Application of cascade-correlation neural networks to nonlinear system identification

Much research in recent years has been done in applying artificial neural networks to the problem of nonlinear system identification. The most common neural network architecture, the multilayer feed-forward network, trained with the backpropagation algorithm, has been shown to be capable of universal function approximation which makes it applicable to a much wider range of problems than other nonlinear identification techniques. While these neural networks show great potential, they still suffer several drawbacks, such as slow convergence toward a solution. New neural network architectures have been proposed in an attempt to overcome these limitations. This study examines one such architecture, Cascade-Correlation, and its usefulness in system identification applications, particularly the nonlinear case.