检测低压用户窃电的计及专家经验的模糊聚类法

Fuzzy C-means Clustering Method in Considering of Experts'Experience to Detect Low-voltage Customers'Suspicion of Electricity Theft

针对电网营销系统中低压窃电工单少的现状,提出不依赖窃电标签的基于专家经验和模糊C均值(fuzzy C-means,FCM)聚类算法的低压用户窃电检测技术方案。首先,提炼现场专家经验,提出预打标签法,对原始用户初步标记;其次,重构特征和归一化操作后,采用FCM算法对用户聚类,训练出聚类中心矩阵和隶属度矩阵;最后,按行计算训练集隶属度矩阵和测试集隶属度矩阵之间的2范数,通过嫌疑系数来判断待测用户近段时间是否存在窃电。以广东某台区低压用户作为算例,验证所提窃电检测方案的有效性和可行性。

Aiming at the situation that low-voltage electricity theft orders are few in the marketing system,this paper proposes a technical scheme for electricity theft detection based on experts’experience and fuzzy C-means(FCM)clustering algorithm,which does not rely on electricity theft tag.Firstly,a pre-labeling method is proposed through extracting experts’experience to mark the original users.Secondly,after reconstructing features and normalization,customer data is clustered by FCM algorithm and then the the clustering center matrix and the membership matrix are trained.Finally,2-norm between the membership matrix of training set and the membership matrix of testing set is calculated by rows,and the suspicion coefficient is used to determine whether the customer is detected electricity theft in recent times.By uses low-voltage customer data in a certain transformer district in Guangdong province as the calculation example,the paper verifies the effectiveness and feasibility of the proposed technical scheme.