The Measurement And Prediction Of Industrial Network Delay Based On ε-svr

The measurement and prediction of industrial network latency is an important method to analyze industrial network performance. The application of network delay measuring and forecasting generates a lot of new techniques and new tools, which providing different methods and solvement to optimize the performance of the network control system, so now how to establish the model of delay measuring and forecasting and how to solve that become a major trend of current study.Based on ε-SVR(Support Vector Regression, ε is not sensitive parameters), this paper studies the related theories and modeling methods of static and dynamic delay measurements and predictions in industrial network, whose main work including the following aspects:In order to collect the data of network delay, the method for measuring network delay is studied and discussed in this paper, focusing on the end-to-end network delay measurement.In addition, this paper uses the Ping to measure round-trip delay in network. The results in end-to-end delay measurement can be used as the samples in establishment of prediction model.This paper process the datas which collected through the Ping method, establishes samples, and proposes the forecasting model of static industrial network delay based onε-SVR to achieve the prediction in this respect. Applicate the support vector regression theory and use the RBF kernel function to establish a static delay predictive model, deduce the decision function, finally, establish the coefficients to be determined in decision function.Take simulation study on establishing static industrial network delay prediction model through using the RBF function of MATLAB. By changing the width of the prediction model center, the number of sensing unit and the size of sample packet, talking to the influence of the parameters about the effect of the static network delay prediction model.Based on ε-SVR static industrial network delay prediction model can only complete predictive analysis with the offline static delay data, not the real-time changes in dynamic network delay. Therefore, this paper analyzes the characteristics of the dynamic industrial network delay data, on the basis of the static industrial network delay prediction model, based on further research to establish a method of dynamic industrial ε-SVR forecasting modelbased on network delay, further research to a method of establishing dynamic industrial network delay prediction model based on ε-SVR. This method separate to constantly changing dynamic network latency data to the same length,give the correction method of sample data, use the gaussian radial basis kernel function to establish the dynamic delay predictive model, deduce the decision function, finally, establish the coefficients to be determined in decision function.In order to verify the feasibility of the dynamic network delay prediction model, the prediction algorithm of dynamic industrial network delay is designed, which is taking simulation trainings for some sample data that obtained through experiments according to the length of selected sample data set. Calculate the delay prediction error as the basis to the simulation results, and verify the feasibility of dynamic industrial network delay prediction model established base on ε-SVR.