Quantitative recognition of electrical parameters of transformer oil based on nondestructive ultrasound and the combined KPCA-WOA-Elman neural network

• Published in: Sensors and actuators. A. Physical. Vol. 363; p. 114764
• Main Authors: Jia, Lufen, Zhang, Yu, Feng, Weiquan, Li, Baoliang, Zhou, Qu
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.12.2023
• Subjects: Dielectric loss factor; KPCA-WOA-Elman neural network; Nondestructive multi-frequency ultrasound; Quantitative recognition; Transformer oil; Quantitative recognition; Dielectric loss factor; Transformer oil; KPCA-WOA-Elman neural network; Nondestructive multi-frequency ultrasound
• ISSN: 0924-4247; 1873-3069
• DOI: 10.1016/j.sna.2023.114764

The dielectric loss factor is an essential electrical parameter used to evaluate the quality of transformer oil and judge the operation status of transformers. In this work, a new nondestructive detection method of dielectric loss factor based on multi-frequency ultrasound technology and artificial neural network is proposed. For the extraction of valid information, dimensionality reduction algorithm Kernel Principle Component Analysis (KPCA) is adopted to pre-process the ultrasound data set before the training of neural network. Particle Swarm Optimization (PSO) and Whale Optimization Algorithm (WOA) are considered for the optimization of Elman neural network (Elman neural network). The analysis of the prediction results shows that the prediction accuracy of WOA-Elman and PSO-Elman are 93.2 % and 85.5 % respectively. Through the comparison of four evaluation indexes, WOA-Elman neural network behaves better on the prediction of the dielectric loss factor. Consequently, the proposal of the prediction model Elman neural network based on Kernel Principle Component Analysis and Whale Optimization Algorithm (KPCA-WOA-Elman neural network) lays a foundation for the effective prediction of the quality of transformer oil online. [Display omitted] •Based on the attenuation characteristic of ultrasonic wave in medium, the multi-frequency ultrasonic technology is applied to the detection of transformer oil.•The correlation between dielectric loss factor and multi-frequency ultrasonic signal is analyzed.•The dimensionality reduction algorithms KPCA was adopted to pre-process the 242-dimensional data into 36-dimensions and WOA (Whale Optimization Algorithm) was adapted to optimize Elman neural network, with a prediction accuracy of 93.2 %.