Prediction accuracy is an important index of power load forecasting. In order to enhance the accuracy of prediction, this paper presents support vector machine prediction model based on wavelet transform and the mutant fruit fly parameter optimization algorithm (WFOAAM-LSSVM). The load data are pretreated by wavelet transform, and the load curves are decomposed into different scales, in order to strengthen the regularity and randomness of historical data. In order to overcome the problems of low convergence precision and easily relapsing into local extremum in basic fruit fly optimization algorithm (FOA), and on the condition of basic FOA’s trapping in local extremum judging from the population’s fitness variance and the current optimal, it carries out optimal individual disturbance and Gauss mutation operation and optimizes mutated replicates again to jump out of local extremism and continue to optimize. The accuracy of prediction model is obviously enhanced. The next few days of historical load data of a certain area of Henan Province in 2015 is predicted by using WFOAAM-LSSVM, and the prediction results of support vector machine model and the particle swarm optimization model of support vector machine model are compared. The results show that WFOAAM- LSSVM has high precision in short term load forecasting, and it has a very good practical significance.