A Hybrid Neural Network Approach for Congestion Control in TCP/IP Networks

Abstract

In this article, a hybrid approach and model for congestion control in TCP/IP networks using improved neural networks and genetic algorithms is presented. In fact, the primary network traffic flow obtained by monitoring the router buffer has been investigated and the effective parameters have been identified and selected from the remote point of view with the help of the Arma model time series models. The selected model has been used to detect the threshold to increase and decrease the router buffer, Integrated data free of noise and redundancy was presented as input to the neural network algorithm, and at the same time, the genetic algorithm was used in case of crowding, The genetic algorithm has been used to improve the cold start challenge and the transmission rate and the congestion rate, and the neural network algorithm has been developed by relying on the four components of the input rate, the service rate, the percentage of the empty queue and its total in two steps of learning and testing. The proposed method has been investigated from the point of view of root mean square error, the average absolute value of error percentage, coefficient of determination in diagnosis, and correlation error with the help of an applicable and dynamic model. The results obtained with the help of simulations performed in MATLAB and Rapidminer tools show the improvement process compared to RED and DRL-AQM methods.