An artificial intelligence (AI)-driven method for forecasting cooling and heating loads in office buildings by integrating building thermal load characteristics

Due to the thermal inertia of building envelope and random uncertainty of occupant behaviors, real-time and accurate forecasting for building cooling and heating loads is not easy to implement. It requires not only favorable modeling methods, but also attention to the intrinsic cause analysis of loads. However, most previous studies have focused more on the improvement of forecasting algorithms, lacking the comprehensive analysis and scientific research from the load formation mechanism. To address this, a novel artificial intelligence (AI)-driven forecasting method integrating building thermal load characteristics, is proposed. This modeling approach takes into account both strong reflection ability of resistance-capacitance (RC) for the thermophysical characteristics of building envelope and the superior handling capability of dynamic Bayesian network (DBN) for occupant uncertainties. Therewith, the independent building envelope model and occupant behavior model are established. Coupling the predicted values of these two sub-models, the building total load forecasting model is established using fuzzy radial basis function (RBF) neural network. A case study of an office building in Tianjin is utilized to validate the proposed method, revealing excellent predictive performance for both the sub-models and the total load model. The proposed method can effectively predict thermal load both from the formation mechanism of internal and external disturbances. It was found that the AI-driven model incorporating building thermal load characteristics, achieves significantly higher precision compared to the general data-driven model, with a reduction of 14.74% in coefficient of variation of root mean squared error in summer and 59.78% in winter. •>A new AI-based method incorporating building thermal load characteristics is proposed.•Independent models for building envelope and occupant behavior are established.•Time-series weather and occupant parameters are selected as input parameters.•Two independent sub-models are effectively coupled for improved performance.•Thermal load caused by external and internal disturbances is accurately predicted.