Home

Project no. 004074 Project acronym: NATURNET

1. Figure 4 7 An example of a model fragment causing ambiguity 4 3 2 Discontinuous change due to conditional influences The discontinuous change due to conditional influences rule detects opposing influences of which one is removed by a conditional inequality If this influence is dominant removal will make the other influence dominant causing a change in derivative from increasing to decreasing or the other way around This can be illegal since it may break the continuity rule the derivative should first become zero There are three possible ways this construction can occur Figure 4 8 Firstly there is a net negative influence with a conditional inequality on its target quantity and a net positive influence without one Secondly the net positive influence can have the inequality but the net negative can have none Finally both influences can have an inequality but they are on different values i e the influences do not disappear at the same time Figure 4 9 shows two model fragments on which the first variant of the rule applies conditional_influences_discontinuous_change Element Quantity1 Entityl MF Quantity2 Entity2 MF Quantity3 Entity3 MF2 Quantity2 Entity2 MF2 MF netNegInfluence Element Quantityl Entityl Quantity2 Entity2 MF 30 40 Project No 004074 NATURNET REDIME D4 4 netPosInfluenceOn Quantity3 Entity3 Quantity2 Entity2 MF2 MF inequality Quantity2
2. eceeceeceeeceeeeeeeeeeeeteeeseeeteneteeeseeeeeeeteneeaaes 14 3 2 4 REASONING ASSUMPUONS 242205242 siedte E A aE E a a 16 3 2 5 Derived relations cimil ld dia 18 3 2 6 Exogenous behaviours iirc ae cicck ee dei ieee Hel da 18 S2 MS A tots a al tea e ol tk ahah ad SMe eid ale aii RM a 19 32 95 POPMIMAtlOns t2 25 etelec dra 20 S229 OFCOFING sie secede cane dads A P A EEE A AEII A N AEA T A E A T 21 3 22 10 2SUCCESSOF States E A ENEE ES 24 3 3 Conclusions and future research cccecccceeceeeeceeeeceeeeeeeeeaeeeeaeeececeeseaeesaeescaeeseaeeesaeetsaeeesueessaseesass 25 A AAA T 26 4 1 Troubleshooting Viewer c oncninnnnninnnnnnncnncocnrcnrrrnrn rr 26 4 2 Diagnostic Rule ForMat ciones Ke 28 4 3 Diagnostic rulos iio a dd A Se ee ie ae is 29 4 3 1 Ambiguity due to non corresponding quantities ee eee eeeeeeeeeeeeeeeeeeeeaeeeaeeeaeeeaeeeaeeeaeenatesas 29 4 3 2 Discontinuous change due to Conditional influences cee eeceeeeeeeeeeeeeneeeeeeeaeeeaeeeaeeeaeenaeeeas 30 4 3 3 Conflicting causal relations c ee eeceeeeneeeeeeeeeeeeeeaeeeaeeeaeeeaeeeaeeeaeeeaeeeaeeeaeeeaeeeaeesaeeeaeeeaeeeaeesas 31 4 3 4 Invalid decrease increase in bottom top point VAlUC eee eee eeeeeneeeeeeeeeeeeeeeeeeeeeeeaeeeas 32 4 3 5 Invalid decrease increase in bottom top ZEPO eee eeeeeeeeeeeeeeeeeeneeeeeeaeeeaeeeaeeeaeeeaeeeaeeeaeeeas 32 4 3 6 Same quantity with different quantity Spaces eecceeecceesceeee
3. Figure 3 18 Trace Exogenous derivative assignment 3 2 7 Influence resolution Influence resolution is an important part of the reasoning engine It is traceable using option nr 8 Typical example traces are shown in figures 3 19 and 3 20 Y Simulate Engine trace d heat36 plus a ES Eos did ERE Noa al Firstly the Influences and Resulting causal model is related values effects are constructed are listed displayed Figure 3 19 Trace Single influence on quantity 19 40 Project No 004074 NATURNET REDIME D4 4 Simulate Engine trace Joes birth9 immigrationS and death emigrat for ambiguous causal effect d size9 neg 2 esl eae esl ti Bs NS 0 Multiple influences can Ambiguity causes A possible resulting have an unknown branching in the effect is tried and combined effect statesearch displayed Figure 3 20 Trace Multiple influences on a quantity A recent addition to the influence resolution procedure is the calculation of 2 order derivatives 10 As shown in Figure 3 21 these are displayed in a similar style using the same option Simulate Engine trace a X EEEE Cal ea ES EOS Figure 3 21 Trace 2 order derivatives DOs BINS el dl 3 2 8 Terminations Using option nr 9 a trace is shown of the search for terminations Each changing quantity magnitude inequality or q
4. NATURNET REDIME l N I Project no 004074 Project acronym NATURNET REDIME Project title New Education and Decision Support Model for Active Behaviour in Sustainable Development Based on Innovative Web Services and Qualitative Reasoning Instrument SPECIFIC TARGETED RESEARCH PROJECT Thematic Priority SUSTDEV 2004 3 VIll 2 e D4 4 Intelligent Help System Due date of deliverable 31 06 2007 Actual submission date 05 09 2007 Start date of project 1 March 2005 Duration 30 months Organisation name of lead contractor for this deliverable UvA Revision Final Project co funded by the European Commission within the Sixth Framework Programme 2002 2006 Dissemination Level PU Public PP Restricted to other programme participants including the Commission Services RE Restricted to a group specified by the consortium including the Commission Services co Confidential only for members of the consortium including the Commission Services 1 Authors Jochem Liem Floris Linnebank Anders Bouwer Bert Bredeweg Elinor Bakker Project No 004074 NATURNET REDIME D4 4 Abstract This document describes the intelligent help systems added to the Garp3 qualitative modelling and simulation workbench These systems aim to reduce the support required by modellers from QR experts when modelling Firstly to support the regular interaction with t
5. More options can be added to check the details of each inference Still users may choose to inspect only one specific inference type without inspecting the general inferences Each inference type is printed in a specific colour to allow easy identification Colours have been chosen to correspond with button colours as much as possible Table 3 3 gives an overview of the inference types buttons and colours Please note that the background of the tracer window is black this allows many different colours to be used that stand out very clearly 9 40 Project No 004074 NATURNET REDIME D4 4 Nr Button Name Remarks Colour Errors amp Warnings Always on Displays error and white warning messages 4 Ds Show general reasoning Displays reasoning process bite status information overview 2 a rat oe Displays addition of any model B f PP ingredients ragments Show search for 3 Hi assumable model Displays reasoning aan fragments engine default assumptions process inequality reasoning ony search ie madel Displays reconsideration of l 4 Eo fragments still applicable model Aamen turquoise after transition 9 Show added Displays inequalities and 5 Ral dependencies and check exogenous derivative settings on conditional inequalities and value branching Show inequality l E 6 Ral reasoning details Displays inequality reasoning coralred Show inegualit Displays derivable relations Qa
6. S06 semi urban erosion process 100 0 Different consequence value assignments on Erosion rate S06b erosion inactive 100 0 Different consequence value assignments on Erosion rate SO6b erosion inactive 98 8 Non corresponding proportional quantities potentially cause ambiguity U07a mosquitos configuration 93 8 Non corresponding proportional quantities potentially cause ambiguity RO8b assumed res inflow determined by manure 87 5 Non corresponding proportional quantities potentially cause ambiguity U12 well being improvement process 87 5 Non corresponding proportional quantities potentially cause ambiguity U02 controlled drainage process 87 5 Non corresponding proportional quantities potentially cause ambiguity R17 resource inflow for crop production E Non corresponding proportional quantities potentially cause ambiquity RO8c manure influences fertility Non corresponding proportional quantities potentially cause ambiguity RO7a vegetation growth process Description The following non corresponding proportional quantities potentially cause ambiguity Manure gt Fertility in RO8c manure influences fertility Fertility gt Resource inflow in R17 resource inflow for crop production Figure 4 2 An example of a model fragment editor 27 40 Project No 004074 NATURNET REDIME D4 4 060008 FAQ The Qualitative Reasoning and Modelling Portal O e 6 OQ http st
7. inequalityGreaterThanBottom Quantity2 Entity2 MF increase _ in top point value Element Quantityl Entityl MF Quantity2 Entity2 MF Quantity2 netPosInfluence Element Quantityl Entityl Quantity2 Entity2 MF topPointValue Quantity2 Entity2 amp inverse netNegInfluenceOn _Quantity3 _Entity3 Quantity2 Entity2 _MF2 inverse inequalitySmallerThanTop Quantity2 Entity2 MF Figure 4 12 The decrease increase in a bottom top point value rules ent mana o avos ecological integrity of river basin config Neon sustainable actions Influences Sustainable actions Imhm 6 Wax BHigh B Medium A River basin River basin Medium O Blow Blow a Forh M Zero P BZero X resiosica integrity development rate S Ecological integrity o Bpmgh i E 6 High 6 a bes Good A cfd Moderate 2 Emin A Viode Y Poor W Zero v Figure 4 13 A model fragment potentially causing an invalid increase in top point value of Ecological integrity 4 3 5 Invalid decrease increase in bottom top zero The invalid decrease in zero and the invalid increase in zero rules detect instances where influences potentially cause quantities to increase in zero as the maximum value or decrease in zero as the minimum value of a quantity space Figure 4 14 32 40 Project No 004074 NATURNET REDIME D4 4 shows the rules and Figure 4 15 shows a m
8. Entity2 Type Value MF inverse inequality Quantity2 Entity2 _Type2 _Value2 MF2 conditional_influences discontinuous change Element Quantity1 Entityl MF Quantity2 Entity2 MF Quantity3 Entity3 MF2 Quantity2 Entity2 MF2 MF netNegInfluence Element Quantityl Entityl Quantity2 Entity2 MF netPosInfluenceOn Quantity3 Entity3 Quantity2 Entity2 MF2 MF inequality Quantity2 Entity2 Type Value MF2 inverse inequality Quantity2 Entity2 _Type2 _Value2 MF conditional_influences discontinuous change Element Quantity1 Entityl MF Quantity2 Entity2 MF Quantity3 Entity3 MF2 Quantity2 Entity2 MF2 MF netNegInfluence Element Quantityl Entityl Quantity2 Entity2 MF netPosInfluenceOn Quantity3 Entity3 Quantity2 Entity2 MF2 MF inequality Quantity2 Entity2 Type Value MF2 inequality Quantity2 Entity2 Type2 Value2 MF inverse equivalentInequality Quantity2 Entity2 Type Type2 Value Value2 Figure 4 8 The discontinuous derivative change rule influenced influenced Entity Entity Entity Entity ee as S influenceone 1 laminfluenced Zp A Zimh g Zp i Zimh 6 TD ir y y Fhion DD L 3 y BHigh A WZero y BMedium A BZero v BMedium Blow Blow y W Zero Zero Figure 4 9 Two influences that potentially cause a discontinuous chance 4 3 3 Conflicting causal relati
