Network Security Situation Prediction Implemented by Attention and BiLSTM

Abstract

With the increasing diversification and complexity of network security attacks, it is becoming more and more difficult to predict the network situation. In order to improve the effect of situation prediction, this paper constructs a network security situation prediction model for a Improved Particle Swarm Optimization and Attention fusion Bidirectional Long Short-Term Memory (IPSO-ABiLSTM). First, there is no real situation value for the UNSW-NB15 data set, and a situation value is generated based on the impact of the attack. Secondly, the particle swarm algorithm is improved. The IPSO algorithm makes the algorithm's global and local search capabilities more balanced and faster to converge. Finally, optimizing the hyperparameters of the BiLSTM network fused with the attention mechanism to obtain the final model, and combined with PSO-BiLSTM network, PSO-LSTM network, BiLSTM model for performance comparison. The experimental results show that the IPSO-ABiLSTM in this paper has a fitting degree of up to 0.9922, and the error value is relatively smaller, which verifies the effectiveness of the model proposed in this paper in the network security situation prediction problem.