RL-driven Problem Decomposition for Computationally Efficient AC Optimal Power Flow

Solving the AC optimal power flow (AC OPF) problem poses significant challenges in power system operations because of its inherent nonlinearity and complexity. This paper introduces a novel strategy to solve the AC OPF problem by utilizing a new problem decomposition framework combined with reinforcement learning (RL)-based cutting planes. The problem is decomposed into two sub-problems, DC OPF and AC power flow (AC PF) calculation sub-problems. To yield the AC-feasible solution, linear inequality constraints (i.e., cuts) are obtained by an RL agent and added into the DC OPF sub-problem. Then, the AC PF calculation is performed using the solution to the DC OPF sub-problem (i.e., power generation profiles) and voltage magnitude reference values, which is the output of the RL agent. Additionally, the action selection method is employed for the RL agent’s training efficiency. Case studies under various simulation scenarios are conducted to show the effectiveness of the proposed strategy compared to the conventional strategies. The simulation results indicate that the proposed strategy significantly enhances computational efficiency and solution feasibility compared to the conventional methods.