Application And Research On Telephone Traffic Load Prediction With Support Vector Machine

In recent years, with the rapid development of mobile communication industry in our country, the GSM network is under extensive construction. The network is expanding constantly, and the layout needs to be modified at the same time. The design of the network is based on the prediction of telephone traffic load, therefore, conducting the research on the model of telephone traffic load will help us to learn the trend of the load. We can also supervise the expansion of the network and the modulation of carrier frequency with the help of the research, which can advance the network's operation efficiently and reduce the cost of its construction and operation.General stability and residual irrelevance are needed in the traditional method of predicting time series. But the actual telephone traffic load series sometimes can not satisfy the needs which are mentioned above, and it has the property of being nonlinear and unstable. Support vector regression machine (SVR) is a good method of machine learning to solve the problem of nonlinear regression. SVR is used to model the monthly telephone traffic load and monthly busy hour telephone traffic load in three areas of Xinjiang. The hyper-parameter of SVR is optimized via the DE-strategy and the MAPE criteria is defined as the objective function. A good forecasting property is obtained by the method. The effect on the traffic data modeling with different embedding dimension is discussed and two modeling criteria are showed in this paper.The results of the experiment indicate that the proposed method has a better generalization property compared with other forecasting method. Generally speaking, The strategy of perturbing the current best value is better than perturbing the random value.