Research On Method Of Security Situation Awareness For Industrial Internet Based On Recurrent Netural Network

As an important cornerstone of the new industrial revolution,the Industrial Internet not only promotes the development of manufacturing production methods and enterprise organization models in the direction of digitization and intelligence,but also makes the originally isolated industrial control systems face complex network security issues due to access to the Internet.Security situational awareness technology helps to comprehensively understand the security status and development trend of the Industrial Internet,and establish an active defense mechanism to avoid network risks in a timely manner.This thesis conducts a systematic and in-depth study on the two core technologies of Industrial Internet security situational awareness: situational awareness and situational prediction,and proposes an Industrial Internet security situational awareness method based on recurrent netural networks.The main research contents and innovations include:Aiming at the problems of low classification accuracy and poor evaluation result in Industrial Internet security situation assessment,a dual channel based Industrial Internet security situation assessment method is proposed.Firstly,based on the principles of danger and similarity,indicators such as threat and vulnerability are selected so that they can reflect the current security status of the Industrial Internet,so as to constructthe Industrial Internet security situation indicator system.Then,the integrated neural network is used to extract different feature information,the information extracted from the two is fused,and classify them through the fully connected network.Finally,the security situation value is calculated by the evaluation model and the security situation index quantitation method to evaluate the security state of the Industrial Internet.The experimental results in the power system attack dataset show that the classification accuracy of the method in this thesis reaches 94.37%and 92.82% in three classification and thirty-seven classification scenarios.Aiming at the problems of low accuracy and long training time of Industrial Internet security situation prediction,a method for Industrial Internet security situation prediction based on a simple recurrent unit is proposed.Firstly,the dataset is reconstructed using a sliding window.Secondly,the spatial features between various security elements are extracted based on the convolutional network,and the temporal features between the information are extracted by the bidirectional simple recurrent network,which improves the utilization of historical information.At the same time,with the powerful parallel capability of the simple recurrent unit,the training time of the model is reduced.Finally,the attention mechanism is introduced to dynamically adjust the weight coefficient in the hidden state of the bidirectional simple recurrent network,highlighting the strong correlation factors,and realizing the prediction of security situation value.The comparative experimental results in the power system attack dataset show that the proposed method has better performance in single-step prediction than multi-step prediction.Compared with a single convolutional network and a simple recurrent network,the average absolute error of this method is reduced by 7.05% and 4.46%,respectively.Compared with the network model using long short-term memory network and gating unit,the training time is reduced by 25.19% and14.98%,respectively.While using the attention mechanism will increase the training time by 1.51%,it can reduce the prediction error by 2.09%.