Abstract: Based on the Sandia method,this paper uses random forest to extract the importance of features instead of expert judgment to generate typical meteorological years（TMY）for three different energy systems（one fully air-conditioned building and two renewable energy systems）in Beijing. Energy Plus is used for simulation and analysis. The results show that this method can quantitatively generate the weight factor subsets suitable for different energy systems,which can be used to generate the corresponding TMY. The representativeness of TMY can be further improved by dividing time periods according to the characteristics of different energy systems and extracting corresponding weight factor subsets.