Identifying nonlinear dynamic systems using neural nets and evolutionary programming

Nonlinear system behavior is not always well characterized by linearized system models, especially if the system is chaotic. This research studies the use of a neural network algorithm structure to model two nonlinear systems, a quadratic system and a chaotic system. An evolutionary programming approach is employed to train the neural nets so that the training process might better avoid selecting weighting parameters that represent a local minimum rather than a global minimum. This training approach is compared with the more standard backpropagation technique.<<ETX>>