agent-based modelling using matlab modelling using matlab outline outline of topics concepts about...

Agent-based modelling using MATLAB


Shan He

School for Computational ScienceUniversity of Birmingham

Module 06-23836: Computational Modelling with MATLAB


Outline

Outline of Topics

Concepts about Agent-based Modelling

Modelling process of agent-based modelling

A detailed example: BOIDs

Conclusion



What is Agent-based Modelling (ABM)?

I Other names: Agent-based simulation, Individual-basedmodelling.

I A new modelling method: not equation-based, but based on apopulation of agents.

I An agent:I A discrete entity with its own goals and behavioursI Autonomous: with a capability to adapt and modify its

beahaviours

I A new research field: have been widely applied to biology,social sciences and other scientific fields and business.



Some examples

I Agent-Based Models in Biology and Medicine

I Agent-based computational economics

I The Economist: Agents of change

I Nature: Meltdown modelling

I Managing Business Complexity: Discovering StrategicSolutions with Agent-Based Modeling and Simulation

http://www.swarm.org/wiki/Agent-Based_Models_in_Biology_and_Medicine

http://en.wikipedia.org/wiki/Agent-based_computational_economics

http://www.economist.com/node/16636121

http://mark.buchanan.pagesperso-orange.fr/nature_economic_modelling.pdf

http://www.amazon.co.uk/Managing-Business-Complexity-Discovering-Agent-Based/dp/0195172116

http://www.amazon.co.uk/Managing-Business-Complexity-Discovering-Agent-Based/dp/0195172116



The assumptions behind ABM

I Some key aspect of behaviours can be described.

I Mechanisms by which agents interact can be described.

I Complex behaviours are emerged from simple, usually localinteractions.

I The model is built “from the bottom up”.



Modelling process

SimulationDefinitionIdentification

ValidationAnalysis Satisfied?

No

Yes



Definition

I Step 1: Define the agent types and other objects along withtheir attributes.

I Step 2: Define the environments the agent will live in andinteract with

I Step 3: Define agent method for updating the agent’sattribute in response to their interactions with other agentsand the environment.

I Step 4: Define behavioural rules to specify which agentsinteract, when they interact and how they interact. This canbe changed to other methods, e.g., ANN as we shall seeshortly.



Animal swarming behaviour

Pictures from NY Times

Starlings flocking video

http://www.youtube.com/watch?v=eakKfY5aHmY&feature=related



BOIDs: ABM of animal schooling/flocking behaviour

I Invented in 1986 by Craig Reynolds to simulate coordinatedanimal motion such as bird flocks and fish schools.

I A giant step forward in computer animation: animation‘emerges’ from complex system

I First film: Stanley and Stella in: Breaking the Ice

I Used in Tim Burton’s film Batman Returns: bat swarms and”army” of penguins marching through the streets of GothamCity.

I Winner of technical Oscar: 1997 Sci-Tech Awards fromAcademy of Motion Pictures and Science.

http://www.youtube.com/watch?v=pliaEEUzl0U



Batman Returns



How BOIDs is developed?

I Assumption: the global patterns are the emergingphenomena from interaction of birds or BOIDs.

I Let’s follow the definition steps:

I Step 1: Identify the agent types and other objects along withtheir attributes.

I Agent: only Birds, or BOIDs

I Agent attributes: position and velocity




I Step 2: Define the environments the agent will live in andinteract with

I A simple 2D or 3D world, no other objects.




I Step 3: Specify agent method for updating the agent’sattributes in response to their interactions with other agentsand the environment.

I Only consider interactions between BOIDs:- update a BOID’sattribute, e.g., position and velocity in response to theirinteractions with other BOIDs.




I Step 4: Define behavioural rules to specify which agentsinteract, when they interact and how they interact.

I The most difficult step: usually done by trail-and-error.

I Sometimes obtestation can be used to derive the behaviouralrules and then validate by experiments.

I Ideally, the rules should be derived based on experimentaldata – data-driven modelling.

I It is possible to automatically generate/evolve behaviouralrules by computer, e.g., Artificial Neural Network andEvolutionary Computation.



BOIDs: three simple rules

I Repulsion: steer to avoid crowding local flockmates

I Alignment: steer towards the average heading of localflockmates

I Cohesion: steer to move toward the average position of localflockmates



BOIDs: reaction zones



BOIDs: Repulsion

Repulsion zone



BOIDs: Alignment

Alignment zone



BOIDs: Attraction

Attraction zone



Implementing BOIDs in MATLAB

I First question: What data structures should we used?

I Requirement: Can easily store/retrive data, ideally should bedifferent types of data, e.g., integer, string or even a datastructure.

I Cell Array or Structure Array or Object?

I Cell Array is more convenient for coding, e.g, number asindex, but bad for code clarity.

I Object stores data in named fields and also encapsulates theoperations perform on data, but comes with complexity forcoding.

I Structure Array stores data in named fields and relativelyconvenient for coding, therefore it is more suitable for ourBOID model.


Conclusion

Advantages

I It captures emergent phenomena, therefore, it is a better toolfor modelling complex systems.

I It provides a natural description of a system. The behaviouralrules can be expressed by natural language, which is easy tobe understood.

I It is flexible, e.g., heterogeneous agents and theadaptive/evolving rules.


Conclusion

Disadvantages

I Models too simple: it is very easy to introduce too manyassumptions to oversimplified the model.

I Models too complicated: it is also very easy to introduce toomany rules to overcomplicate the model.

I Solution: Use the modelling principle and process in the 2ndLecture.

I Error-prone: many lines of code – very easy to make mistakewhen coding.

I Difficult to determine model parameters.


Conclusion

Other agent-based modelling tools

I MATLAB is limited in 2D/3D animation.

I Other tools are more powerful:

I Netlogo

I SWARM

I Repast

I MASON

I Processing

http://ccl.northwestern.edu/netlogo/

http://www.swarm.org/

http://repast.sourceforge.net/screenshots.html

http://cs.gmu.edu/~eclab/projects/mason/

http://processing.org/

agent-based modelling using matlab modelling using matlab outline outline of topics concepts about...

Documents