Data-driven modeling of building thermal dynamics: Methodology and state of the art

Abstract Data-driven approach is essential to the modeling of building thermal dynamics. It has been widely applied in building operation optimization, energy management, system performance evaluation, and so on. This present paper describes common concepts and fundamental theories of data-driven modeling within the context of building applications. Three types of data-driven models, namely transfer-function (TF) based, resistor-capacitor (RC) based, and artificial-intelligence (AI) based, are critically reviewed, including their formulations, interpretability of physical meanings, and prediction accuracy. Considerations on input and output variables are discussed. Conventional methods and techniques for model training and selection are also presented. Then, the three different models are illustrated through a case study of a real house using on-site monitored data. The case study suggests that the AI model generally outperforms the TF and RC models in predicting indoor temperatures while the RC model is the most appropriate for interpreting the physical behaviours of a building.