s channel input with multimeter during the same 10 seconds interval with that value find Value Expected 2 Apply equation to calculate GAIN 3 Apply GAIN to the system 6
76. he most correct way to represent the steps but as long as the steps are always represented in the middle of the period the time coherency is maintained Oven Power e Measure Of Change UP Step DOWN Step ff A PT y M Steady Changing Power o Steady Changing Steady Changing Changing Time Figure 2 5 Detection of Steps 2 5 4 Appliance Classification It is assumed that there is a priori a list or database with the appliances as well as its characterization in terms of signatures defined by confined areas in the complex plane in this case in which it s only considered ON OFF models the OFF signature of an appliance is the symmetric the ON signature The algorithm takes each step and finds if it belongs to any of the areas of appliances signatures being discarded otherwise The output is then the appliance and the state change i e an ON or an OFF 2 5 5 Appliance Tracking Here the information about the appliances changes is used to update and maintain its state on a database of the registered appliances guaranteeing coherency of data avoiding and detecting anomalies such as repeated events such as two ONs or two OF Fs in a row 2 5 6 Data Output Tabulate Statistics Lastly once we have the information about how the appliances are behaving i e their power and their times of changing state and consequently the periods on which they are consuming e
77. he presence or not of current It works as a comparator with hysteresis G 0 1 Physical Considerations The circuit was developed to interface directly with uMRF s being designed to be supplied by a 3 3V source It was found that the supply of uMRFs when connected to the circuit generated noise that made the circuit unusable causing false detection of states To eliminate this problem the circuit is supplied externally by a 3 3V supply deployed by a 9V battery and a UA78m33 linear voltage regulator The schematic of the circuit is represented in figure Which fully assembled version of system is shown in figure G 3 As it can be seen the sensor was attached to the cable via nylon clamp 120 GND MCP6022 E P R1 1M D1 3V3 SINE MCP6022 E P gles T2 2u 3 1k R GND DH Be R10 Lo tr GND t ircui irect Current Sensor Ci 1rec f the Ind 121 Schematic o Figure G 2 Figure G 3 Fully Assembled Version of Indirect Current Sensor Circuit 122 Bibliography 10 11 12 13 14 15 16 17 78M6618 Evaluation Board User Manual Rev 1 1 Maxim Integrated Products May 2011 78M6613 Single phase AC Power Measurement IC Rev 3 11 Maxim Integrated Prod ucts Mar 2011 78M6613 Split Phase Evaluation Board User Manual Rev 1 0 Maxim Integrated Prod ucts Jun 2011
78. he slave indicating that it has been registered Once the slave receives the reply of the registering phase its is ready to work normally Network Maintenance The network has detection of failures for slaves by an I m Alive mechanism This detection of failures mechanism is implemented by the imposition on the slaves of a periodical transmission to the master an I m Alive message to indicate that it is working The master detects if there is any problem with a slave by detecting failure in receiving the I m Alive messages from the slave WSN System Function The master is in the center of the network It has full knowledge of all the slaves in the network as well as the states of the sensors monitoring the appliances The master is responsible for coordinating the network and to communicate with the PC informing it of all the states of sensors and signalize when a state change occurs in any of the appliances The slaves are at the ends of the network and attached to the appliances being monitored The slaves must inform the master every time their sensors detect a state change in an appliance In figure 4 14 it is represented the network and a sequence of events that occur when one appliance changes state Slave 1 detects a state change in the appliance that it monitors and informs the master of that occurrence After checking the information s coherency the master informs the PC of the change of state in the sen
79. his limitation it is proposed an inexpensive individual load monitoring with the deployment of a WSN based with Indirect Power Sensing 14 Chapter 3 System s Architecture and Hardware Development This chapter starts by giving an high level overview of the system that was developed be ing identified its building blocks and basic function It is discussed the implementation of the hardware platform needed to perform the Energy Monitoring and the selection of components The discussion starts with the introduction of some considerations about the requirements it proceeds to the comparison of the available materials devices and components and finally its set up and deployment 3 1 System s Architecture The developed system can be seen as a group of individual building blocks that perform the essential operations for the deployment of a Fine Grained Electrical Energy Monitor These blocks are represented in figure 5 1 and will be described in this section Main Power Source and Total Load The system targets a real household The power source it is assumed to be 230 V single phase at 50H z There are various independent appliances connected to the internal distribu tion infrastructure Data Acquisition In order to monitor the total load and get the needed values of complex power and voltage a data acquisition system needs to be deployed This system will be an integrated solution that senses the voltage and the current and pr
80. his statistics shall be e Energy spent by each appliance with its distribution over time and cost e Calculations on variations of energy consumption during day week month e etc Graphical User Interface GUI A graphical user interface gives visual feedback of the electric installation to the user This GUI takes as input the real time stats to display e Graphs with circuit voltage current and total complex power e Real time state of each appliance e Power spent by each appliance 16 e Power Cost weight of each appliance in the total power Historical data and statistics to display e Detailed Electrical Bills e Graphs of power consumption versus time e Any other relevant statistic worth of being displayed Indirect Power Sensing and WSN To help the Load Disambiguation Algorithm with similar loads it was deployed a system based on Indirect Power Sensing that provides via WSN information about individual appli ances to the system The WSN measures physical quantities that indicate that the appliance is working The WSN system is deployed to detect changes i e to detect if an appliance changes state and then to communicate it to the computational system to augment the completeness of its report 3 2 Deployment of Data Acquisition System To perform NIALM one has to have access to various parameters of the facility s electric installation as it was stated in Chapter 2 these parameters are RMS Voltage and I
81. http en wikipedia org wiki Lux MAGNETIC CURRENT SENSING AN 209 Honeywell Sensing and Control 2000 Manual da qualidade da energia el ctrica edp Dec 2005 A Marchiori D Hakkarinen Q Han and L Earle Circuit Level Load Monitoring for Household Energy Management Pervasive Computing IEEE vol 10 no 1 pp 40 48 Mar 2011 ISSN 1536 1268 DOI 10 1109 MPRV 2010 72 MCP3909 dsPIC33FJ128GP206 3 Phase Energy Meter Reference Design Microchip Technology Inc 2009 MCP3909 Energy Metering IC with SPI Interface and Active Power Pulse Output Microchip Technology Inc 2009 Optimizing Performance from Rogowski Coil Current Transformers 2011 W Pettigrew Selecting the Most Effective Current Sensing Technology Power Elec tronics Europe vol Issue 8 Dec 2007 PIC18F87J72 Single Phase Energy Meter Reference Design Microchip Technology Inc 2011 PIC32 Family Reference Manual Microchip Technology Inc 2011 PIC32MX5XX 6XX 7XX Family Data Sheet Microchip Technology Inc 2011 Feb 2012 Premier Farnell plc Online Available http pt farnell com Nov 2012 Random Integer Generator Online Available http www random org A Ruzzelli C Nicolas A Schoofs and G O Hare Real time recognition and profiling of appliances through a single electricity sensor in Sensor Mesh and Ad Hoc Commu nications and Networks SECON 2010 7th Annual IEEE Communications Society Conference
82. if power_window STD lt STD Threshold if last window steady false i New Steady State Detectedx state power power_window get_mean_Power state time System get_Time new_Steady_State state last _window_steady true else power_is_changing last_window_steady false 38 39 41 43 45 47 49 51 53 55 57 59 61 63 65 Step Calculation for each new Steady State Step old Steady State new Steady State new_Step_Detected Step Appliance Classification x for each New Step Detected Comparison by euclidean distancex Closest App Signature step find closest signature if step is within tolerance Closest App Signature t A Change of State in an Appliance has been detected x Appliance Closest App Signature getAppliance Event Closest App Signature getEvent newEventDetected Appliance State else No Operation 4 1 3 Data acquisition By sensing the installation s current and voltage it is computed the RMS Voltage and Power which are sampled for further processing 4 1 4 Normalize Normalize takes Voltage V and Power P as inputs and normalizes P through Vnom A Prorm P 4 1 41 where Vnom is the nominal voltage of the load being monitored Following there is a pseudo code representation 4 1 5 Step Detector The step detector takes the normali
83. igned because it is multi state being only possible to described by the FSM model 4 3 Preliminary Algorithm s Software Implementation Now that the algorithm was verified in MATLAB it is necessary to implement it in a general purpose programming language The programming language chosen was JAVA for its easiness integration and portability The implementation here is a bit different than that on Matlab since the data is processed iteratively as it is pushed into the processing instead of being processed as a whole 45 4 3 1 Step Detector The heart of this implementation relies in the stepDetector class interface where are implemented the structures and functions to perform step detection Firstly the structures Processing Window This is the window for signal processing where the power flow is pushed to be analysed by the operations of standard deviation and mean Steady States Queue This is a FIFO queue for steady states where they are pushed after being identified The data in this structure are states that are defined by a steady state power signature and a epoch time stamp Non Steady States Queue Alongside the queue for steady states there is a queue for non steady states which the only data of value to push in is the time stamp of the identification of a start of a changing period Steps Queue This is the queue where the detected steps are pushed into for passing to further processing The methods Push i
