Abstract:

Addressing the issue that the large amounts of data generated by devices during communication transmission are prone to becoming targets for hackers and malicious users， thereby generating abnormal traffic， and that the sparsity of traffic data makes it difficult to capture the associations between global features， which in turn affects the detection effectiveness of abnormal traffic， a spatiotemporal fusion detection method for abnormal traffic in industrial Internet of Things （IoT） based on the improved bidirectional long short-term memory （BiLSTM） neural network algorithm using mean squared error （MSE） is proposed. Firstly，the industrial Internet traffic data is converted into numerical data through the One-Hot coding method， and the SE mechanism in MSE is used to adjust the weight of traffic characteristics to capture the correlation between global characteristics.Secondly，using the forward and backward LSTM of BiLSTM neural network， the spatiotemporal fusion features of network traffic are extracted.Lastly， and the spatio temporal fusion features are input into the softmax classifier to identify traffic and achieve anomaly detection. The experimental results show that when the number of iterations reaches 30， the loss value of the proposed method can reach below 0.4， when the number of iterations reaches 60， both F1 and Matthews correlation coefficients can reach 60， proving that this method has good overall performance.

Key words: multi head squeezing incentive mechanism, BiLSTM neural network, feature fusion, abnormal traffic detection, Softmax classifier