一种基于离群算法的窃电行为检测的研究

针对窃电行为现场查证具有难以克服的现实困难，提出一种基于离群数据挖掘的窃电行为检测方法。该离群算法基于密度聚类算法，采用基于用电量波动的不同方向识别不同的用电模式，基于用电频率、离群距离以及异常规则关联度的计算挖掘潜在离群数据点，并通过基于评价矩阵确定离群阈值对离群数据点存在窃电行为的可能性进行确定性分析，实现对窃电行为的数据化检测。最后通过仿真测试证明该算法在针对混杂不同用电模式的用电数据的窃电检测方面相对于其他数据挖掘算法具有更好的性能表现。

Research on Detection of Electric Larceny Based on Outlier Algorithm

In view of the practical difficulties of on-site verification of stealing power, a method based on outlier mining is proposed. The outlier detection algorithm based on density clustering algorithm, the electricity consumption fluctuation pattern recognition method based on different, potential outliers mining frequency, outlier distance and correlation calculation based on abnormal rules, and through the evaluation matrix to determine the threshold for outlier outliers exist the possibility of stealing behavior of uncertainty analysis based on the implementation of detection of acts of stealing data. Finally, the simulation test shows that the algorithm has better performance than other data mining algorithms in the detection of power stealing data for different power modes.