Reinforcement learning approach of a multi-model controller

Abstract

A control scheme based on reinforcement learning network is presented in this paper. The proposed controller is based on the multi-model approach to improve the system performance of a control system. The multi-model control scheme depends on the multiple representation of the process using different models; these models are used to generate the control signal needed to make the system follows a prescribed desired trajectory. The control signal is composed of two components; the first results from a proportional linear feedback component, while the second represents the corrective component generated from the neural network to locate the best model representing the system at certain operating environment. The controller is robust as it can accommodate high and sudden deviation from the prescribed trajectory. Simulation results are included to illustrate the potential of the developed controller.

Link to Full Article