Home

the thesis

1. costFunction Key value map each item stands for a part added to the cost function Adjusts the soft constraints norm Determines the norm of this cost function part weight Holds the value of weight of this part refMin refMax When set it sets constant minimal maximal constraint of the variable weighted in the cost function The band between values is not penalized locations outside are weighted by the value of weight field refMinSchedule refMaxSchedule Same as preceding but with a vari able constraint The value is a reference into data part of the structure where the constraint is stored constraints Hard constraints refMin refMax When set introduces a constraint on the specified vari able which must not be violated by a solver refMinSchedule refMaxSchedule Same as preceding but instead of constant uses variable as a constraint The value is a reference into data part of the structure where the constraint is stored 6 CHAPTER 1 INTRODUCTION e models Key value structure each item of it holds the information about one model in the system e controlSamplingPeriod Sampling period of the model discretization in seconds e defaultInitialState Vector of the state which the model is initialized to In the current HVAC there is also a script called MPC Viewer which reads the results and plots them on a web page which serves as a diagnostic tool for the
2. 2 4 6 Running the simulation When you set all parameters to the desired values you can launch the solver This action does not directly call some numerical routines Instead it enqueues the simulation into the queue common for all users where it waits for some free for use computation core You can watch the log informing you about the simulation state or you can experiment with another setting or another simulation 2 4 7 Results When the simulation finishes results can be viewed Results are presented in a form of charts where related series are displayed in the same figure You can view only particular data series you can compare two or more series which are not related The result view can show results of more than one simulation at the same time and allows to do these comparisons over more simulations 20 CHAPTER 2 USER MANUAL 2 4 8 Simulation visibility There can be simulations of more distinct systems or of a different kind in the application For the user not to be lost while seeing all simulations in the system there are some relations between users and simulations possible Moreover it is a good habit to found some security rules even simple to increase user comfort specifically others will not silently change your simulation parameters and clear results and so on When you log in you see only simulations which were assigned to you by an admin istrator plus you see results and working copies you have made Aft
3. buildings GB Patent No 210880 April 1927 D Gyalistras and M Gwerder Use of weather and occupancy forecasts for optimal building climate control opticontrol Two years progress report Technical report Terrestrial Systems Ecology ETH Zurich Switzerland and Building Technologies Division Siemens Switzerland Ltd Zug Switzerland 2010 Michael Heim Exploring Indiana Highways Exploring America s Highway 2007 ISBN 9780974435831 Raad Z Homod Khairul Salleh Mohamed Sahari Haider A F Almurib and Farrukh Hafiz Nagi Gradient auto tuned takagi sugeno fuzzy forward control of a hvac system using predicted mean vote index Energy and Buildings 0 2012 Krist f Kov cs Cassandra vs mongodb vs couchdb vs redis vs riak vs hbase vs membase vs neo4j comparison http kkovacs eu cassandra vs mongodb vs couchdb vs redis Online Accessed May 1 2012 BIBLIOGRAPHY 71 29 30 31 32 33 34 35 36 37 WH Kwon AM Bruckstein and T Kailath Stabilizing state feedback design via the moving horizon method International Journal of Control 37 3 631 643 1983 Jingran Ma S J Qin Bo Li and T Salsbury Economic model predictive control for building energy systems In Innovative Smart Grid Technologies ISGT 2011 IEEE PES pages 1 6 jan 2011 Y Ma A Kelman A Daly and F Borrelli Predictive control for energy efficient buildings with thermal storage Modeling stimulat
4. data is transferred using HTTP which carries YAML text data Structure serialized into YAML is a list of dictionaries each dictionary has following fields e type Identifies which kind of information the item holds e value Holds content of the information 8 12 APPLICATION COMPONENTS 63 These key set may be extended by some extra fields when the information the item kind needs it name When the information is targeted to some field in the simulation data struc ture the target is specified here target When the target needs to be specified even more and the rest of the information is very similar to another one this field is the target fine specifier The top level list items can be of following types script in its value defines MATLAB script to be executed for task processing controller in its value selects which simulation controller from the batch of preloaded data will be used model in its value selects which simulation model from the batch of preloaded data will be operated predictionHorizon in its value carries length of the prediction horizon floating point number in the units of hours outputseries in its value key stores the primary key of the output data row in the database The output data will be saved in this row as CSV initialConditions in its value has a dictionary which in its initial key holds initial measurement values to use with Kalman filtration for intial
5. distribution in current buildings better which can be done by engagement of Building Automation Systems BAS or only by introducing new algorithms for existing ones without an interference with physical aspects of a building There were efforts of many academic and industrial teams to implement various advanced control algorithms in the area of HVAC 25 81 85 B0 27 1 1 Advanced control techniques for HVAC There has been two main directions in the heating control research recently One group involves method of the artificial intelligence specifically neural networks genetic algorithms fuzzy methods and others The second group which is in concern of this work employs techniques based on principles of the classical control theory called Model Predictive Control MPC MPC is in theory easy to formulate as it consists in direct application of optimization techniques on the given control problem It is widely used in many industry areas see for instance for further reading The idea behind MPC is to express a goal of the control by the terms of a cost function value of which can be called penalization Minimization of the cost function results in a control strategy which also fulfils constraints given as a part of MPC problem formulation This optimization happens on the specific constant time interval which is called prediction 2 CHAPTER 1 INTRODUCTION horizon with the length specific for the control problem solved Specifi
6. 1 1 1 1 1 1 1 1 1 1 1 2 1 2 1 2 1 2 1 2 1 2 1 00 00 04 00 08 00 12 00 16 00 20 00 00 00 04 00 08 00 12 00 16 00 20 00 E EEE aleja Displayedseiesy Dejvice A2 North block heating 11 14 14 14 14 14 2 1 24 24 2 1 24 24 00 00 04 00 08 00 12 00 16 00 20 00 00 00 04 00 08 00 12 00 16 00 20 00 t 1 a a o Displayed series Weather Prague Dejvice B Amb_Temp_Min E Amb Temp Max 1 1 1 1 1 1 1 1 1 1 1 1 2 1 2 1 2 1 2 1 2 1 2 1 00 00 04 00 08 00 12 00 16 00 20 00 00 00 04 00 08 00 12 00 16 00 20 00 Figure 2 14 The result page 34 CHAPTER 2 USER MANUAL To describe buttons under the Results title Solution logs Drops down a menu whose items correspond to shown simulations and clicking one of them causes displaying modal window with the solution log of the chosen simulation e Series comparator Turning on this toggle button checkbox appears before each title of displayed data series in the legend There also appears button Compare selected beside the button just turned on Checking some data series and clicking the Compare selected button the user fire the Series comparator modal window displaying selected series all in one graph as shown in the Picture e Displayed figures This is a dropdown menu where the user can restrict the set of displayed figures only to those important for him or her The dropdown contains identifiers of all figures divided into groups as shown in the
7. 1 2012 Mysql http www mysql com Online Accessed May 1 2012 Orbited Real time communication for the browser http labs gameclosure com orbited2 Online Accessed May 1 2012 Postgresql http www postgresal org Online Accessed May 1 2012 69 70 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 BIBLIOGRAPHY Postgresql Window functions http www postgresgl org docs 8 4 static functions window html Online Accessed May 1 2012 sdpt3 http www math nus edu sg mattohkc sdpt3 html Online Accessed May 1 2012 Sedumi http sedumi ie lehigh edu Online Accessed May 1 2012 Snakeyaml http code google com p snakeyaml Online Accessed May 1 2012 Sqlalchemy the database toolkit for pythonn http www sglalchemy org Online Accessed May 1 2012 Turbogears web framework http www ape project org Online Accessed May 1 2012 Wsgi web server gateway interface http www wsgi org Online Accessed May 1 2012 Yalmip http users isy liu se johanl yalmip Online Accessed May 1 2012 Yami Yaml ain t markup language http yaml org Online Accessed May 1 2012 yamlmatlab Matlab reader and writer of yaml formated files http code google com p yamlmatlab Online Accessed May 1 2012 Richard Godfrey Crittall and Joseph Leslie Musgrave Heating and cooling of
8. BACTool 3 for example which evaluate a mean behaviour of the controlled system over a long time in order of months or a year 1 4 Motivation of the application When the engineer wants to run a simulation of a control problem he or she has many possibilities how to do that The system plentifully mentioned throughout this work MATLAB is only one example 1 5 BASIC REQUIREMENTS FOR THE APPLICATION 7 The problems of these systems are often their complexity paradoxically their feature richness need to learn special kind of de facto programming language which cause the novice to be easily lost within Mentioned systems are often hard to maintain even for a lecturer during a speech presenting some control problem and their appearance can be confusing for a speech participants who can only watch what the lecturer does in addition There are some applets and simple tools easily found on the internet dealing with control engineering branch even with MPC but they are all actually textbook examples with no background in some real specific system much less with the ability for user to define his her own control problem in some non trivial way Another aspect is that specifically MPC simulations are known to be time consuming even on some better hardware Some tool could offer to perform a long lasting simulation from some very low end netbook but with the actual computation distributed to some really powerful hardware shared with ot
9. Description system There can be a description associated to each simulation The description is stored as a plain HTML file in some static folder of the application web server accessible from the internet The file self should make sense as a whole because it can be displayed in the when user clicks the button with the i letter The same file is used as a source of concrete descriptions for setting blocks in simula tion forms See the grey boxes on the Picture 42 CHAPTER 2 USER MANUAL There is a method load of jQuery library more info in Section 3 7 1 allowing AJAX load of a file content of which is parsed as HTML and used as a content of given DOM element Nodes in the description HTML file are labelled using HTML class attributes corresponding to the names of the adjustments blocks being described load function is then used to locate the parts of the file into the appropriate places There are many AJAX requests initiated but all but the first one will be replied from the web browser cache The URL path to the file is stored in the ObjectDescription model along with path to the resolver Java Script file which will be loaded before any description The file provides the resolveHelp helpClass helpIdent function through which all requests for description are passed through making it the extension point of mapping between described blocks and the description HTML file To describe particular simulation create appropriate HTML fil
10. Picture test vah x0 opravene refs na energii TABS control of block B3 of CTU building in Prague Dejvice Y Dejvice B3 Room temperature Y Dejvice B3 Cost function F Y Dejvice B3 South block heating Y Dejvice B3 North block heating Y Weather Prague Dejvice test costu inf TABS control of block A4 of CTU building in Prague Dejvice Y Dejvice A4 Room temperature Y Dejvice A4 Cost function Y Dejvice A4 South block heating M Dejvice A4 North block heating F Y Weather Prague Dejvice Figure 2 15 Dropdown menu for selecting displayed figures The behaviour of figures is very similar to the Data cut out picker described in section 2 6 10 with the same meaning of buttons above the plot area the red data cursor and the same system of panning zooming The new thing is the Displayed series dropdown menu which allows the user to choose which data series will be displayed in given figure If there is a description page defined clicking the name of arbitrary data series in the plot legend navigates to the Outputs section of that page 2 6 WALK THROUGH THE APPLICATION 35 Series comparator 1 1 a a Displayed series E test vah x0 opravene refs na energii B3 S RW 01 03 01 42 x 35 99 W test costu inf A4 S RW 31 12 1 1 1 1 1 1 2 1 2 1 2 1 3 1 3 1 16 00 00 00 08 00 16 00 00 00 08 00 16 00 00 00 08 00 Close Figure 2 16 Series comparator 2 6 13 Django
11. a Working copy e Work with copy from original Create a copy of the template which was used to create current simulation e Go to original Open the template which was used to create current simulation e Working copies Show all working copies from current simulation e Compute Enqueue the simulation for processing e Computation log Show live computation log e Administer this simulation Navigate to the administrator page for current simulation 26 CHAPTER 2 USER MANUAL TABS control of block A3 of CTU building in Prague Dejvice Save Work with unsolved copy _ Work with copy from original Go to original Administer this simulation User description Shortly describe what you are trying to achieve User description Simulation Initial condition Overheated north v Overall weights Overall weight comfort vs economy 67 33 Figure 2 8 The manager view Underneath the bar you can find the text area to fill in your description Fill in something which describes what you changed in the form or what you are trying to show using your simulation Remember that first 128 characters of this field will appear in User description column of table displaying working copies Simulation listings and Task queue and you will use this description part to identify the simulation 2 6 8 Simulation form the manager view Picture of the view was placed in the previous section As said earlier the spe
12. a few powerful servers computing the solution and much higher number of some thin clients used only for communication within these servers and controlled technology One problem seen in this approach is that there is no failover safeguard e g what to do if the internet connection is broken One possible solution is to implement some less 67 sophisticated control algorithm directly into the thin client which in case of connection failure would temporarily adopt the control even with the worse performance until the connection is fixed But this is entirely out of the scope of this work There are many messaging techniques established AMQP protocol is one repre sentative which uses one or more so called message brokers which are servers main taining message gueues and connections with communicating clients message producers and consumers The system is designed to distribute thousands of messages per second brokers have often built in persistence and journaling facilities to overcome restarts and powerdowns so messages are not lost so easily Some messaging systems have also higher level abilities out of the box such as clus tering which increases the messaging system availability On the use of Django and its ORM The author was disappointed many times by the Django framework as by the reading of many articles about it had appeared that the framework is really outstanding But everything has its limits Django often does more
13. and set Presentation to Engineer which will consequently allows you to edit largest possible set of pa rameters The parameters will be still there if the choosen presentation will be finally Manager The key is that Manager presentation will hide the parameters but use them all as a resulting problem set up unless something some weights gets overriden by overall weights Just for case remove all associations the simulation may have to users and groups Click Save and continue editing at the bottom of the page then View on site which will lead you to the simulation form of the problem Set simulation parameters as you like run several iterations of its computation to check whether you set it correctly Then clear the simulation results Remember to Save or Compute the simulation to persist changes you made Optionally if you plan to make the simulation presentation to be Manager go to the administration page of the simulation and select Manager presentation Then again View on site Now choose the default state of Overall weight sliders then again do not forget to persist the changes Finally go to the simulation administration again and check Is template there Additionally you can associate it some descriptor You now have your brand new template Associate it to some users log in using their accounts and see whether it works as expected 2 4 THE APPLICATION IN GENERAL ORDINARY USER POINT OF VI
