带追溯功能的窃电侦查多维离群点分析方法及应用

Electricity Stealing Identification by Multi-Dimensional Outlier Detection and Tracing Analysis

针对现有方法对历史月已开始的潜在持续性窃电的搜索能力较差的问题,提出一种带追溯功能的窃电侦查多维离群点分析方法。首先从集抄数据中提取出月用电量水平、日温度相关性、月温度相关性、日用电量波动率、环比增长率、谷电比重6项指标,采用多维离群点分析方法,找出当月开始窃电的疑似用户;随后,对其余用户进行历史月的多维离群点分析,找出存在历史疑似窃电起始月的用户,从最近的疑似窃电起始月至当前月,运用累计和值法对这些用户实施日用电量序列的向上水平迁移分析,将不存在向上水平迁移的情况判定为潜在的持续性窃电。对某小区的实例分析结果表明,所提方法可在不明显增加误判率的同时显著提升对窃电可疑用户的搜索率。

To improve the search rate of potential and continuous electricity stealing customers who started electricity stealing in history month,a multi-dimensional outlier detection and tracing analysis method is presented in this paper. Firstly,six indexes are extracted from the daily consumption series and a cluster-outlier interactive algorithm is applied to find suspects in the objective month. The six indexes include the consumption level index,the daily temperature relevance index,the monthly temperature relevance index,the daily consumption variation index,the chain growth rate index and the valley consumption ratio index. Then,the cluster-outlier interactive algorithm is applied again to other customers to find suspects in history months and an accumulative sum method is adopted to these suspects to find whether there exists upward level change in daily consumption series from the starting stealing month to the objective month. The suspect will be judged as potential and continuous electricity stealing customer whose daily consumption series exist upward level change. An example is provided towards a residential community. Results show that the proposed method is of higher search rate and meanwhile the misjudge rate does not improve apparently.