A Multi‐Model Electricity Consumption Forecasting Framework for a Middle‐Income Household: Integrating Seasonal Decomposition, Explainability and Uncertainty Quantification
International Journal of Energy Research · 2026-01 · 2026 (1)
DOI: 10.1155/er/1494226 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Energy Research, online erschienen am 03.08.2026. Verfasst von Stephen O. Oladipo und 1 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Accurate electricity consumption forecasting is essential for effective energy management and grid stability. However, most existing studies focus narrowly on point prediction, often neglecting seasonal dynamics, model interpretability and uncertainty quantification. This study proposes a comprehensive forecasting framework applied to real‐world data collected from a middle‐income household in Gauteng Province, South Africa, between March 2019 and August 2021. The framework combines Seasonal‐Trend decomposition using Loess (STL), K‐Means clustering, Lorenz‐style cumulative power curves, and SHapley Additive exPlanations (SHAP) in a single workflow. Seven models comprising statistical, ensemble, and deep learning approaches were developed and compared. The goal is to find the most accurate and uncertainty‐aware approach for forecasting electricity consumption under different climatic conditions. Model performance was measured using root mean square error (RMSE), mean absolute error (MAE), standard deviation (SD), coefficient of variation of root mean square error (CVRMSE), and mean absolute deviation (MAD). Uncertainty was evaluated using prediction interval coverage probability (PICP) and mean prediction interval width (MPIW). STL decomposition revealed strong seasonal variability. Winter recorded the highest consumption levels. Lorenz curve analysis further revealed unequal distributional patterns of consumption across seasons. SHAP analysis revealed that ambient temperature and relative humidity were the most influential predictors. Among all models tested, SimpleRNN gave the best overall accuracy, recording the lowest test RMSE of 0.2177 kWh and the narrowest MPIW of 0.6553 kWh. Random forest (RF) produced the highest PICP. The results establish that integrating seasonal analysis, clustering and explainability within a single coherent framework significantly enhances forecast reliability and transparency of a household‐level electricity demand forecasting, advancing sustainable development goals (SDGs) through evidence‐based energy planning.
Zitieren
APA
Oladipo, S. O., & Akuru, U. B. (2026). A Multi‐Model Electricity Consumption Forecasting Framework for a Middle‐Income Household: Integrating Seasonal Decomposition, Explainability and Uncertainty Quantification. International Journal of Energy Research, 2026(1). https://doi.org/10.1155/er/1494226
Vancouver
Oladipo SO, Akuru UB. A Multi‐Model Electricity Consumption Forecasting Framework for a Middle‐Income Household: Integrating Seasonal Decomposition, Explainability and Uncertainty Quantification. International Journal of Energy Research. 2026;2026(1). doi:10.1155/er/1494226
BibTeX
@article{Oladipo2026,
author = {Oladipo, Stephen O. and Akuru, Udochukwu B.},
title = {A Multi‐Model Electricity Consumption Forecasting Framework for a Middle‐Income Household: Integrating Seasonal Decomposition, Explainability and Uncertainty Quantification},
journal = {International Journal of Energy Research},
year = {2026},
volume = {2026},
number = {1},
doi = {10.1155/er/1494226}
}
EUVIMED