Agent Based Modelling

Agent Based Modelling (ABM) is a technique to model Complex Adaptive Systems (CAS) by specifying each actor in a scenario, the actors internal behaviour, interaction with other agents and interaction with its environment. With even simple rules of behaviour and interaction, in an entire system of many agents unexpected behaviour can emerge. Actors may represent many different physical entities - for instance in biology an actor may be an organism, or in micro-biology a cell. In sociology, actors are often individual humans, in economics they may be corporations whereas in computer science the actors may be automata.

In our model, we seek to use ABM in order to model a variety of actors within the electricity network, individuals, corporations, regulators, automata and smart devices. ABM has two great advantages in this environment

  • It can easily accomodate a range of heterogeneous agents as long as their interaction is specified
  • It can easily accommodate a wide range of parameters for each agent, allowing wide variation of attributes even within a given agent 'type'

One example of using ABM is given below, many more are available here implemented in NetLogo

Schelling segregation

A classic experiment from Thomas Schelling which sought to explain racial segregation. Schelling sets up a 2D grid environment with heterogenous agent types (represented here by colour of dots). The initial type of an agent and their position on the grid is random. Schelling further gives agents a simple rule based only on their immediate neighbours

  • The agent wants a certain percentage of immediate neighbours to be 'like' them (i.e. the same coloured dot), if this percentage is satisfied do nothing, otherwise move to a randomly selected empty square.

From this simple local rule, we see a macroscopic clustering and segregation emerge. It is also interesting to note that a mild preference still leads to strong segregation. In the example video below, a concept of “generations” is also introduced where agents die after a number of years (set between 80 and 100) and are replaced by new agents of a random type. We see that over many such “generations”, the segregation effect becomes stronger. Generations occur roughly every 7-10 seconds in the video and are characterised by bursts of activity followed by a stable period for each generation. This implementation is in the Repast Simphony package.

 
examples/abm.txt · Last modified: 20/07/2010 11:22 by cascade     Back to top

Recent changes RSS feed Creative Commons License Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki