基于季节分解和SARIMA-GARCH模型的铁路月度客运量预测方法

Method on Monthly Railway Passenger Traffic Forecast Based on Seasonal Decomposition and SARIMA-GARCH Model

首先,根据铁路月度客运量时序图呈现的趋势性、周期性和随机波动性,运用季节分解法将其分解为趋势循环分量、季节因子分量和不规则分量,直观量化地表征出所蕴含的特征信息。接着,引入季节时间序列模型(SARIMA)对平稳化和单整检验后的月度客运量序列的趋势性和季节性进行建模,通过季节差分序列的相关图筛选确定出最佳模型阶数,得到SARIMA基础预测模型。然后,为提高模型对波动性的刻画精度,消除异方差影响,再对基础模型的回归残差进行ARCH检验,构建出广义自回归条件异方差(GARCH)模型,并检验所建SARIMA-GARCH融合模型的稳定性。最后,将融合模型与常规SARIMA、ARIMA和NAR动态神经网络模型的短期预测值进行精度对比验证,并对其中长期预测性能做测试分析。结果表明,SARIMA-GARCH模型短期预测性能优于SARIMA、ARIMA和NAR动态神经网络模型。

Firstly,the trend,the periodicity and the random volatility,exhibited by the monthly railway passenger traffic time series,were decomposed into trend cyclic component,seasonal factor component and random component with the seasonal decomposition method.In this way,the feature information contained in the series of the passenger flows could be presented visually and quantitatively.Secondly,the SARIMA product model was introduced to model the trend and seasonality of monthly passenger volume series after stationary and single-integration tests.Then the optimal model order was identified by correlation graph of seasonal difference series to obtain the SARIMA basic prediction model for seasonal time series.Thirdly,in order to further improve the accuracy of stochastic volatility and eliminate the influence of heteroscedasticity,the ARCH effect test on the regression residual of the basic model was performed to construct the GARCH model,and test the stability of the SARIMA-GARCH fusion model.Finally,the accuracy of the short-term prediction values obtained by the proposed fusion model was compared with that of the conventional SARIMA,ARIMA and NAR dynamic neural network models.Furthermore,the long-term prediction performance of the model was also tested and analyzed.The results show that the short-term prediction performance of SARIMA-GARCH model is superior to that of the conventional SARIMA,ARIMA and NAR dynamic neural network models.