14. and toolboxes but lags behind others in matter of modern features like pointers references and is quite slow in many cases 3 9 1 How to call MATLAB from outside The fact is that while there is a variety of ways to call third party libraries from MATLAB there is also a lack of official and comfortable ways how do so reversely at least without some additional toolbox MATLAB has the Java Virtual Machine built in as its core component and the Java language is well accessible including most of third party libraries But again there is no clear and official way of calling MATLAB from Java There are possibilities how to bypass this shortage MATLAB can be launched with the statement wanted to evaluate entering it to the command line but this is very slow one MATLAB start per one statement execution Periodical file read and write opera tions are also possible but it is slow as well and not least ugly Moreover there is no easy mechanism how to transfer data between MATLAB and the outside world even Java Matrix can be passed to some Java function also some string but starting from structures and cell arrays developer is out of luck and needs to specify protocol saying how to marshal and unmarshal every data structure which is somewhat complex There is an undocumented feature the Java Matlab Interface JMI 8 which is used by many people to achieve the MATLAB call There are some third party libraries which t
15. check the initiated connections using the Java Script debugger to eliminate possible waste of server resources Chapter 3 Developer point of view 3 1 Architectural overview The heart of software is an application written in Django see section 3 8 1 web frame work for Python programming language see section 3 4 Data persistence is backed by PostgreSQL relational database section 3 10 The application distributes computations on several cores at the moment running on MATLAB Cores poll the database for tasks using HTTP protocol and saves results there as computation finishes 3 2 Top level component interconnections The components mainly communicate over HTTP in the current configuration However the system uses well established and proven technologies which can easily scale The current configuration is in the picture 1 AI participants of the system workers web browsers communicate with the application through HTTP using only one web server The web application connects to the database running on the same machine via the proprietary protocol of the database 43 44 CHAPTER 3 DEVELOPER POINT OF VIEW User computers hl Web server d Computatioal cores workers SQL database Figure 3 1 Current system configuration It is possible to operate the system in the form proposed in Figure The design figures on running lightweight web server on each worker machine and so creates more potential exte
16. e Disturbanceltem Represents the Disturbances block of the simulation form e CostFunctionItem Represents the Cost function block of the simulation form e InitialCondition Holds information about one possible setting of Initial condi tion Selection of the state is indicated by existence of relation between this object and MPCProblem e IitialConditionSet Object associating the MPCProblem with all its possible Initial states 3 12 6 MATLAB poller Although recent versions of MATLAB have some operations implicitly paralelized there is no core support for threads in MATLAB language self There is Paralel Computing Toolbox which could do the job but it costs something and it is too strong tool for the purpose This imperfection can be part way eliminated using threads of built in JVM The feature is not omnipotent There is still only one thread of MATLAB interpreter but the Java thread may do some non computational work like to communicate which was used in the application There is a small class written in Java which is launched in MATLAB and polls the message queue every 5 seconds The poller is as small and simple as possible hence minimizing needs of debugging in two environments application and MATLAB side The communication handle was finally designed as a HTTP channel The reason was mainly the fact that HTTP is old but yet widely used It is fast and libraries implementing it mainly create some kind of abstracti
17. from simple edit to graph picker and back as you change the Use field There is a minimum of A2 north supply water set to be fixed with value specified to be 15 C Maximum value is set to be variable and you see that this choice changed an edit field for fixed value to combobox for data series selection in the case on the picture the series Reference max room temperature is used and finally there is graph picker small button with bar graph icon with text showing date and time of the starting point currently selected 08 11 2011 14 06 in our case By clicking the button the picker is launched which will be explained in detail in the following section The Cost function block is functionally same as the Constraints block except for having a Norm combobox and Weight edit field in addition This block introduces soft constraints of the problem There are simulation inputs expressed in the Disturbances block can be only variable because constants would not have much sense for this purpose 2 6 10 Data cut out picker This modal window as shown in picture 2 10 allows you to explore how does the data series look like zoom to see its details and pick the time interval which will be used as the input to the simulation Graph navigation You can move the graph using mouse by drag amp drop Zooming is performed using mousewheel Three leftmost buttons on the graph toolbar change the zooming behaviour they set the direction
18. importance as the code library They frequently allows the user to apply the Don t Repeat Yourself DRY principle more easily which means in consequence less copy paste operations and less bugs There are several framework types diversified by several criteria Most dominant and spoken are perhaps Model View Controller MVC frameworks named by three main components which together constitute the application Frameworks can deal with sessions authentication even creating PDFs but they could or should contain these tools in general e Model layer This is where the metadata is defined including relationships be tween the data and probably with the validation rules The model is alpha and omega of the MVC application If a developer define it well and used framework is well behaved then any new operation with the data under that model is very likely to be implemented quite fast because the model logic will handle many important actions sometimes sim ple sometimes complicated ones often tasks which are repeated on many places without a framework and hence are very error prone e Object relational mapper ORM The model is often described in an object oriented way which is guite in a conflict with widely used relational database tech nology We do not want to claim that ORM must be used only that using of it could make a developer s life easier Of course ORM must be well designed otherwise it could turn itself
19. in a system this implementation will not be overly painful Django has a built in module ContentTypes which can provide such functionality It abstracts polymorphic relation using third table where the name of a targeted table is stored At first sight the feature is very useful but is exactly the one of the examples of constructs not suiting for the relational databases which is maybe one of the reasons that the ORM facilities are even more restricted for polymorphic relations than for normal ones 3 12 5 MPC simulation InitialCondition InitialConditionSet ResultitemSeriesGroup 0 ResultitemSeries SS CostFunctionltem 0 Disturbanceltem 0 HiWeightltem Figure 3 4 Class diagram of a mpcproblem module This is a structure of classes storing data about the concrete simulation Their purpose is following e MPCProblem root of the concrete simulation Holds some general informations like prediction horizon length model and controller name and script name to be launched for computation e ResultItemSeriesGroup The object grouping somewhat related data series to gether e ResultItemSeries Output of the simulation where the data is stored in a textfield one series per object as a CSV e HiWeightItem Overall weight for the simulation 3 12 APPLICATION COMPONENTS 61 e ConstraintItem Object reflecting the Constraints block of the engineer simula tion form
20. in which the view is zoomed in and out Follow the icons The buttons with magnifier are for zooming without using the mousewheel The main reason for introducing them was that the graph component used has a bug reading the 2 6 WALK THROUGH THE APPLICATION 29 mousewheel in some browsers so when the mousewheel will not do as you expect it to du use these buttons instead The purpose of the button with crossed circle is to reset view to the default in case you loose yourself during exploring This can easily happen due to mousewheel bug mentioned above which by an accident sets the zooming parameters to such values that the data cannot be displayed anymore using zooming and panning operations There always is a red data cursor which allows the user to measure exact values found in the figure Measured values are updated on every mouse move and displayed in the left top corner of the plot area Time period selection Turning on the toggle button with pencil causes that every mouse click in the graph area will move the starting time point to the location when the user clicked Time selection is represented by green area in the graph its length is automatically set to be the length of prediction horizon For better explanation see picture There is also possibility to choose the exact starting point using the datetime picker as seen on picture Data cut out definition 1 1 a a e 4 08 11 2011 14 06 Syne group 6 0 68 12 18 201
21. into a relational database There are specialized key value storages log systems and particularly messaging queue systems which will serve way better The article is a bit too strict saying that even session data does not belong to the database although doing this is a very common habit 3 12 4 Generic simulation jj Li Polymorphic association GenericProblem A LN SR Polymorphic association A MPCProblem OtherConcreteProblem Figure 3 3 Polymorphic associations in the application All associations are one to one This is de facto a core of the application so a simplified class diagram representing the relationships between parts of the application and a parental relationship is seen in the picture During the development of the application there was an exertion to think about it as being extensible to look at it as a generic simulation platform Moreover it was interesting to see how the ORM works and how it will perform So the GenericProblem object was established which is base for all simulations holding the data which is common for them trying to loose couple the concrete simulation object from the task queue and other parts of the system which it has nothing to do with 60 CHAPTER 3 DEVELOPER POINT OF VIEW The intention was to achieve such an architecture that as someone in future decide that he or she would like to found a fuzzy or another regulator or completely different simulation
22. name So in general when there is a key named parent in the YAML dictionary the dictio nary which contains it is considered to be a parent the rest of the data is looked up for another key of the name specified by the value of parent field and its value is used as a child When there is an import key the behaviour is almost the same with the difference that the content of the file with name contained in the import value is considered to be a child See YAMLMatlab page for further details 3 7 Web browser part of the application Web pages evolved a lot during past years They are no longer only HTML documents rendered from static files The trend is to make web pages to make user feels like he or she works with the real desktop application This involves client side scripting to be used intensively Today the most widespread language to be run in a web browser is Java Script It is simple language similar to Java in syntax but has its own manners Java Script is more and more powerful which allows the developer to write really sophisticated applications like word processors and spreadsheet processors and run them in the browser Java Script itself is quite lightweight and not feature rich hence many extensions were 3 7 WEB BROWSER PART OF THE APPLICATION 49 written to make routine tasks easier and less error prone and to span differences between various browsers Amongst others the most widely known extensions are jQuery and
23. of the window Users you now added are ordinary users In the listing click on each username of newly created users and in Groups block select the group you just created Then click Save If you like to make the users being administrators check Staff status and Superuser status for them But do not activate this now we need just ordinary users for the next instruction list Publishing templates to users 1 In Django administration click Generic problems then click on the simulation you want to associate The page you just navigated to can be also navigated from the simulation form using Administer this simulation or from the listing of simulations clicking the button with the icon of gear In the administration page of simulation choose group you created in steps of previous list in the Viewer users block Additionally you can afterwards create yet another users and associate them some simulations individually in the Viewer groups block At the end of each modification click Save Log out and then log in as some user you created Check that he or she can see simulations you associated him or her and that he or she can operate with them Remember that if you have turned on the is public flag of some simulation this simulation will appear in the listing of templates of every user if it is a template even if you have not explicitly associated it In addition if you set is public flag for working copy solved it will appea
24. simple in Java terms and constructs In Python software is designed against a contract This means struggling with the exact opposite thing It allows to do maybe too much which can recoil on the developer in the means of code mess and difficult refactoring because it is easy to break the contract even without knowing about it Both languages have their pros and cons every one suit to some sort of problems Python is more feature rich and more flexible It is a better language to learn something with because there is no need to deal with very constrained design possibilities and a beginner can simply concentrate on technology he or she wants to master 3 5 Brief introduction of YAML format YAML 22 is according to its official site a human friendly data serialization standard for all programming languages In its simple form it is very similar to the Python or Javascript format of dictionaries and lists definition In the full form it can be annotated by metadata to determine which class was serialized But full featured YAML was not used throughout this work except for storing date and time records The format is well established and human readable which suits it particularly well for academical usage where is often needed to change the raw data by hand In addition there are parsers for many languages 3 6 YAML INTERFACE FOR MATLAB 47 Its weakness can be storing some larger datasets especially large number of classes ann
25. small changes YAML files have special structure convenient for the purpose of the controlling prob lem To read the files there is YAMLMatlab see Section interface used which implements a kind of inheritance and allows to merge several files into one data struc ture Files are organized beginning with the most general level where for instance the variable classes are defined with their default items for example maximal temperature The fields of these classes are then successively specified by another files as needed For example when there is some element of the class with the maximal water temperature set and this value is different from the one defined in the class which the element comes under the value is overwritten by the one of the more specific class After YAMLMatlab merges the files the structure is as follows only important fields stated e variables Defines all possible variables which may occur with their default values constraints units etc e data Actual data processed by the controller Important part of the on line implementation but not used in the application because of inefficiency e controllers Contains key value map which holds settings of all controllers defined in the system Each value of this map is another map containing these key value pairs predictionHorizon Prediction horizon length in seconds solver Specifies which solver is used for computation and its settings
26. than the developer would like sometimes contrarywise lacks very basic feature frequently used elsewhere The ORM is subjectively very feature less slightly more complex queries are simply impossible to build with it But again Django is the best framework for beginners has very clear philosophy and absolutely marvellous documentation Where the documentation eventually says nothing there tells the source code which was frequently studied throughout the development We should not forget that Django was founded as a rapid development framework for newsrooms which is quire a specific area Then it did its job quite well in the application for control engineering As for the next development it is suggested to rewrite the application to utilize the TurboGears framework It is much more complex and complicated than Django but offers the freedom to the developer It contains SQLAIchemy 18 which is considered to be one of the best ORMs and SQL builders available It is designed to be open and it is not rare that it offers integration more libraries doing the same thing more templating engines to name one which allows the developer to make the choice To not only prise it TurboGears has quite ugly documentation compared to its complexity and not only because of that it has a mild learning curve On the development itself The application took approximately four months of hard work to write to the current appearance This is can be co
