A Calendar-Aware Frequency-Decoupled Framework for Day-Ahead Substation Load Forecasting Using SHAP-Based Interpretation
Applied Sciences · 09.06.2026 · 16 (12) · 5815
DOI: 10.3390/app16125815 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 09.06.2026. Verfasst von Beixuan He und 5 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: State Grid Jiangsu Electric Power Company.
Abstract (Originalsprache)
Accurate substation-level Short-Term Load Forecasting (STLF) is essential for secure day-ahead power-system operation, yet localized demand is often affected by meteorological variation and discrete calendar shifts such as statutory holidays and makeup workdays. At this spatial scale, end-to-end forecasting models may over-smooth abrupt local changes and fail to represent peaks and valleys accurately. To address this issue, this study proposes a Calendar-Aware Frequency-Decoupled Framework (CA-FDF) for 24 h ahead substation load forecasting. The load series is decomposed by the Discrete Wavelet Transform (DWT), and the low-frequency component is tracked by a regime-aware Recursive Least Squares (RLS) filter. The residuals are then refined through explicit calendar-state encoding and day-ahead weather forecasts. A Multi-Layer Perceptron (MLP) learns latent weather representations, while SHapley Additive exPlanations (SHAP) interpret calendar- and weather-related effects. Experiments on hourly operational data from 29 anonymized substations in East China show that CA-FDF achieves a Mean Absolute Percentage Error (MAPE) of 1.92% and outperforms representative baselines under the same day-ahead setting. The results indicate that frequency-decoupled residual refinement improves localized load forecasting, with calendar-aware correction contributing the largest gain.
Förderung
- State Grid Jiangsu Electric Power Company
Zitieren
APA
He, B., Cai, C., Diao, R., Han, J., Qian, B., & Wu, S. (2026). A Calendar-Aware Frequency-Decoupled Framework for Day-Ahead Substation Load Forecasting Using SHAP-Based Interpretation. Applied Sciences, 16(12), 5815. https://doi.org/10.3390/app16125815
Vancouver
He B, Cai C, Diao R, Han J, Qian B, Wu S. A Calendar-Aware Frequency-Decoupled Framework for Day-Ahead Substation Load Forecasting Using SHAP-Based Interpretation. Applied Sciences. 2026;16(12):5815. doi:10.3390/app16125815
BibTeX
@article{He2026,
author = {He, Beixuan and Cai, Chao and Diao, Ruisheng and Han, Jun and Qian, Bohan and Wu, Siheng},
title = {A Calendar-Aware Frequency-Decoupled Framework for Day-Ahead Substation Load Forecasting Using SHAP-Based Interpretation},
journal = {Applied Sciences},
year = {2026},
volume = {16},
number = {12},
pages = {5815},
doi = {10.3390/app16125815}
}
EUVIMED