A hybrid method of dynamic cooling and heating load forecasting for office buildings based on artificial intelligence and regression analysis

• Published in: Energy and buildings Vol. 174; pp. 293 - 308
• Main Authors: Zhao, Jing, Liu, Xiaojuan
• Format: Journal Article
• Language: English
• Published: Lausanne Elsevier B.V 01.09.2018; Elsevier BV
• Subjects: Air conditioners; Air conditioning; Artificial intelligence; Buildings; Case studies; Cooling; Cooling and heating load forecasting; Correlation analysis; Dynamic loads; Heating; Heating load; HVAC equipment; Input factors selection; Mathematical models; Model accuracy; Office buildings; Partial least squares regression; Regression analysis; Sensitivity analysis; Support vector machines; System effectiveness; Ventilation; Wavelet analysis; Wavelet transform; Wavelet transforms; Weather forecasting; Support vector machines; Cooling and heating load forecasting; Input factors selection; Wavelet transform; Partial least squares regression
• ISSN: 0378-7788; 1872-6178
• DOI: 10.1016/j.enbuild.2018.06.050

•Wavelet transform, SVM and PLS are used for building the load forecast model.•Sensitivity analysis and correlation analysis are used to select the input variables.•The significant meteorological factors to cooling and heating load are obtained.•Dynamic load predictions for different time horizons are evaluated. Dynamic cooling and heating load forecasting of heating, ventilation and air conditioning (HVAC) systems is a basis for optimizing the operation of HVAC systems and can contribute to achieving the effective management for the HVAC systems. This paper proposes a load forecasting method for office buildings based on artificial intelligence and regression analysis, including wavelet transform, support vector machines (SVM), and partial least squares regression (PLS). An office building located in Tianjin, China is taken as the building case study to validate the proposed model. For selecting the input variables, the methods of sensitivity analysis and correlation analysis are used. The results of different prediction horizons, mainly including 1 h ahead, 2 h ahead, 3 h ahead and 24 h ahead forecasting, are finally provided. In order to illustrate the accuracy improvement of the proposed model, the other three models are built to compare with the proposed model. Further, the influence of weather forecast precision on the proposed model is analyzed in this paper. The results indicate that the proposed method can realize dynamic load forecasting with high accuracy for different time horizons.