27. the application abilities Fig ure displays how the room temperature tracks the reference as the overall weight setting is changed by showing the results of simulating one optimization step of MPC controlling mathematical model of one part of the building see Section 1 2 for details The simulation was performed using the Manager interface described in Section 2 6 8 with settings comfort economy being 1 99 50 50 99 1 and variable disturbance ambient temperature used which was fluctuating around 10 C It can be seen that as the weight trade off changes from the emphasis on saving money to the emphasis on heat comfort the resulting room temperature starts to track the reference more and more precisely which indicates increasing action of the heating system as expected EB 50 50 A2 south room temperature 01 03 13 14 x 21 1 E 50 50 A2 south room min reference temperature ES 50 50 A2 south room max reference temperature 1 99 A2 south room temperature EB 99 1 A2 south room temperature 34 4 3 1 08 00 1 1 1 1 1 1 2 1 2 1 2 1 3 1 1 16 00 00 00 08 00 16 00 00 00 08 00 16 00 m Figure 4 1 Comparison of simulation results obtained using three differ ent weight settings In the legend the ratio expresses the setting comfort economy References were the same in all three cases 65 66 CHAPTER 4 CONCLUSION On the connection between the web browser and the server The applicatio
28. to an error in the computation log Aborted There is also one more state Aborted There are more reasons for the task to be aborted the task is aborted when it is engueued longer than its engueued timeout or is processed longer than its processing timeout but as for user the most obvious reason is hitting the Abort button on the live simulation log page see Section 2 6 11 2 4 THE APPLICATION IN GENERAL ORDINARY USER POINT OF VIEW 19 Aborted _ Tr Figure 2 2 Possible states of a simulation 2 4 5 Simulation presentations The simulation can be expressed to the user in several ways Currently there are two Manager presentation This one shows only user description initial conditions and overall weights It is meant for people without an expert knowledge of the building just to see how the trade off between comfort and cost influences resulting system behaviour ofthe heating Overall weights will be used to create Cost function weights as explained above The mapping is created by an administrator at the time the simulation is inserted into the application Other settings mentioned in Section 1 3 1 are hidden but are still present and serve as a predefined simulation settings Engineer presentation This interface is more complex and shows user description initial conditions constraints cost function and disturbances This is a place for sophis ticated experiments with the simulation
29. 1 14 37 x 25 53 1 3 2011 1 5 2011 1 7 2011 1 9 2011 1 11 2011 1 1 2012 1 3 2012 00 00 00 00 00 00 00 00 00 00 00 00 00 00 Close Figure 2 10 Data cut out picker 30 CHAPTER 2 USER MANUAL 1 1 a a o 02 10 2012 01 08 Syne group 6 0 66 02 10 2012 01 08 x 1 12 8 2 2012 9 2 2012 10 2 2012 11 2 2012 12 2 2012 13 2 2012 14 2 2012 15 2 2012 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 Figure 2 11 Data cut out picker zoomed to better see the selection 1 1 a a o Y 02 10 2012 01 08 Syne group 000000 o February 2012 o 02 12 2012 23 12 x 1 73 Su Mo Tu We Th Fr Sa 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 23 29 Time 01 08 Hour 8 2 2012 9 2 2012 10 2 2012 Minute 2012 14 2 2012 15 2 2012 00 00 00 00 00 00 00 00 00 00 00 Now Done Close Figure 2 12 Data cut out picker the calendar picker Data series synchronization There is one more noticeable feature in the picker win dow the starting point synchronization It is performed using colored toggle buttons located in the top right corner of the window If no button is selected the picker will neither do nor be under any synchronization If the user checks one or more of the toggle buttons then as soon as you close the window with these buttons checked every other series cut out component on the simu lation form even hidden one with the same colour combination checked will be set exac
30. 30 2012 22 17 33 CLOSURE Worker session finished Uploaded 13 output series Script finished ie ton n ee Figure 2 13 Computation log When you click the Compute button you are redirected to the page showing Live computation log Here you can watch what is happening with your simulation the table containing log is polled every few seconds The same log view is obtained on the page of results after the computation finished after clicking Computation log Again it is best to describe this using the example See Figure The meaning of columns is obvious Possible log record types are e Info Just says something to let user know that the process did not hung e Error Informs that the worker script raised an exception which was unhandled The exception is serialized and shown in the log in form similar to the way how it is displayed in MATLAB 32 CHAPTER 2 USER MANUAL e Debug This will probably contain dump of some MATLAB variable or structure which will be displayed after clicking View record in the Message area e Wsnapshot Signals that whole workspace has been just saved The Message area then contains the link where mat file can be downloaded and further be used to track scripts action in case of an error or for an interest e Closure This is mostly the last message of the log It says how exactly the computation was finished 2 6 12 Results page When the task is done it is possible to show its res
31. CZECH TECHNICAL UNIVERSITY IN PRAGUE FACULTY OF ELECTRICAL ENGINEERING Diploma Thesis BuildingLAB Predictive control of buildings Praha 2012 Author Bc Pavel Tomasko Supervisor Ing Jiri Cigler esk vysok u en technick v Praze Fakulta elektrotechnick katedra dic techniky ZAD N DIPLOMOV PR CE Student Bc Pavel Tom ko Studijn program Kybernetika a robotika Obor Syst my a zen N zev t matu BuildingLab prediktivn zen budov Pokyny pro vypracov n 1 Seznamte se s p stupy k integraci prediktivn ho regul toru MPC do st vaj c ch syst m zen topen ventilace a klimatizace HVAC 2 Navrhn te a implementujte vhodnou topologii v po etn ho syst mu pro MPC kter bude umo ovat komunikaci s HVAC syst mem is u ivatelsk m rozhran m pro lad n MPC regul tor 3 U ivatelsk rozhran navrhn te a implementujte tak aby umo ovalo simulaci chodu libovoln budovy z p eddefinovan sady budov pro r zn nastaven prediktivn ho regul toru r zn klimatick podm nky a prom nn po adavky zen D le toto rozhran bude umo ovat posuzovat kvalitu zen na z klad m ry poru en po adavk zen a spot eby energie 4 Integrujte cel syst m a ov te spr vnou funkci na pilotn m projektu prediktivn ho zen budovy VUT v pra sk ch Dejvic ch Seznam odborn literatury irok J Oldewurtel F
32. Cigler J 8 Privara S 2011 Experimental analysis of model predictive control for an energy efficient building heating system Applied Energy 88 9 1 9 doi 10 1016 j apenergy 2011 03 009 Vedouc Ing Ji Cigler Platnost zad n do konce letn ho semestru 2012 2013 prof Ing Michael ebek DrSc prof Ing Pavel Ripka CSc vedouc katedry d kan V Praze dne 4 11 2011 Declaration I declare that this diploma thesis was created entirely and only by me and that I used only materials cited in an attached list P ID E Prague MS RAMA Wik ee uala signature Prohl en Prohla uji e jsem svou diplomovou pr ci vypracoval samostatn a pou il jsem pouze podklady literaturu projekty SW atd uveden v p ilo en m seznamu V A V Praze dne AGS 2072 dr u CT A podpis iii Acknowledgement Many thanks primarily to my supervisor Ing Ji i Cigler who made his best effort every time I asked him for a help for some materials or whenever I had some other problem and invested a lot of his time for this work to be successfully done Many thanks also to my consultant Ing Jan Siroky who did his best too was always very co operative never refused to help me and invested a vast amount of time to assist me with this work as well I cannot omit to express my thanks to Ond ej Fiala who literally saved my life using his professional knowledge when I was really deep in the dead end Thanks to my
33. EW 17 2 4 The application in general Ordinary user point of view 2 4 1 Introduction and base terms The application basically allows the users to work with some simulation to set it up launch it and explore its output It can especially be used to iteratively run the same simulation tune selected parameters and track results evolution 2 4 2 Overall architecture The cornerstone of the system is the web interface expressing the simulation backend in a non expert accessible way Next noticeable part of the system is the task queue mak ing the application asynchronous The optimal control problems solved by the system are typically quite time consuming with a need of a special software MATLAB in our case This led to an establishment of a mechanism which in fact delivers the power of MATLAB to the user s web browser and in addition allows the user to run more tasks simultaneously Tasks are temporarily stored in FIFO queue and are taken by one of computation cores as the core is free User can then immediately continue by setting up the next task even though the first one has not yet ended 2 4 3 Templates and Working copies The system is built on the concept of templates and working copies An administrator creates a set of templates and associates them to particular users who can then use them The ordinary user can then take a look at the templates which have been associated to him or her choose one set its parameters a
34. MPC zen je t eba jim n jakou cestou p ibl it jeho metodiku Pokud si toti sami tuto my lenku neov nebudou ochotni nasazovat tento zp sob regulace v praxi p esto e bylo prok z no e dok e uspo it a 30 n klad na topen Proto je c lem t to pr ce vyvinout n stroj kter lidem p ibl strategii MPC we bovou aplikaci BuildingLAB Tento text se zab v designem takov ho n stroje kter umo uje u ivateli simulovat b h MPC K demonstraci funk nosti jsou pou ity modely budovy VUT FEL v Praze Dejvic ch Aplikace je pops na z pohledu u ivatele formou u ivatelsk ho manu lu a n sledn je pops n jej v voj z pohledu softwarov ho n vrh e Contents 1 Introduction 2 1 1 Advanced control techniques for HVAC 1 2 Controlled plant os oS oa oe ee eee REESE ERE SORE CE EES 1 2 1 MPC problem formulation KP A E Bese Ge Ge Bas 1 3 1 The off line simulation of the system fete Be od Oo HOE ee er PO o ee User manual OD pe ya AO OA 2 4 The application in general Ordinary user point of view 2 4 1 Introduction and base terms lt lt lt lt lt lt lt PP T EE ee 2 4 4 Simulation States 9 a soda ner ew ER AR ER pa SEE RO RO BO O RO 2 4 6 Running the simulation lt 2 zz 44404 lt 4454 eae 24 T OCU tia a s p a S Ga a sp a 5 ed OE AO EA eee 2 4 8 Simulation visibility
35. MooTools both a little similar in purpose but with slightly different philosophy 3 7 1 jQuery This Java Script extension was chosen for the application because is probably the most widely known and looking to its manual and tutorials signalled that this is the right choice Moreover the extension is used by Bootstrap library for its plugins interactivity It has small code footprint and is light as a basic package But even the basic one comes with really helpful utilities e Document object model DOM manipulation Offers methods which can very simply do operations with the HTML document tree like add delete move adding and removing CSS classes in a cross browser compatible way It abstracts out lots of browsers quirks and truly simplified the application development These features were used throughout the application for example for hiding and showing the arrow indicating tasks in the queue e Advanced event management jQuery simplifies the work with events creating user defined events and creating event handlers In the application there is lot of event bubbling used when events traverses the DOM tree up bubbles and triggers event handlers on its way registered for type of that event It is possible to stop this propagation and emit another event for instance Or only prevent the event to trigger its default behaviour for example link navigation or form submission Instead of hooking each element where something
36. N 37 Django administration of users and groups These modules are present in the interface by default Refer to the Django manual for details Welcome Pavel Go to the lab Change password Log out Home gt Lab gt Generic problems gt GenericProblem object Change generic problem View on site Owner admin F Is public Is template Presentation manager v Name TABS control of block A2 of CTU buildi Descriptor _ _ User comfort description 4 Viewer users pepicek jsiroky jcigler E brnator Hold dowm Control or Command on a Mac to select more than one Viewer groups a Hold dovm Control or Command on a Mac to select more than one Maximum 3600 time to wait for dequeuing secs Maximum 300 time for solution secs Delete Save and continue editing Save Figure 2 18 Django administration interface editing simulation 38 CHAPTER 2 USER MANUAL 2 7 Loading data into the application This section describes the process of inserting new simulations and creating new data series in the repository Actions described in this section cannot be done using the web interface An administrator must have an access to the underlying database through the Django application console 2 7 1 Loading simulations For creating new simulations the YAML files describing simulation for on line control are used These
37. RE status 0 ORDER BY id ASC LIMIT 1 AND status 0 RETURNING Every row is one enqueued item which can have status 0 enqueued and 1 dequeued The inner SELECT orders rows by their automatically generated primary key which is increasing sequence at least in PostgreSQL and in efffect returns id of item which has not been dequeued yet and has lowest id Outer UPDATE takes the id returned and marks the row as dequeued changing its status to 1 Whenever there is an inner SELECT the entire query may not be longer atomic as a whole So we added yet another check of status field to ensure that the row was not dequeued by another process just between the end of the SELECT and the beginning of the UPDATE Unfortunately RETURNING clause is not part of SQL standard and to make the ap plication to be potentially compatible with other databases finally was used similar double checked solution simplified UPDATE queue SET status 1 workerlock wlock d WHERE id SELECT id FROM queue WHERE status 0 ORDER BY id ASC LIMIT 1 AND status 0 SELECT FROM queue WHERE workerlock wlock d and then UPDATE queue SET workerlock NULL WHERE workerlock wlock d wlock d is Python syntax and means that this string will be replaced by variable wlock This variable is a random number from large interval which a dequeuing process will use to actually mark the queue item for later selecting its content The workerlock has set a UNIQUE constraint to
38. a database and the idea is grown throughout the whole framework There are some packages for Django which introduce other kinds of database Specifically for MongoDB there is a project Django MongoDB Engine 5 which according to its documentation does quite a nice job nevertheless it have not con vinced us to use it This is a final choice for the application storage because we know the technology we are quite used to think its way and it can in fact handle all needs of the application to specify some storing the application state storing simulations storing results task queue data repository 3 11 What happens after clicking the Compute button For a reader to become familiar with the core of the system it is sketched what happens with the simulation form when the user clicks the Compute button 96 CHAPTER 3 DEVELOPER POINT OF VIEW 1 At first the button behaves as the Save This involves data validation data types numeric value limits then persisting of the simulation data New Queueltem object is created which points to the simulation object using generic relation This means that object is now enqueued and waits for worker to take it Some worker Java poller running in MATLAB with the name matching task aim pattern reports its idle state during some poll sending appropriate HTTP request which dequeues the task and returns the task primary key as a response This is actually almost the whole fu
39. admin_routines py which loads helper functions into the console namespace 7 Load the specification file spec working yml created in MATLAB by command with open r spec_working yml r as f specdict yaml load f 2 7 LOADING DATA INTO THE APPLICATION 39 8 10 11 12 13 14 15 For given model controller pair which is to be inserted launch command gensim specAssimilator specdict controller_name model_name where controller_name and model_name respectively are names of the model and controller in the specification files gensim variable now holds the instance of Django model GenericSimulation of the simulation just inserted Load appropriate definition of simulation outputs for plotting with open r some_plot_definition yml r as f plotdef yaml load f Enter command plotAssimilator plotdef gensim concrete_problem_object which associates given plotting metadata with the simulation Repeat actions from step 8 to insert all model controller pairs wanted After the insertion execute specDictClean specdict which clears all specifics of the on line run data series reference settings etc Save the cleaned structure by command with open r cleaned_for_matlab yml w as f yaml dump specdict f In MATLAB load the file cleaned for matlab yml using command spec ReadYaml main_file yml 0 1 0 1 Please consider switches on the e
40. administration interface Welcome Pavel Go to the lab Change pass word Log out Site administration Recent Actions Groups Add Change My Actions Users Add Change 4 admin 2 brnator O me Change prsou Object descriptions Z ee oe htrhtr GenericProblem object Generic problem Ae htrhtr 2 GenericProblem object jeneric problem 2 GenericProblem object Generic problem 4 GenericProblem object Generic problem 2 jcigler Figure 2 17 Django administration interface main page The administration interface is accessible from the User menu the interface is captured on picture The function of the interface is expectable and consistent Clicking the label on the main page specifically Groups Users Generic problems Object descriptions the user will get a listing of object of that kind present in the system There are also 36 CHAPTER 2 USER MANUAL actions Add and Change with clear meaning The interface does not allow to add simulations because their structure is too complex to be expressed in the interface in some well arranged manner Django administration of simulations As one example for all you can look over administration page of one simulation see the picture The meaning of all buttons is obvious the meaning of editable items was briefly explained in section Just to specify Owner Changes the user which is considered to be a creator of the simulation being edited I
