Outliers Detection and Repairing Technique for Measurement Data in the Distribution System

In any power network, data perform a very critical role in the operation, management, and regulations of power systems. However, most of the data contain anomalies, which may have an impact on the outcomes of data-driven applications. Therefore, to avoid problems during operations it is very important to detect these outliers and remove them from the data. This research investigates the anomalies cleaning approach for measuring data in the distribution system in order to enhance data quality. This approach includes a set of association rule (AR) that are built automatically using past measuring data. This study demonstrates a data-mining approach based on a mix of density-based spatial clustering of applications with noise (DBSCAN) clustering and auto-generated association rules using historical data. Following that, a novel cost function based on Mahalanobis distance is developed and used for data restoration; this function describes the similarity between different data points. Finally, simulation results show that the suggested model outperforms existing detection and repair strategies. The evaluation section of this research demonstrates that as the number of historical data increases, so does the resilience of the suggested method.