葛研究員が吉川・山口賞を受賞
研究員の葛乾 (Ge Qian)さんが，本学土木系の特に優れた博士論文に対して贈られる吉川・山口賞を受賞しました．おめでとうございます．
Ge Qian さんの博士論文「DATA-DRIVEN AND MACROSCOPIC APPROACHES FOR ESTIMATING ORIGIN-DESTINATION TRAVEL DEMAND」の主要な成果は次の二点です：(1)静的枠組で従来型交通調査から得られた流動表（OD表）をビッグデータを用いて最新のOD表に更新する方法論を構築したこと；(2) OD交通量が時刻に応じて変動する動的枠組みのもとで，巨視的アプローチとデータ指向型アプローチを統合して計算効率性の高い動的OD交通量推定手法を開発したこと．(1)に関しては，エントロピー最大化原理を援用した静的OD表の更新方法の開発を行い，2008年東京都市圏パーソントリップ調査と2012年メッシュ単位滞在人口統計データを用いた実証分析より提案手法の妥当性を確認しています．(2)に関しては，都市域全体を複数の貯水域から構成されるInput-Outputシステムと見立て，異なる貯水域間の流動（エリア間OD交通量）を交通流理論に整合的に記述するモデルを開発しています．これらの研究成果は，交通工学分野の最高ランクジャーナルであるTransportation Research Part C 及び Part Bにそれぞれ掲載されています．
This dissertation concentrates on the problem of inferring origin-destination (OD) travel demand from multiple sources of data. This dissertation attempts to extend previous studies in using new data, building macroscopic dynamic network loading for complex multi-reservoir urban transportation network and developing novel method that incorporates perceived and observed data for estimating path flow. We present a maximum entropy based updating method to yield static OD matrices using aggregate mobile phone data. We adopt this dataset for the following reasons: low cost, easy to collect, and privacy-free. The proposed approach calculates trip flows of each OD pair using two sequential sub-models. Performance of the proposed methodology is validated through a numerical example and confirmed by case study using the data of Tokyo. However, the aggregate mobile phone data is not sufficient for dynamic OD demand estimation (DODE) problem. We build a dynamic network loading model upon the macroscopic characteristics of traffic flow depicted by Macroscopic Fundamental Diagram (MFD). The dynamic net- work loading model for multi-reservoir system (MRDNL) is specified in terms of a system of partial differential equations following the conservation law. Spatial discretization method and numerical scheme are also developed for tracking the vehicles’ behavior guided by this model. To identifying reservoirs in large network more efficiently without using the demand data, we present a community detection method that yields neighborhoods in which the links are closely related and share similar traffic characteristics. Finally, we emphasize on developing a methodology framework for estimating dynamic OD demand using traffic counts with incorporation of observed path cost.
