基于运营系统的出租车出行需求短时预测模型  

Short-Term Prediction Model of Taxi Passenger Demand Based on Operation Systems

作　　者：林永杰[1] 邹难[2] 

机构地区：[1]美国西北大学土木与环境工程系,美国埃文斯敦60208 [2]山东大学控制科学与工程学院,山东济南250100

出　　处：《东北大学学报（自然科学版）》2016年第9期1235-1240,共6页Journal of Northeastern University(Natural Science)

基　　金：国家科技支撑计划项目(2014BAG03B04);中央高校基本科研业务费专项资金资助项目(2014JC036)

摘　　要：出租车系统作为城市交通运输系统的重要组成部分,其宏观规划和调度管理的合理性决定了出租车服务质量.本文主要研究出租车乘客出行需求估计及预测,为出租车规划和实时调度提供数据支持.首先,分析了出租车定位系统和计费系统,改进了传统出租车需求网格划分方法,考虑了地形、建筑群和道路网络特征,保持了网格自身出行特性的完整性.其次,根据实时收集的出租车数据,建立了易于计算的出行需求估计方法.最后,以实际数据为基础,对影响短时出行量的主要变量进行了相关性分析,提出了基于人工神经网络的短时需求预测模型,根据相关性分析确定了模型结构.以实际获取的出租车数据为例,验证了提出的需求估计和预测模型.结果证明:相比于传统自回归滑动平均模型,提出的人工神经网络模型其平均绝对误差百分比提高了32%.此外,人工神经网络模型的绝对误差百分比超过50%的概率低于10%,而自回归滑动平均模型高达23%.As a critical component of urban transportation systems, the service level of taxis is significantly affected by taxi planning and dispatching. The objective of this study is to estimate and predict taxi passenger demand to support for planning and dispatching. Firstly,the data collection of the in-vehicle taxi GPS system and fare collection system are analyzed in the paper. In terms of data analysis,the traditional grid partition of taxi demand is improved by adding other factors, such as topography, buildings, and road network. The developed partition preserves the completeness of passenger demand in a grid. And then,an easy- to-use estimation method of grid- based demand is presented by the usage of real-time taxi GPS system and fare collection system. Finally, an artificial neural network （ ANN） model is developed to predict short-term taxi demand. The structure of the ANN model is designed based on the functional characteristics of the input-output pairing correlation. Taking the field data from taxi operation system as an example, the performance of proposed estimation and prediction models is evaluated and validated. The results reveal that the proposed ANN prediction model significantly outperforms the existing auto-regression-moving-average （ARMA） model in terms of the reduction of 32% on average absolute percentage error. Moreover, the probability of absolute percentage error greater than 50% for both ANN and ARMA models is 10% and 23%, respectively.

关 键 词：出租车运营 乘客需求估计 网格 短时预测 人工神经网络 

分 类 号：U491.54[交通运输工程—交通运输规划与管理]