Optimized Active Disturbance Rejection Control for Switched Systems: A Triggered-Learning Approach

This article studies active disturbance rejection control (ADRC) for uncertain switched systems under triggered learning. The novel triggered-learning ADRC (TL-ADRC) framework optimizes the ADRC performance of switched systems through reinforcement learning (RL) and enables “on-demand” updates of neural networks with guidance from a pre-designed trigger. The innovation of this article is mainly reflected in four aspects: (i) The RL-based gain automatic update mechanism (i.e., the dual-gain optimization mechanism (DGOM)) optimizes the performance of extended state observer (ESO) and controller; the optimal policy is obtained from the experience-based deep deterministic policy gradient (DDPG) with self-learning ability. (ii) The adaptive performance-triggered strategy guides the update of dual-gain; the on-demand triggering judgment is achieved by comparing cost functions that reflect the tracking control performance. (iii) The proposed theoretical analysis method proves that the learning mechanism can enhance the closed-loop system performance. (iv) The constructed ADRC-based switching law accelerates the convergence of system tracking error. Finally, a comparative simulation example demonstrates the effectiveness of the proposed method.