Toward explainable and interpretable building energy modelling

Abstract

Building energy modelling is essential for the operation and optimization of energy systems. Existing efforts have focused on the model fidelity, which however should not be the only concern. Due to the lack of adequate understanding and trust in the model, many accurate energy models have not been deployed in the real world. In this article, we introduce explainability and interpretability into the building energy model. We first use partial dependence plots to explain and quantify the importance of features at a fine-grained level and reveal different importance change patterns with different feature values. We also use a surrogate-based approach to interpret the internal mechanism and decision-making process of the model. We show that the local mechanism interpretation is feasible when the surrogate is both accurate and explainable. Explanation and interpretation are visualized, to provide system users with intuitive and informative insights with multi-disciplinary discussions. Our research promotes the operation and optimization of building energy systems and supports the widespread adoption of energy systems based on machine learning.