Explainable AI framework for reliable and transparent automated energy management in buildings

Abstract

The increasing integration of Artificial Intelligence (AI) into Building Energy Management Systems (BEMS) is revolutionizing energy optimization by enabling real-time monitoring, predictive analytics, and automated control. While these advancements improve energy efficiency and sustainability, the opacity of AI models poses challenges in interpretability, limiting user trust and hindering widespread adoption in operational decision-making. Ensuring transparency is crucial for aligning AI insights with building performance requirements and regulatory expectations. This paper presents EI-Build, a novel Explainable Artificial Intelligence (XAI) framework designed to enhance the interpretability of intelligent automated BEMS. EI-Build integrates multiple XAI techniques, including Shapley Additive Explanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), Anchors, Partial Dependence Plots, Feature Permutation Importance, and correlation-based statistical analysis, to provide comprehensive explanations of model behavior. By dynamically tailoring the format and depth of explanations, EI-Build ensures that insights remain accessible and actionable for different user profiles, from general occupants to energy specialists and machine learning experts. A case study on photovoltaic power generation forecasting applied to a real BEMS context evaluates EI-Build’s capacity to deliver to deliver both global and local explanations, validate feature dependencies, and facilitate cross-comparison of interpretability techniques. The results highlight how EI-Build enhances user trust, facilitates informed decision-making, and improves model validation. By consolidating diverse XAI methods into a single automated framework, EI-Build represents a significant advancement in bridging the gap between complex AI energy models and real-world applications.

Topics & Concepts