Research On Network Traffic Prediction Model Based On LSTM And Wavelet Transformation

With the continuous development of the Internet,the network services are more diverse,and the total amount of network services is growing every year.In order to allocate network resources more reasonably and guarantee the quality of service better,an accurate and effective network traffic prediction model is necessary.However,the network traffic under the current network presents a variety of nonlinear characteristics.This makes the traditional statistical model no longer suitable for describing the characteristics of network traffic.The linear regression method is difficult to accurately predict the network traffic with nonlinear characteristics.Machine learning methods which develop these years provide a new way to solve the problem.After the summary of traditional network traffic models,this paper discusses and studies the support vector machine and kinds of neural networks which are suitable for network traffic prediction,and puts forward two new prediction models based on chaos theory and wavelet analysis theory.The main work and contributions of this paper are as follows:Firstly,a long short-term memory traffic prediction model based on phase space reconstruction is proposed.The model utilizes the method of phase space reconstruction in chaos theory,restores the traffic sequence to the high dimensional space,and constructs the sample data through the reconstructed traffic sequence.In order to learn the nonlinear characteristics of traffic,a long short-term memory recurrent neural network is designed and used to construct the prediction model.The experimental results show that the prediction model based on long short-term memory neural network has higher prediction accuracy than the models based on BP neural network and Elman neural network.Secondly,a hybrid traffic prediction model based on wavelet decomposition and reconstruction is proposed.The model utilizes the method of discrete wavelet transformation in wavelet analysis theory,disperses the trend and fluctuation characteristics of traffic sequence into different component sequences.A component model is constructed for each component sequence according to the support vector machine and neural networks.The methods of segmenting data,training model and predicting subsequent sequence are designed for the component models.The optimal component models are selected by contrasting prediction error,and the prediction of them are integrated by wavelet reconstruction to form the final result.The experimental results show that the hybrid traffic prediction model has high prediction accuracy,and the selection method of component models can effectively improve the prediction accuracy of model based on wavelet decomposition and reconstruction.These two models are constructed based on short-term network traffic and long-term network traffic respectively.The former is suitable for short-term traffic with sudden changes,which has a certain reference value for real-time network traffic prediction;the latter is suitable for traffic with periodic changes and fluctuations,which has a certain theoretical and practical value for long-term network traffic prediction.