基于生产经营状态识别的低误报率窃电检测二次筛查方法

数据驱动的窃电检测方法主要根据电量及派生指标识别低电量异常,容易受干扰影响误报。利用工商业用户生产经营状态指标大致固定的特点,提出基于生产经营状态识别的窃电二次筛查方法。首先,将检出的低电量异常用户每天的三相功率作为负荷特征,用以标识其当天的用电行为模式及生产经营状态。然后,将每天的负荷特征进行近邻传播聚类。当低电量异常时段负荷特征与正常低电量生产经营状态聚为同类时,认为是用户状态正常转换导致的异常,可排除窃电嫌疑。基于实际窃电数据的测试表明,所提方法可降低误报率。

Second Inspection Method for Electricity Theft Detection with Low False Alarm Rate Based on Identification of Production and Operation Status

Data-driven power theft detection methods mainly identify low power abnormalies according to power and derived indicators, which could result in high false positive due to interference. Taking the advantage of the characteristic that the indicies of production and operation status of industrial and commercial users are generally constant, a second inspection method for electricity theft based on identification of production and operation status is proposed. First, the daily three-phase power of the detected abnormal low-power users is used as the load characteristic to identify the power consumption behavior mode and the production and operation status of the day. Then, the daily load characteristics are clustered by affinitty propagation. When the load characteristics of the abnormal period of low power are clustered with the normal low-power production and operation status as the same group, it is considered that the abnormality caused by the normal transition of the state can be ruled out of the suspected electricity theft. Based on the experiments of actual electricity theft data, the proposed method can reduce the false positive rate.