Optimal control of nonlinear plants using artificial neural networks.

Abstract

In this thesis, algorithms for application of Artificial Neural Networks in solving nonlinear optimal control problems are developed. The conventional Multi-layers Feed-forward Neural Network is employed as the state feedback optimal controller. The newly developed Block Partial Derivatives concept is used to compute the gradient needed for neural network training. State tracking, regulation, terminal control, minimum control effort, minimum time, and output tracking problems are attempted. The performance of the proposed algorithms is investigated through application in simulated plants. Results obtained agree with the ones found through other standard techniques.