Robust fault-tolerant control for wastewater treatment processes via integrated adaptive dynamic programming with fault-aware experience replay

Abstract

Wastewater treatment is a crucial measure to address water scarcity and water pollution, and various control methods have been proposed for wastewater treatment processes (WWTPs). However, actuator faults can compromise existing control strategies and destabilize the system. In this article, an integrated adaptive fault-tolerant control method with improved experience replay is developed for WWTPs. First, an integrated adaptive dynamic programming with fault-aware experience replay (IADP-FAER) is designed. Based on the IADP-FAER, a robust fault-tolerant control framework is established for WWTPs. It utilizes a dynamic weight allocation mechanism to seamlessly integrate the priori knowledge of the original controller into the adaptive dynamic programming (ADP) design, thereby avoiding control conflicts and providing a stable learning environment under actuator faults. Second, the cost function considering actuator faults is constructed for the fault-tolerant controller. The exploration policy obtained from the cost function can effectively suppress adverse effects of actuator faults. The critic and action networks can be established using radial basis function neural networks, which aims to approximate the cost function as well as the exploration policy. Their parameters are adaptively adjusted via the policy iteration algorithm so that the control performance of the system is continuously optimized. Third, to further reduce the tracking error and improve the data usage efficiency, a fault-aware experience replay (FAER) is adopted to further adjust the parameters of the networks. By focusing on data with fault features, it allows the controller to quickly adapt to actuator faults. Finally, through simulation experiments, the feasibility of the fault-tolerant control is confirmed.