9. Report D4 1 15 Bredeweg B Bouwer A Jellema J Bertels D Linnebank F and Liem J Garp3 A new Workbench for Qualitative Reasoning and Modelling 20th International Workshop on Qualitative Reasoning QR 06 C Bailey Kellogg and B Kuipers eds pages 21 28 Hanover New Hampshire USA 10 12 July 2006 16 P Salles B Bredeweg and N Bensusan 2006 The ants garden Qualitative models of complex interactions between populations Ecological Modelling Volume 194 Issues 1 3 pages 90 101 17 Linnebank F 2004 Common Sense Reasoning Master thesis University of Amsterdam Amsterdam The Netherlands 18 Simmons R 1986 Commonsense Arithmetic Reasoning Proceedings of AAAI 86 p 118 124 Philadelphia California 40 40
10. Simulate Engine trace QuE 2 Re oa as SI 0 1 a Three combinations of Note that without ordering the 5 terminations are returned terminations in this example have 31 by the ordering procedure possible combinations Figure 3 26 Trace Ordering results 23 1 40 Project No 004074 NATURNET REDIME D4 4 3 2 10 Successor states The application of transition scenarios is very similar to the application of normal scenarios This part of the reasoning process is displayed at a general level using option nr 11 and at a detailed level using options 2 and 4 Note that epsilon derivative continuity constraints and 2 order derivative continuity constraints are displayed at this point just as in the termination procedure Option nr 4 is also important in the context of the search for successor states because it shows the checking of candidate model fragments that were active in the previous state A typical trace is shown in Figure 3 27 fi fal Y Simulate Engine trace n is immediate Raol Lal Ra Re Figure 3 27 Trace Apply transition content da Ss el sl After the complete state description of a state has been determined it is compared to existing states Again option nr 11 controls if this process is traced If an equal state is found a transition to that state is returned If no equal state is found the listing of the new state is displayed A trace of the compar
11. assignments 4 3 9 Conflicting derivative value assignment and causal relation The conflicting consequential derivative value assignment and causal relation rule detects derivative value assignments on quantities that are also influences by a causal relation in some way It searches for a consequential derivative value assignment on Q as a key ingredient and either a proportionality or an influence on Q Since the net result of the causal relation can be positive stable or negative it potentially clashes with the derivative value assignment Figure 4 22 shows the rule and Figure 4 23 shows a model fragment with a derivative value assignment and an influence on Qone as a possible example of the problem detected by the rule derivative_value_assignment_plus_causal Element Quantity Entity QuantityX EntityX MF1 Quantity Entity MF1 derivativeAssignment Element Quantity Entity amp eProportionality QuantityX EntityX Quantity Entity MF1 eInfluence QuantityX EntityX Quantity Entity MF1 Ja Figure 4 22 The conflicting derivative value assignment and causal relation rule PB Quantity1 Entity Entity Oe p Owe Zimh 6 Zlmh High BHigh Medium A 8 Medium A Blow Blow WS Zero gt y BZero 7 Figure 4 23 A model fragment with a derivative value assignment and an influence on Qone 4 3 10 Multiple consequential value assignments on quantity The multiple consequential value assignments
12. nen y after the model fragment 7 reasoning details earch and aferintience indianred derivable relations resolution Show influence resolution E ae Displays calculations on the 8 50 using influences and sida proportionalities 9 be Show search for possible Displays terminations and gold x terminations causes Show ordering and 10 Di removal of possible Displays ordering process orange terminations Displays application of Show search for il l 11 di successor states using rd kempanisan wii brown ordered terminations existing states and lists all details of newly found states Table 3 3 Inference types buttons and colours Regular buttons are also present to perform some general tasks These are shown in Figure 3 4 and perform the following functions select all options deselect all options inverse the selection save the trace to a file clear the window and close the window 9 l Figure 3 4 Regular task buttons 10 40 Project No 004074 NATURNET REDIME D4 4 3 2 Example traces In this section the value of the tracer will be demonstrated using example traces from existing standard models Stove1 and Ants Garden 16 The different trace option buttons will be referred to by their number assigned in Table 3 3 Please note that a small red triangle is always present in these examples This cursor is part of the tracer window and should not be confused with the arr
13. plus emigrationl plus current dependencies immigrationl gt emigrationl constant dependencies testing combination from_greater_to_ equal immigrationl emigrationl onsistent combination testing combination from_greater_to_equal immigration emigrationl to_point_below immigrationl inconsistent combination testing combination to_point_ below immigrationl inconsistent combination Finished determining mathematical consistency for terminations from_greater_to_equal inmigrationl to_point_below immigrationl EEES Two out of the three The current Two terminations possible combinations are situation is are tested in this inconsistent given example Figure 3 25 Trace Mathematical ordering 22 1 40 Project No 004074 NATURNET REDIME D4 4 Mathematical ordering uses the inequality reasoning engine to determine valid combinations of terminations Note that this procedure can reveal a lot of extra information when options 2 4 and 5 are also selected because the details of each combination are added to a temporary environment and tested for consistency A typical trace of the mathematical ordering process is shown in Figure 3 25 The trace of the ordering procedure finishes with a listing of al valid combinations of terminations that will be used as in transition scenarios in the search for successor states A typical example of such a listing is provided in Figure 3 26
14. proportionalities 4 3 8 Consequential derivative value assignment causes conflict The multiple consequential derivative value assignments conflict rule searches for quantities that have multiple derivative value assignments on them as a consequence If these model fragments fire at the same time an inconsistency would result since a quantity cannot have multiple derivative values at the same time Figure 4 20 shows the rule and Figure 4 21 shows 3 model fragments with a value assignment on different derivatives as a possible example of the problem detected by this rule By having different assumptions for each value assignment or having the derivative value assignments as conditions which makes the simulator assume that the quantity has that particular value this issue can be resolved derivative value assignments consequences Element Quantity Entity ModelFragments derivativeAssignment Element Quantity Entity amp differentDerivativeAssignments Quantity Entity ModelFragments Figure 4 20 The multiple conflicting consequential derivative value assignments rule 34 40 Project No 004074 NATURNET REDIME D4 4 EF Quantity EF Quantity1 EF Quantity1 Entity Entity Entity Entity Entity Entity Qone aone Q ne Zimh Zimh Zimh BHigh A BHigh BHigh 6 Medium 8 Medium a medium MD A Blow 3 y Blow i Blow BZero W Zero WZero y Figure 4 21 Three model fragments with different derivative value
15. 13 was converted and significantly improved to fit this new template The result is a hyperlinked webpage version of the Garp3 user manual that consists of 351 different pages Each of these pages describes the current window the tasks that can be performed and the available short cuts Furthermore each page links to pages describing the menu options additional features related tasks used icons and used definitions in the glossary In total there are 1330 different explanations of uses of the 818 icons in different contexts From each Garp3 window the modeller can use the contextual help system to open the documentation page that is relevant to the workspace this person is currently working in This type of support is particularly useful to modellers who are not yet familiar with the Garp3 software as it describes all the knowledge needed to interact with the application 2 1 Contextual help To create the contextual help system a help button was added to each of the windows in Garp3 Figure 2 1 shows such a help button in the upper right corner The help button was designed to look like a wise owl to have the conceptual association with the help system Pressing the owl opens the help page associated to the screen in a browser This is implemented by adding the unique id of the current window to the URL and having a script on the webpage redirect the browser to the correct webpage based on this id RstarEcologicalModellin
