Power transformer plays an important role in the process of transmission and distribution, so it is necessary to diagnose the insulation and predict the service life of the power transformer. Support vector machine (SVM) is widely used in the field of multi-source information fusion for data prediction. The traditional SVM algorithm projects the data into high-dimensional space and constructs a hyperplane for classification. The new data is projected as the prediction set when building the model, and the prediction value is obtained according to the hyperplane classification. The Recovery voltage method (RVM) is a non-destructive test method. Its test characteristic quantity can reflect the aging and moisture of oil-paper insulation from many aspects. In this paper, the improved SVM algorithm called Shuffle-SVM is used to train multi-source data obtained from the RVM test are put into model training to predict the polymerization degree (DP) to detect the insulation state and calculate the service life of the transformer.
 

oil-paper insulation,time-domain dielectric,Shuffle-SVM algorithm,Multi-source information