41. age of having more workers able to deal with your tasks 1 Start some simulation The knowledge of how to do it can be found in previous lists When you are in Live computation log do not wait for the simulation to finish Set up another simulation using the main menu Better said Repeat from the first step until you have enqueued all simulations you wanted You probably noticed white play arrow which appeared on the right side of the Simulations menu Whenever the arrow is there it means that you have some tasks in the gueue or some tasks just being computed on workers or both When all your tasks are done the arrow disappears This will happen when you navigate to another page or automatically the arrow state reloads itself every 30s If you have feeling that the computation takes too long you can go for Process ing Task gueue to see where in the gueue you are You can look for your user name in Owner column or just look for content of Task column which has a blue colour This means that the task is accessible by you hence it is probably your task unless you are an administrator in that case all tasks are yours in principle When all your tasks are finished you can display their results all at once to be able to compare them Go to Simulations My results Sort the list clicking the Last touched column twice in order to sort it descending by the time when tasks were finished Choose tasks for result
42. and it is in its responsibility to use some default The initial condition will be mostly an equilibrium or another interesting state Constraints This block contains hard constraint settings of the controller It is a list of variables for which you set the constraint Each item of this list has minimal and maximal value which can be each fixed or variable Cost function This block is a bunch of cost function contributor variables For each there can be set e Norm 1 norm quadratic norm and infinity norm e Weight floating point number higher than zero e Minimal and maximal value each can be either fixed or variable as in the Constraints block At the time of writing this document variable hard constraints are not yet implemented in underlying MPC scripts 2In current implementation of underlying MPC scripts there is an effort to normalize weights so values between 0 and 1 should make the best sense 1 5 BASIC REQUIREMENTS FOR THE APPLICATION 9 Disturbances List of input disturbance variables Each has variable content the data series cut out Overall weight This is simply a set of One big cost settings for the whole control problem When the simulation has the Manager interface set it is remapped to the weights in Cost function block which are overwritten by these new values User description This is place where you should briefly and clearly describe specifics of your current simulation setting to disting
43. avoid collision If the random number would be the same as the one already in the database the UPDATE will fail and the item will be dequeued next time The time when workerlock is not set to NULL is so short and the interval from which the random number is chosen is so large that probability of collision is very low There is a possibility of aiming a task to particular worker using pattern matching against its name Dequeuing SQL query includes LIKE clause against an expression saved as a part of QueueItem object The pattern field implicitly contains meaning that anybody can dequeue the task But for example aborting a task employs this technique to abort only the task chosen The mechanism is also good for debugging There is a setting in Django configuration file introduced which allows the administrator to set whether user with particular name when logged will automatically aim his or her tasks to particular worker 3 12 APPLICATION COMPONENTS 59 Why not to do this in a relational database However a relational database seems to be a best suitable tool to implement the queue And it in principle may be in practice relational databases can be fast at INSERT can be fast at DELETE but probably will not be fast at both If the application was large scale there would be problems with responsiveness Opinions like this can be found in many articles In one article for all in is said that any ephemeral data should not go
44. branch But imagine how many branches there can be how large part of them will be identical and the work to do when there is a need to change this common part later A programmer can use some SQL builder where he or she could solve the example easily with code similar to query A objects Auer query order_by day ieh query filter salary__gt 1000 ie Another actions with query In principle the builder will incrementally create the SQL tree and then trans form it to SQL string just before the database is to be hit The ORM will automatically instantiate objects of the Model including their relationships Ideally the builder plus ORM should constitute a transparent system without user having to know that he or she actually uses a relational database e Well established controller architecture This is the mechanism which maps concrete URL to a method which will handle the request call appropriate methods of model and renders the right response Form management Gathering informations from users is also one of the crucial things Framework handles the form with everything it involves It creates appro priate fields and grabs data from the HTTP request dealing with decoding them from URL or POST data It will validate data the user entered to check for its correct data types for various kinds of attacks and so on Frameworks can also go far that they generate the form automatically using infor mations fr
45. cally for the HVAC the cost function is roughly speaking equivalent to energy consumed plus penalty for some comfort requirement violation The key point which really differs the technique from others is just mentioned MPC factoring of given constraints into the control strategy hence it predicts potential future saturations and picks a control trajectory which both minimizes the cost function and does not violate the constraints all at once The technique heavily relies on availability of a controlled plant model which makes the process of plant model identification of a great importance To cover uncertainties and unpredictable disturbances of the controlled system there is a need of feedback to be integrated So in the on line implementations the methodology is applied as explained but only the first point of the optimization result is used as an input to the controlled system Then in the next sampling period the optimization is performed again with the new measured data and the whole process is repeated 1 2 Controlled plant Specifically for this work MPC is used for control of heating in CTU FEE building in Prague Its description follows Text and illustrations of this subsection are adopted from 36 Figure 1 1 The building of Czech Technical University in Prague Dejvice The building of the CTU uses Crittall 24 type ceiling radiant heating and cooling system In this system the heating or cooling beams are embedded
46. cific of the manager view is its simplicity So the only parameters the user can change here are Initial condition and Overall weights The latter is set using the slider and graphical illustration of trade off between comfort and economical aspects of the problem 2 6 WALK THROUGH THE APPLICATION 27 2 6 9 Simulation form the engineer view TABS control of block A2 of CTU building in Prague Dejvice Save Work with unsolved copy Work with copy from original Goto original etc Administer this simulation Description Simulation abstract or something suitable to be displayed as an introduction on the simulation form User description s ly describe what you are trying to achieve User description Simulation Prediction horizon 96 0 hours You can set prediction horizon and initial conditions Initial condition X Constraints A2 north supply f water temperature Use min Fixed v Temperature of water supplying the north part ofthe heating system AZ south supply Fixed min 15 0 water temperature Use max Fixed v Fixed max 50 0 Cost function A2 north difference supply minus return Norm 1 Norm x water temperature Equivalent of heat dissipated in the south wing of A2 Weight 1 0 A2 north difference supply minus return water temperature Use min Fixed x A2 north room Fixed min 0 0 temperature A2 south difference Use max Fixed supply minus re
47. comparison turning on appropriate checkboxes on the left Click Show results at once which will navigate you to the Results page where you can see results of all simulations selected Now use technigues you learnt in previous listings To revise Hide resulting figures of all simulations you are not interested to After this step you will probably end up with simulation result block with one figure per one simulation Turn on the comparator and select series to compare Then launch the comparator dialog 2 3 STEP BY STEP GUIDE FOR AN ADMINISTRATOR 15 2 3 Step by step guide for an administrator Adding users and groups The concept of users and groups in Django and so in the application is probably the same you are used to Groups are utilized basically to associate more templates to more users in some comfortable way 1 Select username Django admin from the main menu where username is the name of the user currently logged in and there must be the sign SU after that name which indicates that the user is a superuser otherwise the Django admin choice will not be displayed In the admin interface choose Groups Create one group using Add user at the top right corner of the window Go to admin home link at the top left corner of the window and click Users Now you see the listing of all users registered in the system including additional information about them Add several users using Add user at the top right corner
48. could happen handler can be attached to the common node of the DOM tree and captured events can be processed on this one place Events are used almost everywhere in the application where some button is possible to be clicked There was a strong effort to use event bubbling to optimize number of handler bindings and thereupon optimizing the application responsiveness e Active Java Script And XML AJAX The term AJAX is nowadays used more likely for any connection initiated from the browser which does not come from any browser plug in e g Flash or Java applet and does not involve any page reload or navigation jQuery again encapsulates browser differences in this technology and offers methods that allows content to be loaded into some DOM node or to load JSON content both things in one or two lines of code AJAX transfer is used for dynamic loading of description blocks in the simulation form and on polled pages There is no page automatically reloaded Instead only part of the page is reloaded which means better user experience page and browser 50 CHAPTER 3 DEVELOPER POINT OF VIEW suffers with almost no flickering and menus are operable whole time no matter how many times the content is reloaded 3 7 2 Bootstrap This is a library giving the application its appearance 4 Basically it is de facto one clever CSS file with possibility to be extended with some jQuery interactive components The reason it was chosen for t
49. ded is a better and much more flexible choice Use some kind of non relational database Nowadays these databases compete with the relational ones As their name says they do not have relations which was recognized in the project design so far though There is a possibility to mix relational and non relational databases together but this would perhaps need lots of time to manage for the first time and lots of skills to fine tune and debug On top of that more servers would be needed to install and administer We concentrated especially on MongoDB during the initial research Other possibilities can be found at 28 where the systems are also compared Non relational databases are extraordinary technology and certainly will be considered next time but finally the idea of using them was rejected for the application Use relational database The purity of relational databases and the fact that they are on the market for many years make them to be used very often even in new projects which would be even beter suited with another kind of storage PostgreSQL 13 and MySQL L are amongst others an outstanding technology both of these incredibly available for free The final decision is to use the relational database only We have been already familiar with PostgreSQL an SQL at the time the project started and one more for the relational database the framework chosen Django was designed with the idea that it will be used with such
50. e the ObjectDescription object and associate it to the simulation being described The resolver function is called for every described element with following attributes e helpClass Cathegory of description Can be one of following constraints Description of some tab of the Constraints block costfunction Description of some tab of the Cost function block disturbances Description of some tab of the Disturbances block simulation settings Description of Simulation settings block for the engineer interface simulation settings manager Description of Simulation settings block for the manager interface simulation description Short description of a simulation as a whole hiweight Description of the overall weight e helpIdent specifier of the concrete object being described Resolver matches the helpClass and helpIdent string pair to the selector for mentioned jQuery load function which it yields as a return value See the Query documentation for details There was an issue observed When the path to the HTML description is stored as a path to the directory where an index html file is stored and the path does not with a slash server will return a non permanent redirection causing the browser to reload the same content of the description file many times debugged with the Firebug extension of the Firefox web browser It is recommended to
51. e and close the picker Time points are synchronized by pushing the value selected at the moment the picker is closed to all others synchronized by the same coloured button combination Not pulling it from lastly selected So if you choose the starting time now you will loose your selection Repeat previous step for all series which are somewhat related to each other all references Now open some reference graph picker for which you just selected the sync com bination You can deal with the graph as the ones on Results page If you need to zoom to the part which you want to choose click button with the pencil and then in the plot area Green selection appears with its beginning in the point where you clicked Or choose the point from the Calendar picker clicking the edit box with the date and time When you close the picker it will publish its value to all subscribers with the same sync signature 14 CHAPTER 2 USER MANUAL Time point selections are persisted including the state of synchronization buttons so do not worry if you need to do this clicking procedure you probably will have to do it only once Persist the form using the Save button Do the same steps for disturbances if there are more than one Launching the simulation is exactly the same process as described above Running multiple simulations checking the queue If you need to compare more simulation results you will probably feel the advant
52. e application 2 6 1 Top menu BuildingLAB Simulations Templates Welcome ab My results Please log in to in order to work with the application or look for public templates working copies Figure 2 3 The top menu This menu is present on all pages except the description page It contains a couple of useful links e Simulations Templates List templates visible for currently logged user See Section 2 6 2 My copies List all working copies visible to the current user More in Section 2 6 3 My results List all simulations marked as Finished Read more in Section e Processing Task gueue Display tasks waiting for processing and tasks just being pro cessed Workers Show online computation cores e User menu Not logged in or username Log out Django admin for administrator only 22 CHAPTER 2 USER MANUAL Available templates sr Pavel Tomasko admir Delete Pre Simulation name Type Copies Time created M B TABS control of block A1 of CTU building in Prague Dejvice o mpc 1 April 20 2012 2 48 p m 7 B TABS control of block A2 of CTU building in Prague Dejvice o mpc 1 April 20 2012 2 48 p m m a TABS control of block B2 of CTU building in Prague Dejvice o mpc 0 April 20 2012 2 51 p m 6 TABS control of block B2 of CTU building in Prague Dejvice o mpc 1 April 20 2012 2 48 p m m m TABS control of block A4 of CTU building in Prague Dejvice o mpc 0 A
53. ee 3 11 What happens after clicking the Compute button 2 3 12 Application components 3 12 1 Data repository z 28 amp 8 Azs ole 3 12 2 Output series objects 3 12 3 Task queue cons esas 65 3 12 4 Generic simulation 3 12 5 MPC simulation 02 3 12 6 MATLAB poller 2 4 44 24 64 3 12 7 The MATLAB eluel 3 12 8 Data transfer protocol for an assignment Bibliography Content of the attached CD 43 43 43 45 45 46 47 48 49 50 50 50 53 53 54 54 55 56 56 57 57 59 60 61 61 62 65 71 Chapter 1 Introduction Buildings as such consume approximately 40 of total energy Nearly half of that is put into HVAC Heating Ventilation and Air Conditioning hence governments of many states concerned about impacts of this fact on the environment issuing standards regu lating the consumption and promising further savings in this field for the future 36 There are new houses built incessantly which comply to these standards but the number of new buildings compared to the old ones is small and this ratio is changing only slowly thus engineers are looking for new ways to keep the trend of increasing the savings by changing of structures which already exist It is possible to adjust or rebuild a part of the construction to achieve the goal but this is very costly Another way of save up is to make the efficiency of energy
54. er e to be able to return to the assignment change some parameter and launch the simulation again 8 CHAPTER 1 INTRODUCTION Engineer level which will serve to validate mathematical model of the building The user with an expert knowledge of the building will be able to show that the heating system behaves correctly or poorly with the given conditions This interface will be more complex with the functionality approximately same as the previous one but in addition e There will be a possibility to combine various input data series with more freedom and control over the settings e It will be possible to fine tune the simulation settings especially the weights of particular variable penalization for the MPC strategy AI results should be some way archived in the user s workspace for further return to the simulations already finished in the future 1 5 1 Structure of solved problem For the purpose of the application design and its clearance there was a data structure sketched which captures the entities important for the application The structure as mapped to the application has following parts Prediction horizon This is a constant floating point value in the units of hours pre defined to be between 8 and 168 Initial conditions Setting expressed by combobox filled with initial condition sets defined by administrator There is also one item crossed out which means that no initial state will be provided to the simulation
55. er login you can typically see the list of templates of which you have a read privilege because an administrator created a viewer relation between you or your user group and these templates As you create a working copy you are becoming an owner of the copy which gives you write access and allows you to enqueue it for computation 2 5 The application in general Administrator point of view 2 5 1 Simulation visibility One of the most noticeable differences that you are behaving as an owner to all simula tions present in the system That means you can view edit launch and abort foreign simulations 2 5 2 Administrator interface The next big difference is that an administrator has access to the admin area where some additional settings of the system can be changed e Add delete and change users and join them to groups e Add delete and change user groups e Change some properties of simulations Rename and delete Assign template to users or groups Change user description Create delete and assign simulation descriptor simulation parameter legends general introduction Change public flag of the simulation turning it on make the simulation visible for all users even anonymous 2 6 WALK THROUGH THE APPLICATION 21 Change template flag of the simulation turned on makes the simulation behave like a template Change simulation presentation Manager Engineer 2 6 Walk through th
56. from a rescuer to a nightmare The ORM basically maps every fetched row from the database to one mapped object instance and vice versa But it can sometimes map more instances per row or more rows per instance too All queries are possible to be coded by hand in raw SQL but this is said to be very error prone often dangerous last but not least a tedious work A programmer must properly handle quotation carefully test all queries to be sure he or she has not forgotten any punctuation somewhere In raw SQL there is no kind of SQL builder support e SQL builder This should be an implicit part of most of the ORMs It is the Swiss knife for querying against the Model Specifically imagine the developer knows that he or she would like to query against table A so in SQL he or she would start 52 CHAPTER 3 DEVELOPER POINT OF VIEW SELECT FROM A Then a program flow passes through some condition where is decided that there is a need to sort the rows by say day column so appropriate ORDER BY clause is added SELECT FROM A ORDER BY day Now the code passed to the place where was decided that the rows should be filtered to contain only those with salary bigger than 1000 How to do that There are basically three approaches maybe more The last query can be some way split then inserted the WHERE clause to obtain SELECT FROM A WHERE salary gt 1000 ORDER BY day Brand new SQL statement can be made for every possible
57. future soft realtime larger scale projects and finally the tool can not only represent control technology but a technology as is in another words present the knowhow of a control technology plus that we are able to bring it into the life using nice and effective ways On the connection between the application and workers The connection is pretty much of the same kind as the one with the browser Worker poll the server every five seconds every poll the one from the browser as well often initiates several SQL queries and updates The passive task queue would not succeed in any slightly more rush environment simply because SQL servers are not technologically designed to back responsive queues actually neither the unresponsive ones but the queue implementation as is definitely suits quite well to the relational storage nice theory ugly practice to be accurate To the future growth of the application it is highly recommended to use some kind of messaging system Reasons are roughly the same as these mentioned at the push technology Messaging systems are widely used in business applications to communicate between various parts often to asynchronously distribute time intensive tasks such as document transformations video processing and so on Moreover there was an idea spoken that the task queue system developed can be used to distribute the computational power for on line controlling purposes This would allow the concept of
58. h a tool which allows the user to simulate the MPC control strategy Mathematical models of the CTU FEE building in Prague Dejvice will be used as the demonstration of the application operability The application is described from the view of the user in the form of a user manual and then its development from the software designer point of view is captured Abstrakt zen je pom rn t ko posti iteln discipl na zvl t pokud je vysv tlov na lidem kte pracuj v jin m oboru Je asto t k m kolem vysv tlit jak pracuje regul tor p edat vhled o tom pro zasahuje dan m zp sobem a definovat celou oblast jeho odpov dnosti In en i pracuj c v oblasti HVAC Heating Ventilation and Air Conditioning asto cht j aby se nap ventil tory to ily v n jak specifick jimi definovan asov interval Kdy jsou t z ni pro tak odpov e je to ov en nastaven kter funguje et energii apod V tomto p pad by to ale m l b t pr v regul tor kter rozhodne kdy jsou ventil tory zapnut a kdy vypnut N kdy je prost t k pochopit a vst ebat filozofii teorie zen Probl m s porozum n m zp tnovazebn mu zen se je t prohlubuje kdy jsou pou ity pokro il technologie Jednou takovou technologi je zvl t v posledn ch letech MPC Model Predictive Control Aby lid pracuj c v oboru HVAC ale i mana e i porozum li cel mu procesu
59. he application is that it is good looking very easy to use and lightweight It rids the developer of a necessity of having deep knowledge of CSS and various browser differences or bugs so he or she can concentrate on the application functionality itself The library is good for quick starts however it is said to be non stylish and not much extensible inside its code Maybe it is a necessary compromise because for example its Java Script addon files are really minimalistic 3 7 3 Flot and graphs in the application in general Flot is a tool for rendering graphs using Java Script in the browser 7 It is very feature rich and no such a good alternative was found on the internet It has outstanding docu mentation including examples and fairly understandable API Its functionality was widen a bit to be able to reload displayed data on the fly as the plot area is resized or panned for Data cut out picker see section 2 6 10 In the picker whenever user changes zoom or panning and does no such action for next 300ms an AJAX request is sent for new data The request includes new time starting point new time end point and size of a plot area in pixels Server then resamples the data to return approximately the number of data points which is equal to the number of the pixels which saves bandwith When the user requests a view which is zoomed out sufficiently to call for every n th data point from the database decimation will take place already
60. he simulation was last saved 2 6 3 Working copy list The list includes working copies which are owned by the current user The columns are same regardless the place from where the list was navigated to Working copies can be navigated from the template list from My copies and My results items from the Simulations menu and by Working copies button in the simulation form Checkboxes on the left side allows the user to select simulations for bulk operation Delete button serves for the obvious reason Just remember that the deletion will also cover results of the selected simulations and the action cannot be undone 2 6 WALK THROUGH THE APPLICATION 23 All copies Show results at once Delete User description Pre Template Owner Time created Status m B m TABS control of block A3 of CTU building in Prague o Jan Siroky April 23 2012 8 42 Solved Dejvice siroky am El money E TABS control of block A2 of CTU building in Prague o Pavel Tom ko April 23 2012 Not Dejvice admin 12 42 a m solved a El dve rozdilne reference m TABS control of block A2 of CTU building in Prague o Ji Cigler jcigler April 23 2012 Not Dejvice o 10 25 a m solved Figure 2 5 Working copies listing There is one more button for bulk operation Show results at once which is particularly useful for comparing results of more simulations The simulation form of the working copy is viewed clicking on a label or an icon i
61. hen redirected to 11 12 CHAPTER 2 USER MANUAL the form of that copy You will notice that a button bar under the simulation title changed because you are now able to do more with the simulation Fill in the description Use something short and clear say emphasized economy Choose some initial condition and use slider to set desired trade off between comfort and economy slide it right Now you want only to save your changes use Save button Navigate Simulations My copies from the main menu and you will see your newly created working copy there described with text you entered You can also navigate the Template listing page again and you will notice that the number on the button in the Copies column of the template you just copied increased by one Click the button with the number and you should again see your working copy in the listing Now use browser Back button or click on the working copy User description to get back to the working copy form You want to launch the simulation process now click Compute This button will save potential changes and then enqueue the simulation for processing You will be redirected to the Live computation log You can wait for simulation to finish or continue your work To abort the simulation click the Abort button You will be redirected back to the simulation form but the simulation will be marked as Aborted and will be read only This is because there may be some mes
62. her users of the application This system could moreover allow the user to run more simulations at the same time which distributed on several computation cores are finished faster than in the case they are run on a standard desktop all at once 1 5 Basic requirements for the application The system will be operated using a web browser and generally speaking it will allow to launch simulations on some remote computational core currently MATLAB Specifically it will simulate one optimization step of a predictive controller on a defined model controller building MPC pair from the set loaded into the system Subsequently it will give an ability to clearly display the simulation results and especially to explore changes in a behaviour of the controller in the dependency on its settings in an illustrative way The work with the software can be divided into two base levels Manager level which will make ordinary people to become familiar with the practice in heating systems design with the MPC method by use of a simple interactive interface which will in fact allow them only e to display a prepared assignment template e to change one or at most a few simple parameters of a simulation for example one number which for given simulation determines a trade off between heating comfort and heating cost e to launch the simulation e to display resulting data series in relation with the system inputs and compare more results with each oth
63. hings clear In the example above there will be one slider created which will be labeled Overall weight comfort vs economy and will have a default value 50 This slider will be identified as HW1 in MATLAB After the simulation is started slider value 0 is transformed to 1 and stored as a weight of B3 S RoomCost If the slider is on value 100 the resulting value will be 0 Numbers inside this interval will be linearly interpolated Value of B3 S HeatCost is obtained by the same way but the slope the slope of the interpolating function is opposite to the slope of the for B3 S RoomCost 2 7 4 Loading data into the repository 1 Run the Django maintenance console and load admin routines py as described in the previous instructions 2 Some way load the data into the Python console workspace There are many ways how to do it it depends on the format the data is stored in Before loading to the application database the data must be formed to list of lists Inner list must have exactly two items first of them is time in milliseconds UNIX time 1000 second one is the value of the variable in that time Example data 1 0 2 0 2 0 3 0 3 0 2 0 3 Load the data into the system modrep RepositoryItem new data Name to identify the series There is a routine called saveSomeData in admin routines py which creates sim ple sinusoid Study its code for a basic idea about how to create and store the data Series 2 7 5
64. ies which you like to compare Remember that checkboxes will remain checked when figure with them is hidden if you check them and then hide the containing figure to do another comparison the series from the last comparison will still appear in the comparator window If you do not like manual unchecking you can refresh the page which will restore it to the state just after it has finished Click Compare selected button and the Series comparator modal window will appear The figure inside behaves like any other figure and displays the series you selected in the previous step Using the engineer presentation 1 Follow the first step of previous list with the difference that you choose some tem plate marked by blue M Now you are working with the Engineer presentation of the simulation Choose some initial state Simple edit fields does not deserve some additional comments But it would be boring if all references and disturbances would be constant So change Room temperatures from the Cost function block to both have have variable minimum and maximum choosing this in appropriate comboboxes choose the appropriate data series from another just appeared combobox for them it will be named in some clear manner typically it will contain the word reference and specifier min Click the button with bar graph to initiate the Graph picker At this time only turn on some coloured synchronization button say green on
65. instructions figure on the fact that the system is cleaned from simulations as next to last step will overwrite the specifications on workers but the new ones Updating the old definitions on workers is unfortunately not yet handled Assimilation of the data can be done in following steps 1 In MATLAB load YAML files as would be done by on line This is done by command spec ReadYaml main_file yml 0 1 with main file yml being the root file of on line implementation definitions 2 Enter command WriteYaml spec_working yml Spec which writes the data just loaded into the new file spec working yml Now all simulation definitions imported and mixed in are written in this one file 3 Now it is time to decorate the file by the initial states and weight mappings If adding of these is needed refer to Sections and 4 There is a file in buildinglab folder of the application called admin routines py which contains helper functions to insert new simulations Open the operating system console and navigate to the folder In first rows of admin_routines py is a line beginning with pth Ensure that the string which follows contain an absolute path to the root of the application the folder which contains the buildinglab folder 5 Execute command manage py syncdb with appropriate changes of the command on Linux This will start Django maintenance Python console 6 In the console issue command execfile
66. into the concrete ceiling A simplified scheme of the ceiling radiant heating system is illustrated in Fig ure The source of heat is a vapor liquid heat exchanger which supplies the heating water to the water container A mixing occurs here and the water is supplied to the respective heating circuits An accurate temperature control of the heating water for respective circuits is achieved by a three port valve with a servo drive The heating water is then supplied to the respective ceiling beams There is one measurement point in a reference room for every circuit The set point of the control valve is therefore the control variable for the ceiling radiant heating system in each circuit 1 2 CONTROLLED PLANT outside temperature of the input temperature water to the ceiling pipes An ceiling reference 6 radiant too heating temperature of the ouput water from container inside temperature temperature of the output water from the ceiling pipes heat contai exchanger ner Figure 1 2 Simplified scheme of the ceiling radiant heating system 1 2 1 MPC problem formulation Direct formulation of the problem of building heating control is N 1 min Rkur s Qrlyr Ze e k 0 subject to Fyre Gaur lt he k41 Azp Buk RVOE k 1 Ny Yk Czk Dur Wo k 1 N Lo Vinit Tg S Zk STk k 1 N With these quantities e x system state of dimension ny e u system input cont
