XClose

The Bartlett Centre for Advanced Spatial Analysis

Home
Menu

CASA Working Paper 10

WP10

1 May 1999


Multi-Agent Simulation: New Approaches to Exploring Space-Time Dynamics Within GIS

As part of the long term quest to develop more disaggregate, temporally dynamic models of spatial behaviour, micro-simulation has evolved to the point where the actions of many individuals can be computed. These multi-agent systems/simulation (MAS) models are a consequence of much better micro data, more powerful and user-friendly computer environments often based on parallel processing, and the generally recognised need in spatial science for modelling temporal process.

In this paper, we develop a series of multi-agent models which operate in cellular space. These demonstrate the well-known principle that local action can give rise to global pattern but also how such pattern emerges as the consequence of positive feedback and learned behaviour. We first summarise the way cellular representation is important in adding new process functionality to GIS, and the way this is effected through ideas from cellular automata (CA) modelling. We then outline the key ideas of multi-agent simulation and this sets the scene for three applications to problems involving the use of agents to explore geographic space. We first illustrate how agents can be programmed to search route networks, finding shortest routes in ad hoc as well as structured ways equivalent to the operation of the Bellman-Dijkstra algorithm. We then demonstrate how the agent-based approach can be used to simulate the dynamics of water flow, implying that such models can be used to effectively model the evolution of river systems. Finally we show how agents can detect the geometric properties of space, generating powerful results that are not possible using conventional geometry, and we illustrate these ideas by computing the visual fields or isovists associated with different viewpoints within the Tate Gallery. Our forays into MAS are all based on developing reactive agent models with minimal interaction and we conclude with suggestions for how these models might incorporate cognition, planning, and stronger positive feedbacks between agents.

This working paper is available as a PDF. The file size is 279KB.

Authors: Michael Batty, Bin Jiang

Publication Date: 1/4/1999

Download working paper No. 10.