Research On Network Security Situation Prediction Method Based On LSTM And Its Variants

With the development of intelligent network,people’s life is becoming intelligent,convenient and efficient.At the same time,the network security situation is under severe test.Network security situation prediction technology aims to solve the security problems of cyberspace and provide theoretical support to network security administrators.Network security situation prediction is to mine and reasonably analyze a large number of network security elements,and finally achieve the purpose of preventing external network attacks and network intrusion.Combined with neural network technology,this dissertation proposed a network security situation prediction method based on long short-term memory neural network(LSTM)and a network security situation prediction method based on its variant Gated Recurrent Units(GRU).Combined with the idea of optimization of hyperparameters,different algorithms are utilized to optimize the hyperparameters of the model,The experimental verification and comparison are carried out as follows.1.A single-model prediction method based on LSTM is proposed.LSTM is an improved recurrent neural network(RNN),which solved the problem of RNN in dealing with longdistance dependence.This dissertation used the advantages of LSTM in dealing with longdistance dependence to elevate the accuracy and efficiency of situation prediction model.To elevate the efficiency of the proposed method,The improved Sparrow Search Algorithm(SSA)is used to optimize the hyperparameters of the prediction model,the proposed method trained the situation time series data through the optimized prediction model and predicted the final security situation value.Finally,the experimental simulation and analysis are carried out,and the classical KDDCUP99 dataset is used as the data to verify the proposed method,the performance of the optimized model and the unmodified model are compared.2.In order to improve the single-model method,a multi-model method based on AttentionCNN-BiGRU is proposed,which also used the idea of hyperparameters optimization and utilized the improved PSO algorithm to optimize the hyperparameters of the prediction model.GRU used fewer parameters and shorter training times than LSTM in the training process,and there is no great difference in model performance,Bidirectional GRU(BiGRU)is used to learn and train long-distance dependent situation data in this dissertation;According to the influence of each situation attribute on the security situation value,the Attention mechanism is adopted to give different weight values;In addition,CNN neural network is combined for multi-model situation prediction.Finally,the experimental simulation and analysis are carried out.The KDDCUP99 dataset and the security data from CNCERT are used as the experimental data.The multi-model method is verified on different datasets.Results showed that compared with the single-model method,the multi-model method can complete the task of situation prediction with better capacity and performance,but the training times of the multi-model method needs to be improved.