Sliding Mode Learning Control for Interconnected Nonlinear Systems via Data-Driven Self-Triggered Fuzzy Approach

This study combines the fuzzy logic and the data-driven technology to solve the reinforcement learning-based sliding mode control problem of interconnected nonlinear systems with unmodeled dynamics. By assigning cost functions associated with the sliding-mode function for all auxiliary subsystems, the original control problem is equivalently converted into designing a group of optimal control policies updating in a self-triggered manner. To derive the optimal policies, a single-critic network architecture under the framework of reinforcement learning is constructed. Meanwhile, a fuzzy logic-based control policy is designed to handle the dynamical uncertainty issue aroused from the unmodeled dynamics. Furthermore, a data-driven method is used to reconstruct the unknown system dynamic through a three-layer neural network. Eventually, effectiveness and superiority of the proposed control strategy are demonstrated via experiments on optimal control of an interconnected two-stage chemical reactor system.