Identification and Control of Dynamic Systems via Adaptive Neural Networks

摘要

Various applications of multilayer perceptron neural networks are studied inorder to identify and control certain types of dynamic systems. Two different methods for updating the weights of the neural network are explored. A variable structure control formulation and a gradient descent approach are considered. Both methods are digitally simulated and their performances in parameter identification are compared. An adaptive model reference control scheme is presented along with an indirect self-tuning control scheme based on a multilayer perceptron neural network. Examples of the simulated responses for both schemes are obtained using different linear plans.