Bayesian Neural Network-based IP Bearer Network Performance Prediction

This thesis is based on IP bearer network operation and maintenance systemdevelopment lags behind the status quo, according to the actual needs of the IP bearernetwork, in order to solve the IP bearer network communication business and theoperation and maintenance of system resources can not be shared, business data cannot predict, especially the operation and maintenance personnel can not advancetroubleshooting, as perceived by the user can not get the raise other issues, on thebasis of today’s IP bearer network system performance prediction module concept isintroduced into the IP bearer network. Trying to create a good beginning for thedevelopment and progress of the operation and maintenance of the system of IP bearernetwork performance prediction module core algorithm.First, the thesis focuses on the understanding of the structure and characteristicsof the IP bearer network. On this basis, in-depth analysis of the core issues-performance prediction module for IP bearer network. Combined with practicalapplication, designed IP bearer network performance prediction module. Given thethe corresponding overall framework, as well as data management and dataorganization structure and detailed framework of the various sub-module functions,technical indicators and lay the groundwork for the IP bearer network performanceprediction.Secondly, based on the neural network especially Bias MLP neural networklearning, found that the use of Bayesian MLP neural network in dealing with a largenumber of real-time data with high accuracy and a reasonable fit, and also be able toensure the IP bearer network of strong generalization ability, which makes theprediction results easy to achieve the expectations of the operators. To do this, Try touse Bayesian neural network I to forecast performance IP bearer network. Andembarked on the algorithm and process design of Bayesian MLP neural network,Then through the experimental use of Bayesian MLP neural network to predict the mobile communications business through rate, congestion, load three sample dataanalysis. Prove Bayesian MLP neural network can be used for the IP bearer networkperformance prediction.To further verify the accuracy and reliability of the Bayesian neural networks. BPneural network and RBF neural network two kinds of prediction methods are used topredict experiment for the same sample data respectively. In order to ensure theexperimental rigor, with the two methods of analysis: qualitative analysis andquantitative analysis, ultimately proves Bayesian MLP neural network in the IP bearernetwork performance prediction is indeed accurate and reliable.Finally, do a simple summary of the content, and make a prospect for the nextexperiment.