基于空间密度聚类和异常数据域的负荷异常值识别方法

For mass power load data, a method based on density-based spatial clustering and outlier boundaries is proposed to identify the load outlier. Firstly, the density-based spatial clustering method is used to classify load curves into normal and abnormal power consumption patterns. Also, the load curves with normal power consumption pattern are classified into different load levels. Then, the outlier boundaries are built using the confidence interval of load expected value and the inter-quartile range of the deviation between load sample and sample average at different load levels. Considering the contingencies of atypical power consumption behavior, the obtained outlier boundaries are corrected by time offset of power consumption, and outlier boundaries for abnormal power consumption patterns are built. Finally, the proposed method is tested in the example with the load data sets of residential and industrial users. Compared with the traditional method, the precision of the proposed method is improved over 10% on average, and the comprehensive evaluation index is improved over 4% on average.