A Genetic Algorithm Based EvolutionaryudComputational Neural Network forudModelling Spatial Interaction Data

摘要

Building a feedforward computational neural network model (CNN) involves two distinctudtasks: determination of the network topology and weight estimation. The specification of audproblem adequate network topology is a key issue and the primary focus of this contribution.udUp to now, this issue has been either completely neglected in spatial application domains, orudtackled by search heuristics (see Fischer and Gopal 1994). With the view of modellingudinteractions over geographic space, this paper considers this problem as a globaludoptimization problem and proposes a novel approach that embeds backpropagation learningudinto the evolutionary paradigm of genetic algorithms. This is accomplished by interweaving audgenetic search for finding an optimal CNN topology with gradient-based backpropagationudlearning for determining the network parameters. Thus, the model builder will be relieved ofudthe burden of identifying appropriate CNN-topologies that will allow a problem to be solvedudwith simple, but powerful learning mechanisms, such as backpropagation of gradient descentuderrors. The approach has been applied to the family of three inputs, single hidden layer,udsingle output feedforward CNN models using interregional telecommunication traffic data forudAustria, to illustrate its performance and to evaluate its robustness. (authors' abstract)