Research On Transformer Fault Diagnosis Based On Improved Fuzzy C-means Algorithm

The power transformer is one of the core apparatus in the electric power system.Therefore,transformer fault diagnosis technology is very important and meaningful in guaranteeing the safe and stable operation of the power grid.Although most faults can be diagnosed by adopting the traditional Three-ratio Method which is based on Dissolved Gas Analysis(DGA),the accuracy rate of diagnosis is not satisfactory,especially in the vicinity of thresholds of ratios.Fuzzy C-Means Algorithm(FCM),with its advantages in dealing with fuzzy problems,is of high application value in the field of transformer fault diagnosis;but its result is dependent on initial value,which is easily converged to local optimum.Therefore,in this paper two improved algorithms are researched based on FCM Algorithm;simulation results show the superiority of these two methods.The main research work of this paper is as follows:(1)The causes and types of transformer fault are summarized.The Three-ratio Method is introduced and analyzed in details,with its disadvantages commented on.(2)A "Possibility Density Function Initialized Fuzzy C-Means Algorithm" is researched.According to the required number of clusters,select as initial clusters the individuals whose neighboring area exist the most sample points(after getting rid of the influence of previously selected individuals);and then Fuzzy C-Means Clustering is applied.The result of simulation in Matlab R2010b indicates that the accuracy of fault diagnosis is increased by adopting this Algorithm than by using the Three-ratio Method or basic FCM Algorithm.(3)A "Dynamic Genetic Algorithm Optimized Fuzzy C-Means Algorithm" is designed.In this Algorithm,the variable-length-coding Genetic Algorithm is integrated with Fuzzy C-Means Clustering;and a new cluster validity function is introduced.The result of simulation in Matlab R2010b indicates that the accuracy of fault diagnosis is increased by adopting this Algorithm than by using the Three-ratio Method or basic FCM Algorithm,and the number of clusters is not required to be known in advance.