Deep Reinforcement Learning-based Online Decision Making for Emergency Control of Power Systems

The increase in the proportion of new energy generation and the development of ultra-high voltage transmission have ensured the supply of electricity, but also made the dynamic process of the power system more complex, and the difficulty of maintaining the stable operation of the power system has greatly increased. Currently, the emergency control used in the actual power grid is mainly based on the strategy table prepared offline, and its adaptability to changes in power system conditions is insufficient and prone to mismatch and other problems. In this paper, a deep reinforcement learning-based decision-making method for emergency control of power systems is proposed. The emergency control problem is modeled as a Markov decision-making process, and the Proximal Policy Optimization algorithm is used to train the agent, which can quickly get the control action online based on the system state information. A branching architecture embedded with safety masking is used in this paper to solve the problems of dimensional disaster and illegal actions that may exist during the training process. Finally, a case study based on the IEEE 39-bus test system is conducted to verify the effectiveness of the method proposed.