67. ion and experiments Control Systems IEEE 32 1 44 64 feb 2012 V Peterka Predictor based self tuning control Automatica 20 1 39 50 1984 J Richalet A Rault J L Testud and J Papon Algoritmic control of industrial process In Proceedings Symposium on Identification and System Parameter Estimation Tbilisi 1976 IFAC J Richalet A Rault J L Testud and J Papon Model predictive heuristic control Applications to industrial processes Automatica 14 5 413 428 1978 Clara Verhelst Filip Logist Jan Van Impe and Lieve Helsen Study of the optimal control problem formulation for modulating air to water heat pumps connected to a residential floor heating system Energy and Buildings 45 0 43 53 2012 Jan irok Frauke Oldewurtel Ji Cigler and Samuel Pr vara Experimental analysis of model predictive control for an energy efficient building heating system Applied Energy 88 9 3079 3087 2011 Frank Wiles Three things you should never put in your database http www revsys com blog 2012 may 01 three things you should never put your database Online Accessed May 1 2012 12 BIBLIOGRAPHY Content of the attached CD There is a CD attached to this work which has following content e BuildingLAB web BuildingLAB web application e BuildingLAB worker BuildingLAB worker software e Text Electronic version of this text
68. itor can track nearly every variable throughout the code thus for instance automatic method renaming can be performed and it can be sure at least if the reflection was not used at this point that the language will allow proper tracking of the method being renamed or refactored Java classes have all precisely defined interface which is one of its biggest strengths In Python this is hardly possible because there is no certainty that method of the particular name will be on the same place next time Application can delete the instance variable or method on the fly and it is gone without generally no chance to know about whether it happens at the development time Python has no interfaces at all and although they can be implemented as a kind of an extension this would be probably called as non Pythonic approach The type system here is called duck typing its principle says roughly If you see an animal which quack like a duck and swims like a duck then you can say it is a duck 26 The fact that a programmer can define a method whenever wants and is not forced to design against an interface has an implication of faster and cheaper development than in Java But it is not so simple Both languages typing systems need a developer to have some design skills In Java application is designed against an interface Developer will struggle with its strictness and simplicity because he or she will have to abstract a real world which is not so
69. line particularly when explained to people who have another profession It is often difficult task to explain how the controller works to create an insight into its action and to define the whole field of its responsibility HVAC Heating Ventilation and Air Conditioning engineers often have reguirements like the one that they want their ventilators to work from this time to that time When they are asked why the answer often is that it is a tested setting which works saves energy etc In this case it should be instead the responsibility of the controller to decide when the fans are powered on and when they are not Sometimes it is hard to accept and absorb the control theory philosophy The problem of understanding the feedback control starts to be even bigger when advanced technologies are applied One of these technologies starts to be MPC Model Predictive Control especially in the last few years In order to make not only the people working in HVAC branch but also managers understand the whole process of control it is needed to some way expound them the MPC strategy Unless they check the idea themselves they will not be willing to implement the strategy in practice even though the advanced MPC was proven to be capable of saving up to 30 of the total heating bill Hence the goal of the work is to develop a tool that will bring the MPC strategy closer to people the web application called BuildingLAB This text deals with design of suc
70. ll its data structures and IO operations with it are fast Get identifiers of simulation model and controller to know which part of the data just loaded to use In the data subtrees specified by these identifiers create or rewrite values according to the information in the assignment If there are overall weights present in the assignment perform remapping to lo level weights According to the assignment download appropriate time series from the server These series are downloaded one by one by their dedicated HTTP channel in CSV format using MATLAB function urlwrite saving it as a temporary file and then the series are loaded into the workspace using dlmread Set initial conditions for the simulation if provided in the assignment Call the simulation backend which formulates the problem using YALMIP and then solves it Upload resulting time series to the server using the same principle as their read Series are sent in CSV format using POST request The script occasionally emits log messages about its progress as another special HTTP requests If there is an exception raised the script will serialize it as YAML data and sends it to the server as another log message 3 12 8 Data transfer protocol for an assignment First data transferred to MATLAB after the task processing is launched is the assignment which is a list of items which are essential for simulation to begin As said in section 3 12 6
71. mentation At the moment the development of the application started there was an MPC runtime HVAC control system established The controller in this runtime is implemented in MATLAB and takes into account e Disturbance predictions This is a weather forecast which includes mainly an ambient temperature and also partly a solar radiation power e Considers future references Controller has a knowledge of future requirements of a temperature which allows it to prepare controller plant for the reference change before it starts e Keeps constraints of a plant Takes into account maximal ratings of all controlled variables minimal and maximal water temperatures maximal speed of temperature change The system reads input data first Data is used to formulate the optimization problem for which YALMIP toolbox 21 is employed Then a solver is called to obtain the solution There is mainly sdpt3 solver used also SeDuMi solver was engaged but evinced worse results mainly with longer prediction horizons Computed results are eventually written to disk 1 3 CURRENT CONTROLLER IMPLEMENTATION 5 The MPC runtime stores the data and even communicates with PLC using YAML more in Section 3 5 files stored locally on the disk of the machine which performs on line computations One of the key points of the application being developed is to emulate this communication and use the system as just described for off line simulations with only
72. mputation It is worth mentioning that this page is not realtime e g when you find the status here saying that the task is in processing this label will not change until you manually refresh the page There are more pages containing some information which is expected to change anon but are not refreshed automatically These pages have the footer saying when the page was generated which was an effort to hold the displayed data consistency without the need of having each odd page auto reloaded 24 CHAPTER 2 USER MANUAL 2 6 4 Result list This page looks the same as the one showing working copies of the user but the content is filtered to show only Finished and Aborted simulations 2 6 5 Task queue Task queue Status Task Owner Last touch Finished TABS control of block A4 of CTU building in Prague Dejvice Pavel Tomasko admin 29 04 2012 23 59 20 In processing on prague_serf3 TABS control of block A4 of CTU building in Prague Dejvice Pavel Tom ko admin 29 04 2012 23 59 25 In processing on prague_serf2 TABS control of block A4 of CTU building in Prague Dejvice Pavel Tom ko admin 29 04 2012 23 59 30 In processing on prague serf1 TABS control of block A4 of CTU building in Prague Dejvice Pavel Tom ko admin 30 04 2012 00 02 02 Enqueued TABS control of block A4 of CTU building in Prague Dejvice Pavel Tom ko admin 29 04 2012 23 59 40 Enqueued TABS control of block A4 of CTU building in Prague Dej
73. n the User description column Every simulation has associated the user description text block which serves as a fine identification of every simulation If you do not fill in the User description field and create for example ten copies from the same template you will easily loose control over which is which Pre column was described in section 2 6 2 Clicking an item in Template column will navigate to the template of the given copy Remember this difference from the User description column which leads to the actual copy Especially when the user does not annotate his or her simulation with user descriptions it is easy to click the Template column instead of the User description column and copy the template instead of the existing copy by an accident So try to annotate your simulations when possible Right of the Template column is place for additional operation buttons which are the same as described in section 2 6 2 The information about who created the copy is found in the column Owner The user who is not an administrator will probably find here himself or herself in most cases Time created column says when the particular copy was made Status column indicates what is happening with the simulation Used terms which might appear in there are described in section with the difference that the state Solving is named Processing on worker name instead with worker name being the iden tification of the core which deals with the co
74. n uses polling a lot which is known to be a methodology wasting a lot of both network and machine resources The reason it was used is that the application is supposed to be very small scale approx 20 users at once in average plus some workers so the overhead is acceptable HTTP protocol is further designed to be a reguest response which suits very well for polling It is ineffective to reload part of a page every few seconds as it is done on several places in the application The application is finally fairly soft realtime which makes the user feeling better but induces an idea of rewriting at least part of it to use some HTTP push comet technology rather than polling There are several possibilities Orbited 12 APE 2 which can handle the socket abstraction very well It finally seems like the server is initiating the connection response in the browser is nearly immediate resource usage is much more effective Push technologies are proven to be fully functional by companies such as Facebook Meebo and others The reason that the push technology was not even tried is that it looked likely to be too heavy weight for the application and so the author rather concentrated on more important aspect of the application like its design and philosophy of the framework The push technology is finally highly recommended because at least the knowledge gained during the project would have been excellent it would establish a base for some
75. nction of the poller which then polls with busy status notification Dequeuing also means that there is an association between Equip object and the de queued item which serves for the purpose of determining which worker is processing the task When worker polls with the busy state it serves et first to inform the system that worker is still alive Every busy poll hit can also dequeue item which is marked as control At the moment control messages are used only to abort computations When the abort item is dequeued by poller it performs JMI call causing currently running MATLAB script to interrupt as the user hits Ctrl C and then starts polling with the idle status Because an abort request is just another item of the queue all control items overtake other items in the task queue MATLAB script is called which uses urlread to download the assignment from the server using supplied primary key On the server the data structure of simulation is recursively traversed which results in a simulation assignment The assignment is serialized as YAML data and returned as a response 5 The script looks up the assignment for information about specific script which should be executed and executes it Now follow actions specific for the current implementation of MPC The script flow is explained in section 3 12 7 3 12 Application components 3 12 1 Data repository Simulations can be feed with either constant or variable data In
76. nd launch it To use a template means that the user creates a working copy of the template This copy becomes his or her private working place where he or she can set various parameters run the simulation using these parameters and finally explore the results produced by the simulation 18 CHAPTER 2 USER MANUAL NN Results Explore results Templates A Choose template to work with Selected Working template copy Create copy to be N private for user Working Launch computation copy Figure 2 1 Simplified application workflow 2 4 4 Simulation states Each simulation goes through a simple lifecycle which has following states Unsolved At the time user creates the working copy the state is set to Unsolved This is the only state which allows user to edit the parameters of the simulation Other states may signal presence of partial results and log items which would not be consistent with saved changes in simulation settings hence other states disable saving and launching of the simulation Enqueued When the simulation is launched its state is changed to Enqueued This means that the task was pushed into the task queue but not yet received by computation core Solving As the core receives the task its state is changed to Solving Finished After the simulation run the state is set to Finished This happens also in case of an error raised during the simulation You can find out whether the task finished normally or due
77. nd of the command it is different from the one in step 1 Especially the last one is important because it disables the YAML importer form performing inheritance If the switch is not present or is 0 inheritance will pollute your cleaned data by defaults from higher level definition classes which will issue no error but will probably lead to strange simulation results Save spec variable as file Spec01 mat save Spec01 spec Finally distribute the file SpecO1 mat to the worker machines into the folder blab scripts located in the root of worker SW equipment 40 CHAPTER 2 USER MANUAL 2 7 2 Adding the initial states to the definitions There is a field added to the YAML definitions added to the root of a selected model The model section of YAML may then look like unimportant fields omitted dejviceTempAl BLinitialStates fields A1 S Room A1_S_RW A1 N Room A1 N RW A1_S_SW A1 N SV states name Cooled down value state 19 5 24 19 24 initial 19 5 24 19 24 26 25 7 name Overheated south value initial 23 5 28 21 5 26 35 25 9 name Overheated north value initial 20 5 26 22 5 30 27 8 36 1 name Overheated both circuits value initial 23 5 28 23 0 26 37 2 38 7 Values under state key will be used as initial measurements supplied to the simulation backend as variables named by the content of fields Values under initial will be u
78. nsidered a good job for a person who has never written the web application before The total time investment was approximately three times longer though For the author it took this long to absorb all technologies and mainly to find the way to the goal which suits him The development was also a very good lesson of self confidence which is maybe the most important outcome of this work for the author 68 CHAPTER 4 CONCLUSION Bibliography 10 11 12 13 Advanced message queuing protocol http www amqp org Online Accessed May 1 2012 Ajax push engine complete comet solution http www ape project org Online Accessed May 1 2012 Bactool http www bactool ethz ch Online Accessed May 1 2012 Bootstrap from twitter http twitter github com bootstrap Online Accessed May 1 2012 Django mongodb http django mongodb org Online Accessed May 1 2012 Fapws fast asynchronous python web server http www fapws org Online Accessed May 1 2012 Flot Attractive javascript plotting for jquery http code google com p flot Online Accessed May 1 2012 Java to matlab interface http undocumentedmatlab com blog jmi java to matlab interface Online Accessed May 1 2012 Matlabcontrol http code google com p matlabcontrol Online Accessed May 1 2012 Mongodb web site http www mongodb org Online Accessed May
79. nsion points for e g doing some preprocessing in Python code but not wasting the resources of the primary web server The lightweight web server can be one of many available which allows the use of WSGI 20 for example Fapws 6 User computers Web server SQL database LWSGI Computatioal cores workers Figure 3 2 Full system configuration LWSGI stands for Light Web Server Gateway Interface 3 3 SOFTWARE PACKAGES USED IN THE APPLICATION 45 3 3 Software packages used in the application The application uses mainly following software packages products e Python 2 7 2 main programming language of the application e Django 1 3 1 e PostgreSQL 9 1 7 e Numpy 1 4 1 for data series cut out resampling and time shifting e Apache HTTP Server 2 2 e MATLAB R2011b e SnakeYAML 1 9 e Java Virtual Machine 1 6 0 17 part of MATLAB e GNU screen for ability to keep workers running with detached terminal 3 4 Python programming language This section could easily become a chapter even a book This section will mainly discuss the language typing in short because it was the most noticeable aspect of the language throughout the project Just to be complete Python is feature rich language which allows both rapid pro totyping and writing complex applications It is an interpreted language and is easily readable by humans During the research there was a lot of time spared by searching through the web