16. 4 10 Bouwer A Liem J Linnebank F and Bredeweg B 2007 Analysis of Frequently Asked Questions and Improvements to the Garp3 Workbench Naturnet Redime STREP project co funded by the European Commission within the Sixth Framework Programme 2002 2006 Project no 004074 Project Deliverable Milestone D4 2 3 11 J Liem B Bredeweg and A Bouwer 2005 QR Subportal First Release Naturnet Redime STREP project co funded by the European Commission within 39 40 Project No 004074 NATURNET REDIME D4 4 the Sixth Framework Programme 2002 2006 Project no 004074 Project Deliverable Report D3 3 12 Bouwer A Liem J and Bredeweg B 2005 User Manual for Single User Version of QR Workbench Naturnet Redime STREP project co funded by the European Commission within the Sixth Framework Programme 2002 2006 Project no 004074 Project Deliverable Report D4 2 1 13 E Bakker A Bouwer J Liem and B Bredeweg 2006 User Manual for Collaborative QR model building and simulation workbench Naturnet Redime STREP project co funded by the European Commission within the Sixth Framework Programme 2002 2006 Project no 004074 Project Deliverable Report D4 2 2 14 Bredeweg B Bouwer A and Liem J 2006 Single user QR model building and simulation workbench Naturnet Redime STREP project co funded by the European Commission within the Sixth Framework Programme 2002 2006 Project no 004074 Project Deliverable
17. 4 4 3 2 2 Loading the scenario inequality reasoning When option 2 is switched on the trace will show the addition of all scenario ingredients The result for the Stove model is shown in figure 3 8 One of the useful points here is that when a quantity is first added to the state description its alternative internal name is given This name is used in all subsequent tracer statements Simulate Engine trace peoples va bli BAA mil SEE blocks indicate each ingredient Internal quantity beginning and type has a names are given ending section Figure 3 8 Trace Scenario application As shown in Figure 3 9 the definition of quantities can be brought to the surface in this context if option nr 5 is also turned on Simulate Engine trace ao X erature25 il temperature25 add Pea Ras 4 4 Fe he Rea et as BD i Be cs i Quantity and The quantity space The ordering The quantity quantity space constrains the quantity of points is Space is zero supply inequalities value and derivative defined plus max Figure 3 9 Trace Quantity definitions in the scenario 13 40 Project No 004074 NATURNET REDIME D4 4 And as shown in Figure 3 10 the definition of quantity values is also brought to the surface in this case Simulate Engine trace aA UR E a del SS The quantity Multiple inequalities value supplies assign the qu
18. _Quantity3 _Entity3 Quantity2 Entity2 _MF2 inverse inequalityGreaterThanBottom Quantity2 Entity2 MF increase _ in top _zero Element Quantityl Entityl MF Quantity2 Entity2 MF Quantity2 netPosInfluence Element Quantityl Entityl Quantity2 Entity2 MF zeroTopValue Quantity2 Entity2 inverse netNegInfluenceOn _Quantity3 _Entity3 Quantity2 Entity2 _MF2 inverse inequalitySmallerThanTop Quantity2 Entity2 MF Figure 4 14 The decrease increase in a bottom top zero point value rules Population Any population S Death SY Number of Zp 3 Pus A i 6 WZcro gt arge g Medium Asma B Zero Figure 4 15 A model fragment in which Number of potentially decreases in zero v 4 3 6 Same quantity with different quantity spaces The same quantity different quantity space rule searches for quantities that have different quantity spaces in different model fragments This issue is detected because quantities with the same name but with different quantity spaces cannot be unified It is probably not what the modeller had in mind Figure 4 16 shows the rule and Figure 4 17 shows two model fragments for which the issue occurs same quantity different_qs Element Quantity Entity MF1 MF2 quantity Element Quantity Entity nonEqualQS Quantity Entity Quantity Entity MF1 MF2 Figure 4 16 The same quantity different quantity spaces ru
19. aff science uva nl jliem QRM help faq ambiguiy v gt G a Disable Cookies J CSS E Forms Images Information Miscellaneous Outline Resize J Tools 2 View Sourc isu Stumble a I like it e Sendto Channel E 2 O fa Aloy of Favorites Da Friends Tools 8 The simulation contains too many states What can do about this There may be several reasons for this o When quantities can change independently from each other this may lead to exponential growth of the state transition graph because of all the possible combinations of quantities that can increase remain steady or decrease either synchronously or asynchronously The number of successor states is 24q 1 where q is the number of unrelated i e non corresponding quantities changing in an interval this calculation does not take quantities changing in different directions into account Software Check if there are bogus states and try to fix the model in such a way that it prevents them from Causal Deps being generated For example quantities may be fluctuating when they shouldn t Check if there are quantities which should behave synchronously and add correspondences or other means to enforce that behaviour Model Fragments o When there are model fragments that have too many interpretations or model fragments that 0 erroneously exclude each other this may also cause too many states to be generated Check if model fragments fire
20. ainty without recreating the simulation engine Using the simulation engine to diagnose and repair issues may be pursued in future work The main reason for choosing the approach presented here is that the rule elements in a diagnostic rule can be distributed over multiple model ingredients that do not necessarily fire all Therefore each rule element in a diagnostic rule returns a chance between 0 and 100 The algorithm calculates the chance of the construction in the model really causing problems during simulation by adding the chances returned by each of the rule elements and dividing this by the number of rule elements The syntax of the diagnostic rules is based on the Prolog syntax The syntax differs since the semantics of the rules differs from Prolog In contrast with Prolog there is no backtracking after a rule element fails or applies The reason for this is that each diagnostic rule is based on a single key ingredient Given this key ingredient the algorithm determines whether the rest of the rule is true The diagnostic rule syntax consists of seemingly regular Prolog calls However the character for and is replaced by amp the character for or is replaced by and the character combination for if is replaced by This syntax distinguishes normal Prolog calls from the diagnostic rules Figure 4 4 shows a diagnostic rule using this code Before the is the name of the rule Element is
21. antity inequalities value to an interval A single inequality assigns the quantity value to a point Figure 3 10 Trace Value definitions in the scenario Another level of detail is added if we turn on option nr 6 as well Now every newly derived fact is shown preceded by its parent relations This results in an elaborate trace such as the one pictured in Figure 3 11 In this example it is derived that the temperature is above absolute zero absnil because it is above melting point freeze_melt which is above absolute zero i Simulate Engine trace m al temperature26 lt condense boil temperature26 Bajos to Reale Bet Oe AS 0 1 The parent The new fact relations are is displayed at displayed first the bottom Figure 3 11 Trace Inequality reasoning 3 2 3 Model fragment selection inequality reasoning The model fragment selection process is traced using primarily options nr 2 3 4 and 5 As in the case of scenario content addition the procedure is outlined using a start and end block statement A typical trace of the stove model is given in Figure 3 12 Here we see candidate selection based on simple ingredient conditions entities configurations 14 40 Project No 004074 NATURNET REDIME D4 4 isa type and parent model fragments Selected candidates then get checked for consistent value and inequality conditions after which the candidates are accepted or rej
22. as expected and adjust them if necessary Values o When there are multiple initial states which all lead to many successor states it can help to simplify or constrain the scenario so that only one initial state is generated o When far too many states are generated it can help the investigation process to temporarily add Save and Print some constraints to the model to limit the number of states in a clear cut way For example add correspondences to make a set of quantities behave synchronously as described above or add an a inequality statement that only allows certain states but not others Inequalities Scenarios Simulation Done Figure 4 3 FAQ entry describing how to resolve the ambiguity issue 4 2 Diagnostic Rule Format Besides the issues of formalizing domain knowledge modellers encounter simulation issues caused by specific constructions of model ingredients For example certain constructions can cause model fragments not to fire or irresolvable causal relations may exist or ambiguity many generated states and even inconsistencies non generated states We call the rule descriptions that detect the above constructions diagnostic rules To automatically detect issues in models the diagnostic rule format was developed Each diagnostic rule is formalised as a set of rule elements It is not possible to determine whether an issue is really caused by the construction described in the diagnostic rule with absolute cert
23. cho Fundo case model shows that the trouble shooter can be useful to identify issues in complex models 38 40 Project No 004074 NATURNET REDIME D4 4 6 References 1 Eugenia Cioaca Bert Bredeweg Paulo Salles Tim Nuttle A Garp3 Model of Environmental Sustainability in the Danube Delta Biosphere Reserve based on Qualitative Reasoning Concept 21st International Workshop on Qualitative Reasoning QR 07 Chris Price eds pages 7 16 Aberystwyth Wales U K 26 28 June 2007 Elena Nakova Bert Bredeweg Paulo Salles Tim Nuttle A Garp3 model of environmental sustainability in the River Mesta Bulgaria 21st International Workshop on Qualitative Reasoning QR 07 Chris Price eds pages 87 95 Aberystwyth Wales U K 26 28 June 2007 Richard Noble Tim Nuttle Paulo Salles Integrative qualitative modelling of ecological and socio economic aspects of river rehabilitation in England 21st International Workshop on Qualitative Reasoning QR 07 Chris Price eds pages 96 101 Aberystwyth Wales U K 26 28 June 2007 Paulo Salles Bert Bredeweg Ana Luiza Rios Caldas Tim Nuttle Modelling sustainability in the Riacho Fundo water basin Bras lia Brazil 21st International Workshop on Qualitative Reasoning QR 07 Chris Price eds pages 147 160 Aberystwyth Wales U K 26 28 June 2007 Andreas Zitek Susanne Muhar Sabine Preis Stefan Schmutz The riverine landscape Kamp Austria an i
