A Combined Model Based on ISSA-CNN-BiGRU-MH-Attention and Its Application in Power Load Forecasting

Existing prediction algorithms face challenges in terms of accuracy and training speed, which hinders high-efficiency, high-accuracy power load forecasting. This paper proposes a combined prediction model to address these issues. To overcome the Bidirectional Gated Recurrent Unit (BiGRU)'s limitations in capturing long-term dependencies and handling complex time-series data, a Convolutional Neural Network (CNN) module is introduced to extract local features and enhance the model's feature representation. Additionally, a Multi-Head Attention (MH-Attention) module is incorporated to dynamically assign weights to different time steps, improving adaptivity and focus on key features. For hyperparameter optimization, an Improved Sparrow Search Algorithm (ISSA) is proposed, which addresses traditional SSA’s tendency to fall into local optima and slow convergence by incorporating an adaptive update mechanism and hybrid heuristic strategy. The model is validated using a power plant dataset from Quanzhou, with results showing excellent forecasting ability: R2=0.9955, RMSE=56.9596, and MAE=34.6080. Comparison with other models demonstrates improved performance, with R2 increasing by 0.3%-0.65%, RMSE decreasing by 6.38%-35.96%, and MAE reducing by 26.54%-49.52%. These results confirm the model’s effectiveness and superiority.