Research on Electricity Sales Forecasting Method Based on GRU-Attention Considering Multi-Market Factors

Accurately forecasting electricity sales is crucial for power system planning and market decision-making. However, due to the multidimensional coupling effects of market prices, load characteristics, economic conditions, and meteorological factors, traditional methods struggle to balance predictive accuracy and interpretability. To address this issue, this study proposes a hybrid deep learning forecasting method based on multi-market factors. First, the shapley additive explanations (SHAP) algorithm is employed to quantify the marginal contributions of various factors and identify key driving variables. Then, a gated recurrent unit (GRU) combined with an Attention mechanism is constructed, in which the attention mechanism dynamically focuses on critical time steps to capture long-term dependencies in multi-factor time series. Finally, an empirical analysis is conducted using historical data from Province G. The results indicate that the proposed model significantly outperforms traditional time series methods and mainstream machine learning approaches in both trend fitting and volatility prediction. The evaluation metrics, including mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE), all achieve optimal levels, with the mean absolute percentage error reaching 1.52%. Prediction errors are more concentrated and extreme deviations are substantially reduced, demonstrating both high accuracy and interpretability.