84. in the middle of the steady states as yellow stems and the negative steps turn OFF are represented as black stems The step threshold used in this case was of 100 W 42 3 oven Real Power and Standard Deviation 1600 1400 Real Power Standard Deviation 1200 1000 800 Power W 600 400 200 3 53 3 54 3 55 3 56 3 57 3 58 3 59 3 6 3 61 Seconds s x104 Figure 4 4 Pass of Standard Deviation window through the Real Power of an Oven 3 oven Aparent Power and Standard Deviation and Steady States E A 1400 Aparent Power Standard Deviation 1200 O Steady States 1000 o 800 o a 600 400 200 0 3 53 3 54 3 55 3 56 3 57 3 58 3 59 3 6 3 61 Seconds s x 10 Figure 4 5 Representation of Steady State Identification 43 3 oven Aparent Power and Standard Deviation and Steady States Aparent Power Standard Deviation TURN ON event 2 TURN OFF event 1 3 54 3 56 3 58 3 6 3 62 3 64 3 66 Seconds s 4 Figure 4 6 Representation of Step Identification 44 4 2 3 Results Some of the appliance files of the REDD database were selected to validate the algo rithm After analysing the individual power flows of the various appliances it was possible to characterize some The parameters of the analysis were e Time interval 3134061 seconds e Window Size 30 seconds e Standard deviation threshold 50 Watts e Step Threshold 100 Watts Table 4 1 shows th
85. istema dever identificar as cargas individuais na casa o que faz com que o hardware tenda a ser simples e minimal recaindo a complexidade para o software O sistema desenvolvido dever ser de baixo custo e de f cil instala o O sistema que foi desenvolvido foi capaz de identificar individualmente e com grande sucesso os aparelhos presentes num ambiente controlado Ao contr rio de outros sistemas deste tipo o sistema apresentado nesta disser ta o capaz de distinguir cargas eletricamente id nticas atrav s do uso da t cnica de Medi o Indirecta de Pot ncia implementada atrav s de uma Rede de Sensores Sem Fios Keywords Abstract Non Intrusive Appliance Load Monitoring Wireless Sensor Networks Load Disambiguation Indirect Power Sensing Nowadays we witness a great pressure and effort in order to stimulate energy conservation either due to economic constraints environmental concerns or regulatory pressures At the top of the energy consumption is the con sumption of energy in the form of electricity The first step to improve energy utilization efficiency is energy awareness So if there is some manner to provide the average person information about how he is consuming electricity he may take it by its own interest and In fact if electricity consumers have access to detailed information about how when and by which devices energy is consumed they can make good and well supported decisions on how they can change
86. ith the total current to be measured passing through it being its output described by the transfer function of equation Lsn vi Rsh Figure B 1 Current shunt simple model V Rent juLsl 3 B 1 It must be taken into account when designing the sensor a phase mismatch due to the parasitic inductance As there is no galvanic isolation of the hot parts of the circuit the shunts are normally used in low ground referenced currents in the order of the tenths of Amperes 32 It must be 99 also noted that these kind of sensors dissipate high amounts of power which cause the need for thermal dissipation increasing the sensor s footprint and cost Figure B 2 shows some examples of shunt resistors i 4 be k NA Figure B 2 Examples of Current Shunt Resistors 9 B 2 Current Transformer One of the most used and most reliable solution for measuring current is using current transformers These rely on the transformer s magnetic coupling between two coils being the primary provided by the user being commonly only one turn A simple model of a current transformer is shown on figure B 3 And the input output current relation is represented by 12 Figure B 3 Simple model of a Current Transformer equation n 19 o B 2 ae B 2 that can be generalized to equation _ B 3 12 n2 The number of turns on the secondary must then be designed to suit the application s need As one may see this sensor provi
87. itivity to the environment As the circuit output must be digital at the previous circuit s output it is put a schmitt trigger in order to force the digital levels This sensor circuit is to be attached to uMRFs where it will output its digital value and from where it will get its power supply of 3 3V The schmitt trigger was deployed by one gate of HEF4098 Quad Nand Schmitt Trigger by driving one of its inputs to 3 3V and the other to the sensor output The resulting schematic may be seen in figure D 2 111 Figure D 2 Light Sensor Circuit Schematic Circuit Analysis By analysis of HEF4093 s datasheet it can be found that its hysteresis if supplied by 3 3 V is Vin 2 243V Vin 1 415V So the sensor has to exceed V rr when detecting light and has to go below Viz when the appliance is off To guarantee correct operation and simplify calculations it will be considered a greater threshold for V 4 and a smaller for Viz at the sensor s output Let them be Vin 2 5V Vin 13V Now to make the correspondence with light it must be first calculated the current V I otoH gt D 2 photoH R Rp for high threshold and Vit 1 otoL gt D 3 photoL R Rp for low threshold After some calculations and conversions to light according to TEPT5600 s datasheet 43 the light thresholds for R Rp 10kQ are Photonign T0lux Photojow 40 lux For R Rp 20k0 the light thresholds are Photonign 40 lux
88. ix E Current and Magnetic Field 115 Appendix F UART Protocol 117 Appendix G Indirect Current Sensor 119 Bibliography 123 List of Figures ee ee 4 2 2 Finite State Appliance Models af generic 1000 W two star appliance no o io a eae Bevel on figure of T ett oe Db A E ee ee CE a ee 11 Fo obs sie vay A peste ad Navas ep dices 12 ee Die Ged Ree Ge dt ae ee ge Ge a 16 Der Beil Bo aces O A 19 paa Galore ha Gee cen Gg AS Ae eee eee ee A eS RO 20 3 4 Load System Configuration 2222222 e 21 E dee dis L Sea eee bite TE cea be 22 3 6 chipKit Max32 development board a 0 wu u u u a eee He ar wa 23 pe ol ee AE READ WS ae ee ED 24 3 8 SPL Packet su sera nam Ke pe Ra Ree CR ee RE aa ee a 25 Pee Ge es ak eee en ee ae ene See eee ee a eee 26 3 10 Gateway Read lt se c aos a aane afi E E a e Oa 26 E a ae aa erat Gs de O 28 3 12 UART protocol basic packet 2 2222 2 ee ee 29 3 13 Gateway Dataflow 2 0 ee 30 3 14 uMREs wireless nodel5 conductor cable b Magnetic Flux vs TA SIS Gee E aa 517 5530 Series Linear Hall Effect Sensors ee 34 ae Block Diagram for indirect current sensing 2 2 22 rn nn nn 35 Phe Re eave a 35 bite ee Ve aa ease 37 PR ee GL eRe eRe des 40 264 bard hag AD a we hand oS Al o 4B 4 5 Representation of Steady State Identification 08 43 4 6 Representation of Step Identification 2 2200 44 iii 4 11 Step Detection entity
89. lationships may be expressed as Voltage Input V Current Input V Veons ant Lconstant 5 11 tant VRMS LSBs tant TRMS LSBs wall 69 The meter constants ought to be calculated under Steady conditions and with very precise loads It was observed that voltage varies with the load in the circuit In addition the test bed is connected to the installation of a large building where numerous appliances are always working and switching so one cannot guarantee the steadiness of voltage For this reason a simple measure of Vrms or Irms does not suffice to accurately find the meter constants The meter constants will be computed with the linear regression of various points of current and voltage observed with different loads Method of calibration for Voltage Meter Constant A list of ordered pairs VRMS LS Bs Voltage Input V Voltages List needs to be ob tained by the following steps 1 Set System s load 2 Get VRMS register from ADE7953 during an interval of 10 seconds at a sampling fre quency of 10H z From the values obtained compute the mean to get one VRMS LS Bs value Parallelly measure mean value of Vrms at the Electric Installation input with the multimeter during the same 10 seconds interval to get one value of Voltage Input V 3 Store ordered pair in Voltages List The previous steps are to be repeated until the needed number of values are obtained Next a linear regression of Voltage Input V VRMSI
90. le 5 6 it can be concluded that all the steps were detected for all the appliances which can be interpreted as a measure of correctness of the step detector Due to the nature of the system it is necessary to detect the appliances in real time thus the system cannot delay its processing in order to wait that the transients of the appliances to end There is so an inherent notion of the signal being sufficiently stable to be detected as 75 a steady state This notion is expressed by the detection of steady state upon a threshold on the standard deviation of the processing window This can affect the step detection in a way that the steps detected may not be exactly equal to the appliances steady state consumption though they shall be close The appliances that should be mainly resistive such as the radiator heater and the lamp have very fast transients and the detections are very close to the steady state signature being the most straight forward to detect Appliances that by nature of the physical task present slow transients such as the fan and the fan heater which have motors that require long times to stabilize the spinning speed present greater deviations from steady state signature Another effect worth to notice in this type of appliances is that the detected signatures may differ greatly if a turn ON is performed by cold start i e the motor is stopped where the transient will be the slowest and the deviation will be big or if
91. liances using inexpensive sensors Indirect Power Sensing is based on the fact that an appliance emits measurable signals when it consumes energy By sensing these signals one could estimate power consumption 21 This concept was introduced by Younghum Kim et al in 2009 with Viridiscope a Fine Grained Power Monitoring System for Homes 21 where he uses some inexpensive sensors such as light sound or magnetic sensors to estimate the power consumption of individual appliances These were attached in a wireless sensor network that would report the gathered information to a PC Fusion Center that makes appropriate calculations to estimate the power being spent Another use for this concept was documented in ANNOT Automated Electricity Data Annotation Using Wireless Sensor Networks where Schoofs et al where the data gathered by the sensors is used to annotate the data about power consumption in order to facilitate the installation and training of a NIALM system 39 It is hereby proposed the use of WSN with Indirect Power Sensing to augment the NIALM system by gathering information to help distinguishing similar identical appliances and identify small loads Nodes have to be attached to the appliances of interest to identify if the appliance is ON or OFF and report it to the main monitoring system The information gathered by the WSN is absolute The deployment of a WSN implies little intrusiveness But nodes are small and can be integrated seaml
