考虑动态波动性的轨道交通站点短时客流预测方法

摘要

Previous methods on forecasting passenger flow of rail transit lacks consideration of dynamic volatility,and cannot predict the range of short-term passenger flow.Taking typical rail transit stations in Beijing as a case study,an ARIMA-GARCH model is established to simulate the prediction interval (PI),and fit the stochastic volatility of shortterm passenger flow.The effect of "sharp peak and heavy tail" is analyzed by using t distribution.The asymmetry volatility effects are addressed by using T-GARCH and E-GARCH models.Results show that the integrated ARIMA-GARCH models can significantly reduce the mean prediction interval length (MPIL) in forecasting passenger flow by more than 20％,and improve the prediction interval coverage probability (PICP) by about 1％.It is also found that volatility of passenger flow in weekdays is larger than weekends,while no evident volatility exists during non-peak hours.Note that,an ARIMA-GARCH model will not significantly reduce mean absolute prediction error (MAPE).However,the hybrid models can accurately forecast the range of passenger flow of rail transit under the premise of ensuring single-point forecasting.%轨道交通站点客流预测研究缺乏对短时客流动态波动性的考虑,不能预测短时客流区间.以北京市典型轨道交通站点为例开展实证,构建ARIMA-GARCH模型对误差项建模分析,拟合短时客流的随机波动特征.不同于以往的ARIMA-GARCH模型,研究还通过t分布揭示了客流的“尖峰后尾”效应,通过2种非对称GARCH模型识别了短时客流的非对称波动特征.模型结果表明,相比传统ARIMA模型,ARIMA-GARCH混合模型降低了20％以上的客流平均置信区间长度(MPII),同时提高了1％左右的置信区间覆盖率(PICP);周内客流波动性大于周末客流,而非高峰时段的客流不具有波动性.值得指出的是ARIMA-GARCH模型没有明显降低客流预测的平均绝对误差,尽管如此,混合模型可以在保证客流单点预测的前提下,准确地预测地铁客流区间.