Causally Certified Energy Forecasting for Industrial Buildings: Beyond Accuracy toward Trustworthy Operational Decisions

Dongsoo  Moon; Gwangsu Lee

doi:10.29284/51ys0784

Authors

Dongsoo Moon Department of Computer Education, Sungkyunkwan University, Seoul 03063, Republic of Korea
Gwangsu Lee School of Computer Software, Daegu Catholic University, Gyeongsangbuk-do 38430, Republic of Korea

DOI:

https://doi.org/10.29284/51ys0784

Keywords:

causal certification; intervention analysis; trustworthy machine learning; explainable AI (XAI); composite scoring; energy informatics; industrial buildings; peak demand; short-term load forecasting; tail risk

Abstract

Electricity demand in industrial buildings exhibits nonstationary behavior, and errors during peak regimes can destabilize operational decisions. This study presents a fair, operation-oriented benchmarking of statistical, machine learning, and deep learning fore- casters using a standardized preprocessing and validation protocol. Using 15-min data from 30 sites, 1- and 24-h-ahead forecasts were evaluated with a weighted CompositeScore integrating MAE, tail risk (P95AbsErr), and peak-regime reliability (PeakMAE95). Among 11 candidate models, gradient-boosted tree ensembles provided the most stable accuracy–risk trade-off. Compared with the persistence baseline, LightGBM reduced MAE and PeakMAE95 for the 1-h horizon by 19.6% and 32.1%, respectively, while a robust XGBoost variant achieved reductions of 30.5% and 36.1%, respectively, for the 24-h horizon. To ensure operational trustworthiness, a Causal Certification framework was applied to SHAP analysis. The selected models demonstrated high causal agreement between feature importance and intervention based sensitivity, with Spearman rank correlations of

References

[1] Antonesi, G.; Cioara, T.; Anghel, I.; Michalakopoulos, V.; Sarmas, E.; Toderean, L. A systematic review of transformers and large language models in the energy sector: Towards agentic digital twins. Appl. Energy 2025, 401, 126670. DOI:10.1016/j.apen- ergy.2025.126670.

[2] Baron, S. Explainable AI and causal understanding: Counterfactual approaches considered. Minds Mach. 2023, 33, 347–377. DOI:10.1007/s11023-023-09637-x.

[3] Bhandary, A.; Dobariya, V.; Yenduri, G.; Jhaveri, R.H.; Gochhait, S.; Benedetto, F. Enhancing household energy consumption predictions through explainable AI frameworks. IEEE Access 2024, 12, 36764-–36777. DOI:10.1109/ACCESS.2024.3373552.

[4] Chen, Y.; Gong, W.; Obrecht, C.; Kuznik, F. A review of machine learning techniques for building electrical energy consumption prediction. Energy AI 2025, 21, 100518. DOI:10.1016/j.egyai.2025.100518.

[5] Chen, Z.; Xiao, F.; Guo, F.; Yan, J. Interpretable machine learning for building energy management: A state-of-the-art review. Adv. Appl. Energy 2023, 9, 100123. DOI:10.1016/j.adapen.2023.100123.

[6] Fan, F.; et al. Transmission power-oriented forecasting towards stability-constrained operation of renewable energy power plants with energy storages. CSEE J. Power Energy Syst. 2025.

[7] Guo, H.; Huang, B.; Wang, J. Probabilistic load forecasting for integrated energy systems using attentive quantile regression temporal convolutional network. Adv. Appl. Energy 2024, 14, 100165. DOI:10.1016/j.adapen.2024.100165.

[8] Koukaras, P.; Mustapha, A.; Mystakidis, A.; Tjortjis, C. Optimizing building short-term load forecasting: A comparative analysis of machine learning models. Energies 2024, 17, 1450. DOI:10.3390/en17061450.

[9] Neubauer, A.; Brandt, S.; Kriegel, M. Explainable multi-step heating load forecasting: Using SHAP values and temporal atten- tion mechanisms for enhanced interpretability. Energy AI 2025, 20, 100480. DOI:10.1016/j.egyai.2025.100480.

[10] Sultana, N., 2025, Balancing Accuracy and Efficiency in Multi-Step Building-Level Electric Load Forecasting: Deep Learning vs. Classical Time-Series Models. Energy and Built Environment.

[11] Teixeira, B.; Carvalhais, L.; Pinto, T.; Vale, Z. Explainable AI framework for reliable and transparent automated energy management in buildings. Energy Build. 2025, 347, 116246. DOI:10.1016/j.enbuild.2025.116246.

[12] Van Zyl, C.; Ye, X.; Naidoo, R. Harnessing eXplainable artificial intelligence for feature selection in time series energy forecasting: A comparative analysis of Grad-CAM and SHAP. Appl. Energy 2024, 353, 122079. DOI:10.1016/j.apenergy.2023.122079.

[13] Xu, C.; Sun, Y.; Du, A.; Gao, D. Quantile regression based probabilistic forecasting of renewable energy generation and building electrical load: A state of the art review. J. Build. Eng. 2023, 79, 107772. DOI:10.1016/j.jobe.2023.107772.

[14] Yuan, M.; Xie, J.; Liu, C.; Xu, Z. Short-term load forecasting for an industrial building based on diverse load patterns. Energy 2025, 334, 137481. DOI:10.1016/j.energy.2025.137481.

[15] Zhang, Z.; Gang, W.; Zhang, Y. A peak shaving strategy based on occupant allocation for campus buildings under extreme hot weather conditions. Energy Build. 2025, 345, 116119. DOI:10.1016/j.enbuild.2025.116119.

[16] Zhao, W.; Fan, L. Short-term load forecasting method for industrial buildings based on signal decomposition and composite prediction model. Sustainability 2024, 16, 2522. DOI:10.3390/su16062522.

[17] Zhen, C.; Niu, J.; Tian, Z.; Lu, Y.; Liang, C. Risk-averse transactions optimization strategy for building users participating in incentive-based demand response programs. Appl. Energy 2025, 380, 125009. DOI:10.1016/j.apenergy.2024.125009.

[18] Zheng, P.; Zhou, H.; Liu, J.; Nakanishi, Y. Interpretable building energy consumption forecasting using spectral clustering algorithm and temporal fusion transformers architecture. Appl. Energy 2023, 349, 121607. DOI:10.1016/j.apenergy.2023.121607.