The Research Of Network Traffic Prediction Method Based On Improved Echo State Network

With the development of Internet technology and scale,people's requirements for network management are getting higher and higher.The network traffic is one of most important parameters for evaluating network load and running status.Continuous monitoring and accurate predicting of the network traffic sequence are very important to network management and control.Because the network traffic sequence has characters of timeliness,nonlinearity,mutability,chaos and diversity,traditional prediction models such as Neural Network,Grey Model and Support Vector Machine are difficult to meet the requirements of existing network traffic prediction tasks.Echo State Network(ESN)overcomes shortcomings of traditional Recurrent Neural Network with its unique reservoir structure,and has powerful nonlinear processing ability and high training speed.Although ESN has achieved some success,there are certain problems in network traffic prediction.Firstly,ESN randomly generated reservoir structure and weights affects its ability to handle the nonlinear network traffic sequence and meet the real-time requirement of network traffic prediction.Secondly,mutability and chaos of network traffic have a negative impact on the prediction of ESN.Finally,ESN key parameters selected by experience are inadequate for network traffic prediction tasks.In this paper,for the above problems,we study ESN with multiple loops reservoir structure,network traffic denoising and parameter optimization method based on Grey Wolf Optimizer.The details are as follows:In order to handle the timeliness and nonlinearity of network traffic sequence,a novel ESN with multiple loops reservoir structure(ESN-MLRS)is proposed.The multiple loops reservoir structure avoids the randomness of reservoir in the classic ESN,strengthens the connection of neurons in the reservoir and improves the nonlinear approximation ability of ESN.The simulation results demonstrate that ESN-MLRS has stronger processing ability to nonlinear network traffic sequence and better real-time performance.Furthermore,to handle the mutability and chaos of network traffic sequence,a new network traffic prediction method based on Local Preserving Projection denoising(LPP)and ESN with double loops reservoir structure(ESN-DLRS)is proposed.The network traffic denoising algorithm based on LPP denoises the raw network traffic sequence to reduce the mutability and chaos of network traffic sequence.In addition,a network traffic prediction model based on ESN-DLRS is constructed,which takes both the denoised network traffic sequence and the raw network trafficsequence as input to improve its prediction accuracy.The simulation results demonstrate that the proposed method can achieve better performance on network traffic prediction compared with other similar methods.Finally,in order to handle the diversity of network traffic sequence,a network traffic prediction parameter optimization method based on Grey Wolf Optimizer(GWO)is proposed.This method uses GWO algorithm to iteratively optimize the parameters in the network traffic prediction method based on LPP and ESN-DLRS,adaptively selects the optimal parameters,which improves the self-adaptive ability of the proposed prediction method for network traffic prediction and avoids the limitations of empirical selection.The simulation results demonstrate that the parameters selected by this proposed method are more suitable for network traffic prediction tasks and can effectively improve prediction accuracy.