92. m with identifying electrically identical loads was successfully tackled with the Indirect Power Sensing WSN It was shown that it was possible to detect the internal power state of some appliances by using simple sensors that sense physical quantities in the vicinity of the appliances It was possible to detect lamps with light sensors placed near the luminous source It was also possible to detect large loads by sensing the magnetic field created by the current flux in their supply cable with a hall effect sensor attached to it without the need of breaking the circuit The gathered information is used to find the state of the appliances in the circuit being able to tell what appliances are ON or OFF in real time and historically This information is believed to be useful for improving the user s awareness of how he is spending electrical energy and hopefully for providing insight to reduce his consumption habits in order to reduce the expenditure The overall cost of the data acquisition system system is low Although the ADE7953 Eval Board was expensive its cost doesn t reflect the material value and it can be made very cheap Thus considering the availability of small low cost computers such as the Raspberry Pi in my opinion a system like the one here developed even if without the Indirect Power Sensing module could be very easily deployed in every home without a representing big cost to the costumer and it would pay itself in short te
93. n case of a write All the data is sent or received MSb first a A representation of a packet may be found in figure 24 Active BUS nn za Start of Packet End of Packet Figure 3 8 SPI Packet PIC32 SPI Interface The PIC32 provides a full function SPI hardware driver that can provide master or slave communications For this implementation this interface is configured as master providing a clock of 1 Mhz to the slave and will be responsible of managing the communications Assum ing that the system needs the data from Voltage Active Power and Reactive Power registers at a frequency of 1 2kHz being the registers of 24 bits and having the communications an overhead of 24 bits the minimum clock of the SPI has to be 173kHz This configuration is more than sufficient to the specified requirements UART Interface An UART interface is used to connect the gateway to a PC This UART interface is implemented with PIC32 s UART driver connected to FTDI FT232R USB UART converter that connects with PC v a USB and is seen by Operative System OS as a virtual COMM RS232 port The interface is configured to operate asynchronously at a baudrate of 115200 bps Assuming that the gateway would work with an optimized protocol that sends to the PC the needed data continuously with a minimum overhead of 2 bytes one before the data and one after having the system to get data from three 24 bit registers at 1 2kHz the data has to arrive at
94. ned before the ADE7953 Evaluation Board doesn t provide a direct way to sense current One has to design sensory circuitry to allow current sensing meeting the platform s 20 UTILITY S Ge POWER SOURCE Neutral Phase 19 Current Transformer CR8349 IAN VP 230 Vrms 00000008 Figure 3 4 Load System Configuration requirements Only one current channel needs to be used Channel A was selected since it is the most versatile 4 Current sensing will be then made through the current transformer mentioned in subsubsection 3 2 1 The project of a transimpedance circuit to measure current is done in appendix C where the sensor circuit was deployed with a Royrden 2 50 1 The circuit then connects to the Ta and Ta of ADE7953 Evaluation Board 3 3 Communicating with the Data Acquisition System 3 3 1 ADE7953 Communication Model Memory Model Before implementing any kind of communication with the ADE7953 it is important to understand the underlying concepts All of the features can be accessed through addressable registers via 3 distinct communication interfaces Registers are 8 16 24 and 32 bit long The data in 32 bit registers is the same of that in its correspondent 24 bit registers through signal extension These differences are justified with the 24 bit registers being more efficient in terms of communication speed while the 32 bit registers provide easier programming with the long format
95. nergy or not we can produce a big number of statistics and other valuable information The system shall be able with little effort to 12 e compute how much energy is being spent by any individual appliance at any time e compute the cost of any appliance within the total bill e tabulate the variation of energy consumption with time of day temperature One of the aims of this block is to produce a detailed bill of appliances consumption such as the one that is common at the phone bills Here we would have the times at which the appliances were on for how long and how much they cost 2 6 Known Limitations There are some capital limitations in what concerns NIALM They result primarily from the fact that this technique is only applicable to a restricted set of appliances It s not suitable for detecting very small appliances continuously variable appliances and appliances that operate constantly Apart from that there is another fair limitation that is that it can t distinguish between electrically similar appliances This limitation is quite interesting because one may have similar appliances in different rooms of a house such as lamps and one may be interested in knowing how much energy each one consumes 2 7 Indirect Power Monitoring WSN To ease some of the limitations above mentioned it is possible to deploy an Wireless Sen sor Network WSN with Indirect Power Sensing in order to get information about individual app
96. nstant Complex Power Thus it is necessary to get deploy a system capable of being attached to a home s electric circuit and monitor these parameters Considering the type of analysis that it s going to be developed which is load disam biguation via step detection on steady state signatures the requirements of this system are providing a measurement of complex power and a measurement of voltage at a sampling fre quency in the range of 2 10 Hz as suggested in 18 However the system developed in the scope of this thesis aims to be a versatile platform capable of sophisticated analysis to be used in scientific research in the field of Load Monitoring so it is reasonable to consider that it necessary to develop a system capable of providing data for processing transient analysis and harmonic analysis on the various components of an electrical installation These re quirements imply that the system has to provide measures of voltage current active power and reactive power at high sampling rates thus a minimum sampling rate of 1 2 2k Hz is suggested for high harmonic and transient analysis 5 In order to accelerate the development process it was decided to use an integrated solution for monitoring electrical parameters This option is viable since as it will be seen further on there are available in the market diverse equipments that satisfy the specifications and that are affordable Market Survey and Device Selection System s
97. nto Window This is a method to push one power sample to the window in order to slide the power flow Once the window is full at each push the oldest sample is discarded Steady Power Discrimination This is a method to discriminate if the power in window is steady by comparing the standard deviation to a pre determined STD Threshold States Identification This is a method that by observing the succession of steadiness or not of the power flow identifies and pushes a steady state a power sample with the mean of the window at the time of the detection and the time stamp to the Steady State Queue The non steady states are also identified and pushed to the Non Steady State Queue with a dummy power sample and the time stamp Step Detection This is a method that takes at each time the two least recent steady states and finds the distance between them comparing it to a step threshold and classifying if they are steps pushing a power sample of the difference between the steps and a time stamp for that sample 4 3 2 Preliminary Appliance Classifier Having the steps been identified another important part of the system is the appliance classifier For that there is another class interface implemented that is responsible by iden tifying the appliances that are turning ON and OFF given the step changes There is an Appliance structure for registering what appliances are in the system in which the classifier relies to perform the classifi
98. o research on energy disaggregation 22 The data provided consists of whole home and device specific power consumption from various houses over several months For each house it was recorded the whole home electricity signal at high frequency 15k Hz For every recorded home the dataset provides data about up to 24 individual circuits These circuits are normally for categorized appliances Data is provided at 0 5Hz The dataset also provides data about up to 20 individual appliances that were monitored via plug level 41 monitors This data was recorded at 1H z 22 The only data that is useful in this project is that of the low frequency In case of whole home data it is available the Real Power at a frequency of 1Hz samples In the case of single appliance data it is provided again Real Power but at 3H z The sampling frequency is different from the frequency it was recorded because it is the result of some processing and not raw data For each monitored home there are individual files with data from aggregated load called mains and files for individual appliances 4 2 2 Testing the Algorithm in MATLAB The platform chosen to assess the algorithms validity was MATLAB due to its flexibility in programming and by making it easy to handle large quantities of data Single Appliance Characterization A script that characterizes an appliance by executing the disambiguation algorithm upon the data flow of power of a single appliance was de
99. oceedings of the IEEE vol 80 no 12 pp 1870 1891 Oct 1992 ISSN 0018 9219 DOI 10 1109 5 192069 Honeywell Hall Effect Sensing and Application 1998 123 20 21 26 27 28 29 30 31 32 33 34 35 36 37 38 K A D Karen Ehrhardt Martinez and J A Laitner Advanced Metering Initiatives and Residential Feedback Programs A Meta Review for Household Electricity Saving Opportunities American Council for an Energy Efficient Economy Research Report E105 2010 Y Kim T Schmid Z M Charbiwala and M B Srivastava Viridiscope design and implementation of a fine grained power monitoring system for homes in Proceedings of the 11th international conference on Ubiquitous computing ser Ubicomp 09 Orlando Florida USA ACM 2009 pp 245 254 ISBN 978 1 60558 431 7 DOI acm org 10 1145 1620545 1620582 Online Available http doi acm org 10 1145 1620545 1620582 J Z Kolter and M J Johnson REDD A Public Data Set for Energy Disaggregation Research in SustKDD Workshop on Data Mining Appliations in Sustainbility 2011 W Koon Current sensing for energy metering 2001 C Laughman K Lee R Cox S Shaw S Leeb L Norford and P Armstrong Power signature analysis Power and Energy Magazine IEEE vol 1 no 2 pp 56 63 Mar 2003 ISSN 1540 7977 DOI 10 1109 MPAE 2003 1192027 Sep 2012 Lux Wikipedia Online Available
