Evaluation of Breakdown Voltage and Water Content in Transformer Oil Using Multi Frequency Ultrasonic and Generalized Regression Neural Network

Power transformer is regarded as one of the crucial part of electrical power transmission and distribution system. The quality of transformer oil can directly affect the operation of the power transformer, and breakdown voltage (BDV) and water content are the two main parameters of transformer oil quality. Monitoring the BDV and water content of transformer oil is considered as an important method to evaluate the safe operation of power systems. This work proposes the measurement of BDV and water content in transformer oil using multi frequency ultrasonic and generalized regression neural network (GRNN). The BDV and water content of all 210 samples were firstly tested according to the traditional testing methods and the multi frequency ultra-sonic technology, separately. And then the 210 samples were randomly divided into training sets and test sets. The obtained multi frequency ultrasonic data were set as the input of GRNN, and the BDV and water content as the output of GRNN. Moreover, the 20-fold-cross-validation was incorporated to obtain the best smoothing factor sigma for GRNN. Finally, the GRNN model was trained by the training sets with sigma = 4.54 and was evaluated with the test sets. All results show that the lower BDV or the higher water content of the sample will cause greater ultrasonic sound attenuation, and the prediction accuracy of the prediction model for BDV and water con-tent in oil is up to 95%. It provides a new method for evaluating the health of transformer oil.