Reinforcement Learning-Based Event-Triggered Optimal Control of Power Systems With Control Input Saturation

In this article, a reinforcement learning (RL)-based event-triggered guaranteed cost control (GCC) is developed for power systems with control input saturation (CIS). An event-triggered mechanism with a balance factor is developed to optimize both control performance of power systems and computation efficiency. To obtain an online solution to the corresponding modified Hamilton–Jacobi–Bellman equation for the power system, a critic neural network is employed. In contrast to RL-based methods employing dual-network setups, the single neural network not only conserves computational resources but also eliminates the need for an initial admissible control. For accelerating convergence toward optimal solutions, a new adaptive weight tuning law is constructed. Control performance of power systems with CIS and limited computation power is ensured by the proposed RL-based optimal GCC strategy. Simulation results validate the effectiveness of the proposed methodology.