EVOLUTIONARY NEURAL TREES FOR MODELING AND PREDICTING COMPLEX-SYSTEMS

Modeling and predicting the behavior of many technical systems is comp licated because they are generally characterized by a large number of variables, parameters and interactions, and limited amounts of collect ed data. This paper investigates an evolutionary method for learning m odels of such systems. The models thus evolved are based on trees of h eterogeneous neural units. The set of different neuron types is define d by the application domain, and the specific type of each unit is det ermined during the evolutionary learning process. The structure, size, and weights of the neural trees are also adapted by evolution. Since the genetic search used for training does nor require error derivative s, a wide range of neural models can be constructed. This generality i s contrasted with various existing methods for complex system modeling , which investigate only restricted topological subsets rather than th e complete class of architectures. An improvement in the predictive ac curacy and parsimony of models is reported, against backpropagation ne tworks and other well-engineered polynomial-based methods for two prob lems: MacKey-Glass and Lorenz-like chaotic systems. The authors also d emonstrate the importance of the selection pressure towards model pars imony for the improvement of prediction accuracy. (C) 1997 Published b y Elsevier Science Ltd. All rights reserved.