Abstract: A hybrid prediction model based on support vector machine (SVM) was developed for building energy consumption prediction of rural residences in hot summer and cold winter regions. An initial dataset was constructed by collecting building parameters, meteorological parameters, behavioral parameters, equipment parameters, and annual energy consumption data from typical rural residences. A progressive screening framework incorporating significance analysis, collinearity analysis, random forest sensitivity analysis, and backward stepwise regression method was employed to select 10 key variables from 29 candidate variables, significantly reducing model complexity. A hybrid SVM prediction model was established by integrating theoretical calculation data from white-box models with measured data from black-box models, and a joint strategy combining grid search and cross-validation was adopted to optimize key model parameters for performance enhancement. The validation results demonstrate that the proposed model achieves a coefficient of determination (R²) of 0.914 and a coefficient of variation of root mean square error (CVRMSE) of 0.163, maintaining prediction accuracy while realizing optimal balance in model complexity. The variable screening and data fusion strategies developed in this study are proved to effectively address the prediction challenges caused by missing design parameters and insufficient energy consumption data in rural residences of this region.