Outlier Detection for Load Monitoring Data of Transformer Based on Kernel-Density-Fluctuation Outlier Factor

In this paper, an outlier detection method based on kernel-density-fluctuation is proposed, which is suitable for preprocessing transformer load monitoring data and can improve the quality of data sets used for data mining tasks. At first, some basic concepts involved in this newly proposed algorithm are introduced, and then the definition and calculation method of the k-density-fluctuation outlier factor are illustrated, which is used to quantify the possibility that this object is an anomaly. Finally, the algorithm rearranges the objects in the origin data set according to the value of their k-density-fluctuation outlier factors to discover the outliers. The simulation results of the algorithm running on the artificial data set and the natural data set to verify that the performance of the proposed algorithm in detecting isolated points is more ideal than the current used algorithms such as LOF in preprocessing transformer load monitoring data.