Multi-scale Risk Assessment Model of Network Security Based on LSTM

摘要

As the problem of network security becomes more and more serious, how to accurately perceive the current network security situation and discover the attack behavior in time has become the focus of research in the field of network security. This paper proposes a multi-scale risk assessment model for network security based on Long Short Term Memory neural network (LSTM). The model utilizes the wavelet transform to decompose network traffic time series into sub-sequences of various scales. The LSTM network is used to predict the sub-sequence of wavelet decomposition. By comparing the difference between the actual sub-sequence and the predicted sub-sequence, the model determines whether there is "anomaly" in the network traffic time series. The anomaly detection results of each sub-sequence are summarized into a network security risk value to assess the risk of network security. The introduction of the multi-scale technology improves the detection accuracy of network traffic time series anomaly detection and effectively enhances the reliability of the risk value.