Abstract: With the frequent occurrence of extreme weather, the electricity consumption of temperature-sensitive loads is increasing year by year. As a high-quality regulation resource of virtual power plant (VPP), temperature-sensitive loads urgently need to be analyzed for the impact of meteorological changes on them. Due to the influence of abnormal weather such as extreme high temperature and large-scale cold waves, temperature-sensitive loads fluctuate violently. Conventional analysis and prediction methods are not adaptable to the extreme meteorological scenarios. Aiming at the problem of insufficient sample data and prediction accuracy of temperature-sensitive loads under cold wave weather, this paper proposes a daily maximum load prediction method for temperature-sensitive loads under the condition of small sample in cold wave weather. In this method, the TimeGAN is used to expand the small sample data during the cold wave period, and then the CNN-LSTM network is used to predict the daily maximum load during the cold wave period. Finally, the model is verified by the load data of a province in China during the winter period in the past two years. The results show that the prediction results of the proposed model are better than those of other models, with the prediction accuracy of the daily maximum load on the verification set being 99.5%.