Go to Paper
Return to GeoComputation 99 Index
Parameter Estimation in Neural Spatial Interaction Modelling by a Derivative Free Global Optimization Method
FISCHER, Manfred M. (Manfred.M.Fischer@wu-wien.ac.at), Department of Economic and Social Geography, Wirtschaftsuniversitat Wien, Augasse 2-6, A-1090 Vienna, Austria; and REISMANN, Martin, (firstname.lastname@example.org), Institute for Urban and Regional Research, Austrian Academy of Sciences, Postgasse 7/4/2, A-1010 Vienna, Austria
Key Words: neural spatial interaction modelling, parameter optimization, evolutionary computation, backpropagation of conjugate gradient descent errors, real world application performance test
Parameter estimation is one of the central issues in neural spatial interaction modelling. Current practice is dominated by gradient-based local minimization techniques. They find local minima efficiently and work best in unimodal minimization problems, but can get trapped in multimodal problems. Global search procedures provide an alternative optimization scheme that allows escape from local minima. Differential evolution recently has been introduced as an efficient direct search method for optimizing real-valued multi-modal objective functions (Storn and Price 1997). The method is conceptually simple and attractive, but little is known about its behaviour in real-world applications. This paper explores this method as an alternative to current practice for solving the parameter estimation task, and attempts to assess its robustness, measured in terms of in-sample and out-of-sample performance. A benchmark comparison against back propagation of conjugate gradients is based on Austrian interregional telecommunication traffic data.