Comparison of four algorithms based on machine learning for cooling load forecasting of large-scale shopping mall

Abstract Short-term forecasting of air-conditioning cooling load of shopping mall is hard with much accuracy due to its chaotic and non-linear characteristic. Four forecasting algorithms based on machine learning are illustrated in this paper including Chaos-SVR, WD-SVR, SVR and BP, whose predicting performances are compared. For Chaos-SVR, the selection of lag time and embedding dimension during phase space reconstruction are described, while for WD-SVR, the modeling process of DB2 is proposed. Furthermore, the optimization of the hyper-parameters for SVR model is also presented. It’s shown that these four approaches have different characteristics which are suitable for different types of cooling load time series.