Multi-agent system modelling for urban systems: The series of SIMPOP models

Abstract

The SIMPOP simulation model was the first application of a multi-agent system (MAS) in geography and has been followed by a series of related applications. This chapter summarizes a few specific features of this series of models and the main insights that have been gained from this experiment. As the first objective of these models was to reconstruct the evolution of urban settlements at broad scales in geographical space and time (i.e. at national or continental levels and for decades or centuries), we explain the selection of the stylised facts making up knowledge about the dynamics of complex urban systems. They are used in the simulation models so as to reconstruct the interaction networks structuring the systems of cities. The originality of SIMPOP is thus to simulate the emergence and further hierarchical and functional differentiation of interdependent cities from interactions between them, so that the agents in this model are immobile entities, representing complex aggregation of individuals at meso-level. The quality of MAS is underlined for its flexibility in modelling spatial interactions with varied geographical configurations, and for its ability to deal with objects occurring on different scales, between geospatial entities that have expanding range of activities. We explain how we set about restricting the impact of the difficulties inherent in this type of modelling, which have been the subject of frequent criticism, in particular for their excessive complexity, thought to make any validation procedure unfeasible. In particular, we endeavour to describe in detail the various stages of model construction, on the basis of stylised facts obtained via numerous observations and comparisons, and set out to perform a multi-scale validation by testing the plausibility of the results delivered by the model at different aggregation levels. However, despite these promising methods of validation, further improvements are necessary for fully exploiting the capacities of simulation by using more powerful computing devices and validation methods. In this direction, the generic model SIMPOP will be completed and-transferred to an open and scalable simulation platform, and specific versions will be developed and tested for the main regions of the world.