A Novel Acoustic Temperature Measurement Technology of Transformers Based on Ultrasonic Sensing

Abstract

This article addresses the challenges of monitoring the internal temperature of oil-immersed power transformers, where the metal shell barrier limits the effectiveness of conventional temperature measurement methods, and existing detection techniques lack accuracy and practicality. To tackle these issues, an ultrasonic temperature measurement method is introduced in this article, combined with machine learning algorithms for enhanced temperature inversion and monitoring. First, an experimental platform is established to collect acoustic data at various temperatures, analyze interference noise from transformer core vibrations, and filter out magnetically induced noise. After that, the key time-domain features that reflect the change in dynamic waveform with temperature are extracted. Finally, the random forest (RF), k-nearest neighbor (KNN), and support vector machine (SVM) are used to build a transformer temperature identification model, and feature optimization is achieved with the help of two feature dimensionality reduction methods, RF, and correlation-based feature selection (CFS). Experimental results show that the RF, GS-KNN, and ZOA-SVM models achieve recognition accuracies of 88.57%, 94.29%, and 91.43%, respectively. These findings highlight the proposed method’s ability to accurately diagnose internal transformer temperatures, which is of some significance in engineering practice.