Reactive Power Optimization for Voltage Stability in Energy Internet Based on Graph Convolutional Networks and Deep Q-learning
Sheng Guo, Junwei Cao · 2021
The rapid response of reactive power compensation is crucial to guarantee the stable operation of energy Internet (EI) with variable loads and distributed power generations. Therefore, this paper proposes an intelligent approach for reactive power optimization in EI based on graph convolutional networks (GCN) and deep Q-learning (DQN). With the adjacency matrix that represents topology of EI, the GCN in the proposed approach fuses the monitoring data of EI nodes for a comprehensive feature extraction. Furthermore, reactive power optimization of EI during voltage sags is solved by DQN method in which GCN is used as the Q network. Thus, the optimized output of reactive power compensation device can be put into EI to ensure the voltage stability. The case study on the simulation data of an EI system that considers photovoltaic and battery storage system verifies the effectiveness of the proposed approach. The result shows that the proposed approach achieves fast response to the faults and sudden increase of load in EI, and gives more accurate reactive power compensation than the common control method of reactive power compensation device.