In the design stage of green buildings, there is a lack of a set of systematic performance-based evaluation methods for high-performance cold sources to evaluate whether the selected cold sources match the load characteristics of buildings, which leads to inappropriate selection of cold sources in the design stage. Due to this problem, this study proposes a multi-dimensional performance-based evaluation method. Data were collected from a public building in the cold region of China, and a set of calculation tools were developed to simulate the actual operation performance of cold source units. Considering its energy and economic efficiency, the multi-dimensional comprehensive evaluation index is put forward. Finally, taking a green building located in Tianjin, China as an example, the feasibility analysis of the multi-dimensional performance-based evaluation method is carried out. Results indicate that: 1) the prediction error of hourly performance coefficient prediction model based on Fuzzy clustering and Back-Propagation (BP) neural network algorithm is less than 10%, which can accurately predict the actual operation energy efficiency of cold source; 2) compared with the actual cold source, the improved cold source of the case building has the seasonal power consumption decreased by 34.0%, and the average load rate increased by 16.9%, which proves the feasibility of multidimensional performance-based evaluation method. The method proposed in this study can provide reference and help for the selection and performance evaluation of cold source units in green buildings at the design stage.