Under the background of big data

a series of power theft behaviors of residents not only cause significant economic losses but also bring incalculable security risks. According to the unique characteristics of residential users

a dual clustering detection model of power theft is proposed in this paper based on analysis of users' power consumption behavior. First

by establishing a preliminary diagnosis model of power theft based on improved K-means clustering

shortcomings of the original K-means algorithm

such as determination of the K value and initial cluster center

low calculation efficiency

and vulnerability to noise

are improved. Second

based on power consumption characteristics of residential users

their power consumption behavior patterns are classified

11 types of residential power consumption patterns are summarized

and the final set of power thieves is determined by curve similarity analysis. Simulation results show the proposed model can significantly improve detection rate

false detection rate

and other evaluation indicators for the dataset of residential users at the same time

which provides a reference for power theft detection in power supply companies.