Research And Design Of China Telecom's Behavior Model Forecast Based On Cloud Computing

In the era of large data,the amount of data stored by telecom operators is exploding,and the traditional computing and storage model has been unable to adapt to the demand of communication system.Cloud computing technology as a large-scale,high availability and high scalability and other characteristics of massive data processing technology can be used to solve the problem.It also provides solutions for a new generation of telecommunications services platform.The thesis focuses on the application of cloud computing in telecom Tianyi cloud and user model prediction.The thesis expounds the new problems and new challenges faced by the communication operators in the era of large data,and introduces the development history and current development of cloud computing and its research direction and research contents in the communication system.The thesis studies and analyzes the application and function of telecom Tianyi cloud.Tianyi cloud 3.0 combines the telecommunication network and service capability to provide the cloud service with different characteristics of the customer demand.On this basis,the thesis analyzes the nature of Tianyi cloud and the shortcomings in the prediction of user behavior pattern.Based on the study of the implementation mechanism and important implementation details of the Hadoop and the real-time cloud computing framework Storm,the thesis designs a distributed clustering algorithm for popular use.Based on the research of data mining technology,the distributed clustering algorithm under Hadoop is designed based on the traditional system clustering algorithm,which selects the shortest distance method as the recursive formula.The distributed algorithm is divided into two groups: lossy clustering and lossless clustering.In this way,the lossless clustering involves a large number of sample data in the process of parameter transfer,the resource occupancy is high,but it can maintain the same clustering effect as the serial algorithm,and the cluster clustering is used to replace the whole cluster Class,effectively reducing the amount of data.Based on the support vector machine(SVM)algorithm,the algorithm of distributed prediction model based on SVM is designed by converting the user model prediction problem into two kinds of classification problems.The algorithm is divided into two modes: offline and online.In the training phase,the normalized sample data is clustered first,and the SVM model is trained for each cluster satisfying the specific condition.The offline mode involves massive historical data,and uses Hadoop with distributed ways;the online mode corresponds to the application phase,the pattern processing data is relatively small and real-time requirements are high,so the real-time prediction algorithm under Storm is designed.According to the normalized user data,the distance from the cluster center is calculated and the trained SVM model corresponding to the minimum cluster is used to predict the user data.The model algorithm is simulated by Zhenjiang area data,and the performance of the algorithm is studied from the aspects of accuracy and timeliness.It has certain application foreground.