A mid-long term temperature-lowering load forecasting method based on support vector machine (SVM) and multiple regression is proposed. A variety of socio-economic data is taken as the input parameter of the SVM model and the maximum temperature-lowering load as the output parameter. Grid search algorithm is used to optimize the parameters of SVM; liner, quadratic and cubic regression are used in multiple regression; finally, the two methods are integrated using optimal combined forecasting method. The SVM and multiple regression model is tested using 2008-2011 data as training sample, and 2012-2014 data as test sample. The 2015 and 2020 annual maximum temperature-lowing load are forecasted as well. The result shows that the error between the predicted value and the real value can be kept in 5%, which shows the model is effective to do mid-long term temperature-lowering load forecasting. Currently, the prediction model has been applied in Guangdong power grid.