| With the rapid development of Internet,the scale of Internet service providers and network users is increasing.Internet has profoundly changed the way of production and life of the masses.The development of the Internet also brings more challenges to the quality of network services and maintainability of manageability.Network delay is not only an important index to measure the performance of the network,but also an important factor affecting the experience of the user network.The pre management and control of network delay is of great significance in maintaining network state,dynamic bandwidth allocation and improving user experience.Based on the IVCE platform,in order to meet the requirement of improving the prediction accuracy of network delay,a complex network time delay prediction model combined with time series model and neural network model is designed and implemented.The main contents of this paper are as follows.(1)On the basis of the traditional fixed order AIC criterion of time series analysis model,in view of the self similarity of network delay and the pertinence of the long phase,a new time series order criterion,AICcc criterion,which considers the correlation degree,is proposed.On the basis of AIC criterion,this criterion increases the correlation degree of the model,and strengthens the fitting accuracy of the change trend of the network delay data before the improvement,and realizes the unification of the three kinds of fitting accuracy,complexity and correlation degree of the optimal model.In the test data set,the model selected by the AICcc criterion is compared with the AIC criterion,and the average of MSE is reduced by 6.45%.The average correlation degree is increased by 3%and the modeling time is not increased.The simulation results verify the feasibility of the proposed AICcc criterion in improving the prediction accuracy of network delay.(2)A complex prediction model of network delay based on the combination of time series model and neural network model is proposed,in which the time series model predicts the long-term autocorrelation components of time delay,and the neural network model predicts the short term fluctuation component of time delay.The composite model has better fitting accuracy than the single time series model for random fluctuation and abrupt change of data.According to the experimental data,the composite model proposed in this paper reduces the prediction error by 17.4%compared with the single time series model.It shows that the proposed complex model can better mine the network time delay variation.(3)A network delay prediction model based on IVCE is designed and implemented.The module consists of four functional modules:network delay measurement module,delay result recovery module,time delay result aggregation module and delay result prediction module.It realizes a series of work from time delay data acquisition,data cleaning and formatting,aggregating data according to specified conditions,setting up pretest models and analyzing modeling performance.In this paper,the network delay is predicted and the results are analyzed in four different scenarios.Compared with the general time series model,the model RMSE is reduced by 10%to 30%,and the correlation is increased by 0.1 to 0.2.It can be considered that the model proposed in this paper effectively improves the accuracy and relevance of network delay data prediction. |
