Research And Application Of Time Series Analysis In Mobile Communication Data

As an important business of the telecommunications industry, mobile communication provides information services to the community, whose adjustment will impact on economic security, people’s living, telecom operators own development. Mobile communications are mostly time series, if we can extract useful information from it, it will be helpful to the development of telecom operators.Support vector machine is based on the structure minimization principle of statistical learning methods, getting rid of the traditional empirical risk minimization principle, and, in theory,it can reach the global optimum.What is more, SVM can solve the problem of small sample size, nonlinear and high dimension.The forecast is both the nature and the important role of time series analysis,while the support vector machine is a good prediction method. Then the time series analysis and support vector machine were carried out in-depth study, the main research work of this paper is as follows:(1) According to the theory of time series, this paper gives a forecasting model based on mobile communication time series data, analyzed the data pre-processing and forecasting learning process.(2)The method based on support vector machine is applied to mobile communication series prediction. Then the solution to two key issues in specific applications for SVM is given:First problem is SVM kernel function selection, I studies the construction method of the kernel function, in view of the localized kernel function learning ability and generalization performance is weak, strong generalization performance of the global nuclear function, learning ability weak, in order to get the strong learning ability and strong generalization ability, these two types of kernel function is combined to construct the mixed-kernel function. And the calculation method used to control the ratio between local kernel function with the global kernel function according to the features of the mobile communication timing sequence is given; The other problem is SVM parameters selection, which is actually a parameter optimization problem.We choose particle swarm optimization algorithm to optimize the parameters of SVM. Howere, traditional particle swarm algorithm is easy to fall into local optimal solution when solving complex problems.To overcome this defect, this paper introduces the concept of population density, simulate the real foraging behavior of biological populations furtherly.(3) Finally, this paper uses IPSO-MSVM prediction model, traditional ARIMA model, the standard SVM prediction model, PSO-SVM prediction model to predict with the same training set--he Mobile SMS traffic from January2012to January2013. As a result, it show that the mixed kernel SVM based on the improved particle swarm optimization has obvious advantages over many other predicting algorithms both in forecasting accuracy and forecasting precision.