The FE Approach for Data Cleaning of Phase Measurement Units

Abstract

With the rising use of Phase Measurement Units (PMUs) in smart grid applications, it is important for PMUs to function in extreme circumstances, resulting in outliers and missing dataset. Traditional approaches take an inordinate amount of time to clear outliers and fill missing data to assure better accuracy. This study offers a flexible ensemble approach (FEA) to construct a precise, rapid, and sustainable data cleaning procedure with Apache Spark. To discover outliers in the suggested system, an ensemble model based on a soft voting technique employs PCA in combination with the K-means, GMM, and iForest approach. The suggested method fills the data with an improved gradient-boosting decision tree for each obtained PMUs characteristic after outlier detection. The test results demonstrate that the proposed model acquired good accuracy during comparing with LOF and DBSCAN techniques. To evaluate the suggested technique's data-filling outcomes against modern methods such as decision tree and linear regression techniques, the MAE and RMSE criteria are applied.