The invention discloses a power load abnormal data recognition and modification method based on nonparametric regression analysis. The method comprises the steps of 1, performing power utilization mode classification on power load data to obtain a common power utilization mode data set and a special power utilization mode data set; 2, extracting a load feature value at each moment from the obtained common power utilization mode data set by adopting a nonparametric regression analysis method; 3, forming an abnormal data field by using the extracted load feature values according to the selected confidence level; 4, performing load abnormal data recognition on load data in the common power utilization mode data set and the special power utilization mode data set by using the abnormal data field formed in step 3; and 5, modifying the recognized load abnormal value by using an improved introduced load level mapping relation and a weighted mean method considering the influence of feature values. The method can recognize and modify power load abnormal data including big industrial power load data, and simultaneously can overcome the defect of the load abnormal data recognition and modification theory on the aspect of power load data processing.