Design of the Neural Model Structure Based on Genetic Algorithms

Genetic algorithm based neural model structure design of a non-linear dynamic system is described. The genetic algorithm represents an optimisation procedure, where the cost function, which is minimized consists of the non-linear dynamic process neural model simulation and a selected performance index evaluation. Using this approach the neural model of the process has been optimised from point of view its internal architecture. A multilayer perceptron (MLP) artificial neural network has been used, where the training was realized by the Levenberg-Marquardt method.