Clustering and Cleaning Method based on Practical Fault Data in Distribution Network

Abstract

The issue of large amounts of interference, errors, and invalid waveforms in the electrical parameter data collected after DTU (Data Terminal Unit) devices generate false remote control signals poses a huge threat to the safe and stable operation of the power grid. This paper proposes a clustering cleaning method based on Stacked Sparse Autoencoders (SSAE) to clean up faulty data. By using SSAE to extract features and reduce dimensionality of the waveform data uploaded by DTU devices, the clustering by fast search and find of density peaks (CFSFDP) algorithm is then used to cluster and clean up erroneous and faulty data. The network parameters are reasonably adjusted to ensure a high correct cleaning rate while maintaining a low erroneous cleaning rate. Comparative analysis proves the superiority of the CFSFDP algorithm in isolating interference, errors, and invalid waveforms. Effective filtering of erroneous and faulty data is achieved through pre-cleaning and complete cleaning, which improves the quality of the database data and provides technical support for subsequent data mining research and effective support for accurate fault diagnosis.