24. d allows them to determine why certain behaviour in the simulation state graph occurs The third addition to Garp3 is a proof of concept of a trouble shooter that is meant to help modellers to troubleshoot models without needing to use the tracer which requires some knowledge about how the reasoning works This trouble shooter is the first step towards an automated debugging facility on Garp3 The trouble shooter is integrated with the model building environment of Garp3 This trouble shooter detects possible faults in models based on a set of diagnostic rules It then determines the probability that this fault actually occurs during simulation Selecting one of the possible issues explains what the issue is and gives feedback on how it can be resolved The goal of the trouble shooter is to detect the most frequently occurring issues modellers face without having to use the tracer and to then suggest changes to the model that would resolve the issue http www skype com http flashmeeting open ac uk 3 http hcs science uva nl QRM help support 4 1 40 Project No 004074 NATURNET REDIME D4 4 2 Contextual Online Help The contextual online help system supports modellers by directing them to the documentation that is relevant to their current task For the online help system a web page template was developed for documentation pages that describe each of the windows in the Garp3 software The text from user manual documents 12
25. d or rejected ing for child MFs candidate children of still active MFs ing for new ME a A always indicates an inconsistency Ya w a DiS SN 0 ll Figure 3 6 Trace Simulation with no states From this trace we learn that 2 model fragments were accepted in the state description substance and liquid_phase Also 2 model fragments were already rejected because of invalid conditions gas_phase and solid_phase Most importantly we learn that the example_model_ fragment is the culprit To see which statements in this model fragment cause the contradiction the simulation is run once again with options numbers 1 2 and 5 switched on The crucial part of the resulting trace can be seen in Figure 3 7 Note that the colour coding and indentation ensure that the trace can be easily interpreted Simulate Engine trace Joe BE wle RaRa wa d 0 DAD 0 7 i top level other the inconsistent statements statements are line out left indented relation is shown Figure 3 7 Trace Model fragment with inconsistent consequences The inconsistent relation is marked with a and it states that the amount of liquid is zero However in this example model the scenario not shown states that the amount of liquid is at the point Max Since the facts are contradictory it is now clear why the simulation stops 12 40 Project No 004074 NATURNET REDIME D
26. d to be model flaws It may well be that case that the modeller created certain features on purpose and it may also be the case that the trouble shooter found non existing problems After all the trouble shooter implements a heuristic approach to support modellers in finding problems To ultimately discover whether a fault exists the best method is to run the simulation engine and inspect the simulation results aar Mm 37 1 40 Project No 004074 NATURNET REDIME D4 4 5 Conclusions The intelligent help system in Garp3 consists of three aspects that make the software more intelligent and easier to use The online help system directs modellers to the correct documentation based on their current task For this system a new web based user manual has been developed The tracer shows the inferences made by Garp3 when simulating qualitative models The tracer provides modellers the means to accurately pinpoint why models show certain behaviour The trouble shooter helps modellers to debug models without needing to use the tracer which requires some knowledge about how the reasoning works This trouble shooter is the first step towards an automated debugging facility for Garp3 The trouble shooter detects possible faults in models based on a set of rules It then determines the probability that this fault actually occurs during simulation The trouble shooter also gives feedback on how the issue can be resolved An experiment with the Ria
27. d to refine the diagnostic rules Figure 4 2 shows the results for running the trouble shooter on the Riacho Fundo model version March 16 2007 Thirty issues were found Figure 4 2 highlights one of those namely the rule referring to 87 5 Non corresponding proportional quantities potentially cause ambiguity R08 manure influences fertility This issue was found because of the following situation In the model fragment RO8c Figure 4 26 Rural rf Fertility is set to be proportional to Cattle Manure and in model fragment R17 Figure 4 27 the Crop Resource inflow is set to be proportional to Rural rf Fertility This potentially allows for a whole set of states referring to all the possible combinations of all values for Manure Fertility and the Resource inflow the latter only if Amount of Water remains unset Eros rural land fertility Occupied by Supports Vegetation Rural soil Rur Cattle Vegetation Rural soil ura rf D Cattle Covers Eroded land SHnutrient o ae o Zimhm Zimh 6 E Max BHigh BHigh Medium 8 Medium Blow A Blow BZero v E Zero Figure 4 26 Model fragment RO8c manure influences fertility 36 40 Project No 004074 NATURNET REDIME D4 4 Prosa ans fertility corresp nutrient g Asn fert corr nutrient Vegetation Vegetation Rural rf Rural rf eii Occupied by E eroded land Qutrent Q Fertility Rural soil Covers Zimhm amp 6 Max Crop gt c Ps BHig
28. e Calculus Conditional model fragment Configuration Correspondence Entity Identity Inequality Influence Proportionality Quantity Value Change view Collapse relations Expand all Expand relations Full redraw Hide Show relevant Show subfragments Collapse Expand Save Diagram to EPS file Model to disk Figure 2 2 Help page for the Model Fragment Editor part 1 2 2 7 Icons The Icons section is a single link to a page showing all the icons used in the editor and describing their meaning Figure 2 3 shows the Icons section on the model fragment editor webpage 2 2 8 Related tasks The Related tasks section enumerates all the other tasks that can be performed on the model ingredient that the current screen manipulates For example the model fragment being edited by a model fragment editor can also be added deleted copied or be made inactive Additionally the Related tasks section lists similar tasks for different model ingredients For example instead of editing a model fragment in the model fragment editor a scenario could be edited As in the other sections each item links to a page describing the task in more depth Figure 2 3 shows the Related tasks section on the model fragment editor webpage 2 2 9 Shortcuts The Shortcuts section lists all the key combinations that can be used to initiate certain tasks For example Ct
29. ected In case of consistent but unknown conditions the decision is postponed Note that each proper step in the process is outlined using 2 new lines hereby ensuring optimal readability Simulate Engine trace Eaim dense_boil temperature26 il temperature26 gt temperature26 gas2l zero boil temperature26 ture26 rature26 unt_of_liqui MF liquid pl SSI 0 17 dl Bal fale Candidates are A candidate is An inconsistent A candidate is sought and checked and relation is checked and found accepted derived rejected Figure 3 12 Trace Model fragment selection In our example trace option nr 6 is also on displaying an interesting inequality reasoning detail As can be seen a contradiction is derived The conditional inequality is e Temperature gt Boilpoint condense_boil However it is known that e Boilpoint gt Temperature And this clearly is inconsistent The output of the tracer in this case is 15 40 Project No 004074 NATURNET REDIME D4 4 e Zero gt Zero which is indeed an inconsistent relation The reason for this output is that the applied transitivity reasoning 17 18 is a two step process First the relations are combined into e Temperature Boilpoint gt Temperature Boilpoint Then this result is simplified by subtracting equal parts of either side resulting in e Zero gt Zero Secondly we see that the tracer actually outputs e Zero AmountOfGas g
30. eeeeeeeeeceeeeeeeseneeteaeeeeineeteneeseaees 33 4 3 7 Invalid loop of proportionalt S ar e a a a a a a eE a EEE 34 4 3 8 Consequential derivative value assignment causes CONFLICT eee eeeeeeeeeeeeeeeeeeeteeeeeeaneees 34 4 3 9 Conflicting derivative value assignment and Causal relation c eeeeeeeeeeeeeeneeeneeeneeeaeeees 35 4 3 10 Multiple consequential value assignments on quantity cccceeeceeeeeeeeeeeeeeeeeseeeeeseeeesaees 35 4 4 Troubleshooting the Riacho Fundo Model ooococinccconnccononcnnonncnonncnancconrn conc co nnn non n cn nn nn rara nr nn nncnnnnns 36 5 CONCIUSION Si ie 38 6 RETEKEN COS casaca ca lada rta 39 3 1 40 Project No 004074 NATURNET REDIME D4 4 1 Introduction Qualitative modelling is a difficult task As a result modellers require a significant amount of support during the development of qualitative models During the NaturNet Redime project partners have been creating Qualitative Reasoning QR models about 5 Case studies 1 2 3 4 5 A large effort during the NaturNet Redime project was supporting the partners working on these case study models and modellers outside of the project Modellers have been supported in several ways Firstly several QR training workshops have been organised 6 7 8 9 Secondly the partners working on the case studies had weekly meetings of about 1 5 hours via Skype and VNC or FlashMeeting 10 The partners sent their models and their issues before
