Research On Network Security Situation Prediction Based On Optimized GRU

Network security situation prediction technology can actively prevent and curb the occurrence of large-scale network security events and reduce the losses caused by network attacks.It is an important link and core goal in the field of network situation awareness.In recent years,the continuous emergence of new attack means has brought new challenges to the network security situation prediction technology.How to improve the analysis and fitting ability of the prediction algorithm to the security situation is the most concerned problem of researchers.Taking the network security situation prediction workflow as the research route,aiming at the problems existing in the two stages of network data preprocessing and security situation prediction result output,this paper proposes a network security situation prediction method based on optimized GRU.The main work and results are as follows:Firstly,aiming at the problem that the existing network data feature selection algorithms do not fully consider the positive effect of correlation between features on classification,a feature selection method based on m RMR and chaotic immun clone selection algorithm is proposed in this paper.m RMR is used to pre filter features to reduce information redundancy;Then,through the selection,cloning and mutation mechanism of immune cloning algorithm,the overall change of feature subset is promoted,the population richness is significantly improved,and the positive synergy generated by multi feature combination is fully considered.In addition,chaos operator is used to improve the original mutation mechanism,effectively improve the gene activity of the population and accelerate the generation of benign mutations.Secondly,aiming at the problem that the existing network security situation prediction model does not pay attention to the historical information sequence,this paper proposes a GRU network security situation prediction method based on attention mechanism.The algorithm introduces the attention mechanism into the network security situation prediction,calculates the correlation between the current prediction task and the past memory information,guides the model to predict the network future situation with the help of similar historical experience,and effectively improves the accuracy and fitting degree of the model prediction.In addition,the attention measurement based on mahalanobis distance is used to replace the original attention scoring mechanism,which makes it more suitable for the network security situation prediction scenarios with inconsistent data scales and non independent distribution.Finally,the effectiveness of the algorithm is verified on the network security dataset NSL KDD.The experimental results show that the data clustering effect of the proposed feature selection algorithm is improved,the feature subset size is further compressed,and the algorithm has high adaptability to all kinds of feature samples;The accuracy and other indicators of the situation prediction algorithm occupy an advantage in all the control groups.It has a high adaptability to unknown attack means and can effectively reduce the network loss.The overall trend prediction and fitting ability of the algorithm is high,and the average prediction error is reduced by 22.1% compared with the original GRU model,which proves the effectiveness of the algorithm.