[{"data":1,"prerenderedAt":75},["ShallowReactive",2],{"research-zh-tw":3},{"id":4,"title":5,"approach":6,"awards":9,"diagram":18,"extension":23,"lab":24,"lead":25,"limitations":26,"meta":27,"problem":28,"publications":31,"results":59,"role":72,"stem":73,"__hash__":74},"research_zh_tw\u002Fzh-tw\u002Fresearch.yml","空中計程車（UAM）光害預測",[7,8],"我把問題拆成兩部分。考慮幾何資訊的神經網路代理模型（PyTorch 3D-CNN + MLP）學習機體在無遮蔽環境下的反射；只需訓練一次，就能在不同城市共用。","各城市的建築由幾何遮蔽修正處理。可見性事先計算並快取，預測時只需要快速查表。參考資料由 Unreal Engine 5（Nanite／Lumen）模擬產生。",[10],{"title":11,"event":12,"paper":13,"links":14},"獎勵發表賞（Encouragement Presentation Award）","日本資訊處理學會 第 105 回 ITS（智慧運輸系統與智慧社區）研究會，2026 年","Surrogate-Based Dynamic Risk Mapping of Light Pollution for Urban Air Mobility",[15],{"label":16,"href":17},"得獎名單","itsAwardList",{"inputs":19,"surrogate":20,"occlusion":21,"output":22},"城市與飛行資料|3D 建築、UAM 軌跡與姿態、太陽方向、觀測位置","神經網路代理模型|學習無遮蔽環境下的反射光；訓練一次，跨城市共用","遮蔽修正|由建築幾何建立各城市的可見性快取；預測時快速查表","反射光地圖|各地點的反射光曝露，用於比較航線與時段","yml","大阪大學 山口研究室","空中計程車（Urban Air Mobility, UAM）的機體會把太陽光反射到街道與窗戶。我開發了一種方法，能在數秒內（而非數小時）預測整座城市的反射光，讓人在實際飛行前就能比較航線與時段。","這是晴天條件下的模擬研究。輸出是相對的曝露指標，尚未與人體不適感或安全門檻進行校準。",{},[29,30],"UAM 機體的位置與姿態不斷改變，因此反射光的方向，以及哪些建築會擋住它，每一刻都在變化。","以都市尺度做精確的光線追蹤需要數小時。只用神經網路雖然快，但當建築遮住機體時會高估反射光。",[32,44,51,55],{"title":33,"authors":34,"venue":35,"status":36,"links":37},"A Simulation-based Framework for Dynamic Light Pollution Prediction in Urban Air Mobility","Ying Chieh Wang（第一作者）, Tatsuya Amano, Hirozumi Yamaguchi","IEEE SmartComp 2026・義大利墨西拿・pp. 136–143","已發表・出版",[38,41],{"label":39,"href":40},"IEEE Xplore","ieeeXplore",{"label":42,"href":43},"DOI","doi",{"title":13,"authors":34,"venue":45,"status":46,"links":47},"日本資訊處理學會研究報告（ITS）・Vol. 2026-ITS-105, No. 38, pp. 1–8・2026 年 5 月","已發表・獲獎",[48],{"label":49,"href":50},"議程","itsProgram",{"title":52,"authors":34,"venue":53,"status":54},"A Geometry-Aware Neural Surrogate for City-Scale Simulation of Light Pollution from Urban Air Mobility","Journal of Information Processing（JIP）・期刊擴充版","已投稿・審查中",{"title":56,"authors":34,"venue":57,"status":58},"A Real-Time Urban Air Mobility Simulator Using a Ray-Tracing Surrogate Model","DPSWS 2026（日本資訊處理學會 DPS 研討會）・Demo／Poster 論文","已接受",[60,64,68],{"value":61,"label":62,"note":63},"3 小時 → 1.62 秒","評估時間","1 萬個觀測點、5 秒時間窗；城市完成一次性初始化後的時間（Core i5-13600K + RTX 3090）。",{"value":65,"label":66,"note":67},"10–54%","RMSE 降低","相較未修正的神經網路預測；Rome、Taipei、New York 共 18 組設定。",{"value":69,"label":70,"note":71},"+32–91%","高樓密集區的 precision","New York 相對於未修正神經網路預測的相對改善；遮蔽修正會犧牲部分 recall。","在與指導教授討論研究方向的基礎上，我主導了方法設計、模擬環境、資料處理、模型實作、實驗、評估與論文撰寫。","zh-tw\u002Fresearch","hkqk2f1bVrSynU9mfpulhNpyP20aSsZyQtARuxWo1J4",1789832904438]