Fault Diagnosis Of Key Equipment In Power System Based On Deep Learning

Transformer is one of the key equipments in power system.It has the function of power conversion and distribution,and its operation state is closely related to the sec,urity and stability of power system.The insulation deterioration caused by the long-term operation of the transformer accounts for more than 80%of the total failure of the transformer,and the partial discharge(partial discharge,PD)is the main cause and manifestation of the transformer insulation deterioration.Because different types of PD have different degree of damage to insulation,it is of great significance to accurately grasp the types and characteristics of PD to ensure the safe operation and maintenance of the transformer.This paper uses deep learning to recognize PD pattern.In order to optimize the number of hidden layers,hidden layer nodes and weights at the same time,the brain storm optimization(BSO)is used to optimize the number of hidden layers and hidden layers of the network structure,and then the BSO algorithm and the improved BSO algorithm are used to train the weights.Then,the network is applied to patern recognition of PD.The main content of this paper includes the following areas:Firstly,through the study of automatic coding machine,autoencoder network is constructed,the network application in partial discharge pattern recognition.Compared with other methods,the method of recognition rate is higher than other methods.In addition,the number of hidden layers,the training set and the number of hidden layer nodes is discussed on the network,the number of hidden layers and the training set help to improve the recognition rate,and choose the appropriate hidden layer nodes helps to increase the recognition rate.Secondly,due to the structure of autoencoder network(the number of hidden layers and the number of hidden layer node)is constructed manually,the network structure is optimized by using BSO algorithm,thus realizing the intelligent selection of network structure.Thirdly,the idea of BSO algorithm is introduced into the weights optimization part of autoencoder network,due to some of the pitfalls of BSO algorithm itself,by means of improved algorithm itself of difference BSO algorithm is proposed and the way of combining with other algorithms of simulated annealing BSO algorithm is proposed,and improved BSO algorithm training the weights optimization part of autoencoder network to build the autoencoder network based on improved BSO algorithm.In this process,the evaluation function is redesigned and applied to the partial discharge pattern recognition.In this paper,compared with the method of chapter 3 and the influence of evaluation function and population size on network identification is discussed.Simulation results show that improved BSO algorithm can realize the weight training and the network with new evaluation function can improve the recognition rate of partial discharge,and provide the basis for optimizing the number of hidden layers,the number of hidden layer node and weight at the same time.Finally,according to the analysis on the objective function of autoencoder network,the objective function is the problem of multiple objectives,this paper introduced the cconcept of multi-objective brainstorming optimization algorithm into autoencoder network weights optimization process and built a autoencoder network learning algorithm,the applied the pattern recognition of partial discharge,and enriched the study method.