Deep Reinforcement Learning Based Approach for Active Power Security Correction Control of Power System

Active power security correction of power system is of great significance for the safe operation of power grid under bad conditions. In this paper, an active power security correction control method based on deep reinforcement learning (DRL) is proposed. Firstly, by modeling the Markov process of active power security correction, the state space, action space and reward value of reinforcement learning method for active power security correction are designed. Secondly, using twin delayed deep deterministic policy gradient algorithm (TD3), TD3 agent automatically learns the global optimal decision from the interaction with the environment. Then, a control framework combining “off-line training” and “on-line control” of power system security correction based on TD3 is built. Finally, the effectiveness of the method is verified by IEEE 39 -bus system.