Distribution network voltage forecasting based on graph convolutional networks – long short-term memory

The integration of renewable energy sources increases the complexity of distribution systems and the uncertainty of node voltages, thereby amplifying the importance of precise voltage forecasting to maintain system stability. Voltage forecasting is traditionally performed using mathematical models. However, these models have limitations in analysing the diversity of loads and the overvoltage and congestion caused by renewable energy outputs. Thus, AI-based voltage forecasting using data-based approaches is being widely studied to refine forecasting models based on bus to account for temporal voltage variability. However, the voltage of a bus is closely related to those of adjacent buses, necessitating their consideration in forecasting models. In this paper, a graph convolutional networks–long short-term memory-based forecasting model is developed that considers the interconnection information between buses to determine the characteristics of adjacent buses in the distribution system. The proposed method has been validated via simulations.