Research On Anomaly Detection Method Of Industrial Control Network

With the advent of a new round of industrial revolution,the traditional Industrial Control System(ICS)is gradually becoming highly integrated with information technology,and the intelligentization of industrial production processes has become a mainstream trend.In recent years,ICS has gradually expanded from the initial based on specific operating system and communication protocol to general operating system and standard communication protocol,which effectively reduces the development cost of the system and shortens the development cycle of the system,but at the same time,it also brings some security risks to the system.Because ICS contains the key information in the production process,once information security problems occur,not only will the system function be lost,but also may lead to casualties,environmental pollution and other major accidents.Therefore,how to ensure the information security of ICS has become a research hotspot and difficulty in this field.In this paper,the anomaly detection technology of ICS network is deeply studied.On the basis of summarizing the widely used and representative anomaly detection methods at present,aiming at the possible anomalous communication behavior between enterprise information network and process control network,and fully mining the relevant information of data packets on communication link,a machine learning method for predicting communication anomalies in industrial control network is proposed.Firstly,the principal component analysis method is used to reduce the dimension of the original communication data to ensure the scientificity of data eigenvalue selection.Secondly,a feedforward artificial neural network is constructed,and the fish swarm algorithm is improved by referring to the calculation of particle inertia mass calculation in the gravitational search algorithm.The improved fish swarm algorithm is used to optimize the input weights and thresholds of the extreme learning machine.Finally,anomaly detection of on-line data in industrial control network communication is realized by off-line training classifier.The simulation and test results of classical data sets,industry data sets and industrial field data show the effectiveness and applicability of the proposed method,and achieve the purpose of using machine learning to predict whether industrial control network communication behavior is abnormal.The paper has 27 pictures,18 tables and 50 references.