100. ompo nents market but in less options and quantity as they many times customized for each project though it is possible to find general purpose current transformers with ranging ratios from 1 1000 to 1 2000 in a price ranging from 6 to 15 US dollars 12 Linear Hall Effect Current Sensor There are viable solutions in the general market with prices ranging from 15 to 100 US dollars 36 Rogowski Coil Despite being a very interesting solution for energy metering the availabil ity of the Rogowski coils in the general components market was too scarce so there wasn t enough information to reach conclusions about its price 105 106 Appendix C Transimpedance Current Sensor with Current Transformer The maximum full scale input at the terminals JAp and An of the ADE7953 current channel is 250mV and a common mode of less than 25mV is recommended 4 Current sensing will be made through the current transformer mentioned in subsubsec tion 3 2 1 which has the following characteristics e Te 1520 effective number of turns e IR 15A rated current for which output is linear e DCR 800 DC resistance e Vmax 12 8V gt ms maximum voltage at saturation The transimpedance circuit will be dimensioned to the Current Transformer s CT rated current and a schematic of that circuit is in figure The output is DCR ga Yi E Rburden Vr o 4 1520 Figure C 1 Simple model of a Current Transfo
101. on Jun 2010 pp 1 9 DOI 10 1109 SECON 2010 5508244 124 39 40 41 42 43 44 45 A Schoofs A Guerrieri D Delaney G O Hare and A Ruzzelli ANNOT Automated Electricity Data Annotation Using Wireless Sensor Networks in Sensor Mesh and Ad Hoc Communications and Networks SECON 2010 7th Annual IEEE Communications Society Conference on Jun 2010 pp 1 9 DOI D E Shepard and D W Yauch An Overview of Rogowski Coil Current Sensing Tech nology 1998 Solid State Sensors Linear Current Sensors CS Series Product Catalog Honeywell Sensing and Control SS39IET SS49IE SS5IET Series Linear Hall Effect Sensor ICs Honeywell Sensing and Control Feb 2012 TEPT5600 datasheet Vishay Semiconductors Aug 2011 Wattnode Modbus Instalation and Operation Manual Rev 1 16c Continental Control Systems LLC Jan 2011 M Zeifman and K Roth Nonintrusive appliance load monitoring Review and out look in Consumer Electronics ICCE 2011 IEEE International Conference on Jan 2011 pp 239 240 DOI 10 1109 ICCE 2011 5722560 125
102. ovides directly digitally and at high sampling rates the characteristics of the the electric installation This system has to have PC connectivity to where the data must be delivered Load Disambiguation In the computational system it is deployed an application that takes the data from the data acquisition system identifies what appliances are turning ON and OFF in real time Additional information from appliances of interest is fetched via a WSN in order to distinguish 15 Complex Power Computer Q gt and Voltage NER ar Data Load Disambiguation Statistics and GUI Aquisition Appliance Detection Calculations Voltage and Current Main Power Sensing Source Main power line Data Logging Appliance Database Total Load X E TOS gt Appaii i App2 od Indirect Sensing f a WSN Node y Node2 N Figure 3 1 System s architecture overview between similar appliances The produced information is then used to maintain a database that reflects the state of all the appliances at any time The state changes of the all the appliances are also recorded to make possible the production of information that rely on historical data about the appliances Statistics and Calculations This part of the system takes as input the real time and the historical data on power consumption for each appliance as well as for the total load and computes and calculates important and interesting statistics T
103. ower current and admittance of the aggregated load all these parameters may be expressed in the complex form From this parameter s we may search for step changes as signatures The most intuitive parameter to use for someone with electric circuits background is power as it is widely used and gives a quick perception on energy being spent by the appliance But as this parameter and as current are dependant of voltage and there are some vicissitudes that make these inadequate for use The utility s voltage V t is time varying and the power provider only guarantees it within a fluctuation of 10 of its nominal 230 V pm which will consequently make the power drawn by linear devices to vary over 20 In addition there is also a variation in voltage at the load depending on the load itself due to the power source not being ideal by having an equivalent internal resistance Basing NIALM in the detection of changes around a parameter that can varies 20 due to external factors is not acceptable Admittance is preferable since it is independent from voltage and it is additive since the appliances are all wired in parallel as suggested in figure 2 1 Let P t the power and V t the RMS voltage Admitance Y t is P t Y t V t While this may be the adequate parameter for the signature it lacks intuitive meaning therefore it is be preferable to handle this parameter through a linear transformation called normalized power
104. ows the results of the assessment of the gain error comparing it to the accuracy of the reference errorgain represents the ratio in percentage of the gain error of the ADC to Tactual and accuracy represents the accuracy of reference multimeter regarding IUmeasured It may be observed that the gain error is always smaller than the reference s accuracy which means that the calibration may exceed the reference s performance Once again it is not possible to improve the calibration with this reference IvmeasuredlmV Lespectea ESB Lactuail LSB errorgain accuracy 9 326 136060 135535 0 39 0 77 9 328 136111 135618 0 36 0 77 9 390 136673 136792 0 09 0 77 12 702 324490 325841 0 41 0 50 12 777 326406 326448 0 01 0 50 Table 5 2 Assertion of gain error following the calibration of current channel RMS Voltage and RMS Current Meter Constants The data obtained from ADE7953 registers doesn t have direct physical meaning as it is relative only to the values sampled in the ADCs Thus it is necessary to convert the raw values into meaningful values in Volts Amperes Watts etc These values that convert the register data to physical values are called Meter Constants All the values needed to the project that are related to the electrical installation are related RMS Voltage and Current so it is only necessary to find Meter Constants for these The others can be subsequently derived These re
105. p Detection Calculations Graphical Appliance User Interface Classification GUI Fine Grained Electricity Monitor Figure 4 9 Fine Grained Electricity Monitor and its logical entities Data Acquisition Obtains raw data and manipulates it to normalized power Step Detection Takes normalized power and identifies steps within it Appliance Classification From identified steps classifies it in meaningful appliance events Data Logging Records and saves any data valuable for processing and statistics Statistics and Calculations Takes historical and real time data to produce meaningful statistics and figures Graphical User Interface Shows real time data and historical statistics about power con sumption in order to deliver valuable data to the user Indirect Power Sensing WSN Provides individual state data about appliances of inter est to help in disambiguation of similar loads In the center of the system there is a logical database which holds all the valuable data for the systems function where all the data either externally fetched or internally computed or is stored 50 4 4 1 Data Acquisition ies ager UE e 7 Recorded Data _ x un 7 De re mn E Lu Data Acquisition _ Y Se PE AN VRMS lt EN ER aa Data Acquisition A Normalized q Convertion Irms lt System N Power O Power N l
106. ppliances Other electronic devices with switched supplies may have power at higher frequencies such as in the range of the ultrasounds 20 40 k Hz almost all appliances other than resistive heaters and incandescent lights have power in various harmonics 18 One way of building data with this informations is by using a structure called spectral envelope and gathering the powers of the harmonics A spectral envelope may be a vector of the powers at a predetermined number of harmonics 24 This characteristic can also be used as signature for NIALM by augmenting the power vector for the steady state signatures This signature may be useful to disambiguate some appliances that are electrically identical in the utility s frequency but different at other har monics Direct current signatures In sequence of the non linearities there may be some appli ances that have DC components different from zero This behaviour may be noted on some small heating appliances in which manufacturers may put a diode in series with the heating element to create a two level system having the rectified operation as the low setting This half wave rectification produces a significant DC level in the current waveform 18 While this type of signature is very uncommon it has the potential of aiding in some appliance disambiguation Transient Signatures As stated before transient signatures are electrical signals that are only present in the load for a short dura
107. ppliances is summarized table Appliance Signature W Rise Time s Fall Time s Heater 806 75 21 321 0 14 0 16 Heater2 1210 49 30 312 0 14 0 16 Fan 39 78 0 802 8 46 6 11 Lamp 106 62 4 722 0 22 0 18 Fan Heater 1030 64 5 60 1 56 0 18 Table 5 5 Appliance s Characterization 74 Analysing the results from table 5 5 it is clear that the appliances that should be mainly resistive such as the radiator heater and the lamp have faster transients than those that have inductive element and high mechanical inertia have slower transients such as the fan heater and the fan Furthermore the values obtained for the rise and fall times for the fan are much higher than the others This comes from the fact that it is a very small load in which its transient may be confused with noise present in the system The reactive power measurements of the system seem to be disconnected from reality by a systematic error that is shifting them negatively However this error does not affect the detection of loads as the algorithm only requires that the signatures are separated in the P Q space 5 4 Testing the Load Disambiguation Algorithm 5 4 1 Step Detector For testing the step detector step like waveforms were introduced to the system by switch ing ON and OFF the appliances in the test bed and analysing the step detector output which is provided by the data logging entity and is characteriz