31. es 27 08 2007 Jochem Liem 0 8 Wrote troubleshooting textual rule descriptions 28 08 2007 Jochem Liem 0 9 Finalized chapters 1 2 4 except two rules 5 30 08 2007 Jochem Liem and 6 1 0 Added the last two rules 31 08 2007 Jochem Liem 1 1 Added chapter 3 31 08 2007 Floris Linnebank 1 2 Added final two rules chapter4 final editing 04 09 2007 Jochem Liem 1 3 Final editing 05 09 2007 Floris Linnebank 1 4 Final editing 05 09 2007 Bert Bredeweg http Awww garp3 org 2 40 Project No 004074 NATURNET REDIME D4 4 Contents De INFO OU CON AAA A A A PO 4 2 Contextual Online Help a dd 5 2 Contextual helpt a tadas 5 2 27 The OnlinS help pages a Pt 6 22 1 Task descripta id dia 6 UA EOS A 6 2 2 3 Tasks In DIS WorkSpPace ca ir dida 6 LLAMEN OPtlOMS dais 6 2 2 5 Additional e atUreS caia did 6 2 2 6 Definitions of involved ingredient ceceeccseceeeeeeeeeeeeeeeeeeeeeeeseeseesaeesaeeseesieetieetieesieetieeseee 6 A A ON 7 2 2 8 A A 7 A ON 7 22 10 EXAMple A AT E A E E EE Dai 7 Se Nhe Simulation Trace ar ea rE ra id 9 o Ra AE 10 akol akel alo E ik 1O A A E a E a TE 9 dida eamp e tcs ito anidan aa 11 3 21 Erroneous model no States eiii agence thee et Lice ehs ee eee 11 3 2 2 Loading the scenario inequality reasoning 0 0 ee eee cess eeeneeeeneeeeeeteaeeeeeeessaeeteaeeseaeeseneeenaes 13 3 2 3 Model fragment selection inequality reasoning
32. es or results An example trace is given in Figure 3 17 Y Simulate Engine trace ll oy Ban arta va Dale sa 7 s a All known Equal quantities are inequalities shown as a string of are listed equalities Figure 3 17 Trace Derivable relations 3 2 6 Exogenous behaviours Exogenous behaviour patterns have specific rules that govern their derivatives These rules assign values to the derivatives of exogenous quantities after the model fragment search is done but before entering the influence resolution procedure This process can be inspected using option nr 5 An example trace taken from the Ants Garden model is given in Figure 3 18 Note that this process can be a source of branching in the state search 18 40 Project No 004074 NATURNET REDIME D4 4 Simulate Engine trace lol Exogenous derivative immigrationl of type Sinusoidal State direct from scenario At gt Can be set to 3 values from here jon ch going into branch Taking option 1 of 3 for exogenous derivative immigrationl of type Sinusoidal State direct from scenario At extreme lue of quantityspace add d immigrationl lt d zero wa ir eR ta oleo ps Ol y wl Blocks are used to An exogenous An exogenous Exogenous derivatives are indicate beginning and quantity has its quantity is an important branching end of the procedure derivative assigned analyzed point in the state search
33. g h2o0 temperatu into branch Galea Pa ral eal ea oles SNS 2 dl A All possible reasoning The assumption level is A candidate is assumptions at a displayed The nr of accepted reconsidered and still branching point are listed assumptions in this branch assumable Figure 3 14 Trace Reasoning assumptions detailed view In case a candidate model fragment is inconsistent with the current assumption s it is rejected as is shown in Figure 3 15 Simulate Engine trace SAX veol ANSA e Figure 3 15 Trace Rejected possible reasoning assumption If competing reasoning assumptions are present multiple branches are generated in the statesearch After completing one branch the reasoning engine jumps to the next branch by backtracking The resulting trace statements are shown in Figure 3 16 17 1 40 Project No 004074 NATURNET REDIME D4 4 Simulate Engine trace temperature33 1 84 BLA ml a es ek eg Figure 3 16 Trace Backtracking on reasoning assumptions 3 2 5 Derived relations After the search for model fragments and after the influence resolution procedure the tracer shows all derivable relations if option nr 7 is chosen This listing allows the user to gain an overview of the inequality reasoning results so far This can be very important information when debugging a model that has unexpected inconsistenci
34. gFeb2005 Assume resources saturate Static fragment Model Fragment Editor Build 0 x File Edit Conditions Consequences View 12 g kid Phrroduction Plant e8 s Plant SS e5 Lives in ie P Population yy Y Environment Resource consumption Population pr Environment X S immigration Bdigh a ab ai Qpresource inflow Resource available aMed g L i P Zim mero Hiiomass Mus y mi an G DF We 4 Zimt TS GS igt Le r i ri i 1 Proa per capita BHigh l mort per capita Y ga H ih 2 od P q IN 4p Limited resourse build up F Ad ih Y 30 o Es med a uu Figure 2 1 A model fragment editor showing the contextual help button the owl in the upper right corner of the screen 5 40 Project No 004074 NATURNET REDIME D4 4 2 2 The online help pages The online help pages replace the user manual documents to better support modellers in their modelling activities To this end a new template for the help pages was developed This template consists of 9 different types of content The textual descriptions in the user manual were reused when creating the help pages but significant additions and improvements were made to make the documentation more helpful An example of one of the 351 help pages is shown in Figures 2 2 part 1 and 2 3 part 2 2 2 1 Task description The Task description describes the task the current screen is created to support F
35. h ro n Ae R13streamsediment P ll eee OW Suppers A 8Zero Stream O Stream Supplies 7 S Resource inflow of water Ai High e E Medium Zimin Zimhm Pe at oor LE iO a mi 1g Hi x i ne Na Eg Y azero an ins Ol Ra T A Figure 4 27 Model fragment R17 resource inflow for crop production Another issue that was found is 75 Potentially invalid decrease in bottom zero S02 aggregation process not shown in Figure 4 2 The model fragment involved is shown in Figure 4 28 The explanation is as follows The influence from Particle aggregation rate potentially causes Level of aggregation to decrease in zero when Particle aggregation rate has the value Min This is potentially an invalid state of behaviour that should be prevented from occurring That is when Level of aggregation reaches zero the influence form Particle aggregation rate should cease to exist aar dm bso semi urban soil aggregation Semi urban soil Semi urban soil S Particle aggregation rate SY Level of aggregation Mzp Zsmim 6 BPius 14 Max E a Emin Medium Asma y 8Zero Figure 4 28 Model fragment S02 aggregation process This experiment with the well developed Riacho Fundo model shows that the trouble shooter is a useful addition to Garp3 to identify issues during modelling It can help modellers to critically think and rethink modelling choices made However not all issues found by the trouble shooter nee
36. hand and the UvA analysed their progress addressed their issues and identified other issues in the models Thirdly modellers sent questions via the QRM mailinglist 10 11 to be answered by the UvA team or other modellers Finally the mailing list and the new bug report system are used to report bugs to be fixed by the Garp3 team The intelligent help system described in this document aims to make the Garp3 software more intelligent and easier to use The system should reduce the dependency of modellers on QR experts The intelligent help system consists of three additions to the Garp3 qualitative modelling and simulation workbench The first improvement is an online help system aimed at less experienced modellers This system consists of two parts The first is an online user manual that describes all the workspaces the tasks that can be performed using specific spaces and examples of how they are typically used The second part is contextual awareness within Garp3 Each workspace has a help button that brings the modeller to the associated page in the online help system This allows modellers to get access to the correct documentation without having to search in a document The second addition to Garp3 is meant to support more experienced users during model debugging It is a tracer that shows the inferences made by Garp3 when simulating qualitative models The tracer allows modellers to select the type of inferences they are interested in an
37. he program an online help system was created From each editor in the application a help page on the Qualitative Reasoning and Modelling QRM portal can be opened that explains the functionality of that specific editor Secondly to support modellers during model debugging a tracer was added that shows the inferences made by Garp3 when simulating qualitative models The tracer allows modellers to select the type of inferences they are interested in and allows them to determine why certain behaviour in the simulation state graph occurs Finally a prototype trouble shooter was added to the model building environment of Garp3 as a proof of concept This trouble shooter detects possible faults in models determines the probability that this fault actually occurs and gives feedback on how this issue can be resolved The goal of trouble shooter is to solve the most frequently occurring issues modellers face without having to use the tracer Document history Version Status Date Author 0 1 Structure Abstract and Introduction 19 07 2007 Jochem Liem 0 2 Wrote initial draft of troubleshooting section 31 07 2007 Jochem Liem 0 3 Wrote Contextual Online Help section 21 08 2007 Jochem Liem 0 4 Wrote draft of diagnosis until 4 2 onward 21 08 2007 Jochem Liem 0 5 Feedback on vs3 21 08 2007 Bert Bredeweg 0 6 Processed feedback 22 08 2007 Jochem Liem 0 7 Wrote troubleshooting generaltrulestimag
38. he simulation engine while running a simulation The tracer present in the latest version of Garp3 is meant to support users during model development and debugging lt allows them to determine why certain behaviour in the simulation occurs Also it provides general insight in the reasoning process thereby educating the user The tracer can show all inferences made by the Garp3 simulation engine but users may choose to see only the specific types of inferences they are interested in Figure 3 1 Trace button In previous versions of Garp3 a simulation tracer was present 14 15 this tracer however showed only part of the reasoning process and used internal engine level descriptions of concepts It turned out to be of use only for advanced users and therefore we have done a complete overhaul of the simulation tracer 3 1 Buttons and trace levels The user can choose from 11 inference categories that can be traced The corresponding buttons can be seen in the tracer window as shown in Figure 3 2 E Simulate Engine trace Quay Er p Yi Gal Ra B W oe SIS 0 7 sl Figure 3 2 The tracer Window These inference categories have a loosely hierarchical structure Running a model with just the general button on will give an overview of the reasoning process This mode is particularly useful for a first look at a model and will show obvious problems such as inconsistent model fragments very clearly
