Research On Attack Detection Of Distribution Communication Network Based On Machine Learning

With the rapid development of the distribution grid,more and more informationalized and intelligent devices are connected to the power grid through the distribution communication network.The large coverage area,multiple access devices,and scattered deployment of the distribution communication network make it vulnerable to cyber attacks.With the continuous increasing types and numbers of network attacks,traditional attack detection methods not only have more and more prominent problems in missed detection and false detection,but also have disadvantages such as unclear parameters selection,low information utilization,and difficulty in upgrading.It is unable to meet the security requirements of the distribution services for the distribution communication network.As a popular research direction of current network attack detection technology,machine learning classification algorithm has the advantages of high detection accuracy,high model interpretability,and high adaptability to different environments.Considering the practical needs of machine learning attack detection algorithm in terms of architecture flexibility and programmability,this paper integrates the Software Defined Network(SDN)architecture as an attack detection framework into the distribution communication network system.And this paper proposes an attack detection architecture based on SDN for the distribution communication network.Considering that the random forest algorithm has good classification accuracy,no need to preprocess the data,and highly interpretable model in machine learning algorithms,this paper chooses random forest as the attack detection algorithm.In order to solve the problems of the time-consuming in the construction of random forest model and the low classification accuracy of small proportion categories under unbalanced samples,this paper proposes an improved random forest algorithm combined with the improved ReliefF algorithm.The improved random forest algorithm improves the sampling accuracy of the small proportion categories in the data set by setting the sampling weight function,and combines the improved ReliefF algorithm to select the features of the data set before detection,which further reduces the time-consuming process of the attack detection process.Finally,the simulation of attack detection is carried out in the simulated SDN distribution communication network environment.The simulation results prove that the improved algorithm proposed in this paper has a good attack detection effect in the simulated SDN distribution communication network environment.