A Novel Multidimensional Power Load Forecasting Model with CTMSA-GMSC Integration

View at publisher Summarise with VIRAT
Authors: Zhangyi Zhao, Jing Fang, Peisong Wang, Han Song
Cited by: 0
References: 10

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Precision in power load forecasting is fundamental to modern energy systems but remains challenging due to the complex interplay of multidimensional factors and multi-scale temporal dynamics. Traditional deep learning approaches often struggle to model these intricate couplings and long-term dependencies while strictly adhering to causal integrity. To address these limitations, this paper proposes a novel long-term forecasting architecture named CTMSA-GMSC, which integrates a Causal Temporal Multi-Scale Attention (CTMSA) module with a Gated Multi-Scale Convolution (GMSC) backbone. The CTMSA module employs a Periodicity-Aligned Hierarchical Windowing strategy to adaptively capture temporal patterns-ranging from fine-grained fluctuations to seasonal trends-while enforcing strict causal constraints to prevent information leakage. Complementing this, the GMSC module utilizes dual-branch large-kernel convolutions and a gated mechanism to robustly model complex non-linear interactions among high-dimensional multivariate features. Experimental evaluations on the Panama Electricity Load dataset demonstrate that the proposed framework significantly outperforms state-of-the-art baselines, including the Informer and Standard Transformer, achieving a reduction in Mean Squared Error (MSE) of approximately 32% and 41 %, respectively. Ablation studies further verify that the integration of GMSC is essential for extracting deep inter-feature relationships, ensuring superior robustness and accuracy in longterm forecasting tasks.

Fields of Study & Subjects

Energy Load and Power Forecasting Time Series Analysis and Forecasting Traffic Prediction and Management Techniques

Publication details

Journal / source
International Conference on Electrical Automation and Artificial Intelligence (ICEAAI)
Pages
491–496
Publication date
16 January 2026
Language
EN