108. r and have special care with the sensor placement it is possible to detect if an appliance is consuming energy or not and with sufficient training it could be even possible to differentiate ON states for multi state appliances SS49E Linear Hall Effect Sensor IC SS49E SS39ET SS49E T3 SS49E L SS49E T2 SSS59ET I BEE AM Mad es Figure 3 17 SS39 Series Linear Hall Effect Sensors The sensor selected was the SS49E Linear Hall Effect Sensor IC figure This sensor is a low cost and compact linear hall effect sensor The sensor aims to be used for basic current sensing and for position displacement sensing with a sensibility of 1 4mV Gauss 42 Its main electrical characteristics are e Low voltage capability ranging from 2 7 to 6 5 V e Analog Output with 1 4mV Gauss sensibility e Maximum magnetic range of 1000 Gauss e Wideband frequency response up to tens of kilohertz Output voltage Vout 2 5 V at 0 Gauss Detecting Appliances Power State The objective is detecting the power state of the appliance by detecting the current flowing in the cable The aim is then to develop a circuit which output is a logical digital value indicating the presence or not of current in a cable To perform the above mentioned function it is necessary to perform certain operations Figure 3 18 shows a simple diagram of the implemented circuit for indirect current sensing Further details of the development of this circuit are described in
109. r impedance with the intensity of light in a phenomena called photocondutance These sensors have Cadmium in its composition not being suitable for consumer prod ucts having a RoHS prohibition Photo diode Photo transistor these detectors are semiconductor based with p n junc tions that convert light in electricity capturing photons creating a photo current in 31 phenomena called photoelectric effect The main difference between Photo diodes and Photo transistors its the output input gain which is much greater in Photo transistors The selection of photo transistor s comes naturally as they are RoHS compliant and are easy to use The sensor used for light sensing was the TEPT5600 from Vishay this is an ambient light sensor based on a silicon NPN planar photo transistor It is sensitive to visible light much like the human eye having its peak sensitivity at 570 nm 43 The transistor is placed in a transparent case capturing photons in the base collector Junction imposing a current flowing from collector to emitter photo current Sensor Circuitry The circuit to develop is a digital discriminator of presence of light so it is wanted to put the sensor in saturation It s then necessary develop a transimpedance circuit to convert photo current in voltage This sensor circuit is to be attached to uMRFs where it will output its digital value and from where it will get its power supply of 3 3V The circuit developed is presente
110. r that is agnostic with respect to the network the Indirect Power Sensing WSN is seen as a simple data device 56 Indirect Power Sensing WSN WSN Interface Signalize Event Appliances Sensors Adresses Sensor Addr 10 Sensor Database State ON Find Address Name Heater DATABASE WSN Classifier Tracker State OFF Appliance Sensor 10 o Appliance Stat Appliance Database Appliance Event Signal Figure 4 15 Indirect Power Sensing WSN entity representation Thus it provides a notification mechanism that fires asynchronously notifications every time a state change is detected in any sensor Additionally it also permits the access to information about the sensors available and their status by request At start up the interface builds the sensor database with all the sensors and their current state Then at normal operation it waits for messages from the WSN with information about state changes on sensors Once a message about a change in a sensor is received it is checked if the change is allowed taking into account the current state of the sensor If so the information on the database is updated and a notification for the WSN classifier with the information about the occurrence of state change on a sensor is fired WSN Classifier The WSN Classifier waits for an event notification on state change of a sensor Then by the Sensor address it
111. representing the real and the imaginary power of the ith appliance n P t Y aP e 2 2 i 1 where P t is the total power seen at the utility power source at time t and e t a small noise or error term This model may suggest a straightforward criterion for estimating the state of the appli ances Assuming that all the appliances in the circuit are known and the measured P t is given the solution reduces finding the vector a t that minimizes e t n Pit 2 ae Ps del 2 3 arg min While this may seem very attractive since it is a very well studied combinatorial optimization problem being able to be solved computationally by exhaustive techniques those may be proven impractical for growing values of n This model is not suitable for the project s objective either because the complete set of appliances may not be entirely known nor is acceptable to model a household electric circuit by a finite set of appliances as the presence of unknown appliances may cause the attempt of describing it by a combination of the others 18 Another problem that may prove that the solution of equation 2 3 is not suitable for the project is that a small change in the measured P t can be solved as a big change in the switch process a t with a number of appliances switching simultaneously in order to better describe the measured P t Assume that all P are known for example P 100 P gt 200 P3 300 Py 401 W If
112. resent in the circuit as well as their operating state Statistics and calculations on historical data are not yet implemented But the data logger already records data in a way that make the future implementation very easy 63 64 Chapter 5 Tests Results Validations and Calibrations This chapter presents a set of tests that have been carried out to validate the system In addition it is also presented the calibration methodology 5 1 Test bed Taking into account that the system should monitor a household electrical installation it was developed a test bed that simulates the electric circuit of a house The electrical power source is taken from a standard 230 Vrms power outlet to a 6 outlets power strip The voltage signal is directly fed to the data acquisition system through a con nection in parallel with the power strip while the current signal is fed to the data acquisition system through a current transformer in the phase conductor of the power strip Figure shows the most important components of the test bed s electrical installation Along with the electrical installation it is deployed the WSN based Indirect Power Sensing that aims at disambiguating similar appliances There are sensor nodes with light sensors to monitor 2 equal lamps which cannot be disambiguated solely by the normal algorithm It is also deployed one sensor node with a magnetic sensor coupled to a fan heater The reasons for this are that it showed to ha
113. ring IC to be evaluated ADE7953 Interface Board and ADE7953 Evaluation Board e The interface board handles the communication from PC to the ADE7953 e The Evaluation board includes the ADE7953 the needed isolation for high voltage inputs and anti aliasing filters figure A 2 All features available in the ADE7958 are available through supplied software or directly using SPI I2C or UART communication 14 90 Figure A 2 ADE7953 Evaluation board A 2 Microchip s MCP3909 A 2 1 MCP3909 Energy Metering IC with SPI Interface and Active Power Pulse Output The MCP3909 is an energy metering IC designed for standard metering applications the functional block diagram is presented in figure Its architecture is based on two 16 bit second order delta sigma Analog to Digital Converters ADCs and it has Programmable Gain Amplifiers in order to allow different types of current sensors The IC is capable of making waveform measurements and output them digitally via SPI The IC is capable of waveform sampling Current and Voltage over a bandwidth of 14kHz and output the digital data via SPI 80 F2 SCK F1 SDI FOICS NEG SDO MCLR Do O 0SC1 OSC2 Figure A 3 MPC3909 Functional block diagram 91 A 2 2 Microchip s MCP3909 dsPIC33F 3 Phase Meter Reference Design Microchip supplies various energy metering designs for evaluating MCP3909 s perfor mance being the Microchip s MCP3909 dsPIC33F 3 Phase Meter Reference Design
114. rm if the user would be willing to change its consuming habits based on the report on the consumption of the domestic appliances For more advanced monitoring and appliance classification disambiguation the Indirect Power Sensing WSN can be also deployed at a relatively low cost It is however less cost effective mainly because of the cost of the wireless nodes components namely the micro controller and the radio interface and batteries Thus the sensors used are inexpensive and present nearly no cost footprint to the system Nevertheless the system here developed is far from being a commercial system ready to be manufactured Further research on Appliance Models and automation of Appliance Cataloguing and Naming is needed to provide the end user an easy to use and deploy system with minimal set up effort and calibrations 85 6 2 Future Work Concerning future work there are plenty of ways one can improve the system to perform better and provide more valuable information in terms of either end consumer data or data that enables more sophisticated appliance disambiguation and classification Thus there are certain objectives that are to be deployed immediately and others that may be delayed to further implementation The most immediate need is the deployment of an improved communications interface as stated in Section 3 4 3 of Chapter 3 in order to enable higher data throughput There may arise the need for improving the appliance mo
115. rmer q Vr 1 I Turns x Rourden C A method of calculating Rourden is suggested in 115 The burden resistor depends on the maximum current to be sensed the maximum input range of the ADC channel and the number of turns of the CT 107 Part Value Device Package Description CT1 CR8349 CR8349 CR8349 CR8349 Current Transformer Rl 10 R EU_0207 15 0207 15 RESISTOR European symbol R2 10 R EU_0207 15 0207 15 RESISTOR European symbol R3 10 R EU_0207 15 0207 15 RESISTOR European symbol R4 10 R EU_0207 15 0207 15 RESISTOR European symbol X1 MC2 MC2 MC2 Mini Combicon 3 81mm horizontal 2 polig Table C 1 Current Sensor Circuit Bill of Materials The voltage level at the current channel at maximum current on the input signal should be at half full scale to allow headroom Full Scale 250MV peak VoMax applying equation Rburden 2 Vomae CT RN _ 125mV x 1520 125mV C 2 Imax 2 5333 3 Q C 3 The implementation of a transimpedance circuit for current sensing The circuit to be deployed is that of the figure with the current transformer CR8349 1500 8 a vertical mount pcb current transformer and a Ryurden 2 5333 3 Q Taking into account that the system is to be deployed for accurate measurements it must be taken special care with its design The system must be very stable and accurate It was chosen to deploy Rourden with me
