Forecasting Energy Demand in Buildings: The Case for Trees over Deep Nets

Bujoreanu, Dan Alexandru and Chiagoziem Onwuegbuche, Faithful (2026) Forecasting Energy Demand in Buildings: The Case for Trees over Deep Nets. In: Artificial Intelligence and Cognitive Science. AICS 2025. Communications in Computer and Information Science (2950). Springer, Cham, Dublin, Ireland, pp. 263-274. ISBN 978-303225808-3

Full text not available from this repository.

Abstract

Reliable short-term forecasts of building energy demand are essential for operating flexible, low-carbon power systems. They help system operators and facility managers to schedule resources and avoid costly balancing actions. In practical terms, better predictions improve intraday planning and support the dispatch of behind-the-meter storage.

In this paper we evaluate forecasting methods on a heterogeneous public-building portfolio from Drammen, Norway. A leakage-safe, feature-engineering-first pipeline enables a comparison between tree ensembles, LSTM, CNN–LSTM and a Temporal Fusion Transformer (TFT), and assesses a simple stacking approach. While sequence models are configured for multi-step outputs, we standardise the evaluation on the critical one-step-ahead (H+1) forecast to ensure a direct comparison against non-sequence baselines. Across the held-out test window, tree models provide the best balance of predictive performance and speed, consistently surpassing deep networks while training in seconds rather than hours. The stacking ensemble models do not improve on the best single tree. The study shows that careful feature engineering, paired with tree ensembles, is a strong baseline for building demand forecasting.

Item Type:	Book Section
Uncontrolled Keywords:	Deep learning; Energy forecasting; Ensembles; Tree-based models
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science H Social Sciences > HC Economic History and Conditions > Natural resources > Power resources > Energy consumption Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Staff Research and Publications
Depositing User:	Tamara Malone
Date Deposited:	18 Sep 2026 14:22
Last Modified:	18 Sep 2026 14:22
URI:	https://norma.ncirl.ie/id/eprint/9933

Actions (login required)

View Item