80. om the Model Coding forms in HTML and handling them manually is really slutifying job and frameworks can make it fun At least partially 3 9 MATLAB OJ e The templating engine Applies the DRY principle on HTML Allows to es tablish HTML inheritance and to reuse parts of the code As a result HTML will simplify with a better chance for bugs to be fixed and to make changes This can be said for middle to mega frameworks There are also so called microframe works which has minimum functionality leaving vast free space for developer s creativity to use only pure programming language or to use any libraries 3 8 1 Django Parts of a web framework were partly enumerated in the previous section which was done partly with Django in mind For the application it would be best to emphasize following aspects of the framework e Model Model is an integral part of Django and also probably the most useful and beneficial one It is an embodiment of Django datacentric philosophy and does lots of things under the hood All data entities are defined there including all their attributes and associations using declarative style which is in accordance with write less do more principle For example declaring a field to be a float field is one line of code for the application developer but model library of Django sets up validation for the data type is able to emit form field for this data field generates SQL for creation of table for the cla
81. on and encapsulation above the TCP IP layer making the development easier MATLAB has a routine for performing either GET and POST requests in its base libraries Java also has this functionality built in Moreover if the communication will be maintained using HTTP there is no need to design any server side for MATLAB workers They can directly call the web server with peer written in the same way as the application Workers are in fact behaving as automated web browsers 3 12 7 The MATLAB glue The title of this section comprises the code written in the MATLAB language which connects the application with MPC backend glues them together As written in the previous section communication over HTTP made things simpler so the glue once called from the poller does following steps 1 Downloads the assignment By another words simply instructions about what it shall do The instructions are transmitted in YAML format of the form described 62 2 CHAPTER 3 DEVELOPER POINT OF VIEW in the next section and is interpreted by YAMLMatlab library details in section 3 6 Looks the assignment for the information about which script to run and launches the script The script is specific for every kind of simulation For currently implemented MPC the next steps are as follows 1 Load workspace needed for the computation Workspace is stored locally on the worker as a mat file which is a MATLAB native file format can hold a
82. on the database level otherwise every small movement of the graph could result in transmitting of all data form the database to the web server which is 1 2MB now Decimation in PostgreSQL is achieved by first ordering the data by its time and then selecting each n th row using function computing modulo n It needs a row number to be determined which functionality is unfortunately not a part of SQL standard In the application PostgreSQL window function row_number is used to determine the number 3 8 Web development using a framework This section is only a brief summary The theme could be laid out over many thick books 3 8 WEB DEVELOPMENT USING A FRAMEWORK ol Web development is probably the mainstream development activity done on the in ternet of today This is maybe the biggest reason of why many programmers tend to make their daily routine more pleasant As a side effect there were several techniques methodologies and concept established to reduce code repetition increase component de coupling increase decoupling between data business layer and html presentation Tools for development considering such stuff and even more are concentrated into packages so called frameworks Frameworks not only provide physical codebase for projects but also provide philo sophical knowledge base e g how to organize the code how to organize the data how the validation and request routing should be made which is of the same
83. online process Another group of YAML files is used to store the information about which data is plotted and about grouping of these data into charts The structure of these files is e Root of the file is a dictionary with only one key named PlotDefinition The value under this key is a list which contains a number of dictionaries Structure of each that dictionary is Title Determines the title of the chart Variables Groups variables which should be displayed all in one figure It is a list of dictionaries each of them with following items x ID Identifies which variable specified in YAML definition files will be plotted x Color Color to plot the series expressed as a hexadecimal RRGGBB triplet Description Name of the series to be displayed in the legend 1 3 1 The off line simulation of the system Running the system without the connection to the underlying PLC is used to tune the controller to explore energy savings and to study possible failure of the controller The simulations run are short term covering the length of the prediction horizon This strategy of studying the MPC is based on the idea that when the system behaves well on one prediction horizon typically three or four days long in all characteristic situations heating up after a weekend day night transitions etc this induces that it will be likely to behave well also in closed loop This is in contrast in long term approaches
84. otated by its type Parsing is then potentially slow and annotation data overhead grows The format is a primary format for data storage in original MPC scripts An example of YAML file Taken from 22 Time 2001 11 23 15 02 31 5 User ed Warning A slightly different error message Date 2001 11 23 15 03 17 5 User ed Fatal Unknown variable bar Stack file TopClass py line 23 code x MoreObject 345 n file MoreClass py line 58 code foo bar 3 6 YAML interface for MATLAB At the beginning of the project there was need to purify a way YAML files were read and written by the MPC controller to get rid of some bugs and make it more transparent The implementation used and still uses SnakeYAML This library is written in Java and basically transforms YAML text file into structure of Java lists and dictionaries filled with appropriate data types Java strings doubles integers date objects It can do lot more but for the project simplicity was of a strong importance So the goal was to transform this Java data structure to MATLAB one It was not so easy because there is no direct mapping between list dictionary primitive structure and all needed MATLAB types MATLAB has two kinds of matrices in base common matrices have all elements of the same type and cells can store arbitrary data types Matrices can have more dimensions cell matrices can contain another cell matrices and common matrice
85. parents grandparents and uncle who always supported me a lot and survived my long lasting days of being nervous during the work on the thesis Thanks to my girlfriend Aneta K ehn ov who have not broke up with me although I was very workaholic never stopped to support me and always trusted that I will manage this work Last but not least thanks also to all my friends for their support Pod kov n D kuji t mto zejm na m mu vedouc mu Ing Ji mu Ciglerovi kter se sna il mi pomoci v dy kdy jsem byl v nesn z ch a v noval mi spoustu sv ho asu Mnoho d k pat tak m mu konzultantovi Ing Janu irok mu kter pro m t d lal v e co mohl nikdy neodm tl dost o pomoc a do asistence p i t to pr ci tak v noval spoustu asu Nemohu zapomenout pod kovat panu Ond eji Fialovi kter mi poskytnut m sv ho know how doslova zachr nil ivot kdy jsem uv zl ve slep uli ce a nev d l co d l D kuji sv m rodi m prarodi m a str ci kte m v ichni bez ust n velmi pod porovali a p e ili m dlouhotrvaj c obdob nervozity b hem t to pr ce D kuji m p telkyni Anet K ehn ov kter se mnou z stala i kdy jsem byl velmi workoholick nikdy m nep estala podporovat a v dy v ila e tuto pr ci dot hnu do konce V neposledn ad tak za podporu d kuji v em sv m p tel m Abstract Control engineering is guite an elusive discip
86. pril 20 2012 2 51 p m Figure 2 4 Templates listing 2 6 2 Template list Here the user sees which simulation templates are accessible to him or her If you are an administrator you can use checkboxes on the leftmost side of the table and delete them clicking the Delete button Beware that this removal will cascade on every copy created from this template so users can loose their results If you do not like this behaviour use the administrator interface which will not perform cascaded deletion The Pre column lets you know which presentation type is used for the template M stands for Manager view E stands for Engineer view Clicking the item in the Simulation name column will open given template Next to the Simulation name column is place with buttons for additional actions Button with little gear is visible only when you are an administrator and will navigate to the administrator interface for given simulation The button with the i letter will appear when the simulation has a description page assigned and navigates to that page Application is designed to be able to work with more simulation types than only MPC In case you implement a new type you can distinguish it from others by the Type column The Copies column includes buttons whose label eguals of the number of working copies of that template belonging to the logged in user Clicking one of these buttons displays the list of working copies The last column says when t
87. properly alive The value in this column determines when the application heard about the core for the last time At the moment the worker record is automatically discarded when it does not send any message for 30 seconds 2 6 WALK THROUGH THE APPLICATION 25 Online cores Type Identifier Last sign of life before Status Worker brno_serf1 00 03 c Worker brno serf2 00 03 Worker brno serf3 00 03 Idle Worker brno_serf4 00 03 c Worker prague serf1 00 03 Worker prague serf2 00 03 c Worker prague serfi 00 03 Figure 2 7 Online cores listing For Status maybe the name says everything Possible values are Idle and Busy Here should be noted that when there is a task just being processed by the worker and currently logged user has a right to view it the Busy status will be displayed as a button and will navigate to the just being processed simulation form 2 6 7 Simulation form All simulation forms have some common parts There is an info about its state and underneath is a button bar Its content is different according to the current simulation state There can appear following buttons e Clear solution Clear result and log of Finzshed simulation and set its state to Working copy e Results Open result page of finished simulation e Save Persist changes made in the form e Work with unsolved copy Copy the current simulation and navigate to its form If the the current simulation is solved copy it without solution as
88. r in the listing of working copies plus in the listing results of the associated user The user then will not be able to delete it because he or she will not be owner of this simulation unless you set the user to be the owner which might be confusing for him or her 16 CHAPTER 2 USER MANUAL 4 If you want to cancel all associations to users or groups of more templates you can do that as a bulk operation In the Django admin go to the Generic problem listing turn on checkboxes belonging to the simulations you like to cancel associations for and then use appropriate action from the combobox labeled Actions located above the listing and click OK Setting up a template When simulations are inserted to the system they are set to some default state The script assimilating the simulations will make an effort to set up the simulation for you using supplied YAML files by setting weights norms prediction horizon constant maxima minima and so but will not be able to infer some details for example which disturbance and from which time you like to choose The approach to fine tune the simulation follows E Choose some template you want to set up Go to its administration You learnt the knowledge of how to do so in previous instruction listings You do not need to choose a template every simulation can be turned into a template and back which ability we will now utilize On the administration page uncheck Is template flag
89. repository is finally implemented as s single table in a rela tional database holding time value pairs and having foreign key to teh table containing name of each data series 3 12 2 Output series objects Results of a simulation are stored into the database textfield as CSV data There is ResultItemSeries oject for each data series of each simulation which is prepared with the data textfield empty as a template which is an information about which data the application wants from the computation script after it finishes When the solution is ready after the computation appropriate mappings of ResultItemSeries are updated saving the data The object also stores an information about the series name to be displayed in resulting graph legend and color used to plot the series There is also object named ResultItemSeriesGroup which groups together series where displaying of them in one figure makes sense The object holds name of the group to be displayed as a title of resulting figure 3 12 3 Task queue The queue self is implemented as one database table The key moment is how the queued content is dequeued because this includes select and update in a non standard composition and should be done atomically In pure PostgreSQL there can be used its extension of the SQL language One dequeuing shot would then look like this simplified UPDATE queue SET status 1 WHERE id SELECT id 58 CHAPTER 3 DEVELOPER POINT OF VIEW FROM queue WHE
90. rolled variables of dimension nz e y vector of system outputs ny e v vector of disturbances of dimension n e z slack variable on the zone temperature of dimension n e s t norm order of particular parts of the cost function Zo Linit State at time 0 initial state dimension is the same as for k 4 CHAPTER 1 INTRODUCTION e r r minimal and maximal reference variable for the system outputs dimension is the same as for zz e N prediction horizon positive integer number e k discrete time integer number k 0 N 1 e A B C D system matrices of appropriate dimensions e V W matrices of disturbance input of appropriate dimensions e Qg Rk weighting matrices of appropriate dimensions which express trade off be tween reference tracking and energy consumption Fy Gx hy matrices vector of appropriate dimensions defining constrains of sys tem states and input variables This is application of the technique just mentioned Weighted energy equivalent is pe nalized in s norm while weighted reference band violation is penalized in t norm First sharp inequality reflects the physical limits of the system Second inequality constraints the slack variable of reference band The slack variable is artificially introduced variable which serves for penalizing given variable which is attached to only when it escapes given interval system output in our case 1 3 Current controller imple
91. ry to encapsulate the solution of this problem for example matlabcontrol 9 The final decision is not to take somebody else s solution Details will be cleared later in section 3 12 6 3 10 Data storage There came out some different points of view when the project was consulted with other people Possibilities were generally these e Use text files use YAML files use CSV files Filesystem is very fast but using it for BuildingLAB data storage would be very likely complicated and error prone The application was from the beginning heavily experimental in the sense that there was quite unclear what exactly it should do The approach Take a bunch of files load them all parse them all take value here take value there compute something save another file is suitable for one shot scripts But our case since it has begun is something that has some not entirely trivial state there will be some kind of user authentication structured data possibly large data sets and so 3 11 WHAT HAPPENS AFTER CLICKING THE COMPUTE BUTTON 55 There will be many relations between data anyway and manual handling of these not well defined references which are in addition very likely to change is very hard to implement in filesystem even harder to debug A schema which could handle this was to be designed with a lot of effort but it was cut off because of the decision that some technology which is established for handling the data nee
92. s Moreover there are MATLAB structures better speaking maps which have string keys and each value can be of arbitrary type mentioned or another structure We had to restrict a set of expressible data structures with the needs of our specific application taken into account For the system model there are matrices of maximal order of two and data should look well arranged in YAML column vectors ought to be 48 CHAPTER 3 DEVELOPER POINT OF VIEW column and similarly for row vectors Thus when there is possibility to concatenate given list of lists to a matrix same data type same count of elements in leaf lists common matrix is created This process works up to the depth of two For example A 7 8 9 0 3 2 B 4 5 C 123 will create structure with one key A storing 2 to 3 matrix key B storing two element column vector and C storing three element row vector If the data types are incompatible Date and number in the same list or lengths of inner lists are different reader issues a cell matrix YAML writer was written to work the opposite way data written from MATLAB to YAML and read back should be the same as written There was also a requirement for the reader to support imports and mixins Both features mentioned are the same in principle Mixin is a merge of two dictionar ies which are concatenated to one new dictionary with child dictionary keys overwriting the parent ones when they have a same
