The subway ridership is related with the social activities, such as, commuting, festival, holiday, and so on, which makes the time series of subway ridership presents seasonal characteristic. This characteristic inspires us to decompose the time series into a seasonal component and an epoch component, and employ a combination forecasting method to estimate the future ridership. We first transform the raw ridership into a time series matrix, then decompose the ridership into a seasonal component and an epoch component, and at last combine the individual forecasting results of the seasonal component and the epoch component to make forecast. Contributions of this paper include formulating the combination forecasting problem as an optimization problem, proposing an In-Sample Algorithm (ISA) and an Out-of-Sample Algorithm (OSA), and conducting extensive experiments based on the individual forecasting model named Auto Regressive Integrated Moving Average (ARIMA) model and the data provided by Chongqing Rail Transit. We prove that the decomposition and combination forecasting model possesses smallest variance than individual forecasting models from the theory aspect. The experiments further demonstrate that the ISA algorithm can effectively fit original ridership time series and the OSA algorithm can make better forecasting performance than individual forecasting models. Most importantly, the ISA algorithm and the OSA algorithm both possess advantages of smaller forecasting error deviation and smaller absolute forecasting errors.

地铁客流量与通勤、节日、节假日等社会活动相关，使得地铁客流量的时间序列呈现季节性特征。这一特点启发我们将时间序列分解为季节性分量和纪元分量，并采用组合预测方法来估计未来的乘客量。我们首先将原始乘客量转化为时间序列矩阵，然后将乘客量分解为季节分量和纪元分量，最后结合季节分量和纪元分量的个别预测结果进行预测。本文的贡献包括将组合预测问题制定为优化问题，提出了一个样本内算法（ISA）和一个样本外算法（OSA），并基于自回归综合移动平均（ARIMA）模型的个体预测模型和重庆轨道交通提供的数据进行了广泛的实验。我们从理论上证明了分解组合预测模型比单个预测模型具有最小的方差。实验进一步表明，ISA 算法可以有效拟合原始乘客时间序列，OSA 算法可以做出比单个预测模型更好的预测性能。最重要的是，ISA算法和OSA算法都具有预测误差偏差较小和绝对预测误差较小的优点。