A Simulation-Constraint Graph Reinforcement Learning Method for Line Flow Control

Line flow control plays an essential role in maintaining the stability of power system. Considering the randomness and uncertainties in the grid, a simulation-constraint graph reinforcement learning method is proposed to provide a potential solution for the real-time dispatch via topology control. To improve the training efficiency and deployment stability of reinforcement learning agent in the power system with various constraints and large action space, an adaptive simulation regulation strategy and a step-wise control mechanism are utilized. Power system simulation is introduced to keep the agent from actions with invalidity or severe simulated outcomes. Then, graph neural network is applied to help the agent learn a better representation of power system state by dealing with the information hidden in topology. Finally, a numeric example on IEEE 14-bus system is employed to demonstrate the feasibility of the proposed method.