Adaptive Reinforcement Learning for Fault-Tolerant Optimal Consensus Control of Nonlinear Canonical Multiagent Systems With Actuator Loss of Effectiveness

This article addresses the adaptive optimized consensus tracking control problem of nonlinear multiagent systems (MASs) via a reinforcement learning (RL) algorithm. Specifically, the nonlinear high-order MASs are formulated in a canonical form, with considerations for both actuator effectiveness loss and time-varying bias faults. First, neural networks (NNs) are utilized to approximate unknown nonlinear dynamics, and a state identifier and a fault estimator based on NNs are established, both of which are essential for evaluating state information and bias faults, respectively. Second, to achieve a high-order canonical dynamic consensus and enhance the efficiency of the consensus control strategy, a sliding-mode mechanism is employed to regulate tracking errors. Moreover, we develop an adaptive NN-based fault-tolerant optimal control method by integrating the sliding-mode mechanism with an actor-critic structured RL algorithm. It is proved that the outputs of the MASs precisely align with the desired reference signals, while ensuring the boundedness of all closed-loop signals. Finally, the proposed control methodology's effectiveness is validated through a simulation example.