Development of optimal decision-making system for road-reconstruction considering human mobility by applying reinforcement learning
完了
小川 芳樹
Western Japan experienced record-breaking heavy rain from June 28 to July 8, 2018. Approximately 600 road sections were closed due to flooding in Hiroshima and Okayama Prefecture. The government develops road-recovery plans to return human mobility to a specific level as soon as possible, however, restoring neighborhood roads was delayed for a week after this flooding. This means that their own plan might not be effective to achieve their own objective. There three limitations government has: 1) lack of prior knowledge, 2) absence of evaluation indicators and 3) the difficulty of estimating human mobility. For solving these problems, we select Deep Q-Network which decide optimal strategies by interacting the environment without prior reference. Plus, we utilized origin-destination pairs from mobile phone GPS data, and digital road map to estimate human movement in real situation. The agent in our model is a single road crew. Traffic volume represents human mobility and is utilized to calculate the human mobility recovery rate, which refers to each action’s reward. We want the agent to tend to prioritize damaged road construction with high influence on human mobility recovery. There are two types of information from analysis result: 1) Sequence of operations, and 2) traffic volume in each link at each step. We believe that government officials could apply our results as a reference to set the priority order.
変更のために新しい申請を保存します。 This will save a new application on the system for a modification.
申請中の研究者は表示されません。 / Pending researchers are not shown.
小川 芳樹 / 東京大学空間情報科学研究センター
ジュ スヒョン / 東京大学生産技術研究所
申請中のデータセットは表示されません。 / Pending datasets are not shown.
国勢調査メッシュ (CSV形式) データセット
平成27年国勢調査地域メッシュ統計その１ 世界測地系（CSV形式）データセット
平成27年国勢調査地域メッシュ統計その２ 世界測地系（CSV形式）データセット
平成27年国勢調査町丁・字等別地図境域データ 世界測地系 (Shape形式) データセット
拡張版全国デジタル道路地図データベース 2016年版
拡張版全国デジタル道路地図データベース 2017年版
拡張版全国デジタル道路地図データベース 2020年版
年次報告の内容はメンバーのみ表示されます。