AI-BASED PREDICTIVE INSULATION FAILURE DETECTION IN HIGH VOLTAGE TRANSMISSION NETWORKS USING REAL-TIME SENSOR DATA

Abstract

Background: The problem of insulation breakdown in high-voltage transmission networks is a life-threatening problem, and it can cause considerable downtimes in the networks and equipment breakage. Old practices of maintenance are usually reactive and therefore costly and slow to respond. Predictive maintenance using AI has become a promising technology to predict failures using real-time sensor data.

Goal: The purpose of the study is to design and test an AI-based predictive insulation failure prediction model in high-voltage transmission systems with real-time sensor measurements to improve the early detection of faults and to optimize maintenance decisions.

Method: The voltage, current, temperature and partial discharge sensors data were recorded and processed in real time. SVM, Random Forest, and Long Short-Term Memory (LSTM) machine learning (ML) models were trained and tested on classification and remaining useful life (RUL) prediction. The Auto-encoders and Isolation Forest algorithms were used to detect anomalies.

Results: The LSTM model has performed the best in fault prediction with a rate of 94.6 and the Auto-encoders have a high rate of detection with AUC-ROC of 90.3. The most significant feature in predicting failures was found to be partial discharge.

Conclusion: AI-based predictive models are important in terms of predictive maintenance of high-voltage transmission networks, especially with LSTM and Auto-encoders, which will suggest high-quality early warnings and minimize downtime.