Due to the difficulty of load forecasting of smart grid with distributed power supply, this paper proposes a smart grid load forecasting method based on k-means algorithm and PSO-LSSVM. K-means algorithm is adopted to cluster power load users to reduce the influence of season on load prediction. PSO algorithm is proposed to optimize LSSVM algorithm to improve the calculation accuracy and reduce the calculation time of LSSVM algorithm. PSO-LSSVM algorithm is adopted for load prediction of power users after k-means algorithm clustering, and PSO-LSSVM algorithm is adopted to predict the output of distributed power supply. The power dispatcher adjusts the load according to the required load and the output of the distributed power supply, so as to realize the safe and stable operation of the power grid. Simulation results verify that the proposed PSO-LSSVM has higher prediction accuracy than traditional LSSVM. The proposed method can effectively improve the stability of power system.