Partial Discharge Diagnosis Of Transformer Based On Random Forest

Power transformer is a pivotal device in the power system for transform electricity.Partial discharge is an effective indicator to reflect the deterioration of transformer insulation.it’s significance to identify the discharge type.At present,the partial discharge pattern recognition have a poor effect in the substation because of the large differences between them such as the aging degree of the transformer board,multi-source discharge and noise interference.In order to improve the recognition model applicability,this paper did the following research:This paper obtained discharge data of Inter-turn and Wedge model with different test voltage and different paper aging levels,using common classifier algorithms to generate recognition model with different test conditions data.The accuracy of the classifier algorithm in each case was discussed.The results showed that the pressure level has little effect on pattern recognition,while the cardboard aging has a great influence.Based on the scene noise and the multi-source discharge have a great influence on the pattern recognition of partial discharge,the scene noise data was analyzed to get the statistical method of the simulation noise.Discharge data in laboratory and simulated noise were fused to generate the multi-source discharge data.Using common classification algorithms to generate recognition model with multi-source discharge data,and using another group of multi-source discharge data and scene data to obtain its applicable range and prove it’s applicability.Compared with the current multi-source partial discharge recognition method,this model has the advantage of no pulse separation process and low requirement of sampling rate.The theory and performance of each classifier was compared in the above training process to obtain the scope.In addition,the rationality of the model was improved by the special process of classification.The model can avoid the defect that the classifier can only identify known classification.The recognition category was expanded by establishing a fusion classification model.