Scene-Oriented Modeling And Forecasting Of 3G Mobile Data Services

With the growth of mobile users and the increase of services requirements, how to provide far more services and to guarantee the quality of services using limited frequency and network resources, becomes an important topic in maintaining and optimizing of the third generation networks. Traffic prediction is a very important aspect in network optimization, and is of great significance to optimum allocation of network resources. The accuracy and interpretability of the prediction results have a direct impact on network investment and construction scales. Thus the prediction for the future service volume attracts more and more attentions of the operators.For the third generation networks, on different geographic scenes, network resource allocation, wireless environments, user behaviors and QoS are quite different, for instance, the city center and suburbs in rural areas. On the same scene, user behaviors are also great distinct in different time ranges, such as traffic demands in the scene of large-scale stadiums at competition and non-competition periods are markedly different. Different user behaviors in different scenes lead to different characteristics of data service traffic. This thesis proposes scene-oriented traffic prediction that can reduce the complexity and improve the accuracy of forecasting. In addition, considering historical traffic data is limited, by looking for the scenes of similar dimension and matching similar templates, the paper can improve forecast accuracy.The paper aims to use traffic data that does not distinguish specific service types and specific user types and apply the time domain decomposition method to predicate future traffic. For regular time, the trend components, mutation components, period components and random components of the traffic time series are extracted and forecast models are established. For specific time at different scenes, the regular components and the mutation components are extracted to build prediction templates for typical scenario, such as the template for major competitions. According to the above analysis, we establish different forecast logic, combine prediction components to forecast the traffic in the future, and make reasonable explanations to the prediction results.Finally, we utilize the programming language C++, the.NET platform and SQL Server2005 to develop the prototype of traffic prediction for mobile data services. The prototype includes the function modules, such as data preprocessing, data aggregation, traffic prediction and major event templates, and encapsulate a series of prediction logics. Applying of the prototype to traffic prediction in real worlds indicates that the prediction schemes proposed is effective and can provides higher accuracy of forecasting for mobile data services.