Abstract: In the context of global warming and frequent extreme weather events, temperature risk has become an increasingly prominent threat to economic and financial stability. Existing temperature models face limitations such as high computational demands, slow updates, and difficulty capturing extreme temperatures, making them inadequate for meeting the timeliness requirements of the economic and financial systems in interannual-scale risk quantification and stress testing. Based on datasets including the Yangtze river delta temperature index, surface observations, NCEP reanalysis, and CMIP6 model projections, this study proposes a computationally efficient temperature risk stress testing model. By improving the Ornstein-Uhlenbeck (O-U) model and incorporating key physical drivers using the LSTM method, the model enhances the description of extreme temperatures. Empirical analysis based on the Yangtze river delta temperature index from 2022 to 2024 demonstrates that the model effectively improves long-term prediction accuracy and extreme temperature simulation capability while maintaining low computational costs, with particularly strong performance in spring and summer. This model can serve as a flexible and efficient temperature risk stress testing tool for various sectors such as banking, insurance, and energy, supporting daily loss estimation and scenario simulation.