Emergency Control Strategy of Power System Considering Operation Scenario Variations

Power system emergency control is the core of the second and third defense lines. The deep reinforcement learning (DRL) based control scheme breaks through the limitations of traditional control strategy simulation calculation methods based physical models, and shows great advantages in dealing with high-dimensional nonlinear system control problems. However, there is a lack of research on dealing with the renewable energy grid-connected power system control problem with operation scenarios variations, especially the frequent topological changes. Based on the DRL method, the under-voltage load shedding (UVLS) emergency control scheme considering the operation scenarios variations is proposed. And the graph convolutional network (GCN) is adopted to learn the representation of power grid topological states and solve the problem of topology feature extraction. Moreover, renewable energy grid-connected scenarios with medium proportion, high proportion and extremely high proportion permeability are set up. Case studies on the modified IEEE 39-bus system are employed to demonstrate the performance and effectiveness of the proposed method.