Barcelona lecture-3 Universitat Autònoma de Barcelona Barcelona, October 17, 2006 Advanced M&S Methodologies: Multimodels and Multisimulations* Tuncer Ören, Professor Emeritus M&SNet - McLeod Modeling & Simulation Network of the SCS AVP for Ethics of the SCS University of Ottawa, Ottawa, Ontario, Canada *http://www.site.uottawa.ca/~oren/index-conf/confs_2000s.htm Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 1 -
The concepts in this lecture such as: multimodels, multistage models, multisimulation, & switchable understanding: can be useful to develop advanced conflict management training facilities; however, several of the concepts have more general applicability. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 2 -
Conflicts: Are common. Are even more important after the cold war era. (Ören, March 2000) There is a wealth of knowledge on conflict management: - conflict avoidance and - conflict resolution. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 3 -
A thesis: Appropriate formulation of social phenomena may improve: our perception, conception, and understanding of reality and the way solutions may be developed. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 4 -
Some facts: Training of a pilot is not completed without extensive experience on flight simulator(s). Military decision makers take advantage of war gaming. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 5 -
Preparing the decision makers: War gaming: - it has been in practice since pre-computer days - 1880s - nowadays, every possibility in advanced computerization is used to support it: - computer generated forces, - synthetic environments, - advanced graphics, HLA,... Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 6 -
A question & some implication: Would you like to fly in an airplane with a pilot who did not receive extensive flight experience on an appropriate simulator? Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 7 -
Preparing the decision makers: Simulation games to support: - peace - conflict avoidance - conflict resolution may be very useful. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 8 -
Some possible developments which might be desirable Advanced conflict management simulation environments can be developed, to provide extensive experience to decision makers, for conflict avoidance and conflict resolution as well as for peace keeping. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 9 -
A motivational questions Is it worth attempting to conceive - new modeling formalisms? - new simulation formalisms? Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 10 -
How the concepts are developed? A personal experience: In 1970, I developed the first system theory-based model specification language for continuous systems: Ören, T.I. (1971). GEST: General System Theory Implementor, A Combined Digital Simulation Language. Ph.D. dissertation, University of Arizona, Tucson, AZ. Ören, T.I. (1971). GEST: A Combined Digital Simulation Language for Large-Scale Systems. Proceedings of the Tokyo 1971 AICA (Association Internationale pour le Calcul Analogique) Symposium on Simulation of Complex Systems, Tokyo, Japan, September 3-7, pp. B- 1/1 - B-1/4. (& many other publications) Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 11 -
In the 1980s, I was involved in the development of two integration algorithms for piece-wise continuous i.e., discontinuous systems: Birta, L.G., Ören, T.I., Kettenis, D.L. (1985). A Robust Procedure for Discontinuity Handling in Continuous System Simulation, Transactions of the Society for Computer Simulation, 2:3, 189-205. Ören, T.I., Ma Jihu (1986). An Adaptive Order Discontinuity Algorithm for Simulation of Ordinary Differential Equations. In: Proceedings of JSST (Japan Society for Simulation Technology) conference on Recent Advances in Simulation of Complex Systems, Tokyo, Japan, July 15-17, 1986, pp. 283-288. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 12 -
While working on the development of the integration algorithms for discontinuous systems, I remarked that there are two types of discontinuities that can occur separately or together at a given instant of time, Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 13 -
Continuous change x time At every point, there is no jump of the value of the state variable & the derivative from left and right. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 14 -
Discontinuous change x c3 b a c2 c1 0 t a t b time At t a derivative from left and derivative from right are different; i.e., different models are used (model is updated). Hence there is a model discontinuity which necessitates a model update. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 15 -
Discontinuous change x c3 b a c2 c1 0 t a t b time At t b c1: (supposing that derivative from left and derivative from right are the same) there is no discontinuity. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 16 -
Discontinuous change x c3 b a c2 c1 0 t a t b Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 17 - time At t b c2: (supposing that derivative from left and derivative from right are the same) there is a discontinuity. At t b, same model is used but state variable needs to be re-initialized. There is an initialization discontinuity (a jump discontiniity).
