Deep Q-Network-Based Fast Solution Method for Day-Ahead Unit Commitment

Unit commitment (UC) is an important part of the short-term phase of power system operation simulation. In this paper, we develop the reinforcement learning method to study the fast solution of unit commitment for power system operation simulation. Deep Q-Network (DQN) in the reinforcement learning theory is used to derive the detailed modeling process based on Markov decision process and implement the UC solution algorithm under the reinforcement learning framework. The case study compares the solution results of Q-learning algorithm, dynamic programming algorithm and evolutionary algorithm for verifying the feasibility and correctness of DQN algorithm proposed in this paper.