考虑日期属性和天气因素的铁路城际短期客流预测方法  

Short-Term Forecast Method for Intercity Railway Passenger Flow Considering Date Attributes and Weather Factors

作　　者：滕靖[1,2] 李金洋 TENG Jing;LI Jinyang(College of Transportation Engineering,Tongji University,Shanghai 201804,China;Shanghai Key Laboratory of Rail Infrastructure Durability and System Safety,Tongji University,Shanghai 201804,China)

机构地区：[1]同济大学交通运输工程学院,上海201804 [2]同济大学上海市轨道交通结构耐久与系统安全重点实验室,上海201804

出　　处：《中国铁道科学》2020年第5期136-144,共9页China Railway Science

基　　金：国家重点研发计划资助项目(2018YFB1201401)。

摘　　要：分析2014—2018年上海至南京的单向铁路客流数据发现,日期属性和天气因素会对铁路城际短期客流的波动产生显著影响。为此,结合对非线性时间序列数据处理具有优势的长短期记忆(LSTM)神经网络模型,以及可弥补模型中超参数设置主观性的粒子群优化(PSO)算法,将日期属性和天气因素纳入模型的影响因素体系,提出1种基于PSO-LSTM组合预测模型的铁路城际短期客流预测方法,以解决因短期客流波动性大、随机性强而产生的准确预测难度大等问题。利用2014—2018年上海至南京的单向铁路客流以及上海的天气信息,设置预测输入步长为14 d、输出步长为7 d,对模型进行实例验证。结果表明:与实际客流相比,该模型的最终预测平均误差为6.75%;与删除1个影响因素的PSO-LSTM组合预测模型,以及结合了BP神经网络的PSO-BP组合预测模型相比,该模型具有最优预测精度。By analyzing the one-way railway passenger flow data from Shanghai to Nanjing between 2014 and 2018, it is found that date attributes and weather factors have significant impact on the fluctuation of short-term intercity railway passenger flow. For this reason, combining the Long Short-Term Memory(LSTM) neural network model which has advantages in data processing of nonlinear time series, with the Particle Swarm Optimization(PSO) algorithm which can compensate for the subjectivity of setting hyper-parameters in the LSTM model, a forecast method for short-term intercity railway passenger flow based on PSO-LSTM combinatorial forecasting model is proposed, which incorporates date attributes and weather factors into the influencing factor system, in order to solve the difficulties in accurate prediction due to the large fluctuation and randomness of short-term passenger flow. Based on the one-way railway passenger flow from Shanghai to Nanjing between 2014 and 2018 and the weather information of Shanghai, the model is validated by setting the forecasting input step length to 14 days and the output step length to 7 days. Results show that, compared with the actual passenger flow, the final average prediction error of the model is 6. 75%. Compared with the PSOLSTM combinatorial forecasting model with one influencing factor deleted and the PSO-BP combinatorial forecasting model combining BP neural network, this model has the best forecasting accuracy.

关 键 词：铁路城际客流 短期预测 长短期记忆神经网络 粒子群优化算法 日期属性 天气因素 

分 类 号：U293.13[交通运输工程—交通运输规划与管理]