Multi Agent Systems and Simulation

1
Multi Agent Systems and Simulation

• Why bother each other?
• Andy Evans

2
What are Agents?

• A model architecture developed under the rise of
  Object Orientated languages and fast computers.
• The computer contains an environment.
• Objects of interest move independently around the
  landscape.
• Objects have behavioural rules
• They interact with each other.
• They interact with the environment.

3
Very simple algorithm

• Set up agents with behaviour
• Randomly pick agent an get it to interact with
• environment
• agents within distance X
• agents at other scales
• Repeat until all updated
• Repeat whole thing until some criteria

4
Example Glacial hydrology

• Winter subglacial storage ? Summer proglacial
  flow.
• Agent Water packet.
• Each packet takes a path through the environment
  picking up a chemical and sedimentary history.
• They effect the environment.
• Water packets interact chemically and exchange
  sediment.

5
Advantages

• Theyre a more accurate representation of
  discrete systems (though they can hold
  statistical elements / fields).
• Model scenarios where a statistical models would
  be complex to understand or implement
• Can model vast numbers of complex interactions.
• Can model interactions between vast numbers of
  types.
• High resolution of spatio-temporal alterations
  possible (individual packets can have individual
  rules).
• Disaggregate results to the individual level.
• Track individual packets through a system and
  examine their evolution.
• Good for parallelization.

6
Problems

• No single figure output
• Understanding the outputs / validation and
  sensitivity analysis.
• Problems of modelling complex systems
• appropriate outputs, recognising and avoiding
  artefacts, scale.
• Generating behaviours / coping with missing
  behaviours.
• Which architectures best?
• Best methods for optimal parallelization?
• Are there general code blocks that would be
  useful for everyone?
• Terminology and literature vast numbers of
  fields have come at this independently but the
  above areas are universal.

7
Today

• Hazel Parry An ecological model / What social
  science has to offer ecology.
• Sim Reaney An agent based hillslope model.
• Rob Thomas An agent based sediment model /
  philosophical reflections.
• Chris Keylock A future project.