Research On AOS Active Queue Management Algorithm Based On LSTM Prediction

As an important network problem,congestion often occurs in communication networks.Congestion control is also a permanent topic in network research.For network congestion,it generally occurs in the case of traffic overload,and the long correlation characteristics of network traffic provide convenience for the research of congestion control.Generally speaking,on the premise of knowing the traffic data in advance,the congestion control will be greatly improved.Therefore,we can propose the idea: make full use of the long correlation characteristics,and combine the traffic prediction and congestion control to get better results.This thesis focuses on the research of AOS(Advanced Orbiting System)active queue management algorithm based on prediction,hoping to achieve better results in network congestion control.The main research contents are divided into the following aspects:First of all,network congestion,AOS protocol,traffic self similarity principle and multiplexing model are introduced.The idea of optimizing the queue management algorithm to improve the overall efficiency of the system is put forward,and the advantages and disadvantages of several traditional queue management algorithms are analyzed and compared.Aiming at the disadvantages of existing active queue management algorithms,based on the combination of several algorithms,an improved queue management algorithm is proposed.The algorithm is generated from the perspective of queue length and the duration of data sent by the source.Multiple discard method is used to achieve the effect of congestion control.The simulation results show that compared with the traditional algorithm,the proposed algorithm improves the performance of the system significantly in terms of delay and residual error.Secondly,study the prediction model of various self-similar traffic,Analyze the advantages and disadvantages and its applicability from different perspectives.Including the burst of actual sequence,possible information loss in difference process,nonlinear interference such as noise,feasibility of long-term prediction,et al.Finally,the neural network is used to model and predict.Choose between them,LSTM neural network,which is more suitable for the long-term correlation characteristics of network traffic,is selected for prediction.Finally,AOS active queue management algorithm based on prediction is studied.Combining the above two results,an improved congestion control mode is formed.Through simulation and comparison,it can be concluded that the AOS active queue management algorithm based on LSTM prediction has improved in various performance indicators.It can be used as a theory to support engineering practice.