93. s 2 eos ann aa kk oth eae ees rere 351 Simulation SIB 3a sas ee ee a eR ee HES 2 5 2 Administrator interface 2 6 Walk through the application 2 6 1 Top menuj OK ea ew ee ee BR A Le ee 2 6 2 Template list 2 4 2 2 oe eee deena ed 44 oo Be E a 2 6 3 Working copy list 264 Result Bill 2 s sioc fe Goes au ea ESR ERE be ea Be Ge yl oe ae ee ee eee ee 2 6 6 Online cores 4 s eae eae EAA eR eR Re ee te pao ee ee ee ee ye er eg 2 6 8 Simulation form the manager view 13 OO IO O B UN FE F e VM 2 6 11 Computation log E core O we ea h ere boo 2 7 1 Loading simulations 2 7 2 Adding the initial states to the definitions 2 7 3 Adding weight mapping to the definitions 2 7 4 Loading data into the repository 2 1 0 Description system 3 Developer point of view 3 1 Architectural overview 3 2 Top level component interconnections 3 3 Software packages used in the application 3 4 Python programming language 3 5 Brief introduction of YAML format AE ozve 3 7 1 QUE PRT ec V 3 7 2 Bootstrap 8 56 2 aden phage A 3 7 3 Flot and graphs in the application in general 2 3 8 Web development using a framework 3 8 1 Django x 2 4 ua aa er ea CRIAR de a bee ie A ee 3 9 1 How to call MATLAB from outside 3 10 Data storage decae oe ee aa Be
94. s public When checked the choice causes the simulation along with its potential solution to be visible for anonymous not logged in users Is template Determines whether the simulation is treated as a working copy unchecked or as a template checked Presentation Allows an administrator to choose between available presentations of the simulation form Name Allows editing the name of the simulation Descriptor Combobox determining which description will be used for given sim ulation More on this below User description Allows a change of user description Viewer users Selects which users will see this simulation in their Templates listing in case the simulation is a template or in My copies My results listing in case it is not Viewer groups Selects which user group will see the simulation For all users in given groups applies the same as written for Viewer users Maximum time to wait for dequeuing Sets the time in seconds after which the enqueued simulation will be aborted if no worker dequeues it Maximum time for solution Sets the time in seconds in which the core dealing with the simulation must manage its solution If it does not manage it the task will be aborted Django administration of object descriptions To be complete this allows the administrator to relate the simulations with given de scriptor More on this in separate section below 2 6 WALK THROUGH THE APPLICATIO
95. sages in the computation log and some results already present To check the partial result and the log you can click Result button in the button bar But now click Clear solution and you have your working copy as you would never have started it Displaying results comparing results 1 2 Repeat steps 1 4 from the previous list Now click Compute and wait for the simulation to be finished The Results page appears Pan the figures by mouse dragging use mousewheel to test zoom Switch zoom restriction buttons above the figure and explore how this changed the zoom be haviour click magnifier buttons to perform wheel less zoom Finally click button with the crossed circle to reset the view Drag the figure in the area near its title to change the order of figures on the page This can be useful if you like to quickly see two far figures on the same place Change series visibility using Displayed series dropdown menu You must click the checkboxes to achieve the effect not their label Hide some figures using Displayed figures dropdown located in the button bar under the Results title For example imagine that you like to compare North and 2 2 STEP BY STEP GUIDE FOR AN ORDINARY USER 13 South supply water temperatures so switch off other figures which are not of your interest Click Series comparator on the main button bar Checkboxes will appear in front of each series label Turn on checkboxes of ser
96. sed directly as a vector of initial state Each item of states list will be expressed as one item list in Initial conditions control in the web interface under name determined by the content of name 2 7 3 Adding weight mapping to the definitions In each controller YAML specification which should support the weight mapping for the manager interface there is following dictionary defined example for MPCcontrollerB3 MPCcontrollerB3 BLweightMap MAP1 B3 S RoomCost dstVec 1 0 0 0 src HW 1 srcVec 0 0 100 0 B3_S_HeatCost dstVec 0 0 1 0 src HW 1 srcVec 0 0 100 0 BLhiWeights HW 1 value 50 0 description Overall weight comfort vs economy 2 7 LOADING DATA INTO THE APPLICATION 41 In BLhiWeights dictionary there are Overall weights defined which will express them selves each as one slider Currently there is always only one The value field determines the initial value of the slider Description is a string which identifies the slider in the web interface The value from the manager interface slider is transmitted into MATLAB It is always between 0 and 100 The value is transformed using linear interpolation with the XY grid created by srcVec dstVec respectively src field is the identification of the source slider from the BLhiWeights section Resulting value will be used as a weight for variable with the name being the same as a name of the key the transformation is defined in To make t
97. ss which includes this field and so on Everything tries to do as transparently as possible ORM of Django is simple to use but lacks possibility of performing some more com plicated tasks which often leads to suboptimal approaches with complex queries e Templates Django templating system tallies with the general description above It is maybe too feature less which was probably an intention templates are often written by non programmers but during the application development there were several features really missed like more advanced conditions or very constrained possibilities of calling python methods from the template e Forms The framework is able to generate forms based on the model ideally only needing some more declarative description of how exactly to do that e Authentication Seamlessly adds an information about logged user into the request object which is passed into each view method Also provides autogener ated form for log in and many other useful tools 3 9 MATLAB MATLAB is well known and well established computational system It is not only a clever calculator it is de facto programming language 54 CHAPTER 3 DEVELOPER POINT OF VIEW The language have not changed radically since its beginning so that it looks quite ancient in some aspects Good for compatibility but not for people It is very out standing in many kinds of computations especially the matrix ones has very wide area of libraries
98. state estimation and in its state key holds directly the initial state vector Both keys are optional hiweight has hi level weight in its value refMin refMax Determines the minimum maximum of a variable in the value field Specifies the targeted entity using the target key This key can route the value to constraints or costFunction refMinSchedule refMaxSchedule The same as previous point with the differ ence in value field which is a dictionary pointing into the data repository and specifying which data cut out will be downloaded The dictionary has following keys id primary key of a data series in the repository t0 starting time point of a data series t1 end time point of a data series ts sampling period in which the data are expected to be tsh time point to which the first data record will be shifted and the following ones will be shifted by corresponding value All time entries are in milliseconds 64 CHAPTER 3 DEVELOPER POINT OF VIEW Chapter 4 Conclusion The result of the work is a tool which presents MPC control in a human accessible interface and distributes time consuming simulations on potentially powerful workers for computation During the development there was an effort not only to accomplish the functionality reguest but also to create some way engaging environment for the user of the resulting product A demonstration of operation As an example of
99. the case of the variable data we need some flexible source of it In current MPC scripts implementation the data are stored just another YAML file This is quite useful for short online data series but generally not sufficiently flexible It has mainly these drawbacks 3 12 APPLICATION COMPONENTS 57 e Current implementation of YAML file reading is disgustingly slow It is not a fault of the format self but of the implementation of libraries used to read it mainly those written in MATLAB The MATLAB code is highly recursive and YAML structures could be deep Unfortunately MATLAB does not support references so the recursion means that the data structures are copied over and over again The read takes time which is equal or even longer than the actual problem solution This fact is causing failures in the online implementation but users working with the web application are usually annoyed when their request is fulfilled in tens of seconds not speaking about several minutes which the solution of MPC problem usually takes The effort is to shrink this time to be as short as possible e The YAML format has quite a space overhead if used for storing floating point numbers It is even worse in storing information about date and time In current implementation there is by guess more than half space consumed by useless time metadata The repository holds the data which are used to feed simulations by anything distinct from a constant value The
100. tly the same starting time value 2 6 WALK THROUGH THE APPLICATION 31 This is very useful feature when you need to select the starting point for two or more series which are some way related as for example minimum and maximum value maybe some disturbances and so on You can very well manage with exactly only one button checked Buttons could easily be radio buttons instead of checkbox type ones because the checkbox implementation allows you to set up to 128 synchronization groups for one simulation form which is certainly more than you need in the whole system Checkboxes just for fun On all pictures there is a green button switched on 2 6 11 Computation log Time Type Message April 30 2012 22 16 58 April 30 2012 22 17 01 April 30 2012 22 17 01 April 30 2012 22 17 01 Assignment downloaded starting script BLDejviceMain April 30 2012 22 17 02 DEBUG View record April 30 2012 22 17 03 Task engueued Delivered to prague serfl Task degueued Downloaded 2 input series April 30 2012 22 17 03 MEIST Workspace snapshot taken download here April 30 2012 22 17 03 Calling SolveMPC April 30 2012 22 17 03 TT Workspace snapshot taken download here April 30 2012 22 17 04 April 30 2012 22 17 18 SolveMPC Formulating problem SolveMPC Problem formulation successful calling solver April 30 2012 22 17 32 MTU Workspace snapshot taken download here April 30 2012 22 17 33 April 30 2012 22 17 33 April
101. turn water temperature Fixed 00 A2 south difference supply minus return water temperature A2 south room temperature Disturbances Ambient temperature g forecast Series Dejvice AmbTemp v The temperature outside ail 08 11 2011 14 06 Figure 2 9 The engineer view 28 CHAPTER 2 USER MANUAL The engineer view can be a bit mind blowing place but it is nothing complicated in principle You see controls already known from Manager lab There are many extra controls so it is possible that you will see help bars on the right if the administrator set them up which should explain content of each part of the form Best way to explain how does the form behave is perhaps to use the example as seen on the picture The Simulation block is self explanatory The prediction horizon setting is constrained as written in section 1 3 1 The Initial condition field is set to some concrete value which will be used during simulation initialization More interesting are the following blocks which are tabbed to be able to change more settings of the same kind one tab per one constraint one for each cost function addend and the same for disturbances The Constraints block denotes hard constraints of the solved problem Fields are functioning almost the same as in next two blocks Basically if you see the combobox starting with word Use the variable can be of both constant and variable content The appropriate field will change
102. uish between more simulations created from the same template 10 CHAPTER 1 INTRODUCTION Chapter 2 User manual 2 1 Quick start How to launch the simulation For the system to be usable some user should be logged in Log in by clicking the User menu in the top right corner It is labeled Not logged in 1 22 Select Simulations Templates from the main menu From the list choose the simulation to run In the following form it is possible adjust simulation parameters To launch the simulation click the Compute button Wait for results to be displayed 2 2 Step by step guide for an ordinary user Creating copy running the simulation aborting the simulation 1 Choose Simulations Templates from the main menu The list of templates visible by you will appear Click name of the template you like For beginning choose template marked with green M which will lead you to the simpler simulation form Now you can explore the form read descriptions and so on If there are some controls on the form they are greyed out Moreover there is no Save or Compute button It is like this because you are just viewing a template This template is probably shared by more users and serves as a starting point for their simulations This is why the template is read only Click the button Work with unsolved copy This will copy whichever simulation form you see and sets it up to be your working copy You are t
103. ults It happens automatically when the user has the Live computation log displayed at the moment the simulation finishes which then causes the user to be automatically redirected to this page otherwise the user can open it from some page that contains working copies list by clicking an item in User description column or clicking the Results button in the simulation form This page can show results of more than one simulation In that case the results are viewed one after the other just imagine the page on the figure repeated several times with different data 2 6 WALK THROUGH THE APPLICATION 39 Results Solution logs Series comparator Displayed figures TABS control of block A2 of CTU building in Prague Dejvice dve rozdilne reference ted vaha na penize Assignment Solution log 1 1 a a o Displayed series Dejvice A2 Room temperature ER A2 south room temperature EB 42 south room min reference temperature MM A2 south room max reference temperature B 42 north room temperature EB A2 north room min reference temperature MM A2 north room max reference temperature 14 14 14 14 14 14 24 24 24 24 24 24 00 00 04 00 08 00 12 00 16 00 20 00 00 00 04 00 08 00 12 00 16 00 20 00 1 1 a a eo Displayed series Dejvice A2 Cost function B A2 comfort violation penalty south EB A2 comfort violation penalty north EE A2 heating cost south EB A2 heating cost north 01 02 09 19 x 66 50 1
104. vice Pavel Tom ko admin 29 04 2012 23 59 44 Engueued TABS control of block A4 of CTU building in Prague Dejvice Pavel Tom ko admin 29 04 2012 23 59 48 Figure 2 6 Task queue listing As the title says this is the view of the task queue Specifically you can find here enqueued tasks task in processing and done tasks for five minutes since the time they were finished The meaning of column Status was described in section 2 6 3 The Task column contains items of the form user description simulation name and each one navigates to the given simulation form Simulation has its Owner from the time it was created as a working copy This consequently indicates who enqueued the task The Last touch field informs about the time when the state of the simulation was changed For example when you see Enqueued in the Status column this column gives the time when the task was enqueued and so on 2 6 6 Online cores This is the overview of workers available in the system an their current jobs The Type column is actually matter of possible future It is planned to implement one additional type of equipment connected to the BuildingLAB queue which is meant to be a computation core starter Identifier is the unique string which distinguishes one worker from another The Last sign of life before column is maybe relic from the past when the timeouts have not yet been implemented and there was a need to see whether are workers are
105. why there are people who dislike it and exalt PHP above when we talk specifically about the web development No satisfactory reasons were found maybe the one that the Python language specification changes faster than it would be appropriate resulting in inter version incompatibilities This cannot be judged without a try to rewrite any application for another Python version but one can dare to claim that in the matter of pure Python this could be even automated process Problems come when there are some native extensions which need to be rewritten and compiled again But web developer for common middle scale sites probably will not write native extensions only may have problems such that the database backend is not yet compiled for the newest python version so cannot be used Hard to say hopefully it is not such a big problem To be honest the framework used in this work Django is not yet compatible with the last version of Python although the version is already out for a long time so it is somewhat problematic especially when there is a lot of advanced language construct used which in Django sure is 46 CHAPTER 3 DEVELOPER POINT OF VIEW As a dynamically typed language people who came from the world of Java may initially dislike Python because Java is in principle strict in typing One possible reason of Java popularity can be its type safety and language strength in the sense that a language tool IDE programming ed

the thesis

Contents

Download Pdf Manuals

Related Search

Related Contents