数据驱动窃电检测方法综述与低误报率研究展望

配电系统窃电是造成电网非技术损失的主要原因,是供电企业运营管理中长期面对的痼疾。用电信息采集系统采集的海量用户数据使得开展数据驱动的用电异常检测、准确识别窃电用户成为可能。受用户用电行为多样性影响,数据驱动的窃电检测方法的误报率在某些场景下尚难以满足实践需求,严重制约了该类方法的工程应用。首先,介绍了窃电实现手法;然后,梳理了在实践中得到工程应用的窃电检测方法以及数据驱动窃电检测方法的基本思路和局限性;在此基础上,结合工程应用对窃电检测评价指标的差异性需求,分析指出提取的可用信息不足、特征指标项灵敏性和可靠性不高是阻碍数据驱动窃电检测方法走向工程实用的主要原因。最后,从算法设计、状态空间细分以及特征指标项设计选择等不同层面对低误报率窃电检测进行了展望。

Review on Data-driven Based Electricity Theft Detection Method and Research Prospect for Low False Positive Rate

Electricity theft in the power distribution system is the main cause of non-technical loss of power grids, and it is a chronic problem existing in operation and management of power utilities. The electricity information acquisition system collects massive user data, which makes it possible to carry out data-driven abnormal electricity detection and accurately pinpoint electricity theft consumers. Affected by the diversity of electricity consumption behaviors of users, the false positive rate of data-driven based electricity theft detection method is still difficult to meet the practical needs in some scenarios, which seriously restricts the engineering application of this method. Firstly, this paper describes the implementation measures of electricity theft, and then sorts out the basic ideas and limitations of the electricity theft detection methods applied in engineering practice and the data-driven based electricity theft detection methods. On this basis, combining with the different requirements of engineering application for the evaluation index of electricity theft detection, it is pointed out that the lack of useful extracted information, the low sensitivity and reliability of the characteristic index items are the major reasons that hinder the data-driven based electricity theft detection methods from being practical in engineering. Finally, the electricity theft detection with low false positive rate is prospected from different levels such as algorithm design, state space subdivision, and design and selection of characteristic index items.