Research on Network Transmission Anomaly Detection and Optimization Based on Signal Decomposition and LSTM

In high-performance network transmission, traffic presents significant nonlinear and nonsmooth characteristics, which brings challenges to anomaly detection. In order to achieve accurate identification and trend warning of network anomalies, a network transmission anomaly detection and optimization method based on ensemble empirical modal decomposition and attention mechanism two-way long short-term memory network is proposed. The study adopts the ensemble empirical modal decomposition to extract multi-scale features, combines with Hurst index to locate anomalous segments, and realizes high-precision network traffic prediction and early warning through the bi-directional long- and short-term memory network with the introduction of the attention mechanism. Experimental results show that the method achieves 0.96, 0.95, 0.95, 0.95, and 0.95 in accuracy, precision, recall, and F1, respectively; the detection latency is stabilized at about 280ms, and the occupancy rate of the central processor does not exceed 30%. The study shows that the method can improve the detection accuracy and advance warning ability of anomalous traffic in complex encrypted network environment, and provide efficient and intelligent technical support for network security protection.