Research On Intrusion Detection Algorithm Of Power Cyber Physical System

With the concept of smart grid and the development of energy Internet,the power system with physical equipment as the core has gradually developed into a power cyber physical system with deep integration of power physical network and information network.Electric power CPS can obtain sufficient and detailed information of electrical physical equipment.Through the collected information,it can control the equipment more accurately and ensure the stable operation of the system.The use of advanced information technology can enhance the accuracy of power grid control,but it also increases the vulnerability of the system.When the important network nodes in the power grid are attacked,it will bring serious disasters to power production.Therefore,intrusion detection(IDS)is very important for power CPS security.It can control the network and analyze the incoming network traffic to detect whether it is attacked.IDS is a powerful complement to the firewall,it can find the firewall can not detect the attack.The main work of this paper is as follows.(1)Some subsystems of power CPS only involve general data,but not sensitive data,so the data can be trained intensively.However,these data using intrusion detection based on CNN model has the problems of slow convergence,low detection rate and over fitting.This paper proposes a bg-cnn model for IDS.Batch normalization(BN)layer is added to the CNN model to normalize the data after convolution operation,fix the distribution of the data,and then input the data into the activation function,so as to improve the convergence speed of the model and reduce the training time.At the same time,the full connection layer in CNN is deleted and the global average pooling layer(GAP)is added to reduce the number of parameters in the model and the occurrence of over fitting.The model is compared with CNN,bn-cnn and gap-cnn on the cicids2017 dataset.The experimental results show that bg-cnn has higher intrusion recognition accuracy and faster training speed.(2)In order to solve the problem that some subsystems in power CPS involve sensitive data and can’t train the data centrally,this paper proposes a fbg-cnn model,which combines federated learning(FL)with bg-cnn model.FL can train a shared global model with collective data without moving local device data,and realize the common modeling of multiple stations,In this way,on the basis of protecting sensitive data,the problem of insufficient label data of single power station is solved,and the accuracy of local network model is improved.The model uses the same data set to compare with fed-cnn model,fed-bn-cnn model and fed-gap-cnn model.Experiments show that fbg-cnn has better classification performance than the other three models,and the privacy of sensitive data is well protected.