Power User Load Forecasting Based on Spearman-ICEEMDAN-TCN-iTransformer

Abstract To improve the accuracy and dynamic adaptability of power user load forecasting, this study conducts research on load forecasting based on multi-technology integration. Firstly, Spearman rank correlation analysis is used to construct a multi-modal feature set, and the correlation of multi-modal features is quantified to achieve effective fusion of multi-source heterogeneous data such as meteorological data and electricity price data. Secondly, an improved ICEEMDAN algorithm is introduced to decompose and reconstruct the load sequence. Adaptive noise injection is used to optimize time-frequency feature extraction, and a hybrid TCN-iTransformer architecture is combined to capture short-term fluctuations and long-term trends. This provides theoretical and methodological support for accurate load forecasting in the new power system.