116. rocessSl a 10 20 Known Limitations sus eae Bow le we ee we a we Ee ES 13 ele WG wee A ek RE Re Be ee a aE E 13 2 8 OUMMATY oe Fook 0 aed eo O es eR SR A ee oP LE RO 14 3 System s Architecture and Hardware Development 15 Cob eh AGH he PLS RAaAW REGS EEE E 15 pa bd aa ie a eae ee be GES 17 ee ee 21 EEE E ee ES 23 3 0 Wireless Sensor Network Nodes 2 2 2 2 2 nn nn nn 30 3 6 Deployment of Indirect Power Sensing 2 22 2 m nn nn 31 3 7 SUMMA s eee ek ee mia a a a a he ee oe a 35 37 a 37 REN 41 epa Gd hep ed 45 hye oe ae PR Gt ee ee ae Ao Go ee A 50 65 AA AE Beh oy a Wet oe By an ee 65 She unia ee ee Gs eee ees ag eee ee eee eae ee 67 NA eee ee te 74 nec ee 75 pees Ste DO ek ee Genesee ey ee ae ee 81 5 6 SUMMARY A A ek ee Be ee 83 6 Conclusions and Future Work 85 6 1 Conclusions 2 2 2 Cm m nn nn 85 6 2 Future Work 2 Co nn n nn 86 89 Ne vere oe eo eee a oe ee O eA 89 eh dm es ie ee en ee es a ee a ee 91 ee Be eee eee RD E ey Be ee ee ee 93 ER 96 fe ee eee eee eee ea ergs Re ote pi ee 96 99 B 1 Low Resistance Current Shuntsl 2 2 HE m mn 99 B 2 Current Transformer l 2 222 2 m nn nn 100 B 3 Hall Effect Current Sensor 2 2 2 2 2m m mn 101 Dan ee ante Se ee eter PRN ee eee ek age AR Bert 103 BO Prices Survey ocio e ode se Be a Ra pe keg Re eek ee Rak en 105 Appendix C Transimpedance Current Sensor with Current Transformer 107 Appendix D Light Sensor Circuit 111 Append
117. rsion of each register of the list it sends it to the PC via UART 27 PC Gateway ADE7953 Read_Cycle Addr1 Addr2 Addr1 R Addr2 R Datal Data2 Addri R Datal Data2 F 00000 gt 00000 N e UART SPI Figure 3 11 Gateway Read Alternative 28 3 4 4 UART Protocol The UART protocol developed is based on the one for ADE7953 s own protocol for its UART but without the time constraints 4 A representation of a basic packet of the UART protocol is represented in figure A full description of the protocol implementation is presented in appendix F nn o a Active BUS _ _________ Idle BUS Idle BUS 17 0 15 o 0 7715 23 31 Command Address DATA Start of Packet End of Packet Figure 3 12 UART protocol basic packet 3 4 5 Gateway Firmware Model Figure presents the data path state machine inside the gateway illustrating how the operations are processed and how the communications are handled UART Message Handling The handling of UART messages can be considered a process and flows as follows A start of a new message is detected after the Start of Frame Delimiter has passed The type of message is then fetched A decision upon what type of message is made If a read command is detected the machine then fetches the next 2 bytes as address and finalizes the message If a write command was issued the next 2 bytes are fetched again as address Next it is fetched the data which length here N by
118. sed to compute the energy consumption of any appliance and build relevant statistics about power consumption cost variation etc Being only the total load monitored there s only the need to install sensors at one point of the circuit ideally immediately after the power source measuring voltage and current 2 2 Total Load Model In order to be able to disambiguate appliances in the aggregated load it is a requirement to know how it combines and behaves A straightforward and intuitive representation of load in a typical household electric circuit is power as it appears as a combination of the appliances wired in parallel At a first approximation assuming the loads to be linear the power consumed is additive For simplicity assume the model of Figure where the appliances are labelled as Z It is clear that the total load depends on which appliances are switched ON at any given time So the total load can be described with a boolean switch process a t Suppose there o phase Sem zZ Za 2 oneutral Figure 2 1 Simple representation of home s electrical circuit are n appliances numbered 1 to n and let a t be a n component boolean vector describing the state of the n switches at time t 1 if appliance is ON at time t ai t l ae 2 1 0 if appliance is OFF at time t The total power is then the sum of the individual power per appliance being the power P a vector
119. sed upon the Faraday s Law that states The total electro motive force induced on a closed circuit is proportional to the time rate of change of the total magnetic flux linking the circuit Normally the coil is in a toroidal shape and measures the magnetic field passing through the toroid figure Since the magnetic flux is proportional to the current according to Faraday s Law the sensor s output can be described as the equation A Figure B 9 Rogowski Coil current sensor simple representation 11 di dt As it can be observed a current flowing through a Rogowski Coil induces a voltage that is proportional to the time derivative of the primary s current There s the need of an integration for recovering the original current signal As one can notice Rogowski coils are by definition incapable of measuring DC but oposing to current transformers these can tolerate its presence since theoretically there is Vi L B 5 103 no hysteresis or saturation on the magnetic core the device should be fully linear So it is safe to say that it has a very wide current operating range due to its linearity a very good frequency response may be also seen for the same reason Adding to that there is galvanic isolation making this solution very good for measuring high currents In its early uses 1912 it wasn t very widespread in energy metering as the technologies for handling small output signals and for developing integr
120. siest parts to implement and one of the most important features for a finalized product Other features that may have to be considered concern the automation of the system in order to make it easy to use and install This could be implemented by building a large database of appliances were the deployed system would go and look for signatures for the features it has found This database would be on line to allow remote access and would grow with data gathered by user s experiences as in the RECAP system 38 86 Appendices Appendix A Available Electric Monitoring Integrated Solutions Here it is presented a market survey of the most adequate solutions of integrated systems on electrical energy monitoring A 1 Analog Devices ADE7953 The ADE7953 is a high accuracy electrical energy measurement IC intended for single phase applications Analog Devices sells it a single IC and has also an Evaluation board for consumers to evaluate its performance In the following sub subsections we present the main features of these platforms FUNCTIONAL BLOCK DIAGRAM REF RESET o HPF Ao VDD VINTA ADE7953 HPF ACTIVE REACTIVE AND APPARENT ENERGIES AND VOLTAGEICURRENT RMS CALCULATION FOR PHASE B SEE PHASE A FOR DETAILED DATA PATH AV
121. soo o 52 E TRE eee es ee ete ee te ee 53 4 13 Data logging entity o a aa ee 54 ee ee ge ee eee p ee ae es ee 56 eden bee 57 4 16 Graphical User Interface entity 2 222 on nn nn 59 ee eee 60 4 18 Graphics with traces of RMS Voltage and RMS Current 60 4 19 Graphics with traces of the components of complex power of Total Power and ip do RO Bey Gee ee ee ek ee a koe 60 ee 61 Sod ape a a Se 62 5 1 Systems electric installation testbed Jap block diagram Jb physical instalation 5 2 Light Sensor Placement 2 2 2 2 2 on on nn 72 8 oo he a en eae E E ee ae 73 ee 77 ie ee 78 5 6 Appliances Individual Interval of identification b 79 5 7 Histograms of the delay between the detection by the WSN and by the load disambiguation algorithm o 80 5 8 Comparison of Measured Power with Estimated Power 82 A 1 ADE7953 functional block diagram A 89 A 2 ADE7953 Evaluation board DA mess aan 24 a oak Ge eben 91 A 3 MPC3909 Functional block diagram 30 lt 91 A 4 MCP3909 dsPIC33F 3 Phase Meter Reference Design 29 A 5 78M6613 Functional block ee a de are Se A aa 78M6613SP EVM Split eine Evaluation Board 3 A A 7 WattNode connection schematic 44 A8 Microchip s PICISFS7J72 Energy Monitoring PB cc 97 A A 99 B 2 Examples of Current Shunt Resistors 2222222222222 100
122. sor do Slave1 State Change MASTER Slave1 s Appliance as changed state Figure 4 14 WSN Representation 4 4 66 WSN entity Before describing the WSN entity function specificities it is necessary to present some concepts about the abstraction here developed The WSN monitors individual appliances with sensors that are used to infer the internal power state of the appliance To that matter there is a Sensor Database inside the WSN entity that holds information about the individual sensors present in the WSN The entries of the database have the following parameters e Sensor Address Physical address of the WSN node e Status State of the sensor It can be ON indicating the attached appliance is on OFF indicating that the appliance attached is off or Disconnected indicating that the WSN node is malfunctioning The WSN entity role can be seen as two different tasks The WSN Interface that is responsible for communicating directly with the network retrieves information about the sensors The WSN Classifier Tracker is responsible by treating the state changes on the sensors and relate them with the changes in the appliance database These tasks are linked in the access to the Sensor Database Figure 4 15 illustrates the WSN entity architecture WSN Interface The WSN interface is responsible for managing the communications between the WSN and the Electricity Monitor This interface provides an abstraction laye
123. system is performed in order to guarantee the desired performance The results of tests to the various components of the system are presented and analysed Chapter 6 Presents the conclusions regarding the implementation of the system and future work to be considered in the improvement of the system e Appendices Present complementary information about the development of the system Chapter 2 Analysis of Non Intrusive Appliance Load Monitoring This chapter aims to give the reader a full depth analysis of the Non Intrusive Appliance Load Monitoring NIALM examining all the needed details underlying this technology and all the requirements to deploy load disambiguation of total load This chapter will be greatly based on George W Hart s paper Nonintrusive Appliance Load Monitoring 18 as it is the most comprehensive article in this matter to my knowledge 2 1 Definition Considering a typical household electrical circuit we may find various and different appli ances devices that work independently consuming energy from the mains power It is then possible to measure the total energy being spent in the circuit loading the power source Lets call the resulting load aggregated load NIALM is a technique that by analysis of current and voltage waveform of the total load is able to disaggregate it in the components of each appliance present at any time in the circuit The data provided by load disaggregation can then be u
