Nonlinear Model Predictive Congestion Control Based on LSTM for Active Queue Management in TCP Network
Mengzheng Hu, Hiroaki Mukaidani · Asian Control Conference · 2019
In spite of the rapid development of computer network, congestion control becomes an increasingly important problem due to an enormous increase of traffic. Active queue management (AQM) is an effective congestion control approach in the sense that it can be reduced by discarding packets in the buffer of the routers before congestion occurs. In this paper, for the predictive congestion control, long short-term memory (LSTM) that is well known as the most popular recurrent neural network is applied to compensate for the delay for TCP network. Taking into account of the identification of TCP dynamic system, it is shown that the impact of network latency can be reduced and the proposed control scheme stabilizes the router queue length better than other well-known AQM algorithms.