39. in Menu options can be seen as a different representation of the Tasks in this workspace section Figure 2 3 shows the Menu options section on the model fragment editor webpage 2 2 5 Additional features The Additional features section describes characteristics of the screen that are not tasks Examples are colour coding tool tips action buttons conditions and consequences and context sensitivity Each item links to the glossary entry that explains the concept in more depth Figure 2 3 shows the Additional features section on the model fragment editor webpage 2 2 6 Definitions of involved ingredients The Definitions of involved ingredients section lists the names of each of the model ingredients mentioned in the screen Each item links to the glossary entry that explains the concept in more depth Examples of model ingredients are Agent Assumption Attribute and Calculus Figure 2 3 shows the Definitions of involved ingredients section on the model fragment editor webpage 6 40 Project No 004074 NATURNET REDIME D4 4 Workspace Model Fragment editor Task description In this screen you can edit the contents of Model fragments including Entities Attributes Configurations Quantities and Quantity Spaces Values Dependencies Agents Assumptions Inequalities Values Calculations Influences Proportionalities conditional Model fragments Correspondences and Identities Task con
40. in the top field shows a detailed description of the issue in the bottom field lt mentions the quantities and dependencies that are involved in the issue and in which model fragment they can be found Finally pressing the Show relevant model fragments button opens all the relevant model fragments that contribute to the issue f http hcs science uva nl QRM help bugs 26 40 Project No 004074 NATURNET REDIME RF 2007 03 16 final Model fragment definitions editor Default view Build File Edit View EO 2a inflow eq consumption at ed E02 economic production E02b inflow gr consumption Process fragment EO2c inflow OMTSUTTIPIO et RO7aa regeneration eq degrad jul jul 07a vegetation growth process RO7ab regeneration gr degrad RO7ac regeneration sm degrad De eepe a 502 aggregation process ul A ps infiltration active 04 infiltration process a S04b infiltration inactive roces BRE gt ul A goose erosion active S06 semi urban erosion process A S06b erosion inactive U02a runoff eq drained jul amp u02 controlled drainage process U02b runoff gr drained ja UO2c runoff sm drained Figure 4 1 The model fragment definitions editor has a new yellow owl icon that opens the troubleshooting viewer RF 2007 03 16 final Troubleshooting viewer Build IEA Troubleshooting Issues 100 0 Potential invalid increase in top point value
41. ison process and the state listing can be seen in Figure 3 28 24 40 Project No 004074 NATURNET REDIME D4 4 El eS z 4 Simulate Engine trace existing state found Transition to new population pulation_c Te 5_states o ype garp_interna ulation Quantities birth p birthil and_death population ulation P sake lt 3 oa al AILSA sl One existing state is found A listing of all The new and compared but its state ingredients state number values are different is given is given Figure 3 28 Trace State comparison and state listing 3 3 Conclusions and future research The simulation tracer provides a unique and very useful information source for the modeller Some ideas for future work are the integration of the tracer output in a flexible tree structure which can be navigated using an intelligent search component 25 1 40 Project No 004074 NATURNET REDIME D4 4 4 Troubleshooting The UvA accumulated and documented all their modelling support efforts as mentioned in the introduction There are minutes of the Skype meetings recordings of the Flashmeetings an archive of all the questions on the QRM mailing list and the bug report system is publicly accessible Based on an analysis of this material a Frequently Asked Questions FAQ list has been created 10 The ten most frequently occurring troubleshooting issues were selected from this list These iss
42. le 33 40 Project No 004074 NATURNET REDIME D4 4 Entity ae Entity DA Pron Sone Qtwo Zimh 6 Zim 6 Bion 4 Medium Zlmh 6 Zimh 6 Medium Blow A BHigh BHigh Blow y Zero Medium A S Medium WZero y fow y Blow y W Zero BW Zero Figure 4 17 Two model fragments in which Qtwo occurs with different quantity spaces 4 3 7 Invalid loop of proportionalities The proportionality loop rule detects chains of proportionalities that end at the beginning This construction is irresolvable for the simulation engine as each quantity requires the derivative of the previous quantity to be determined in order to determine its own derivative Figure 4 18 shows the proportionality rule code and Figure 4 19 shows an adapted version of the tree and shade model that has a proportionality loop in it This issue can be resolved by removing one of the proportionalities proportionality loop Element MF1 QuantityEntityModelFragmentList proportionality Element Quantityl _Entityl Quantity2 _Entity2 MF1 full _ proportionality loop Element QuantityEntityModelFragmentList Figure 4 18 The invalid loop of proportionalities rule Tree Tree 1 Tree HOT Dogo Yom Sml 6 Sml 6 Sml 6 Sml 6 Sml lo Barge A BLarge E carne arge Large 8 Medium g par aoe S Medium Medium Asma y ma v Bsmal y Asma y Bsmal y Figure 4 19 An adapted version of the tree and shade model causing an invalid loop of
43. n in Figure 3 23 Simulate Engine trace Elo Applying epsilon ordering 1 Immediate terminations 2 NON Immediate terminations to_point_below amount_of_liquid5 to_point_above amount_of_gas5 a inished epsilon ordering No Immediate terminations found ompleting ordering procedure for the set of NQN Immediate terminations 2 Pa bP ea ee eg ee ale SN wa Immediate terminations and In this example there Ordering continues non immediate terminations are no immediate with the remaining are separated terminations terminations Figure 3 23 Trace Epsilon ordering 21 40 Project No 004074 NATURNET REDIME D4 4 Correspondences are an important source of information during ordering The trace of this sub process is shown in Figure 3 24 Simulate Engine trace ale Checking correspondenc a termination to_point_below amount_of_liquid5 should co occur with to_point_above amount_of_gas5 reason active correspondence in current value Ebo EEC A constraint on Each subprocedure possible combinations starts and ends is found with a notice Figure 3 24 Trace Correspondence ordering i Simulate Engine trace Ja Checking mathematical consistency of binary termination subs Determining mathematical consistency for terminations from_greater_to_equal immigrationl emigrationl to_point_below immigration1 current values immigrationl
44. n the maximum or decrease in the minimum point value of a quantity space but not zero see also next section Figure 4 12 shows the rules and Figure 4 13 shows a model fragment where Ecological integrity potentially increases in the top point value The decrease in bottom point value rule searches for a net negative influence between Q1 and Q2 The lowest value of Q2 should be a point There should not be a net positive influence on Q2 and there should not be a conditional inequality on Q2 that removes the negative influence before it reaches the lowest point value The increase in top point value rule searches for a net positive influence between Q1 and Q2 The top value of Q2 should be a point There should not be a net negative influence on Q2 and there should not be a conditional inequality on Q2 that removes the positive influence before it reaches the top point value The issue can be resolved by having a feedback proportionality on the influence causing it to decrease when Q2 reaches the top or bottom point value A directed value correspondence to Q1 can indicate that the influence has no effect once Q2 reaches zero decrease _ in bottom point value Element Quantityl Entityl MF Quantity2 Entity2 MF Quantity2 netNegInfluence Element Quantityl Entityl Quantity2 Entity2 MF bottomPointValue Quantity2 Entity2 inverse netPosInfluenceOn _Quantity3 _Entity3 Quantity2 Entity2 _MF2 inverse
45. ntegrative case study for qualitative modeling of sustainable development 21st International Workshop on Qualitative Reasoning QR 07 Chris Price eds pages 212 217 Aberystwyth Wales U K 26 28 June 2007 Bouwer A Bredeweg B Liem J and Salles P 2006 1st NNR user group workshop on using QR technology Naturnet Redime STREP project co funded by the European Commission within the Sixth Framework Programme 2002 2006 Project no 004074 Project Deliverable Milestone M7 2 1 Nuttle T Liem J Bouwer A Bredeweg B and Salles P 2006 2nd NNR user group workshop on Qualitative Reasoning and Modelling Naturnet Redime STREP project co funded by the European Commission within the Sixth Framework Programme 2002 2006 Project no 004074 Project Deliverable Milestone M7 2 2 Liem J Bouwer A Bredeweg B Salles P and Bakker B 2007 3rd NNR user group workshop on Qualitative Reasoning and Modelling Naturnet Redime STREP project co funded by the European Commission within the Sixth Framework Programme 2002 2006 Project no 004074 Project Deliverable Milestone M7 2 3 Bredeweg B Salles P Bertels D Rafalowicz J Bouwer A Liem J 2007 Fourth NNR user group workshop on Qualitative Reasoning and Modelling Naturnet Redime STREP project co funded by the European Commission within the Sixth Framework Programme 2002 2006 Project no 004074 Project Deliverable Milestone M7 2
