Short-term Electricity Load Forecasting with CEEMDAN-BiGRU-TCN Model based on Dual Attention Mechanism

A CEEMDAN-BiGRU-TCN model based on the dual-attention mechanism is proposed for the short-term load forecasting problem of power systems. The model combines the advantages of adaptive noise Complete Ensemble Empirical Mode Decomposition (CEEMDAN), Bi-directional Gated Recurrent Unit (BiGRU) and Temporal Convolutional Network (Temporal Convolutional Network (TCN). The CEEMDAN algorithm effectively separates multi-scale features in power load data and reduces non-smoothness, while BiGRU and TCN process long-term and local memory information respectively, and jointly capture complex dynamic changes. The self-attention mechanism and channel-attention mechanism further strengthen the model's attention to key input information and improve the prediction accuracy. Experiments demonstrate that the proposed CEEMDAN-BiGRU-TCN model exhibits excellent forecasting performance when simulating actual load data changes, and outperforms traditional forecasting techniques in several evaluation indexes, providing accurate data support for power dispatch decision making.