Research On Intelligent Diagnosis Method Of Transformer Fault Based On Artificial Neural Network

Power transformer is one of the important parts of power system,and the normal operation of transformer is a necessary condition for the safe and stable operation of power grid.With the rapid development of economy in our country,the number and capacity of transformers have increased obviously,so the research on transformer fault diagnosis method has important practical significance.Because of the ability to realize on-line real-time monitoring,dissolved gas analysis in oil has become a widely used transformer fault diagnosis method.Dissolved gas analysis in oil has a simple theory,low accuracy,and can not solve the problem of complex transformer fault types in practical engineering.Therefore,modern intelligent method has attracted the attention of the majority of researchers.Artificial neural network is an efficient modern intelligent algorithm,which has outstanding performance in the field of transformer fault diagnosis.The training of artificial neural network depends on gradient descent algorithm,but the convergence performance of traditional algorithm is poor.In order to accelerate the convergence speed of model parameters and improve the final diagnosis accuracy,three improved gradient descent algorithms are proposed in this paper.The standard Adagrad algorithm always reduces the adaptive step size,but the improved IAdagrad algorithm introduces the step size lifting coefficient to weaken this characteristic and improve the later learning ability of the model;the attenuation rate of the standard RMSProp algorithm is fixed,and the improved IRMSProp algorithm introduces the attenuation rate to change the system,so that the gradient influence of different time changes with the environment;on the basis of standard Adam algorithm,IAdam algorithm introduces step size amplification factor and reduction factor to make the step size have stronger self-adaptability and further accelerate the convergence speed.In this paper,according to the corresponding relationship between dissolved gas in transformer oil and fault type,a neural network fault diagnosis model with infrastructure of 5-12-6 is established and optimized by three improved algorithms.Firstly,this paper observes the performance of the model under different parameter combinations,selects the appropriate algorithm parameters after comparison,and then uses the sample data to train the intelligent diagnosis model of transformer faults.Verified by the test set,the results show that the improved algorithm shows higher performance in transformer fault diagnosis,faster convergence speed and higher accuracy than before.