Research on Substation Load Forecasting Method Based on GRA-LSTM

To optimize load-side resource allocation, power grid companies require advanced substation load forecasting, especially in regions with significant load fluctuations. These fluctuations are influenced by seasonal demands for cooling and heating, changes in electric vehicle (EV) energy consumption, and the complexity introduced by renewable energy sources. Traditional forecasting methods fall short in accurately analyzing and predicting these dynamics. This paper addresses these challenges by first analyzing the previous year's load data at the substation level, breaking it down into base load, seasonal demands, EV load, and renewable energy output. It then constructs a detailed feature set including weather conditions and holidays, and uses Grey Relational Analysis (GRA) to identify features strongly correlated with load. Long Short-Term Memory (LSTM) networks are employed for predicting various load components, with the results aggregated to estimate total substation load. This method, combining load decomposition, GRA, and LSTM, is validated against historical data from a northern region of China and outperforms traditional forecasting algorithms, significantly improving substation load forecasting accuracy.