A Software Defined Network Traffic Prediction Model Based on LSTM and Feature Compression

This paper proposes a traffic prediction model based on long short-term memory network (LSTM), which uses sliding time window and combines feature compression technology to implement modeling and fitting of key traffic features in SDN environment. By building a real Mininet simulation platform, the newly released data set in 2024 is used for training and testing, and the robustness of the model under various network topologies is verified. The experimental results show that this method is significantly better than GRU, 1D-CNN and other comparison models in prediction accuracy, delay control and abnormal traffic perception. The controller feedback mechanism further demonstrates that the prediction results can effectively guide the scheduling logic, improve link utilization and task completion level. This study provides practical technical support for the realization of intelligent flow control strategies in SDN , and has good engineering deployment operability and expansion opportunities.