Fourier Decomposition Residual-Based Power Load Forecasting

Load forecasting plays a vital role in smart grid operation and power system stability. With the increasing integration of renewable energy and diversified consumption patterns, power load exhibits complex characteristics including non-stationarity, volatility, and coupled periodic-stochastic features. While traditional methods struggle to simultaneously model deterministic patterns and random fluctuations, existing approaches often fail to effectively handle high-frequency noise while preserving long-term dependencies. To address these challenges, we propose a novel dual-component forecasting framework based on Fourier decomposition residuals. Our approach first decomposes load signals into deterministic trends and stochastic components through Fourier analysis, then employs Savitzky-Golay filtering and LSTM networks to process the residual features. By integrating Fourier-based trend modeling with deep learning-based residual prediction, our method effectively captures both periodic patterns and nonlinear fluctuations. Experimental results on real-world power load data demonstrate that our approach achieves superior accuracy and robustness compared to state-of-the-art models, with significant improvements in both MAE and RMSE metrics.