Discontinuous change x c3 b a c2 c1 0 t a t b time At t b c3: derivative from left and derivative from right are different & values of the state variable are different. At t b two types of discontinuity exist, namely model discontinuity and initialization discontinuity. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 18 -
In 1987 and 1991, I generalized the model update concept to model switching, i.e., switching from a model to another: Ören, T.I. (1987). Model Update: A Model Specification Formalism with a Generalized View of Discontinuity. In: Proceedings of the Summer Computer Simulation Conference, Montreal, Quebec, Canada, 1987 July 27-30, pp. 689-694. Ören, T.I. (1991). Dynamic Templates and Semantic Rules for Simulation Advisors and Certifiers. In: Knowledge-Based Simulation: Methodology and Application, P.A. Fishwick and R.B. Modjeski (Eds). Springer-Verlag, Berlin, Heidelberg, New York, Tokyo, 53-76. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 19 -
In 2001, I generalized the model update and model switching concepts to multimodels, multistage models, & Multisimulation. Ören, T.I. (2001 Invited contribution). Towards a Modelling Formalism for Conflict Management. In: Discrete Event Modeling and Simulation: A Tapestry of Systems and AI-based Theories and Methodologies. H.S. Sarjoughian and F.E. Cellier (eds.), Springer- Verlag, New York, pp. 93-106. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 20 -
Switchable model & switchable simulation concepts lead to switchable understanding. Ören, T.I. (2000 Invited Opening Paper). Understanding: A Taxonomy and Performance Factors. In: D. Thiel (ed.) Proc. of FOODSIM 2000, June 26-27, 2000, Nantes, France. SCS, San Diego, CA, pp. 3-10. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 21 -
Multivision understanding as opposed to single vision understanding (which may be dogmatic understanding) may lead to switchable understanding: In switchable understanding any one of the following possibilities may be used alone or in combination: - Have more than one meta-model and switch them based on the context (in multisimulation, several meta-models can be used simultaneously.) - Use perceptions of different granularity. - Use different granularity of mappings between the metamodels and the perceptions for understanding. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 22 -
In the 2000s, I continued working with two talented colleagues and we published the following: Yilmaz, L. and Ören, T.I. (2004). Dynamic Model Updating in Simulation with Multimodels: A Taxonomy and a Generic Agent-Based Architecture, Proceedings of SCSC 2004 - Summer Computer Simulation Conference, July 25-29, 2004, San Jose, CA., pp. 3-8. Yilmaz, L. and T.I. Ören (2005). Discrete-Event Multimodels and their Agent-Supported Activation and Update. In Proceedings of the Agent- Directed Simulation Symposium of the Spring Simulation Multiconference (SMC 05), pp. 63-72, San Diego, CA, April 2005. Yilmaz, L., T.I. Ören, and N. Ghasem-Aghaee (2006-In Press). Simulation- Based Problem Solving Environments for Conflict Studies: Toward Exploratory Multisimulation with Dynamic Simulation Updating. Simulation, and Gaming Journal. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 23 -
Now, let s see: Multimodels and Multisimulation Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 24 -
Multimodels A multimodel can encapsulate different aspects of a model. At any time at least one of the component models is active. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 25 -
A multimodel: At least one component model is active at a time to represent the multimodel. M M1 M2 M3 Control of activation of component models Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 26 -
The concept of coupling of component simulation models was first introduced in the GEST language: Ören, T.I. (1971). GEST: A Combined Digital Simulation Language for Large-Scale Systems. Proceedings of the Tokyo 1971 AICA (Association Internationale pour le Calcul Analogique) Symposium on Simulation of Complex Systems, Tokyo, Japan, September 3-7, pp. B-1/1 - B-1/4. Hence, coupling of active component model(s) of a multimodel was easy to conceive. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 27 -
An example - A metamorphic model: (e.g., egg, larva, pupa, butterfly) There is a predefined sequence for the alternate models. M M1 M2 M3 Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 28 -
A type of multimodel - metamorphic model: (e.g., egg, larva, pupa, butterfly) There is a predefined sequence for the alternate models. Another representation to ease depiction of the coupling of component models M 1 M 2 M M n Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 29 -
A multiaspect model: (e.g., ice, water, vapor) More than one alternate model can exist at the same time with possible flows of entities (e.g., mass) between submodels M M1 M2 M3 Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 30 -
Another type of multimodel - multiaspect modelaa (e.g., ice, water, vapor) More than one alternate model can exist at the same time with possible flows of entities (e.g., mass) between submodels M An alternative representation M 1 M 2 M 3 Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 31 -
Evolutionary models: Evolution is irreversible change in an open system. An evolutionary model can be represented by a series of variant models M i. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 32 -
A taxonomy multimodels is offered at: Yilmaz, L. and Ören, T.I. (2004). Dynamic Model Updating in Simulation with Multimodels: A Taxonomy and a Generic Agent-Based Architecture, Proceedings of SCSC 2004 - Summer Computer Simulation Conference, July 25-29, 2004, San Jose, CA., pp. 3-8. Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 33 -
Based on Additional Criteria Type of multimodel (MM) Structure of Number of submodels active at a given time Only one 2or more Single aspect MM (Sequential MM) Multiaspect MM Variability Static Static-structure MM of structure Dynamic Number of Extensible Extensible MM Submodels (Dynamicstructure MM) (Variablestructure MM) submodels Alterations of submodels Depends on model s stage No Yes Multistage MM Non-mutational MM Mutational MM Evolutionary MM Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 34 -
Based on Additional Criteria Type of multimodel (MM) Behavior (activation) Nature of knowledge Constraint-driven Constraint-driven MM (Adaptive MM) of submodels to activate submodels Pattern-directed (Pattern-directed MM) (Metamorphic MM) Submodel selection is cyclic No Acyclic MM Yes Cyclic MM Goal-directed Goal-directed MM (Exploratory MM) Location of knowledge Within the MM (Internal activation of submodels) Active MM (Internally activated MM) to activate submodels Outside the MM (External activation of submodels) Passive MM (Externally activated MM) Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 35 -
Emergence Emergent state Emergent relation Models with emergent states and / or relations Multimodels with emergent states and / or relations Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 36 -
State machines: Mealy machines Moore machines Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 37 -
S2 S1 S5 S3 S4 S5 emergent state Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 38 -
Multistage state machines: Multistage model based on state machine formalism. Emerging conditions may necessitate: emerging states emerging transitions Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 39 -
Some other multistage modeling formalisms: Petri nets Fuzzy logic models DEVS Endomorphic models (ghost models, introspective models) (Intelligent) Agents Multiagents (similar to a multimodel), metamorphic agent, multiaspect agent Learning models Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 40 -
Multisimulation Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 41 -
Multisimulation: Simulation run - experimentation with a dynamic model Simulation study - a collection of simulation runs Multisimulation - simulation with several realities (experimentation with multistage models) - simultaneous - sequential Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 42 -
Multistage models: M 1 M 1 M 1 M 2 M 1 M 2 M 5 M 1 M 2 M 6 M 2 M 3 M 4 M 1 M 3 M 1 M 3 M 7 M 5 M 6 M 7 M 8 M 1 M 3 M 8 M 1 M 4 M 1 M 4 M 8 Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 43 -
Barcelona Lecture 3: MultiModel - MultiSimulation Tuncer Ören 2006-10-17, Barcelona, Spain - 44 -