The Study Of Basic Believe Assignment Based On Fault Result Of Dissolved Gas Transformer

Transformer is one of the important equipment in substation, accurate and effective fault diagnosis of transformer is very important to the whole power system. At present,there are many methods for transformer fault diagnosis, because the oil dissolved gas analysis(DGA) technology can detect the potential faults of the transformer and can prevent a catastrophic accident. So the technology has been considered to be the most convenient and effective method for transformer fault diagnosis. Because there are many uncertain factors in transformer fault diagnosis, leading to a variety of uncertain information can not fusion, so the traditional fault diagnosis method of transformer is limited. The basic belief assignment function(BBA) is the premise of D-S evidence theory to information fusion, which can provide a more accurate result. This article will take the calculation of BBA mumerical which based on the oil dissolved gas for transformer fault daignosis as the research object to further research and analysis.Through the induction and analysis of various calculation methods of BBA in transformer fault, proposing use the improved three ratio method of DGA to calculate the BBA numerical. Through the analysis of fault feature space that defined by improved three ratio method, according to the principle of finite element method, dispersing the fault space for each fault according to the fault characteristics of regional distribution characteristics in the space. Then use the eight node hexahedron spatial interpolation method of fault feature space of eight key node interpolation to obtain the three-dimensional space continuous transformer fault basic belief assignment function.In order to further improve the accuracy calculation by BBA based on the improved three ratio method, BP neural network is applied to calculate the transformer fault BBA, so transformer fault diagnosis BBA model is established. Use the pretreatment sample data to train the neural network model. The number of training, the linear correlation coefficient and square error three evaluation indexes were tested and analyzed on the network. Finally,determine the transformer fault diagnosis BBA model for 5-15-10 network structure.According to the standard BP algorithm and principle, present the momentum adaptive learning rate method. By using the standard BP algorithm, adaptive learning rate method, and additional momentum method that are traditional algorithm and the new algorithm proposed respectively to train the transformer fault diagnosis BBA model. On the training results, only the new algorithm can meet the requirements and the good performance of the network. Through analyse the output of the new algorithm and the linear regression, show that the new algorithm can be used in the calculation of BBA.Finally, using MATLAB to simulat the model trained by the four for the four algorithms,verified the superiority of the new algorithm.Using the transformer fault diagnosis BBA model trained by the new training algorithm to diagnose ten kinds of common fault diagnosis and calculate the BBA of different faults, improved the accuracy of BBA calculated by improved three ratio method.