ARTIFICIAL INTELLIGENCE-BASED PREDICTIVE MAINTENANCE OF SMART POWER DISTRIBUTION NETWORKS USING MACHINE LEARNING FOR FAULT PREDICTION ACCURACY: A SIMULATION-BASED EXPERIMENTAL STUDY

Abstract

The increasing complexity of modern electrical power distribution systems has significantly enhanced the need for intelligent maintenance strategies capable of ensuring uninterrupted power delivery while minimizing operational costs. Traditional maintenance approaches, including corrective and preventive maintenance, often fail to accurately predict equipment failures because they rely primarily on scheduled inspections or reactive interventions after faults occur. Such approaches frequently result in unexpected equipment breakdowns, prolonged outages, increased maintenance expenditure, and reduced system reliability. Recent advances in Artificial Intelligence (AI), Machine Learning (ML), and Smart Grid technologies have created new opportunities for predictive maintenance by enabling utilities to forecast equipment failures before they occur using operational data collected through intelligent monitoring systems.

This study presents a simulation-based experimental framework for predictive maintenance of smart power distribution networks using machine learning techniques to improve fault prediction accuracy. A synthetic electrical equipment dataset representing transformer health, feeder performance, breaker conditions, load variations, environmental influences, voltage fluctuations, current imbalance, insulation resistance, equipment temperature, vibration level, oil quality, and historical maintenance records was generated using MATLAB simulation. Four supervised machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Extreme Gradient Boosting (XGBoost)—were implemented and evaluated for predicting equipment faults.

Performance evaluation was conducted using Accuracy, Precision, Recall, F1-score, Receiver Operating Characteristic–Area Under Curve (ROC-AUC), and Confusion Matrix analysis. Experimental findings indicate that XGBoost achieved the highest prediction accuracy of 97.8%, followed by Random Forest (96.9%), Artificial Neural Network (95.8%), and Support Vector Machine (94.6%). The results demonstrate that ensemble learning techniques significantly outperform conventional machine learning methods in handling nonlinear electrical system characteristics and complex fault patterns. Moreover, AI-based predictive maintenance substantially reduces false alarms, improves maintenance scheduling, enhances asset reliability, and minimizes unexpected power interruptions.

The proposed framework provides an efficient and scalable solution for future intelligent power distribution systems and contributes to the growing body of research on AI-driven smart grid maintenance. The findings also support the adoption of predictive analytics in modern electrical utilities for achieving higher operational efficiency, improved asset management, and enhanced system resilience.