Enhanced data-driven method for design cooling load calculation averting the overestimation due to design weather data

Design cooling load is always overestimated when using design weather data (DWD), resulting in oversized air-conditioning (AC) system, low operating energy efficiency and highly initial investment. In this study, a data-driven model is developed based on explainable feature selection (EFS) and multi-task learning (MTL), termed as the EFS-MTL model, for averting the overestimation. Design cooling loads under various non-guarantee rates (such as 0.4 %, 1 %, 2 % and 50 h) were calculated using heat balance method for a large number of sample rooms combined by building/room parameters with multi-year hourly-recorded weather data (HWD) of a city. The EFS-MTL model was then trained by the dataset of the sample rooms with the room parameters (as inputs) and the corresponding design cooling loads (as outputs). To illustrate the EFS-MTL model's capacity in eliminating the overestimation, the EFS-MTL model was trained with the datasets created by HWD of Beijing, Changsha and Guangzhou. The design cooling loads calculated by the EFS-MTL model and DWD were compared. Results show that the median relative deviations of the EFS-MTL model range from −1.44 % to 0.27 %. These results demonstrate that the EFS-MTL model provides an effective approach to correctly and fast calculate design cooling load of AC systems without DWD.