Forecasting the Internal Temperature of Metal Oxide Surge Arresters Using a Sliding Window Approach and Decision Tree Algorithm

This study presents an innovative approach for monitoring the health status of Metal Oxide Surge Arresters (MOSAs) using machine learning algorithms. MOSAs are critical for protecting electrical equipment from lightning strikes and switching surges. Traditionally, internal temperature is regarded as the most reliable indicator of a MOSA's condition; however, measuring it requires invasive and expensive sensor installations. To overcome these limitations, this research proposes a non-intrusive method for estimating internal temperature using surface temperature data. A sliding window algorithm processes historical surface temperature measurements to capture the dynamic thermal behavior of MOSAs. A decision tree model then predicts the internal temperature, categorizing it into HIGH, MEDIUM, and LOW states. This cost-effective solution achieved a classification accuracy of 93%, demonstrating its potential to enhance preventive maintenance strategies and improve the reliability of electrical systems by enabling timely intervention without the need for invasive sensing technologies.