LSTM-DAM: Malicious Network Traffic Prediction for Cloud Manufacturing System

Abstract

With the rapid development of Internet of Things (IoT), the applications of cloud manufacturing system are growing dramatically, resulting in increasing network heterogeneity and complexity. Network traffic prediction plays an important role in the stable operation of cloud manufacturing systems and the optimal configuration of network systems. However, existing works perform poorly confronting the data which has long time series properties and complex temporal features. To address this problem, we construct a malicious network traffic prediction model based on long and short-term memory (LSTM) neural network and dual attention mechanism. Integrated with the dual attention units of feature space and time sequence, our LSTM model can realize the dynamic correlation between malicious traffic and features series. We first obtain the weight parameters of the input data based on feature attention mechanism, and then leverage LSTM model with the attention mechanism to form a temporal attention module. These two modules strengthen the influence of key historical information. Finally, the malicious traffic prediction result of cloud manufacturing systems can be obtained from our model. The experimental results on real industrial dataset show that the prediction effect of LSTM-DAM model is better than LSTM and CNN-LSTM. Based on CIC-IDS-2017 dataset, the method also performs well in Internet malicious traffic prediction, representing great generalization ability.