46. odel fragment where Number of potentially decreases in zero The decrease in zero rule searches for a net negative influence between Q1 and Q2 The bottom value of Q2 should be zero There should not be a net positive influence on Q2 and there should not be a conditional inequality on Q2 that removes the negative influence before it reaches zero The increase in zero rule searches for a net positive influence between Q1 and Q2 The top value of Q2 should be zero There should not be a net negative influence on Q2 and there should not be a conditional inequality on Q2 that removes the positive influence before it reaches the zero The issue can be resolved by having a feedback proportionality on the influence causing the resulting influence go to zero when Q2 reaches zero A directed value correspondence to Q1 can indicate that the influence has no effect once Q2 reaches zero Notice that this rule relates to the rule discussed in the previous section They are handled separately because zero has a special status in the reasoning engine and because the Garp3 software has different system preferences to alleviate the constraints on increasing and decreasing while being at extreme magnitudes decrease _ in bottom_zero Element Quantityl Entityl MF Quantity2 Entity2 MF Quantity2 netNegInfluence Element Quantityl Entityl Quantity2 Entity2 MF zeroBottomValue Quantity2 Entity2 inverse netPosInfluenceOn
47. on a quantity rule searches for quantities that have multiple value assignments on them as a consequence If these model fragments fire at the same time an inconsistency would result since a quantity cannot have multiple values at the same time Figure 4 24 shows the rule and Figure 4 25 shows 4 model fragments with a value assignment on different magnitudes By having different assumptions for each value assignment or having the value assignments as conditions which makes the simulator assume that the quantity has that particular value this issue can be resolved 35 40 Project No 004074 NATURNET REDIME D4 4 multiple consequence value assignments_on quantity Element QuantityName EntityName ModelFragments valueAssignment Element QuantityName EntityName amp differentValueAssignments QuantityName EntityName ModelFragments Figure 4 24 The multiple consequential value assignments on quantity rule EF Quantity EF Quantity E Quantity FR Quantity Entity Entity Entity Entity Entity Hoone Hoone Q one Hoone Brigh z Bion Bion ID fiiigh 8 Medium A Medium A Z svedu A Medium A mse TE Y ies P fe Figure 4 25 Four different model fragments with different value assignments causing inconsistencies 4 4 Troubleshooting the Riacho Fundo Model During its development the trouble shooter was tested on the latest version of the case studies models in the NaturNet Redime project The results were use
48. ons The conflicting causal relations rule detects occurrences of influences and proportionalities on the same quantity This is construction is always ambiguous since the proportionality determines the target quantity based on the derivative of the source quantity while the influence determines the derivative of the target quantity based on the magnitude of the source quantity There is no way to determine which causal relation is dominant The rule is shown in Figure 4 10 while an example of this construction is shown in Figure 4 11 conflicting_causal_relations Element Quantityl Entityl MF1 Quantity2 Entity2 MF1 Quantity3 Entity3 MF2 Quantity2 Entity2 MF2 MF1 proportionality Element Quantityl Entityl Quantity2 Entity2 MF1 eInfluence Quantity3 Entity3 Quantity2 Entity2 MF2 Figure 4 10 The conflicting causal relations rule Entity Entity Entity Entity Qtwo Qthree E SS Qtwo Zimh 6 Bich zm 8 Zimh 6 BHigh A m Medium A 8 High A Eion Medium Blow g Medium 8 Medium y bow y azoo Valow v Blow y E Zero E Zero B Zero Figure 4 11 Two model fragments with conflicting causal relations 31 40 Project No 004074 NATURNET REDIME D4 4 4 3 4 Invalid decrease increase in bottom top point value The invalid decrease in bottom point value and the invalid increase in top point value rules detect instances where influences potentially cause quantities to increase i
49. or example the Model fragment editor is used to edit the contents of model fragments Figure 2 2 shows the task description on the model fragment editor webpage 2 2 2 Task context The Task context describes how a screen can be reached from other screens Additionally links to the help pages of these other screens are included For example the Model fragment editor can be accessed from the Build tasks in the Garp3 main menu Figure 2 2 shows the task context on the model fragment editor webpage 2 2 3 Tasks in this workspace The Tasks in this workspace section categorizes and describes the tasks that can be performed in the current screen Examples of categories are Add Edit Delete Save Change view etc In these categories are usually types of model ingredients such as Agent Entity Model Fragment etc The combination of the category and the model ingredient constitutes the complete task e g Add agent Edit Entity Delete model fragment Each of these tasks links to their own separate page describing that specific task In addition this section describes how these tasks can be performed via the user interface Figure 2 2 shows the Tasks in this workspace for the model fragment editor 2 2 4 Menu options The Menu options enumerate all the menu options that can be found in the drop down menu in a specific screen Each item links to a separate page with more information about that task The items
50. ows of the explanation boxes used For clarity an overview of the flow of control through the main procedures of the reasoning engine is given in Figure 3 5 This flow of control is largely followed in the discussion of the example tracer output Apply Select Determine Influence i Model Exogenous Scenario g Resolution Content Fragments Derivatives A t J State ll Description Transition ll Scenario Terminations Terminations Figure 3 5 Main procedures in the reasoning engine 3 2 1 Erroneous model no states A common situation arises when a model fragment causes a contradiction because of inconsistent consequences The resulting simulation has no states and a modeller therefore finds valuable information in the tracer as to what has caused the trouble In the trace shown in Figure 3 6 just the general category nr 1 was used to inspect an adapted version of the Stove model As can be seen this level of detail provides a very informative overview of the reasoning process The reasoning engine trace lists all events significant at the top level and clearly indicates what went wrong The contradiction causes the state search to stop and since this is an important event it is displayed within lines http hcs science uva nl QRM models 11 40 Project No 004074 NATURNET REDIME D4 4 A A TI Simulate Engine trace BAX Model Fragments are ion for scenario accepte
51. portionality Quantity Refinement Value Icons Icons related to this task Related tasks Other tasks for this ingredient Add model fragment Edit model fragment properties Delete model fragment Copy model fragment Set model fragment active Set model fragment inactive Similar tasks for other ingredients Edit scenario Shortcuts Del Delete Enter or Double click on an icon Properties M Add a Model Fragment as condition Space Collapse Ctrl Space Collapse Ctrl S Save model to disk Ctri R Show Relevant Expand H Hide Example Top gt Born Process fragment Model Fragment Editor a9 m JE File Edit Conditions Consequences View ca A Population Q Population Population S q Oo Zinhm aMax BHigh Bhiormal a v The example shows the Model fragment editor with an imported conditional model fragment Population to which the Quantity Born is added as a consequence Also between the Quantities Born and Size is a positive Influence and Size has an Inequality to its own Quantity space to indicate it has to be greater than Zero Figure 2 3 Help page for the Model Fragment Editor part 2 8 1 40 Project No 004074 NATURNET REDIME D4 4 3 The Simulation Tracer The trace window is opened from the main Garp3 screen using the trace button pictured in Figure 3 1 The window shows a trace of the reasoning performed by t
52. rl S is short for Save model and Enter shows the properties of a model ingredient Figure 2 3 shows the Shortcuts section on the model fragment editor webpage 2 2 10 Example The Example section shows a screenshot of the editor and describes the user generated content in that editor Figure 2 3 shows the Example section on the model fragment editor webpage 7 40 Project No 004074 NATURNET REDIME D4 4 Menu options File Save diagrarn to EPS file Save model to disk Edit model fragment properties Edit Edit entity Delete entity Conditions Add entity Add agent Add assumption Add quantity Add attribute Add configuration Add inequality Add calculus Add value Add conditional model fragment Add identity Consequences Add entity Add agent Add assumption Add quantity Add configuration Add inequality Add calculus Add value Add correspondence Add proportionality Add influence View Collapse Expand Collapse relations Expand relations Show relevant Full redraw Expand all Hide Show subfragments Additional features Color coding Conditions and consequences Context sensitivity Instance Naming ingredients Tooltips Action buttons Graphical icons Definitions involved ingredients Agent Assumption Attribute Calculus Conditional model fragment Configuration Correspondence Entity Identity Inequality Influence Pro
53. s cannot be determined by an operator relation plus or minus as this would remove the ambiguity The ambiguity rule is shown in Figure 4 6 A model fragment on which the rule fires is shown in Figure 4 7 Adding correspondences between the non corresponding quantities or determining their value using operator relations can resolve this issue ambiguity Element MF1 QuantityEntityList proportionality between A and B start point proportionality Element Quantityl Entityl Quantity2 Entity2 MF1 No correspondence between the start point and the next quantity inverse qsCorrespondence Quantityl Entityl Quantity2 Entity2 _MF2 The proportionality should not go to a rate variable optional sinverse eInfluence Quantity2 Entity2 _Quantity3 _Entity3 _MF3 Not another start point not proportionality and not correspondence rewrite not A and not B gt not A or B no proportionality and A without a correspondence between _X and Q1 inverse eProportionality QuantityX EntityX Quantityl Entityl _MF4 qsCorrespondence QuantityX EntityX Quantityl Entityl MF5 find all proportional non corresponding quantities with A as a beginning ambiguousChain Element QuantityEntityList Figure 4 6 The ambiguity rule Tree Tree amp size P shade Ps gt Darkness nl 6 Sml Sml 6 Blage 4 Mare Bas 4 8 Medium y BMedium g Medium Asma y Asma y Bsmai Y
