一种基于k-means的两阶段用电异常检测方法

摘要

The theft of electricity or other illegal behaviors of power users will always lead to some abnormal electricity situations, resulting in non-technical losses of power supply enterprises. In actual consumption environment, the power loads are under the influence of so many nonlinear factors that the power use behavior patterns present diversity, and it is difficult to accurately identify the abnormal electricity when using the traditional method. To solve this problem, a two-stage electrical anomaly detection method based on k-means clustering algorithm is proposed in this paper. In the first stage, based on power use historical data, the user typical load curve is extracted as the basic model by k-means clustering algorithm; the gray analysis of the key factors influencing the power load according is used to analyze the influence degree and modify the foundation models further. In the second stage, a comparison is made between the daily load curve and revised typical load curve, and the Euclidean distance between the two curves is utilized as the user electricity abnormal considerations on the basis of sorting for suspected users. The practical case shows that this method is of practical value for the detection of abnormal electricity.%窃电或电力用户其他违规行为及计量错误往往会引起用电异常,导致供电企业的非技术损失.在实际用电环境下,电力负荷受到诸多非线性因素的影响,利用传统的用电异常检测方法难以准确识别用电异常.针对这一问题,提出一种基于k-means聚类算法的两阶段用电异常检测方法.第一阶段利用k-means聚类算法根据用户用电历史数据提取用户典型负荷曲线作为基础模型,进一步利用灰色分析法对影响电力负荷的关键因素根据影响程度进行分析,并对基础模型进行修正;第二阶段,将待测日负荷曲线与经过修正的典型负荷曲线进行比较,将二者之间的欧氏距离作为用户用电异常的考量依据为嫌疑用户排序.实际案例表明,该方法对于检测用电异常快速有效,具有实用价值.