关键词: 流量预测, 通用无线分组业务小区, 自回归模型, 自回归移动平均模型, 乘积季节自回归求和移动平均模型

Abstract: The performances of some classic time series prediction models were analyzed together concerning the traffic prediction of General Packets Radio Service （GPRS） cells. Based on summarizing the steps of prediction by time series model, the performances of Auto-Regression （AR） model, Auto-Regression Moving Average （ARMA） model, Auto-Regressive Integrated Moving Average （ARIMA） model and multiple seasonal ARIMA model were analyzed. At first, the traffic changes of GPRS cells were studied. Then the autocorrelation coefficient and partial correlation coefficient of traffic were analyzed from different angles, and the AR model and ARMA model of the GPRS cells traffic were proposed. Furthermore, according to cell traffic changes in one day cycle, the multiple seasonal ARIMA model of the GPRS cells traffic was proposed. At the end, with the historical data of traffic of a cell, the three models were applied to predict the traffic sometimes in the future, and comparative study of the prediction performances of the three models were made as well.

Key words: traffic predication, General Packets Radio Service (GPRS) cell, Auto-Regression (AR) model, Auto-Regression Moving Average (ARMA) model, multiple seasonal Auto-Regressive Integrated Moving Average （ARIMA） model