Prediction method for residential air-condition power load in summer: Extreme weather scenario prediction based on improved informer model

To improve the forecasting accuracy of residential electricity loads under summer extreme weather conditions, this study proposed an Informer-based load forecasting framework optimized by the Phased Enhanced Sand Cat Swarm Optimization algorithm (PESCSO). The method combined electricity consumption curve clustering with meteorological disaster warning signals to divide summer electricity usage into scenario-specific categories, transforming the global forecasting task into multiple scenario-dependent modeling problems. To overcome the limited efficiency and stability of hyperparameter optimization in complex search spaces, PESCSO was developed by enhancing the initialization strategy, introducing a phased search mechanism, and regulating predatory intensity, thereby improving global exploration, local exploitation, and solution stability. Its effectiveness was validated using the CEC 2017 benchmark functions. PESCSO was then applied to adaptively optimize key hyperparameters of the Informer under different scenarios, improving the consistency between model structure and scenario-dependent load characteristics. Case study results showed that PESCSO-Informer consistently outperformed baseline models under summer extreme weather conditions. Compared with the standard Informer, the mean absolute error (MAE) was reduced by 36.5% and 46.4% under Heatwave red and yellow warning conditions, with R 2 increased to 0.971 and 0.963, respectively, while under Rainstorm warning conditions, the MAE was further reduced by 59.6% with a notable reduction in root mean square error (RMSE). Statistical significance tests confirmed that these improvements were significant at the α = 0.05 level, demonstrating the superior accuracy and robustness of the proposed method under extreme weather load fluctuations. • Developed scenario-specific Informer models for air conditioning load forecasting. • Proposed Phased-Enhanced Sand Cat Swarm Optimization for hyperparameter optimization. • Analyzed summer extreme scenarios, including heatwaves, normal days, and rainstorms. • Validated the method using residential electricity load data from Beijing, China.