Detection of Power Data Outliers Using Density Peaks Clustering Algorithm Based on K -Nearest Neighbors

Abstract

As an important research branch in data mining, outlier detection has been widely used in equipment operation monitoring and system operation control. Power data outlier detection is playing an increasingly vital role in power systems. Density peak clustering (DPC) is a simple and efficient density-based clustering algorithm with a good application prospect. Nevertheless, the clustering results by the DPC algorithm can be greatly influenced by the cutoff distance, indicating that the results are highly sensitive to this parameter. To address the shortcomings of the DPC algorithm and take the characteristics of power data into consideration, we propose a DPC algorithm based on $K$ -nearest neighbors for the detection of power data outliers. The proposed DPC algorithm introduces the idea of $K$ -nearest neighbors and uses a unified definition of local density. In the DPC algorithm, only one parameter ( $K$ ) needs to be determined, thus eliminating the influence of cutoff distance on the clustering result of the algorithm. The experimental results showed that the proposed algorithm can achieve accurate detection of power data outliers and has broad application prospects.