Research On Transformer Fault Diagnosis Optimization Strategy Based On The Multiple Model Combination

As one of the intelligent substation’s core equipment, the power transformer plays a decisive role on promoting the smart grid to the trend high voltage, large span, high connectivity and high initiative. Therefore, in order to grasp the working state of power transformer, the study on different fault diagnosis methods of transformer has become the primary problem for constructing the engineering of intelligent high linkage and large power grid.Firstly, based on the boosting of the judgment technology of the equipment abnormal operation state at home and abroad, this paper adopts three kinds of single fault diagnosis models which are fuzzy C- means(Fuzzy C-Means, FCM),the improved three ratio method and neural network, and use the software of MATLAB for each model simulation at the same time. However, only selecting a single model can not give the accurate diagnosis results of the abnormal operation state of the transformer. Therefore, this paper combines the fuzzy logic, FCM, the Bayesian regularization algorithm, glowworm swarm optimization algorithm and other intelligent methods with the single diagnosis model,getting four kinds of models for fault diagnosis of power transformer, which are the fuzzy boundary, the FCM algorithm and fuzzy coding boundary, the optimization of Bayesian regularization algorithm and glowworm swarm algorithm for the fuzzy neural network. In addition, each of the four kinds of models has its own diagnostic advantages,by combining the advantages the combination optimization model can be obtained.The model adopts L-M(Levenberg-Marquardt) algorithm which has the dual advantages of efficiency and rate in the field of fault diagnosis.Then, comparing the simulation results of the the five kinds of combination optimization models, there is a certain gap between the diagnosis results of different models. On one hand, the effects of improved neural network models have better diagnosis results than other models. On the other hand, the L-M combination optimization model based on four diagnosis models has the highest correct rate of 90.91% and contains high diagnosis speed.The model of improvement glowworm swarm optimization for optimizing fuzzy neural network has the lower accuracy of 86.34%. In short, the former model is more suitable for the transformer abnormal operation state’s analysis and judgment.Finally,combined with the real time data obtaining by online monitoring system of oil chromatogram, the expert system made by GUIDE makes the model of transformer fault diagnosis into practice, and achieves a variety functions of real-time data display, history data review and diagnosis results display, which can be better applied to practical engineering.