Research On Deep Learning Algorithm Optimization For Transformer Fault Diagnosis

The oil-immersed power transformer is one of the most important equipment in the power transmission and transformation project.Its safety and reliability are the basis to ensure the continuous operation and power supply of the grid.How to ensure the safe and stable operation of the transformer is one of the key issues of the current power industry.Transformer under complex and changeable operation conditions for a long time,there will be discharge,overheating and other faults,if the abnormal situation of the transformer can be found in time and processed,we can prevent and reduce the occurrence of faults.Since the 19th century,people have begun to study various methods of transformer fault diagnosis.With the rapid development of artificial intelligence and computer technology,various artificial intelligence algorithms have been introduced into transformer fault diagnosis.Therefore,it is particularly important to study the optimization algorithm of transformer fault diagnosis with high accuracy and more stable performance.Papers in determining the transformer fault type and gas relationship on the basis of the first to data mining and selection of all sample data,using I-K means clustering algorithm and t-SNE dimension reduction algorithm visualization analysis,determine the transformer faults are divided into 7 classes,make "off label" sample data into a "label",lay the foundation for subsequent fault diagnosis;Then,in view of the problem that the training results of neural network algorithm are not ideal,the CPSO algorithm is used to optimize the initial weights and thresholds of the neural network,and a transformer fault diagnosis model based on CPSO-BP is built.Through training and experiments,it is found that the optimization algorithm can significantly improve the accuracy rate.Then the CPSO-BP model was further optimized by ADAM gradient optimization algorithm.The CPSO-BP-ADAM transformer fault diagnosis model was built in the Tensor Flow platform,and the parameters were set.2040 oil chromatography training sets were used for training,and 680 oil chromatography test sets and test sets were used for parameter tuning and testing.The diagnostic accuracy of the model for training set,test set and test set is 99.56%,99.71% and99.18%,respectively,which is higher than CPSO-BP model.Then,compared with SGD,RMSProp and ADAM,the experimental results show that ADAM gradient optimization algorithm can improve the accuracy of neural network.Finally,a large number of tests were carried out on the CPSO-BP-ADAM model,and the results showed that the maximum accuracy and recall degree of the training set could reach0.9975,0.9971,0.9971,and 0.9971.The CPSO-BP-ADAM transformer fault diagnosis model was evaluated by multi-classification cross entropy of training set and test set,diagnosis accuracy and time complexity.The results show that the fault diagnosis model optimized by the deep learning algorithm has high stability,strong learning ability and high generalization,which proves the effectiveness and reliability of the CPSO-BP-ADAM optimization algorithm.