A Neural Network-Based Model for Transformer Fault Detection

Abstract

The demand for uninterrupted and reliable electrical energy has escalated in modern power systems, necessitating the near-fault-free operation of critical components such as power transformers. Reducing the frequency and duration of transformer-related outages has become essential, placing significant emphasis on protective relay systems to operate with high accuracy and minimal false tripping. Traditional maintenance alone is inadequate to ensure absolute reliability, as unexpected faults can lead to temporary power interruptions during repairs. This study proposes a predictive fault analysis approach for power transformers utilizing artificial neural networks, specifically leveraging the Scaled Conjugate Gradient algorithm and K-Nearest Neighbors (KNN) classifier. This model shows a significant improvement in prediction accuracy over current techniques by correctly predicting fault probability, which helps to improve transformer management's reliability and preventive measures, achieving up to 98% accuracy.