Research On Network Security Situation Awareness Technique Based On Deep Neural Network

With the continuous development of information technology,communication network and communication devices are increasingly infiltrating into human production and life,and the rate of risk brought by various kinds of cyber threat attacks is aslo rising rapidly.The security technologies such as existing firewall,intrusion detection system,and the virus killing can only carry out preventive and protective measures against the threat attacks that have occurred,and cannot hold the trend of network system in the macroscopic situation.In this case,network security situational awareness quickly becomes an important technology in network security that can actively defend threat attacks.Because the deep neural network has great advantages in dealing with highly complex environment and nonlinear data,this thesis takes use of two key steps in the process that deep neural network has an awareness in the situation,which means namely situation assessment and situation prediction.The specific work is as follows:Firstly,in view of lacking subjectivity of the network security situation assessment method and nonlinearity of the data obtained through the situation elements,an improved gravitational search algorithm based on linear programming method combined with analytic hierarchy process,solving analytic hierarchy process requires expertise experience to give the problem that the subjectivity of judgment matrix in analytic hierarchy process.Then,by using fuzzy neural network which can deal with highly complex and nonlinear data the accurate evaluation value can be obtained better,and the improved gravitational search optimization algorithm is used to optimize the fuzzy neural network evaluation model,improving the convergence speed of the model and solving the problem that the algorithm is in the optimal position.Finally,through the simulation analysis,the rationality and better consistency of linear programming analytic hierarchy process and the superiority of improved gravitational search algorithm are verified,and this evaluation model can obtain better evaluation results.Secondly,in order to quickuly and accurately predict the situation value in the next moment of network security situation,network security situation prediction method based on the improved Nadam algorithm is proposed to optimize the online update mechanism of the long-short term memory network.On the one hand,considering that the situationtime series cannot make better use of the online evaluation.This thesis proposes a situation time series prediction model based on the long-short memory network of the online update mechanism,which minimizes the cost function and more ehhectively improves the prediction accuracy of the model by updating the model parameters online.On the other hand,in order to improve the slow convergence speed in the process of model network training,the adaptive estimation momentum algorithm of nesterov acceleration gradient is improved by Look-ahead method so that it can accelerate speed of the predictive model.