IGA-BiLSTM: An Improved Method for Network Security Situation Awareness

This paper presents an improved prediction model based on LSTM to solve traditional network security situation perception and prediction problems. Due to the significant advantages of BiLSTM in processing network data with time series characteristics and often related content, this paper will design a prediction model based on the Bi-LTM model. As a neural network model, the setting of LSTM’s superparameters is always a difficult point. On the basis of BiLSTM, this paper improves the original algorithm by using the immune genetic algorithm IGA, which enables the new algorithm to select better parameters and process random data on the basis of processing time series data. In this article, we analyze the UNSW-NB15 intrusion detection dataset, which will be used to train and test our model. Compared with the original LSTM algorithm and Bi LSTM algorithm, the prediction accuracy of the final model is improved. In the binary classification problem, the prediction accuracy has been improved by 1%-2%. In the multi classification problem, the prediction accuracy has improved by up to 3%, while F1 Measure has also improved by 2%. In the recognition of different types of anomalies, the F1-Measure of the IGA-LSTM model also has an improvement of 1%-2%. Finally, for situation prediction, the IGA-LSTM model’s situation curve is closer to the actual situation curve.