Electricity theft detection method based on power fluctuation characteristics

In recent years, high loss line power theft detection technology has achieved certain results, but in the actual line operation state, there may still exist small amount of power theft users. Since these users steal small amounts of electricity, they are easy to be ignored by inspectors, leading to misjudgment of the existing power theft detection technology, so it is necessary to consider the user's own metering data changes in the power theft detection technology. In this study, a Quantile Regression Neural Network （QRNN） power theft detection method for small amounts of power theft is proposed to take advantage of the fact that the power consumption of dedicated variable users is characterized by continuity. First, the cycle difference between normal users and power theft users is analyzed to refine the cycle fluctuation characteristics of user information; then, based on the refined user characteristics, the power sequence is reconstructed and the fitted value is obtained through the QRNN model, and a threshold is set adaptively based on the mean square error of the power sequence before and after the reconstruction, and the power theft identification is carried out by judging whether the error is more than the adaptive threshold. Finally, the proposed method is tested and verified by the collected actual power theft data, and the results show that the method has excellent performance.