Distributionally Robust Proximal Policy Optimization for Unit Commitment with Wind Power Uncertainty

Unit commitment problem plays an essential part in the electricity market. The uncertainty of wind power and power demand increases the difficulty of solving the unit commitment problem. Although the traditional method can effectively solve this problem, the computation time grows exponentially with the increasing scale. To overcome this difficulty, we use deep reinforcement learning (DRL) methods by the means of the formulation of the UC problem as a Markov decision process and propose a DRL-based algorithm to solve the problem. Meanwhile, we use distributionally robust optimization to improve the DRL method robustness. The IEEE 39-bus test case is used to conduct numerical results and demonstrate the method’s effectiveness. Finally, we compare it with some existing solutions to verify its effectiveness and superiority.