一個基於BIM的建築判例推薦系統

Abstract

建築資訊建模(Building Information Modeling, BIM)是建築、工程、營建產業中最受學界與業界關注的發展之一。一個建築的生命週期從規劃、設計、招標發包、施工到營運維護管理，影響整個建築工程之成本與品質以前期階段的規劃、設計最為顯著，變更設計所花費成本也會隨著工程的進行不斷增加。建築設計圖是建築執照審核發放的一個關鍵要素，設計與現行的建築法規發生牴觸而導致審核不能通過，就必須在有限的時間內依照建照審查結果的裁判書提供修正或補齊的資料。本研究提出一個基於BIM，語義網路及文字探勘方法的建築判例推薦系統，用來幫助建築師能找到過往相似的判例以及解決方案，同時為了減少因人為輸入關鍵字查詢誤差而造成推薦結果錯誤之問題，本研究希望透過以檔找檔建立自動化輸入資料與動態調整關鍵字權重建立機器學習機制兩個方法來提升推薦的準確度。根據實驗結果顯示，與舊有的方法比較，推薦準確度從85~87.5%提升為92.5%，著實能改善加強舊有的推薦機制，確實能幫助建築師在有效期限內找到低成本的解決方法，通過建照的再次審查。

Building Information Modeling (BIM) is one of the popular issues in building, engineering, and construction industries. The life cycle of a building starts from planning, design, development to maintenance , it spans a long period of time. The two most critical phases are planning and design. A successful application of construction license relies on a building design which meets the existing building codes . If the application was rejected , an updated application had to be refiled . In this thesis, in order to help architects to find similar legal cases and their corresponding solutions , we develop a recommendation system .This recommendation system considers text-mining algorithms and semantic web techniques. Also , in order to reduce the number of false recommendations caused by human errors , we incorporate two techniques File-by-File and Search Engine Optimization method(SEO) into the system . By examining the experimental results , we notice that the proposed recommendation system has a precision ratio up to 92.5% , The system can facilitate the application process for building designers.