Short-Term Electric Load Forecasting by Cross-Feature Analysis and Multimodal Selection

Li-Ling Peng¹, Tong Li¹, Guo-Feng Fan¹, Xin-Yu Yang¹, Wei-Chiang Hong^2,*
1 School of Mathematics & Statistics, Ping Ding Shan University, Ping Ding Shan, China
2 College of Shipbuilding Engineering, Harbin Engineering University, Harbin, China
* Corresponding Author: Wei-Chiang Hong. Email: email
(This article belongs to the Special Issue: Advanced Artificial Intelligence and Machine Learning Methods Applied to Energy Systems, 2nd Edition)

Computer Modeling in Engineering & Sciences https://doi.org/10.32604/cmes.2026.087031

Received 09 June 2026; Accepted 08 September 2026; Published online 20 September 2026

Download PDF

Abstract

Accurate load forecasting has become a critical foundation for ensuring the stable operation of power systems, optimizing generation scheduling, and supporting the efficient functioning of electricity markets. In this paper, the cross-scale and meso-scale characteristics of the complexity of power loads are analyzed, and the meteorological factors, the impact of the emergence law, and the uncertainty are also considered simultaneously. The short-term coupled forecasting model of power loads based on AI technology is proposed. Firstly, the cross-scale and meso-scale characteristics of short-term power loads are explored, and the nonlinear effects between different scales are analyzed in terms of peaks and valleys. Secondly, the modal decomposition method with the Aquila optimizer algorithm is employed to perform stable decomposition and key feature extraction on the strongly nonlinear load sequence. Subsequently, the Gated Recurrent Unit is used to capture the inherent temporal dynamics and nonlinear evolution patterns in the load data. At the same time, the Bidirectional Long Short-Term Memory neural network model optimized by the Coati optimization algorithm is introduced to achieve the deep integration of meteorological features and load data, effectively handling coupled nonlinear and random uncertainties. Finally, the results for the two types of data are reconstructed. Experimental results show that the proposed model can effectively handle various nonlinear problems. The model demonstrates high accuracy and robustness across different time scales and is helpful for the power industry to better cope with power market fluctuations and changes.