Study of the Whole Building Energy Use Inverse Modeling Performance through Support Vector Machine Regression

The performance of a single-variate support vector machine (SVM) was investigated as a whole-building energy use nonlinear inverse modeling tool. Although the SVM is generally employed with multiple attributes, given the benefits of using a single independent variable and for a fair comparison with another conventional building energy inverse modeling method, the change-point regression, only a single attribute was used as an independent variable. Numerical experiments were conducted based on 32 samples of actual chilled water (CHW) and heating hot water (HHW) use in buildings. The outdoor air temperature and outdoor air enthalpy were used as the main regressors. For daily data, although the average performance of SVM models was only slightly better than that of change-point (CP) models, the difference was more remarkable in some samples than in others. However, for monthly data, there was no improvement of performance.