基于特征选择和改进K-均值聚类的异常用电行为检测算法

窃电等异常用电行为严重影响着电网系统的安全、可靠和稳定运行，传统异常用电检测方法存在模型复杂、准确率低等问题。提出了一种基于特征选择和改进K-均值聚类的异常用电检测算法，首先从用电量变化、线路损耗和电力参数三个维度提取15维特征构成特征向量，然后利用相关向量机(Relevance Vector Machine, RVM)进行特征选择，自动确定最优特征集合，最后提出一种基于信息增益的改进K-均值聚类算法对最优特征集合进行聚类分析，从而实现异常用电检测。基于爱尔兰智能电表公开数据集开展实验，结果表明，所提方法在精准率、召回率和ROC曲线AUC值三项指标方面均能获得良好的表现性能，明显优于传统方法。

Abnormal Power Consumption Detection Algorithm Based on Feature Selection and Improved K-means Clustering

Abnormal electricity consumption behaviors such as electricity stealing seriously affect the safe, reliable and stable opevation of power grid system. The traditional detection model of abnormal electricity consconption behavior based on single dimension features has some problems, such as complex model and low accuracy. A detection algorithm for abnormal power consumption based on feature selection and improved K-means clustering is proposed. First, 15 dimensional features are extracted from the three dimensions of power consumption change, line loss and power parameters to form a feature vector. Then, the relevance vector machine (RVM) is used for feature selection to automatically determine the optimal feature set. Finally, an improved K-means clustering algorithm based on information gain is proposed to cluster the optimal feature set, so as to detect abnormal power consumption. Based on the real data of power users in a certain area, the experiment results show that the proposed method can achieve good performance in accuracy, recall and AUC value in ROC curve, which is obviously superior to the traditional methods.