Quality state grade prediction of transformer oil based on SVM and multi-frequency ultrasound

Transformer oil is an important insulating medium in power transformers, and its quality state could reflect the dielectric strength. In this paper, four indexes were selected to analyze the integrated quality of the transformer oil and judge the operating state of the transformer. The parameters include micro water, dielectric loss factor, breakdown voltage and interfacial tension. The multi-frequency ultrasonic testing equipment was used to test the transformer oil by means of the ultrasonic penetration detection method and reflection detection method. And we obtained 242 dimensional data, which can characterize the comprehensive quality of the transformer oil. The data includes amplitude and phase under different frequencies. Support vector machine(SVM) algorithm was adopted to establish the transformer oil quality evaluation and classification prediction model based on multi-frequency ultrasonic technology. The tested transformer oil ultrasonic data was chosen as input and the oil quality grade as output respectively. Through the training and prediction, the classification accuracy could only reach 74.57%. And cross-validation was used to optimize the parameters of the SVM kernel function, and the optimal parameters c=588.1336 and g=0.0068 were obtained, with the classification accuracy rate reaching 91.52%. It is of great significance to apply the multi-frequency ultrasonic combined with cross-validation of support vector machines into detecting the transformer.