54. t Zero AmountOfGas The reason for this equality showing up is that equal quantities share a pointer an internal reference to the quantity used in calculations This mechanism 10 17 is used for efficiency reasons and the tracer will print all quantities with a shared pointer as a continuous equality 3 2 4 Reasoning assumptions The mechanism for making Reasoning Assumptions 10 is active during the model fragment search In Figure 3 13 its trace statements are pictured at the overview level only option nr 1 for a stove model with unknown temperature 9 Simulate Engine trace Bex n MF iling h2 ing umption c mptions t tem atu oil temperat liquid_ A candidate is checked and the decision is postponed A reasoning Reasoning assumptions No more normal assumption is checked are an important branching candidates can be and accepted point in the state search added Figure 3 13 Trace Reasoning assumptions general view On a more detailed level option nr 3 displays more information on this mechanism in a bright green colour As can be seen in Figure 3 14 possible candidates are reconsidered and a listing is provided of all possible reasoning assumptions before committing to any one 16 40 Project No 004074 NATURNET REDIME D4 4 amp Simulate Engine trace Quay Need to make reasoning assumption currently 0 assumptions were made level 0 soning melt boilin
55. text The Model Fragment editor screen can be accessed from the Garp 3 main screen or from the Model Fragments definitions editor screen Press the button Edit latest Model fragment in the Garp3 main screen or the Edit selected Model fragment in the Model fragments definitions editor to open this workspace Build tasks Model Fragments definitions editor Tasks in this workspace Most tasks in this workspace can de executed using either the button bar the pull down menus or the shortcut keys To Add a new ingredient press the appropriate button in the button bar on the left in the workspace Notice that a selection or de selection of already created ingredients may be required To Delete select the ingredient by clicking on it once and then press the delete key on the keyboard To Edit double click on an ingredient or select the ingredient and press the enter key on the keyboard The Change view and Save tasks are accessible via the pull down menus at the top of the workspace Add Agent Assumption Attribute Calculus Conditional model fragment Configuration Correspondence Entity Identity Inequality Influence Refinement Proportionality Quantity Value Delete Agent Assumption Attribute Calculus Conditional model fragment Configuration Correspondence Entity Identity Inequality Influence Refinement Proportionality Quantity Value Edit Agent Assumption Attribut
56. the key element and the other arguments are returned by the rule After the are a set of rule elements that describe when the issue occurs Each rule element takes and returns a set of arguments A rule element that revolves around a specific model ingredient usually returns a model fragment in which that model ingredient occurs This argument can also be used to indicate that the model ingredient must occur in a specific model ingredient The final argument is a list of model fragments in which the ingredient may not occur 28 40 Project No 004074 NATURNET REDIME D4 4 As can be seen from Figure 4 4 the calculations of the chances are not represented in the diagnostic rules This is on purpose since the calculations would make writing the diagnostic rules more complex The new syntax for diagnostic rules allows us to rewrite the rules during compilation These rewritten rules deal with the chances and calculation of the chance of the total diagnostic rule by adding another argument the chance and equations that calculate the result of all the chances Figure 4 5 shows the rewritten form of the rules shown in Figure 4 4 Note the use of inverse as part of the language Inverse is the not in the probabilistic world If a rule element has 30 chance the inverse of the rule element has 70 chance ambiguity Element MF1 QuantityEntityList proportionality between A and B start point proportionality Element Quantityl Entit
57. ual diagnostic rule code and some model fragments containing the issue are shown Note that the rule elements in each diagnostic rule are actually function calls that hide more implementation details 4 3 1 Ambiguity due to non corresponding quantities The ambiguity rule detects the non corresponding quantities related by a proportionality The construction potentially generates many states one for each combination of values of the non corresponding quantities The total amount of states could become the cross product of the values of each quantity given only these model fragments If each of the quantities is in an interval and changing the number of successor states would be 24q 1 where q is the number of unrelated quantities since at least one quantity changes but each quantity can change both with and without the others 29 1 40 Project No 004074 NATURNET REDIME D4 4 The rule searches for a proportionality key element between Q1 and Q2 without a correspondence between Q1 and Q2 Furthermore there cannot be a proportionality without a correspondence from QX to Q1 i e there cannot be another start point for the chain of proportionalities This is logically written as not proportionality AND not correspondence and rewritten as not proportionality OR correspondence Given this construction the rule searches for a chain of non corresponding correspondences beginning at the key element The target of proportionalitie
58. uantity derivative triggers a termination that is displayed accompanied by an explanation of the reasons for the change Changing quantities may have additional epsilon or 2 order derivative continuity constraints which are also displayed in this context Note that other trace options will not reveal more information in this context because no inequality reasoning is actively used in this procedure An example trace of terminations firing is shown in Figure 3 22 20 40 Project No 004074 NATURNET REDIME D4 4 9 Simulate Engine trace y E Epsilon continuity constraint for d temperature3 reason d temperature3 was plus and transition is immediate constraint d temperature3 gt d zero Termination to interval above condense_bo temperature3 reason d temperature3 plus and gas_phase i erminations for state 4 to_interval_above temperature3 uuu oa Di Oe BIS 0 7 i ok Terminations are shown A listing of all Applied epsilon derivative with conditions terminations ends continuity constraints are consequences and causes the procedure trace displayed Figure 3 22 Trace Terminations 3 2 9 Ordering The ordering procedure consists of a number of sub procedures each of which may or may not pose additional constraints on valid or invalid combinations of terminations All of these procedures are traced using option nr 10 Epsilon ordering is the first applied concept and its trace is show
59. ues are now automatically detected by the trouble shooter which also advises the modeller on how the issues can be resolved by directing them to a specific FAQ entry 4 1 Troubleshooting viewer The new troubleshooting viewer can be opened from the model fragment definitions editor see Figure 4 1 By pressing the new yellow owl button the troubleshooting viewer is opened see Figure 4 2 Opening the troubleshooting viewer automatically runs the diagnostic rules In the top field of the troubleshooting viewer a list is presented of the issues found Each issue includes a percentage i e a chance a short description of the issue and the key model fragment in which the issue was found The percentage indicates the probability that the trouble shooter attributes to the issue it has found A higher percentage means that it is more likely that the issue really occurs The issues are ordered by percentage The short description of the issue gives a context independent summary of what the issue entails Finally the key model fragment indicates the model fragment that includes the key model ingredient that the trouble shooter found to be causing the issue Double clicking on the issue opens the FAQ page on the QRM Portal in a browser on precisely the entry that describes the encountered issue This FAQ entry also describes how the issue can be resolved In the bottom field a context dependent description is given about an issue Selecting an issue
60. yl Quantity2 Entity2 MF1 No correspondence between the start point and the next quantity inverse qsCorrespondence Quantityl Entityl Quantity2 Entity2 _MF2 The proportionality should not go to a rate variable optional sinverse eInfluence Quantity2 Entity2 _Quantity3 _Entity3 _MF3 Not another start point not proportionality and not correspondence rewrite not A and not B gt not A or B no proportionality and A without a correspondence between _X and Q1 inverse eProportionality QuantityX EntityX Quantityl Entityl _MF4 qsCorrespondence QuantityX EntityX Quantityl Entityl MF5 find all proportional non corresponding quantities with A as a beginning ambiguousChain Element QuantityEntityList Figure 4 4 A diagnostic rule describing an invalid increase in a top point value design ambiguity A B J P proportionality A C D E F B N gt inverse qsCorrespondence C D E F _ _ L gt inverse eProportionality G H C D _ _ 1 gt true qsCorrespondence G H C D _ 1 gt ambiguousChain A J K gt M is I K gt O is L M gt Q is N O P is Q 4 Figure 4 5 The rewritten Prolog form of the rule shown in Figure 4 4 4 3 Diagnostic rules The following sections describe each of the implemented diagnostic rules A description of the diagnostic rule is given and the act

Project no. 004074 Project acronym: NATURNET

Contents

Download Pdf Manuals

Related Search

Related Contents