A CrossInformer model based on dual-layer decomposition and interpretability for short-term electricity load forecasting

Short-term electricity load forecasting is a critical issue in power system management, and its accuracy is essential for the stable operation of the grid and energy dispatch. This paper proposes a CrossInformer model based on dual-layer decomposition and interpretability for predicting short-term electricity load. Firstly, the CrossInformer employs an intelligent optimization algorithm for feature selection, which enhances the model’s training efficiency and prediction accuracy. Secondly, an ICEEMDAN-RLMD dual-layer decomposition method is adopted to denoise the load data, thereby reducing the influence of noise on the prediction results. Subsequently, the CrossInformer architecture integrates multi-granularity patch inputs, ProbSparse Attention, self-attention distillation, and the HiLo attention mechanism to improve the model’s capability in capturing both long-term and short-term dependencies in time series data. Finally, to enhance the model’s interpretability, the SHAP method is used to quantify the contribution of each feature to the prediction outcomes. Experimental results demonstrate that the proposed method outperforms traditional approaches in electricity load forecasting, offering superior prediction accuracy and stability.