Neural Spatial Interaction Models and Evolutionary Algorithms

Austrian Science Fund

One of the major intellectual achievements and, at the same time, perhaps the most useful contribution by spatial analysis to social science literature is the development of spatial interaction models. Spatial interaction is a broad term encompassing any movement over space such as, e.g. journey-to-work migration, commodity and information flows. Since the pioneering work of Alan G. Wilson on entropy-maximization in the early 70s there has been surprisingly little innovation in the design of spatial interaction models. The principal exceptions include the competing destinations version of Stewart Fotheringham (1983, now at University of Newcastle), the use of genetic algorithms to breed new forms of spatial interaction models (Stan Openshaw, 1988; now at University of Leeds) and WSG's feedforward neural spatial interaction models with a single hidden layer, three input and one output unit jointly developed with Sucharita Gopal (Boston University). The overall objective of the project is to systematically explore and evaluate the usefulness of evolutionary algorithms as global search procedures for solving the parameter estimation and the model choice problem in neural spatial interaction modelling, aiming at a deeper understanding of why and how the operators and the algorithms as a whole do work in a real world context, using interregional telecommunication flow data.
The objectives will be reached via (1) developing, implementing and testing evolutionary stochastic optimizers to the parameter estimation problem, given an a priori chosen (i.e. fixed) neural spatial interaction model.