124. tal film resistors MR25 because they have good temperature stability and have a tolerance of 1 The range of MR25 resistors doesn t have resistors with values close to the one required So the solution was to combine resistors in order to produce a suitable resistor The resistor was then chosen to be deployed by the combination of four 10Q 1 in parallel Rourden 10 10 10 10 2 52 1 C 4 The circuit needs now to be deployed in a PCB The design was developed in EAGLE CAD with the schematic depicted in figure with the layer board in figure and the bill of materials in table CR8300 Figure C 2 Current Sensor Circuit 108 Figure C 3 Current Sensor Board 109 110 Appendix D Light Sensor Circuit The circuit to develop is a digital discriminator of presence of light so it is wanted to put the sensor in saturation It s then relatively easy to develop a circuit to convert light in voltage sufficing to supply the transistors collector and to put a resistor in the emitter to the ground VDD rr Y Irhoto o Vout R Figure D 1 Light Sensor basic Circuit It s easy to see that the output of the circuit is Vout photo R D 1 Taking that into account and that the sensor placement is uncertain it is necessary to deploy some sort of adjustment mechanism to the circuit In series with R will be a poten tiometer in order to adjust in site the sensor s sens
125. teady state signature of ON state e OFF Signature Complex power steady state signature of OFF state e Tolerance Scalar value of tolerance in for identification of steady state signatures e State Indicator of the appliance s current state ON or OFF e Sensor Address Physical address of the node monitoring the appliance This pa rameter will only exist for the appliances monitored via Indirect Power Sensing WSN and will be explained later 52 Appliance Classification Process A step is fetched from logical database and then compared with all signatures of all appli ances in the appliance database Once the closest signature by euclidean distance is found it is assessed if it is contained within the appliance s signature tolerance as in figure 4 3 If so it is then checked if the event is coherent with the appliance s current state since a turn ON can t occur if the appliance is already ON neither a turn OFF if the appliance is already OFF Provided that all the information is consistent the appliance state is updated and the detection of an appliance event is signalized in the database The appliances monitored by Indirect Power Sensing WSN are discarded from this process This process is illustrated in figure Steps Queue Step1 Step1 Ft Step2 Step3 Appliance e a e Signatures Comparison Name Heater ON 1000W Data Consistency OFF 1000W Check
126. tection in aggregated MET eee eS ee ere ee 48 5 1 Assertion of gain error following the calibration of voltage channel 68 5 2 Assertion of gain error following the calibration of current channel 69 5 3 Results of the voltage meter constant fitting 71 5 4 Results of the current meter constant fitting 4 72 5 5 Appliance s Characterization o o ooo a a 74 EEEE 75 ee ee ee 76 eee AA 80 DEE 81 i Be We ea a ea ek e 82 errr EURER 82 C 1 Current Sensor Circuit Bill of Materials 2 22 o nme 108 vil viii List of Acronyms NIALM Non Intrusive Appliance Load Monitoring WSN Wireless Sensor Network NILM Non Intrusive Load Monitoring FSM Finite State Machine SFD Start of Frame Delimiter EFD End of Frame Delimiter MSb most significant bit MSB most significant byte GUI Graphical User Interface ix Chapter 1 Introduction 1 1 Motivation Nowadays we witness demanding economical and regulatory pressures for the deployment of energy efficient technologies to promote energy s conservation The reasons behind this interest are the reduction of carbon dioxide emissions as in Kyoto s Protocol the reduction of costs and ultimately the ambient conservation by reducing the consumption of fossil fuels One of the most if not the most used energy form is electrical energy thus it arises the need to reduce the electrical energy being spent in industrial commercial
127. tes is found by analysis of the address s MSB and finally the packet is finalized The packet finalization depends also of the receiving of End of Frame Delimiter EFD If any error is detected on SFD EFD or any other field which data is different from what is supposed to the program flow returns to the beginning with the start of a new message After a message is finalized the data is passed to the SPI handling process 3 4 6 SPI Handling SPI handling starts after a complete message is received via UART Upon receiving the process decides what to do depending on the operation to perform If a read was issued the system reads N bytes again the length is found from the address of the register and sends the read data out via UART and then finalizes the command If a write was issued N_ bytes of data coming from the data of the received packet are written to the register and then command is finalized 29 New New Command Message Fetch Read Write Write Command Write Read Write Get Register N_bytes from N_bytes to Size Address Address Fetch Fetch Address Address 2bytes 2bytes Send Register Data Fetch Data via UART N_bytes Figure 3 13 Gateway Dataflow The system has ended the processing of a message and is ready to receive and process another 3 5 Wireless Sensor Network Nodes The project s proposition demands the deployment of a complementary WSN to perform Indir
128. the most interesting to this project The Microchip s MCP3909 dsPIC33F 3 Phase Meter Reference Design is a energy meter with many advanced features figure A A It s based on the MCP3909 for waveform sampling of power line and dsPIC33F for signal processing and calculations The current sensing is done by current transformers UART Interface Current Transformer dsPIC33 MCP3909 Figure A 4 MCP3909 dsPIC33F 3 Phase Meter Reference Design This meter is ready for 3 phase metering with neutral current and it produces the following data for each of 3 phases and to its total e line frequency e Vims e lims e real power e reactive power e power factor e fundamental active power e fundamental reactive power e harmonic component measurement for voltage from the 2 to the 31 harmonic 92 e harmonic component measurement for current from the 2 to the 31 harmonic Each of these measurements and calculations are based on frequency analysis DFT based are updated every 3 line cycles and accessible via UART which can provide at it s best a sampling frequency of 16 Hz of the desired parameters 29 A 3 Maxim s 78M6613 A 3 1 78M6613 Single Phase AC Power Measurment IC The 78M6613 is a single phase AC power measurement IC designed for simplified metering of power supplies and smart appliances Its architecture is based in four analogue input channels a 22 bit sigma delta ADC a 32 bit dedicated Comput
129. their habits to spend less energy and ultimately spend less money The project hereby presented develops the basis for a device that provides detailed information on electrical energy expenditure for a home The infor mation to be provided aims to be disambiguated for virtually every individual appliance present The chosen technique to develop such a device was Non Intrusive Appli ance Load Monitoring The system identifies individual loads from a single measurement point That makes that the hardware tends to be simple and minimal passing the complexity to the software The system aims to be very low cost and easy to deploy The system developed was able to identify with great success individual appliances in the total load on a controlled environment Unlike other systems of this type the system presented in this dissertation has the ability of disambiguating electrically identical loads by the use of the technique of Indirect Power Sensing deployed with a Wireless Sensor Network Contents i List of Figures iii List of Tables vii List of Acronyms ix 1 fini ne Sod ee cs oe Se ee ARA A er AD UP ee 1 E ee E RAS SESE ASHE REGRESS ee Rew EPA E EEE 1 1 3 Document Qutlinel a 2 2 Analysis of Non Intrusive Appliance Load Monitoring 3 Dee ee ee ee Ede ee Ben eee ey Ei en 3 2 2 Total Load Modell 2222 22 Comm 3 2 3 Appliance Models o ee 5 Seca Gate a Dre te e io n Se a de eee ee A SS tn 7 2 0 NIALM P
130. tion of time at the changing states of operation Since they exist only for a short duration they are more difficult to detect and are less informative Transient signatures however may provide valuable information for load disambiguation as they are intimately related to the physical task being performed 24 They allow the distinction of appliances otherwise electrical identical since as stated E most observed loads have repeatable transient profiles or at least sections of it Transients have also the potential of being incorporated in FSMs by labelling each arc with the transient relevant for the corresponding state change Transients may be parametrised by e shape e duration e time constants e parametric variables of fitting curves of observed waveforms These parameters should be allowed to be shifted and transformed in order to fit better the detected events Other Non intrusive signatures There are other types of signatures able to be used by NIALM for e g low frequency ripples or ramps in power consumption but they are not very much significant and won t be analysed here 2 5 NIALM Process Diverse approaches presenting different compromises in terms of simplicity accuracy and easiness of deployment may be employed to carry out load disambiguation Due to its simplicity and capability of modeling a vast range of appliances it was decided to use an algorithm derived from the one presented in section IX of 18 b
131. uring an interval of 10 seconds at a sampling fre quency of 10Hz From the values obtained compute a the mean to get one IRMS LS Bs value Parallelly measure Iris at the phase cable of the Electric installation with the current probe attached to an Tektronix TDS210 Oscilloscope to get Current Input A It was not possible to measure accurately the mean current in the interval so the current input is taken as a sample of instant peak current which is then transformed into a RMS value 3 Store ordered pair in Currents List The previous steps are to be repeated until the needed number of values are obtained Next a linear regression of Current Input A IRMS LSBs must be made to get the Cur rent Meter Constant The linear regression result was a current meter constant of ImeterConstant 2 392182902 x 107 Arms zss 5 15 with a correlation factor R 0 9998 5 16 the offset was b 0 0219 Arms 5 17 The accuracy of the Tektronix A622 AC DC current probe at the scale measured that was 100mV A is accuraCYprobe 3 of measurment 50 mA 5 18 The accuracy of the Tektronix TDS210 oscilloscope on the function used that was Delta Volts is accuraCYosciloscope 3 of reading 0 05 div 5 19 Table shows a comparison between the current measured by the current probe and the oscilloscope Irms with the current measured in the data aquisition system when scaled by the current meter constant IrmsSactual One c
