ADAPTIVE-CONTROL OF DYNAMIC-SYSTEMS BY BACK-PROPAGATION NETWORKS

Artificial neural networks-especially those using the error back propagation algorithm-are capable of learning to control an unknown plant by autonomously extracting the necessary information from the plant. Following the approach of Psaltis, Sideris, and Yamamura, and Saerens and Soquet, a control architecture based on error back propagation has been developed and trained to control a third order linear and time invariant plant with dead-time. Simulation results show that the network is able to invert the plant's behaviour and characteristics, thus learning to control the plant accurately. The time to reach the desired outputs of the plant decreases while learning. It is accelerated by local adaptation of the learning rate.