In view of the shortcomings of the current electricity theft detection model, such as requiring a large number of samples, complex training process, and the detection performance needs to be improved, a detection model based on local matrix reconstruction is proposed. The principal component analysis is introduced to analyze the distribution characteristics of the differences between adjacent data samples; the Euclidean distance and data distribution characteristics are used to detect the differences between different power consumption modes; the local outlier score is introduced to determine the power theft samples. The residential power load data provided by a power company is used for experimental analysis, and the results show that the area under the ROC curve of the proposed model is 0.846 3, which has a relatively stable detection threshold and stronger robustness, and has a certain reference for the electricity theft detection and evaluation of distribution network.