基于不同时间粒度的城市轨道交通短时客流预测

Short-term passenger flow forecast of urban rail transit based on different time granularities

为探究城市轨道交通进站客流量预测精度与时间粒度之间的关系,以西安地铁自动售检票(AFC)系统连续50 d进站客流量数据为研究依据,将地铁运营有效时间划分为5、15、30 min,1、2 h及1 d等不同时间粒度,并对不同时间粒度下客流量时间序列采用Pearson系数法进行相似性度量,然后利用差分整合移动平均自回归(ARIMA)模型对不同时间粒度下的全网进站量进行拟合。以Pearson系数等于0.95作为短时客流预测时间粒度的选取阈值,最终选取在30、60 min及1 d三种时间粒度下用ARIMA模型进行短时客流预测,并与自回归滑动平均(AR)模型、支持向量回归(SVR)模型和BP神经网络预测模型的预测结果进行比较分析。研究结果表明:时间粒度相关性系数变化呈现单波峰形态,在30、60 min及1 d时间粒度下,ARIMA模型预测结果平均相对误差分别为4.12%、3.54%、4.97%;在这4种模型中,ARIMA模型平均预测精度最高,在不同时间粒度下这4种模型的预测误差呈现相同的变化趋势,即平均预测误差由大到小的3种时间粒度分别为1 d,30、60 min。因此,时间粒度大小选取的极端化并不会带来短时客流预测效果的直接提升,优化后的时间序列模型在西安地铁全网进站客流量短期预测方面具有较高的精度,研究成果可为城市轨道交通行车组织优化提供技术支持。

In order to explore the relationships between forecast accuracy and time granularity of passenger flow in urban rail transit, based on the inbound passenger flow data of automatic fare collection(AFC) system for 50 consecutive days in Xi’an Metro, the effective time of metro operation was divided into different time granularities, such as 5 minutes, 15 minutes, 30 minutes, 1 hour, 2 hours and 1 day. The similarity of passenger flow time series under different time granularities was measured by Pearson coefficient method. The autoregressive integrated moving average(ARIMA) model was used to fit and predict the total network entry under different time granularities. Taking Pearson coefficient equal to 0.95 as the threshold of time granularity selection for short-term passenger flow forecast, ARIMA model was finally used for short-term passenger flow forecast under three time granularities of 30 minutes, 60 minutes and 1 day. The forecast results of ARIMA model were compared with those of autoregressive(AR) model, support vector regression(SVR) and BP neural network model. The results show that the time granularity correlation coefficient change presents a single peak form, and the average relative error of ARIMA model for 30 minutes, 60 minutes and 1 day time granularities are 4.12%, 3.54% and 4.97% respectively. The forecast results of the four models show that the average forecast accuracy of the ARIMA model is the highest. The four methods have the same change tendency for the forecast error at different time granularities, that is, the average forecast error decreases according to the three time granularities of 1 day, 30 minutes and 60 minutes. Therefore, the extreme selection of time granularity will not directly improve the effect of short-term passenger flow forecast, the optimized time series model has high accuracy in the analysis and short-term prediction of passenger flow data of Xi’an Metro stations,the research results can provide technical support for the optimization of the operation organization of urban rail transit.