Research On Power Transformer Fault Diagnosis Based On Dissolved Gas Analysis In Oil

In the power system,transformer is one of the most important large-scale electrical equipment.The reliable operation of transformer is also one of the key links to ensure the overall stability of the power system.In recent years,with the development of China's economy,the demand for electricity is increasing year by year,the higher voltage level and the larger substation capacity are also the inevitable trend of transformer development.Therefore,how to ensure the safe and reliable operation of the transformer has important practical significance.At present,fault diagnosis technology as the core of transformer condition maintenance is an important means to ensure the stable operation of transformer.In the field of transformer fault diagnosis,the method of dissolved gas analysis(DGA)can monitor the gas data of transformer oil on-line and real-time,which has been widely used at home and abroad.In this paper,according to the support vector machine can effectively solve the problems of high dimension,non-linear and local optimization,and has the advantage of processing small sample data,the support vector mechanism is used to build the fault classifier model.On this basis,considering that binary tree support vector machine has the advantages of fast test speed and no non divisible region compared with one to many combination,one to one combination and directed acyclic graph,a diagnosis model of binary tree support vector machine is constructed.Finally,in order to optimize the parameter model of SVM and improve the accuracy of fault diagnosis,two main parameters C and g that affect the accuracy of SVM classification are optimized by genetic algorithm.By establishing the initial feature set,and according to the coding rules of genetic algorithm,the punishment factors,kernel parameters and feature subsets of support vector machine are chromosome coded,and the fitness function based on 5 fold cross validation accuracy is established,and the combination of optimal feature subsets and support vector machine parameters is jointly optimized.Then,according to the optimal feature subset and parameter combination training diagnosis model,the problem that the parameters of SVM model have great influence on the accuracy of power transformer fault diagnosis is solved.The diagnosis performance is verified by test set and fault examples.The results show that this method can diagnose transformer fault accurately and effectively,and has higher diagnostic accuracy than the traditional support vector machine genetic algorithm model,IEC three ratio method,back propagation neural network and naive Bayes method.