A method for the detection of electricity theft behavior based on Xavier weight initialization

Abstract

With the advancement of machine learning technology, intelligent detection of electricity theft has become a more widely used technical means. The detection of electricity theft behavior includes data collection, data preprocessing, establishment of electricity theft model and online detection process of electricity theft, in which the establishment of the electricity theft model is modeled by neural network, and the random initialization method and Xavier weight initialization method are used in the modeling process. Experiments show that the Xavier weight initialization method has a good effect on error histogram and network performance, and the mean squared normalization error of 0.0686 is lower than that of the stochastic optimization method.