A study on Travel Demand Forecasting for Various New Transportation Services Utilizing Multiple Data Sources

本研究旨在改善新型交通服务的出行需求预测。本文重点研究了离散选择模型，该模型因其易于跟踪和结构灵活而在出行需求分析中得到了广泛的应用。首先，基于传统的四步需求预测方法，对新建交通系统的需求预测进行了项目后评估。结果表明，需求模型，特别是模式分割模型的时间可转移性较低，通道出行效果的评价可能成为新交通系统需求预测的关键因素。因此，我们利用重复的横断面个人出行数据检验了分解模式选择模型的可转移性，并研究了影响时间可转移性的因素。此外，我们还研究了分区制度在多大程度上影响模式选择模型的估计结果，并在此基础上开发了更多的方法来补充原点和目的地观测结果的准确性。其次，我们研究了各种数据源对需求预测的适用性，以提高其预测精度。开发了一个综合的旅行需求模型，该模型利用了多个数据源（如显示偏好、声明偏好（SP）和汇总数据）的组合估计。将该模型应用于城际高铁的需求预测，结果表明SP和汇总数据的利用可以有效地反映用户对新交通服务的意愿，并捕捉全国范围内的出行需求变化。接下来，我们对SP数据进行了深入分析，提出了建模方法和调查方案，以准确捕捉个体的偏好和出行行为变化。此外，我们调查了从车载设备（如GPS设备）收集的验证车辆数据的可用性，结果表明它在了解路线选择行为的动态方面具有巨大的潜力。第三，我们重点研究了建模方法，以改进ITS和TDM政策的需求预测。基于“有限理性”的概念，建立了嵌入半有序词典式决策规则的出行行为模型。该模型对实际和模拟数据进行了实证应用，与传统的随机效用最大化模型相比，具有较高的数据再现性和预测精度，但模型估计的稳定性较低。因此，我们利用数据挖掘方法开发了估计方法，实证结果表明，在模型中指定阈值参数可能有助于提高模型估计的稳定性。少

This research aims to improve the travel demand forecasting for new transportation services. We especially focused on disaggregate discrete choice models which have been widely applied for travel demand analysis due to its tractability and flexible structure.Firstly, we conducted a post-project evaluation of the demand forecast for a new transit system which was based on the conventional four-step demand forecasting procedure. The result showed that the temporal transferability of the demand model, especially modal split model, seems to be low and the evaluation of the effect of access travel might be a crucial factor in demand forecasting for a new transit system. We, therefore, examined the transferability of disaggregate mode choice model using repeated cross-sectional person trip data and investigated factors which affects the temporal transferability. Also, we investigated to what extent the zoning system may affect the estimation result of mode choice model and then developed the … More methodologies for complementing accuracy of observations on origin and destination locations.Secondly, we investigated the applicability of the various data sources toward the demand forecasting to improve its predictive accuracy. An integrated travel demand model which utilizes combined estimation across multiple data sources such as revealed preference, stated preference(SP), and aggregate data was developed. The model was empirically applied for the demand forecast of intercity high speed rail, and results showed that the utilization of SP and aggregate data might be very effective to reflect the user's intention toward new transportation services and to capture the nation-wide changes in travel demand. Next, we conducted in-depth analysis on SP data and proposed the modeling methodologies and the survey schemes to precisely capture the individual's preference and travel behavior changes. Furthermore, we investigated the availability of prove-vehicle data, which is collected from the on-board equipment such as GPS devices, and the results showed that it has huge potentials to understand the dynamic aspects of route choice behavior.Thirdly, we focused on the modeling methodologies to improve the demand forecasting for ITS and TDM policies. Based on the concept of "bounded rationality", we developed the travel behavior model in which the semi-ordered lexicographic decision rule was embedded. The proposed model was empirically applied for the actual and simulated data and showed high data reproducibility and predictive accuracy compared to the conventional random utility maximization model while its stability in model estimation was quite low. We, therefore, developed the estimation methodologies utilizing data mining method and the empirical results showed that it might be useful to specify the threshold parameters in the model resulting in the improvement of stability in model estimation. Less