There are typically two methods for building energy modeling, which are physical based models and data-driven models. However, the simulation results of physical base energy models often deviate greatly from real cases. While the traditional data-driven energy models are more reliable but only applicable to buildings with historical data record. In this paper, we propose a framework of hybrid building energy models developed based on heterogeneous database which contains integration of building formation, field-tested energy data and simulated energy data. This hybrid energy forecasting model is able to predict building energy in the absence of energy record of the target building. The framework consists of three parts: key variables identification, data integration, heterogeneous database and hybrid energy forecasting model development. A chiller energy forecasting model is developed as a case study to demonstrate the feasibility of this framework. The mean cross testing CV-RMSE and R2 of chiller energy forecasting energy model are 0.17 and 0.86 respectively which are fairly acceptable when historical energy data of target building is not available.
Keywords building energy forecasting, data integration, Bayesian inference