甘肃科学学报

准确预测不同外部条件下城市轨道交通客流对于轨道交通运营组织、运力调整和资源管理具有重要意义。利用轨道交通自动售检票系统(AFC)数据分析不同时空粒度下轨道交通客流分布特征，提取了时段信息、工作日类型和天气3个城市轨道客流影响因素，在改进粒子群算法(IPSO)的基础上优化支持向量回归(SVR)算法，构建了考虑外部条件因素的IPSO-SVR城市轨道交通客流预测组合模型，通过对比分析验证了模型预测的准确性。结果表明：SVR模型的客流预测性能低于基于时序特征学习的LSTM模型及IPSO-SVR混合学习模型，均方误差(MSE)、决定系数(R²)和相对精度(RA)分别为9.14%、0.86和85.47%;IPSO-SVR客流预测组合模型相较于常用的SVR、LSTM模型具有更好的预测效果，MSE、R²和RA分别为5.54%、0.96和94.37%;所选时段信息、工作日类型和天气3个外部影响变量可有效刻画城市轨道交通客流耦合影响特征，进而提高轨道交通客流预测精度。

Accurate passenger flow forecast of urban rail transit under different external conditions is of great significance for the passenger flow organization and the adjustment of rail capacity resources of rail transit management department.The distribution characteristics of rail transit passenger flow were analyzed by using the AFC data of rail transit under different spatial and temporal granularity.Three influencing factors of urban rail passenger flow, namely time period information, working day type and weather, were extracted.The support vector regression(SVR) algorithm was optimized on the basis of improved particle swarm optimization(IPSO),then the combined IPSO-SVR passenger flow prediction model of urban rail transit considering external influencing factors is constructed.Finally, the prediction accuracy of the proposed model was verified by comparative analysis.The results show that the performance of the SVR model is lower than that of LSTM model based on temporal feature learning and the combined IPSO-SVR mixed learning model proposed in this paper, and the mean square error(MSE),coefficient of determination(R²) and relative accuracy(RA) are 9.14%,0.86 and 85.47%,respectively.Compared with the SVR and LSTM models, the combined IPSO-SVR passenger flow prediction model has a higher prediction effect, MSE,R² and RA are 5.54%,0.96 and 94.37%,respectively.The above results also show that the three influencing variables of time period information, working day type and weather can effectively characterize the coupling characteristics of urban rail passenger flow, and thus improve the prediction accuracy of passenger flow.