The abnormal power consumption of power users is the main factor that leads to the high line loss rate and the increase of power supply cost of power supply enterprises. Therefore, under the current situation of smart grid construction, the development of advanced metering infrastructure (AMI) and the popularization of smart electricity meters, how to select advanced detection algorithm to realize the effective detection of abnormal power consumption of power users has become the focus of scholars at home and abroad. On this basis, an abnormal electricity fusion detection model based on RF-GBDT is constructed. In this scheme, data cleaning and missing value filling are used to preprocess the original data set, principal component analysis (PCA) and Pearson correlation coefficient method are used to reduce the dimension of the original data, and feature correlation analysis. The random forest (RF) algorithm is used to randomly select the modeling data set and data feature quantity to obtain a mature classifier. Finally, the feature quantity is input into the gradient to improve upgrade the decision tree (GBDT) to complete the detection and classification of the test data set. The experimental results show that the fusion detection model based on RF-GBDT has better recognition speed and higher detection accuracy.