Reinforcement Learning-Based Tracking Control for a Class of Discrete-Time Systems With Actuator Fault

Abstract

Tracking control problem is a common problem and widely used in many fields. In this paper, the data-based tracking control with actuator fault is studied. To deal with the fault, many detection mechanisms and fault-tolerant tracking (FTT) controllers have been studied. However, the existing detection mechanisms cannot detect the systems with non-zero initial value. Furthermore, the existing FTT controllers require some designed parameters to obtain the fault. To solve above problems, a detection mechanism based on expanded time horizon and FTT controller based on reinforcement learning are proposed in this paper. The proposed detection mechanism is appropriate for arbitrary initial value, which consists of the initial value and past data. Besides, a nested calculation method is presented to obtain the fault information of FTT controller, which only uses the system data. Hence, the proposed FTT controller avoids the design of additional parameters for fault. Finally, the effectiveness of proposed methods is verified by simulation example.