Machine Learning-Based Fault Analysis: Transforming Correlated Fault Data in Distributed Generation Systems

Abstract

Distributed generation systems offer several advantages over centralised power systems. However, faults within the system can lead to stability loss, reclosure failures, and voltage fluctuations, necessitating special attention to safeguard the system’s components. Identifying the location and nature of faults is crucial in preventing adverse effects on the system’s overall functionality. Various machine learning (ML) techniques have been proposed for fault location and categorization in distributed generation systems. While many of these techniques effectively pinpoint the fault’s location, accurately determining the fault type remains a challenge. This work presents a novel approach to enhance fault classification efficiency by transforming correlated fault data using Principal Component Analysis (PCA) in ML. The suggested findings demonstrate that the proposed method can significantly improve the performance of specific fault classification algorithms in power systems. The novelty adopted in reducing the dimensions before subjecting the data to the ML algorithm has given higher accuracy in identifying fault types in a faster time frame, thereby enhancing the security and stability of distributed generation systems. Notably, the suggested method achieved a faster time frame. This paper has explored excellent accuracy with two of the ML algorithms namely the; Random Forest Classifier and K-Nearest Neighbors underscoring their further potential for improving system protection.