Research On Deep Networks Based On Genetic Optimization For Transformer Fault Diagnosis

In the power industry,power transformers undoubtedly occupy a pivotal position in the generation,transmission,distribution and use of power,yet transformer accidents occur from time to time.Ensuring the operation of power transformers can be safe,reliable and stable,which is important to reduce economic losses as well as to avoid human casualties.For oil-filled power transformers,the dissolved gas analysis a general diagnostic method;As so far,the study of how to extract the important characteristic information from the detected Dissolved Gas Analysis data,and then realize the accurate and timely diagnosis of power transformers has become a key issue.In recent years,online monitoring technology has been widely applied,and the storage capacity for operating data records has been significantly increased.The massive monitoring data provides the possibility for the research of transformer fault diagnosis by deep learning.Traditional fault diagnosis methods are particularly strained in handling large amounts of data,so it has been difficult to make a breakthrough in transformer fault diagnosis accuracy.With the dramatic increase in computing power of computers,deep learning methods have special advantages in processing large amounts of data at this stage.To solve the problems of traditional transformer diagnosis methods,based on the convolutional neural network（CNN）’s effective feature extraction ability we can build the fault diagnosis model for power transformers.Discussion and analysis points include the influence of the parameters of the convolutional neural network on the classification accuracy of the model,and optimization of convolutional network structure parameters by comparative experiments.The effectiveness and accuracy of the model are verified by combining with arithmetic examples.To improve the accuracy and stability of fault classification,genetic algorithm can be used to automatically optimize the convolutional network parameters.Genetic algorithm can automatically filter out more suitable convolutional parameters.Each iteration selects the individuals that can improve the accuracy into the next iteration to achieve automatic preference of convolutional parameters,thus improving the classification accuracy of the model.After comparing the examples,the final results show that CNN model based on GA can improve the accuracy of classification and stable the classifying of transformer faults.