Research On Transformer Fault Diagnosis Algorithm Based On DGA Technology

In the power system,large oil-immersed power transformer is one of the key equipments to ensure the safety and stable operation of power system.Therefore,it is very necessary to realize transformer fault diagnosis.Transformer running state can be known clearly by doing this and the power system will work more reliable.Dissolved gases analysis(DGA)technology is commonly used in the field of transformer fault diagnosis.In this paper,transformer fault diagnosis method has been researched on the basis of DGA technology and appropriate transformer fault diagnosis method would be formulated.Firstly,the characteristic gases dissolved in transformer oil had deep connection with fault properties.On this basis,the traditional transformer fault diagnosis methods were used as a starting point.Two kinds of typical traditional methods,gas graphical method and improved three ratio method,were used in transformer fault diagnosis.Due to the essence of transformer fault diagnosis problem was multi-class problem,so two kinds of classification intelligent algorithm were chosen to build transformer fault diagnosis models of single intelligent algorithm.These two kinds of intelligent algorithms were kNN and CART respectively.And they have been improved a little.MATLAB software was used for simulation experiments.The performance of the traditional methods was compared with the single intelligent algorithms and the results were analyzed.Then,in view of the ascension of generalization accuracy was restricted by the single intelligent algorithms' own defects.SAMME algorithm,as one of the extension algorithms of classic integrated learning algorithm AdaBoost,was used to integrate the single intelligent algorithms.SAMME-kNN algorithm and SAMME-CART algorithm,which were optimized by 10-fold cross validation method,were used for transformer fault diagnosis.MATLAB software was used for simulation experiments.kNN,CART,SAMME-kNN and SAMME-CART,all these four kinds of algorithms have been compared with each other and the results were analyzed.CART algorithm was chosen as the proper integration object for SAMME.And it is obviously that integration is beneficial to improve generalized precision of the single intelligent algorithm.Next,for the reason that the SAMME-CART fully integrated algorithm had some imperfections,Genetic algorithm which was a global search optimization algorithm has been used to further selective optimize the SAMME-CART based on the selective integration theory.The best single CART algorithm,SAMME-CART fully integrated algorithm and selective-optimized SAMME-CART algorithm,all these methods were compared with each other.The optimal fault diagnosis accuracy of the single CART algorithm was about 83.33%.The fault diagnosis accuracy rate of SAMME-CART fully integrated algorithm was about 88.67%.The fault diagnosis accuracy of SAMME-CART algorithm which was selectively optimized by the genetic algorithm can reach 91.33%.It was shown that using genetic algorithm to selectively optimize SAMME-CART algorithm was effective for improving the generalization ability.At the same time,the constantly optimized hierarchical relationships were also shown.Finally,GUI function of MATLAB software was used to realize the design of transformer oil chromatography fault diagnosis expert system.All these algorithms involved in this paper were embedded in the expert system.The friendly human-computer interaction interface is established.The real-time data display,historical inquiry and diagnosis display and other functions were realized in the expert system.This research could be better applied to practical engineering by this way.