In order to improve short-term power load forecasting accuracy.A LSSVM short-term power load forecasting model based on search algorithm optimization is proposed.The Monte Carlo rule of simulated annealing algorithm is introduced to improve the algorithm, which improves the stability of the algorithm.The prediction precision is improved with the use of wavelet threshold denoising of power load data, which reduce the influence of uncertain factors on load forecasting.The actual historical load data of a regional power grid in Sichuan are selected for analysis and prediction,and compared with PSO-LSSVM and LSSVM prediction models.Numerical example shows that the BAS-LSSVM prediction model proposed in this paper has improved the prediction accuracy by 1.55% compared with LSSVM.Compared with PSO-LSSVM, the running time of the algorithm is reduced by 70%, and the algorithm is more stable,which prove the practicability and effectiveness of the method.