Hybrid LSTM–Transformer Model for Network Traffic Prediction and Intelligent Congestion Control in 6G Networks

Abstract

The development of sixth-generation (6G) communication networks will ensure ultra-reliable low-latency communication, massive machine-type communication, intelligence at the edge, and connectivity ubiquity. Nonetheless, there emerge major challenges associated with traffic prediction and traffic congestion control owing to the growing heterogeneity and dynamism of network traffic. Currently, the congestion control algorithms are designed in a reactive fashion and lead to packet drops, latency, and poor resource utilization in cases of traffic increases. In this regard, this paper proposes a Hybrid LSTM-Transformer (HLT) solution for traffic prediction and intelligent traffic congestion control for 6G networks. The proposed solution merges the temporal learning capabilities of LSTM algorithms with the capability of Transformers to model long-range dependencies. The traffic patterns that are obtained by prediction will be used within an artificial intelligence (AI)-based congestion control module. Various experiments are performed on benchmark datasets of network traffic such as MAWI, CAIDA, UNSW-NB15, and CICIDS dataset. Through the performance analysis, it is found that the proposed HLT framework performs with more accuracy than other models such as ARIMA, RNN, GRU, LSTM, and individual Transformer. Additionally, it is found that proactive congestion control not only decreases packet losses and end-to-end delay but also increases throughput and utilization of the network.