132. ut slightly modified to be simpler and more generic The signature is based in the normalized power and the appliances are modeled by the ON OFF model The signature steps of the appliances are assumed to be known a priori A global view of the whole process is represented in figure 2 4 Each step will be discussed in the remainder of this section 2 5 1 Data Acquisition An adequate sensory system attached to the main power source of the home measures voltage and current and computes digitally the complex power and the RMS voltage at a predetermined sampling frequency The sampling frequency needs to be wisely chosen since it affects the system s time granularity i e the ability of the system to detecting various appliances and appliance features in a short period of time 10 1 Data Aquisition Measure Power and Voltage RMS Data 2 Normalize Vnom Prorm P Normalized complex power 3 Step Detection Step Changes 4 Appliance Classification Appliances State Changes Detected 5 Appliance Tracking Appliance s status ON OFF times 6 Data Output Tabulate Statistics Appliance s energy cost energy vs time of day etc Figure 2 4 NIALM process 2 5 2 Normalize Taking complex power P and RMS voltage V we compute normalized power Phorm by Vnom 2 Phorm P 2 6 7 2 6 where Vnom is the utility s nominal voltage It is created then a voltage independent signature to viti
133. ve a rather slow transient what made the detection through load disambiguation difficult while the detection through indirect power sensing WSN showed to be more reliable Connected to the data acquisition system and to the Indirect Power Sensing WSN there is a PC which executes the all necessary tasks to perform the Electricity Monitoring 65 UTILITY S POWER SOURCE Phas RADIATOR Neutral DATA Acquisition System b Figure 5 1 System s electric installation testbed la block diagram IP physical instalation 66 5 2 Calibration Calibration is very important in systems that depend on measurements and sensing of environment quantities Furthermore from the nature of load monitoring system that aims to be an energy meter it is expected that it is a device with great accuracy and correctness so that it meets or exceeds the common household revenue meter The main measurement system is the ADE7953 Evaluation Board in which the measures of Voltage and Current need to be calibrated To calibrate a system one needs references The reference used for voltage measurement was Fluke 287 True RMS Multimeter while for current measurement the reference used was Tektronix A622 AC DC Current Probe connected to a Tektronix TDS 210 Oscilloscope 5 2 1 Calibration of Measurement System ADE7953 ADC Gain error Calibration The gain error is defined as the per channel difference between the ADC output code and t
134. veloped in MATLAB and it performs the algorithm as follows 1 Standard Deviation The first step is pass the whole Real Power signal through a finite length moving standard deviation Figure 4 4 illustrates this step In this case it was passed a window of 4 samples through the data As one can see the standard deviation signal only rises when a step change in the power flow occurs otherwise is steady near 0 This is the first argument in favour of the algorithm 2 Steady State Identification After passing the standard deviation a threshold is passed to its result to find where the signal is steady and where the signal is changing Once the threshold is passed it is found the middle of the steady periods and taken a sample That sample is called a steady state It is assumed that on the middle of a period the signal is steady enough Figure 4 5 shows the result of steady state identification with the red bullets indicating the middle of a steady state 3 Step Detection and Classification Now that the steady states are known it is simple to calculate the step changes A differentiation of the consecutive steady states is performed resulting in the step changes Not all step changes may qualify as steps of an appliance as they may be too small A threshold is passed through the step changes in order to eliminate false positives Figure 4 6 represents the step identification Positive steps turn ONs are represented as represented
135. where the appliances being turned ON or OFF are identified and the database is updated Summarizing the dataflow must pass through this steps in order 1 Steady State Discrimination 2 Step Identification 3 Appliance Classification 4 3 4 Validation With the software implementation defined it is necessary to assess its accuracy and cor rectness For this it was taken the aggregated load of some data from REDD and the char acterization of some appliances from table and it was performed load disambiguation in order to find the Refrigerator and Oven appliances The parameters used for the process were Window 15 seconds for single appliance 5 seconds for mains data Standard Deviation Threshold 5W Step Threshold 25W Fixed Appliance Tolerance 40W The results presented are from the first 800000 seconds of the records Table 4 2 compares the number of detections made by detection of appliance in single appliance file with detection of appliances in the aggregated load in the mains file by the algorithm implemented in Java 47 Power Window N_Positions Standard Deviation Non Steady States Queue Steady States Queue Statel Statel State2 Steady Power State3 Discrimitation State3 State4 State4 Steps O N Step Identification State2 Step1 Step2 Step3 Step4 ht Appliance Classification Appliance and Event
136. wn tolerance If it does it is registered a detection of an appliance change of state otherwise is discarded For illustration purposes in figure are represented detected steps and appliance sig natures in the P Q plane Step 1 is detected the closest signature is App2 ON Step 1 is contained in App2 ON tolerance so Step 1 is classified as the Appliance 2 turning ON Step 2 s closest signature is App1 ON but its not contained in the tolerance so it is discarded This algorithm represents the basic operations needed to deploy in order to provide NIALM However it isn t supposed to provide a detailed solution for Load Monitoring be cause it lacks appliance tracking energy calculations and user interfaces Following there is a pseudo code representation 40 Detected Step 2 A VAR App1 ON Detected Step 1 App2 OFF Appi OFF Figure 4 3 Steps and Signatures on the P Q plane 4 2 Preliminary Validation of the Load Disambiguation Algo rithm Before implementing the system based on the algorithm presented it is good practice to assess if it can perform as it is supposed e if it provides a correct output provided a correct input 4 2 1 REDD A Public Data Set for Energy Disaggregation Research Reference Energy Disaggregation Dataset REDD is a publicly available dataset contain ing detailed information on power usage from real household environments It was created to provide valuable data t
137. x MUX SYNC CKCE lt 4 9152MHz CKMPU lt 4 9152MHz Figure A 5 78M6613 Functional block diagram 2 VREF TRT XFER BUSY MPU 80515 RESET V3P3A GNDA V3P3D GNDD AZADC CONVERTER DIV ADC CKADC 4 9152MHz CKEIR 4 9152MHz MEMORY SHARE A000 11FF DIGITAL IO 2000 20FF DATA 0000 FFFF MPU XRAM 2KB EMULATOR PORT ERXTX E TCLK E RST Open Drain EAT ICE_E E_TCLK ERST 4H 94 DIO4 DIOS DIOG DIO7 DIO8 DIO14 DIO15 DIO16 DIO17 DIO19 TMUXOUT 78M6613 V7 MAXIA jp qro Sve Valuatlon va Mm LET j ve High Isolated AC Voltage Interface a 78M6613 EVM Evaluation Board t Figure A 6 78M6613 Evaluation Boards a 78M6613 EVM Evaluation Boardfil b 78M6613SP EVM Split Phase Evaluation Board 3 95 A 4 Continental Control Systems WattNode ModBus WattNode ModBus is a kilowatt energy and power meter designed to monitor up to three phases A schematic of the WattNode connected to one phase is represented in figure Ground Shorting Jumpers Source Face Line zZ Neutral M Transformer ayol Figure A 7 WattNode connection schematic It provides for each phase and for the total load the following measurements e Vrms Irms e true RMS power e reactive power e power factor e true RMS energy e line frequency This measurements are updated at every second via EIA RS 485 with Mo
138. y measure active and reactive power and mainly because they provide data that makes the analysis of high frequency signals such as transients possible the ADE7953 provides 1 23 kHz bandwidth waveform sampling and the MCP3909 dsPIC33F provides harmonic analysis of the signals high frequency waveform sampling could also be attained with firmware modification However Microchip s solution isn t available for purchase as the product has become obsolete therefore the Analog Devices ADE7953 EVM Evaluation Board is the only choice viable 3 2 1 Current Sensing Having selected the electricity monitoring integrated solution it is necessary to acquire all the necessary equipment and components in order to deploy the system The ADE7958 EVM Evaluation Board doesn t provide a direct way to measure current Contrariwise to voltage it is necessary to deploy a current sensor to attach to the monitoring system in order to sense the facility s current Again there are some requirements that need to be met The sensor has to perform well at a certain bandwidth and be able to measure currents in the range of what a common house can consume Regarding the bandwidth the sensor needs to measure in a range that goes from the utility s frequency 50 Hz to various kilohertz to not compromise the systems performance In terms of current range a minimum of 30A of maximum current should be considered as it will cover the majority of consumers in Portugal 6
139. zed power as an input flow and sequences it into changing periods and steady periods The power values are inserted in a finite moving window of length N The operations of standard deviation and mean values are computed through the window A representation is shown in figure Standard deviation is used as a measure of how much the power is changing in the window If greater than the threshold STD Threshold the power is considered to be changing at that 39 Normalized Power Window N_Positions Standard Deviation Figure 4 2 Representation of the filter for step detection moment Once it goes below the threshold the power is considered to be steady A time stamped sample of the mean power of the window is taken and registered as a steady state The difference between steady states is calculated and is considered to be a step Not all steps are to be considered because some of them may be too small so they are compared to a minimum threshold Step Threshold If greater the step is registered and passed for further processing otherwise is discarded Following there is a pseudo code representation 4 1 6 Appliance Classifier Once a step is detected it follows the identification of the appliance that changed state To achieve this goal steps are compared with the signatures of the list of all known appliances to find the closest one Once the closest signature is found it is computed if it fits within the appliance s o

César Emanuel Brandão Gomes Monitorização Fina do - Serv-CPS

Contents

Download Pdf Manuals

Related Search

Related Contents

C&eacute;sar Emanuel Brand&atilde;o Gomes Monitoriza&ccedil;&atilde;o Fina do - Serv-CPS

Contents

Download Pdf Manuals

Related Search

Related Contents

César Emanuel Brandão Gomes Monitorização Fina do - Serv-CPS