A Fault Diagnosis Method For Power Communication Network Based On Convolutional Neural Network

Electric power communication network is the neural network of power system.Once the network fails,it is very important to diagnose the fault accurately and timely.However,as the topology of power communication network becomes more and more complex,network fault diagnosis becomes more and more difficult.The Traditionally strategy of collecting alarm data from network management manually and locating fault by experience or rules is unable to guarantee the accuracy and judgment speed.Once making wrong decision,it will affect the efficiency of maintenance,and the network security status can not be restored in time.In recent years,with the improvement of computing power of hardware and software platforms,the continuous improvement of machine learning algorithm and the accumulation of massive training sample data,machine learning's advantages in solving complex non-linear problems are becoming increasingly obvious.Based on the properties of alarm data in power communication network and the advantages of machine learning method,this paper presents a fault diagnosis method for power communication network based on Convolutional Neural Network(CNN).Firstly,a typical network topology of power communication network is built.The alarm data is collected after manually manufacturing faults.Secondly,those alarm data is processed by the following steps: redundant deletion,field selection,standardization and synchronization.After that,the alarm data turned into alarm transaction.To measure the importance of each alarm to fault diagnosis,this paper proposes a mean-square-error alarm weighting method,which uses the weight of each alarm for each fault to comprehensively calculate its importance to all fault diagnosis.The fault state matrix is also defined by the alarm transaction coding,which can represent both the network topology connection and the network fault state.Then,according to the size of the collected sample data set,this paper designs and implements a fault diagnosis model of power communication network based on Convolutional Ceural Cetwork.The model can extract fault features from the fault state matrix,and classifies the faults by different features,so as to realize fault diagnosis.Finally,the learning rate,minimum batch size andepoch size of the model are optimized,and the best diagnosis accuracy of the model is99.1%.