On Congestion Control Algorithms Based On Neural Network

Network congestion control is a very important research content in network communi-cation.If the communication network can’t be effectively controlled,it will have serious consequences for the whole communication system.In recent years,with the rapid development of the Internet industry,the network has been integrated into every corner of people’s life.Information interaction is the guarantee of the normal operation of the whole network,but due to the increase of the number of users,the increase of the amount of transmission information and other factors,the network communication delay becomes larger and the quality of data communication becomes worse.Only through the effective congestion control algorithm to control the communication network process,can we achieve the best communication effect without congestion and the maximum use of network resources.The main work of this paper is as follows.Firstly,according to the characteristics of TCP dynamic network model,a PID active queue management algorithm based on adaptive neural network is designed.This paper analyzes the problems of the active queue management algorithm realized by PID controller,and introduces neural network on the basis of PID controller,which not only has the advantages of simple design and stable performance of the former,but also makes up for the disadvantages of the former that the parameters can not adapt to the environment adjustment.In order to respond to the changes of network environment more quickly and accurately,the learning rate of neural network is designed to be adaptive,and the size of learning rate is controlled by the error of the length of route buffer queue.The simulation results show that the designed controller can make the buffer queue length converge to the set value quickly,and maintain small oscillation in the transient process,especially in the case of network parameters change,it has better transient performance and steady-state performance than the PID controller and the traditional neural network PID controller.Secondly,for the nonlinear TCP dynamic network,considering the complexity and time-varying of the network model and the influence of some uncertain factors,an active queue management algorithm based on Neural Network PID sliding mode controller is designed.The designed controller combines PID sliding mode and RBF neural network,which not only has the robustness of sliding mode controller,but also can compensate for the disturbance of uncertain factors after the introduction of integral term.Basis on this,RBF neural network can approach the uncertain term and make the controlled network system have strong anti-interference ability.The simulation results show that in the same network environment,the controller designed in this paper can quickly restore the route buffer queue to the set value,showing better robustness and faster system response,and has certain advantages compared with the traditional sliding mode controller.Finally,the research work of this paper is summarized,and the future research direction is prospected.