Abstract: In order to improve the accuracy of short-term power load forecasting which is sensitive to external factors,a short-term power load forecasting method based on improved ABC optimized density peak clustering and multiple kernel extreme learning machine is proposed. A short-term power load forecasting model is constructed that integrates feature extraction,artificial bee colony algorithm（ABC）,density peak clustering（DPC）and kernel extreme learning machine（KELM）. In view of the low convergence efficiency of ABC,a new honey source search and bee evolution method is designed to improve the global optimization capability of ABC. Aiming at the deficiency of artificial setting of DPC truncation distance and cluster center,Bonferroni exponential function and cluster center truncation index are defined,and the improved ABC is applied to DPC parameter optimization process to realize the best cluster analysis of DPC. Aiming at the problems of weak regression ability and difficult parameter selection of KELM,a multi-core weighted KELM is designed,and the improved ABC is used for parameter optimization to improve the prediction accuracy of limit learning machine. The simulation results show that the proposed short-term power load forecasting method is more effective,and the average error is reduced by 8.8%～39.8%.