跳到主要內容 :::

年度	108	專案性質	非實驗性質
專案類別	研究專案	研究主題	自訂
申請機構	東海大學	申請系所	環境科學與工程學系
專案主持人	陳鶴文	職等/職稱	教授
專案中文名稱	以人工智慧建立案例推論技術-污染場址辨識技術與整治工法篩選
中文關鍵字	案例式推論, 索引系統, 決策樹分析, 污染場址
專案英文名稱	Establishing Case-Based Reasoning Technology by Artificial Intelligence - Contaminated Sites Identification and Remediation Method Selection
英文關鍵字	Case-based reasoning, Thesaurus, Decision tree, Contaminated site
執行金額	650,000元
執行期間	2018/12/1 至 2019/11/30
計畫中文摘要	環保署針對污染場址之土壤及地下水污染調查評估及整治作業已有一系列的工作規範，且執行至2018年的統計資料顯示，目前全台約有七千多個土壤及地下水管制場址，但仍有許多污染場址未被發現並管制，每年環保署皆須耗費大量經費處理列管場址及新增列管案例，並進行新污染場址的辨識與列管評估；近年來，人工智慧技術與大數據分析已被開發的相當成熟，若結合過去污染場址案例並透過人工智慧技術與大數據分析工具進行污染場址快速辨識，可節省大量的人力、物力及時間成本。為此，本團隊利用決策樹及機器學習分析法建置其污染場址辨識技術(phase I – phase II)，其架構之標的分別為案例收集與案例詮釋、案例索引建置與案例式推論技術，其中案例收集與案例詮釋則由環保署目前所收集的資料進行初步分析，案例索引建置則透過既有表單項目篩選的方式進行索引建立，決策樹建立分為案例搜集與案例銓釋、建立案例特徵、案例改編、案例測試。在案例擷取部分，本團隊利用大數據分析中的非監督分類法進行案例相似度區分，而後利用機器學習中決策樹分析法或其他方法進行推論疑似場址是否為污染場址，最後利用精準度分析評判其模型準確度並將結果回傳至案例庫進行儲存。以上，此方法可提供決策者判斷其場址是否需要被列管之參考依據。
計畫英文摘要	Taiwan environmental protection agency (TWEPA) has already completed a series of working regulations concerning the investigation analysis and remediation operations of the whole soil and groundwater contaminated sites. Also, the regulations has been carried out since then. According to the statistical data of TWEPA in 2018, there are more than seven thousands soil and groundwater monitoring fields with still many of polluting sites not being discovered and monitored. Therefore, massive budget are being spent by TWEPA to only identify, assess, and monitor new polluting sites every year. Recently, Artificial Intelligence and Big Data Analytics have been fully developed, it is of great advantage to conduct rapid identification polluting sites through these technologies in order to reduce considerable amount of budgets. This project aims to construct soil and groundwater contaminated sites identification technology (phase I - phase II) using Machine Learning and Decision tree. Basic Structures consist of Case Collections, Case Interpretations, Build of Index by Cases, and the main core, which are the case-based inference technics. Within the structure, the first two sections are based on data gathered from EPA, while the third part are based on the Expert Conference to construct the case index. As for the Decision tree, it could be mainly split into five major parts, which are the collection of cases, interpretation of case, characteristic of cases, modification of cases and evaluation of cases. From the perspective of capture of cases, this project is using unsupervised learning from the Big Data Analytics to perform classification on likelihood of cases, further then utilizing the decision trees or other learning techniques from Machine Learning to determine whether or not the suspecting sites are actual polluting sites. Lastly, Precision Analysis could help and judge the model then send the result back into the database. To sum, this method could offer the decision makers on the issue of site monitoring.