Research On Intelligent Method For Transformer Fault Diagnosis Based On SVM

Transformers play an important role in the power system,because of the major energy conversion and transmission functions.Once a failure occurs,there will be alargescale blackout,and if measures are not taken in time,it will cause huge socialand economic losses.Therefore,it is important to diagnose the fault in a timely manner and prevent it from further damaging.As the most popular diagnosis method for transformers,dissolved gas analysis(DGA)has a good fault diagnosis effect.Gas ratio method is most commonly used in DGA-based diagnostic methods.But it has limitation suchas the “no decision”and low accuracy.To overcome such problems,intelligence algorithmfor faults classification have been used by a great number of researchers to develophighly precise diagnostic technique based on DGA data,there are many combinationsof gas ratio methods.The classification accuracy is varied with input variables,which makes the intelligent algorithm less robust.There is no study openon selection of input variables for intelligence algorithms.According to the limitation of the current DGA intelligent diagnosis algorithm,a single characteristics gas ratio cannot fully reflect the transformer fault.In this study,a transformer fault diagnosis method is put forward by combing Rapid Miner and support vector machine(SVM).The diagnostic accuracy of support vector machine is closely associated with the paprameter selection,this thesis proposes a new hybrid alrorithm to optimize the structural parameters of SVM,so as to provides an idea for exploring more effective intelligent diagnostic technology.The main research content of this article is as follow:(1)This peer analyzed the principle of the transformer failure and the gas production mechanism,and introduced the advantages and disadvantages of conventional intelligent diagnostic methods.In order to find input features suitable for intelligent algorithms,using Rapid Miner tools and combining the sample data of transformers,the most relevant input variables of fault type are selected.And use it as input to a diagnostic model based on support vector machines.(2)In light of the difficulty in the parameter selection,and the kernel parameters and penalty parameters of SVM have a significant impact on generalization,this paper optimizes the structural parameters by particle swarm optimization,krill herd algorithm and ant lion optimization.Then,the diagnostic results of improved three-ratio method and the results based on intelligent optimization support vector machine are compared.(3)Although the single intelligent algorithm can improve the accuracy of transformer fault diagnosis based on SVM,the traditional particle swarm algorithm can easily fall into the local optimization.To solve such problem,a new hybrid algorithm is proposed.The algorithm uses the basic particle swarm optimization as the framework,and conbines the beetle antennae search(BAS).To verify the convergence of hybrid algorithm with typical test functions.Finally,this new algorithm is used to optimize the parameters of the diagnostic model,and compared with the single optimization algorithm described above.Simulation and results show that the method does improve the accuracy of fault diagnosis.