Abstract: In view the fact that a single forecasting model could hardly adapt to the load changes in different regions of Southwest China grid, in allusion to characteristics of annual load changes in various regions in Southwest China grid, it was proposed to adopt a load variable weight combination forecasting model, in which the comprehensive index evaluation was taken into account. Firstly, the improved grey relation index, the predictive validity index and the model redundancy test index were led in and taken as the basis for the model selection to form the base model library suitable to historical load changes in different regions. Secondly, the adaptive variable weight operator was introduced to conduct the combinational forecasting for base model to obtain regional annual load forecasting values of Southwest China grid. The consumption of electric power in Sichuan province, Chongqing city and Tibet Autonomous Region from 2006 to 2019 was utilized in the computing example for the test, and testing results show that the proposed forecasting method can effectively forecast the load change in Southwest China grid, and the obtained forecasting result by the proposed method is more accurate than those by optimal weight model, equal weight model and optimal single model.