WHDNet: A Weather-Driven Hybrid Decomposition and Deep Learning Framework for Precision Power Load Forecasting

This paper introduces WHDNet, a two-stage hybrid power load forecasting model that combines weather-driven hybrid decomposition with deep learning techniques, including BiTCN, BiGRU, and an attention mechanism. The model integrates EEMD, wavelet de noising, and VMD to preprocess load data, effectively reducing noise and capturing both short- and long-term dependencies. Experimental results demonstrate that WHDNet outperforms baseline models such as BiLSTM, BiGRU, and BiTCN, achieving a lower MAPE of 0.0534 and a higher R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.7952, indicating superior predictive accuracy. The inclusion of attention mechanisms further enhances the models focus on critical time steps, making it well-suited for dynamic and volatile forecasting environments.