Model input selection for building heating load prediction: A case study for an office building in Tianjin

At present, the high-energy consumption of heating, ventilating, and air conditioning (HVAC) systems, which is caused by inefficient operation, is a matter of great concern. An accurate prediction of building load can help improve the operational efficiency of HVAC systems. In this work, the short-term heating load and ultra-short-term heating load prediction models are established with the purpose of predicting the heating load 24 h ahead and 1 h ahead, respectively. The short-term heating load prediction model can help management staff of buildings obtain hourly heating demand in advance and optimally arrange the operation of HVAC systems. The ultra-short-term heating load prediction model can be used for the prediction of a large load fluctuation, which may occur, and for the improvement of the operational safety of HVAC systems. Wavelet decomposition and reconstruction (WD), correlation analysis (CA), and principal component analysis (PCA) are employed to obtain reasonable model inputs, and two machine learning methods, namely the multilayer layer perceptron neural network (MLP) and the support vector regression (SVR), are used to establish the prediction models. The mean relative error (MRE) of the short-term heating load and ultra-short-term heating load prediction models reach 10.7% and 6.0%, respectively. The importance of the interior and exterior variables that influenced the building heating load is compared and the conclusion is that the building heating load is mainly influenced by exterior variables; however, the addition of the interior variables may help obtain more accurate heating load prediction models.