In a power system

the information on power consumption patterns and electricity demand levels is recorded in industrial load curves

part of which

however

will be abnormal because of unexpected interference. Therefore

a method based on nonparametric regression theory is proposed to detect and correct the outliers in industrial load curves. First

for the lateral continuity of load data in time sequence

a fuzzy statistical method is employed for classifying the load curves by consumption patterns. The load data sets are classified into two data sets

one is of the basic consumption patterns and the other of special patterns. Then

considering the longitudinal continuity of load values in various time intervals

the nonparametric regression analysis method is used to estimate the center vector based on the data set of basic patterns. With the center vector

the outlier boundaries are achieved to detect all the outliers. Finally

the mapping of load levels is modeled to carry out the outlier correction in accordance with the weighted average method. The actual industrial load data are adopted to test the proposed method. The result shows the effectiveness of the proposed method.