An improved input variable selection method of the data-driven model for building heating load prediction

Abstract

In building energy conservation, HVAC systems have great potential, and load prediction plays an important role in optimizing system control and improving system energy efficiency. Therefore, improving the accuracy of load prediction is of great significance. This paper aims to improve the prediction accuracy of the data-driven model of load prediction through optimizing the input variable selection. Factors such as outdoor temperature, wind speed, solar radiation, indoor temperature and user behavior have different mechanisms of action to influence the heating load, and there exists difference among the influence degree of each factor on the heating load at each moment. Therefore, the optimal input variable used by the heating load prediction models at different moments will be different. Based on this, this paper proposes to establish an input variable selection method based on hourly correlation analysis. Taking an office building as the case building, this paper uses the back propagation neural network (BPNN) model and the support vector regression (SVR) model to compare the selection methods of hourly correlation analysis and non-hourly correlation analysis. The results show that the influence degree of solar radiation intensity and the historical load at the moment of 24 h ahead on the load at each moment have great difference. Compared with non-hourly correlation analysis, hourly correlation analysis can reduce the root mean square error (RMSE) of the prediction results of the BPNN model and the SVR model by an average of 10.4% and 12.9%, respectively, and increase the coefficient of determination (R2) by an average of 11.7% and 7.9%, respectively, while all the calculation time is about 0.4 s. This shows that the heating load prediction method based on hourly correlation analysis can improve the prediction accuracy of model while keeping the prediction rate unchanged.