Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/29627
Title: | 自組織特徵映射網路於戴奧辛指紋比對之應用 The Application of SOM on Dioxin Fingerprint Matching |
Authors: | Yu-Hsin Chang 張宇欣 |
Advisor: | 馬鴻文 |
Keyword: | 類神經網路,自組織特徵映射網路,戴奧辛指紋比對,戴奧辛衰變,U-matrix,多排放源共同污染, Fingerprint matching,SOM,half-lives of dioxin,U-matrix, |
Publication Year : | 2007 |
Degree: | 碩士 |
Abstract: | 近年來,隨著台灣工業日益發達,電弧爐、焚化廠和燒結廠等戴奧辛排放源亦增加,相關污染事件時有所聞,然而目前追蹤污染源的方法,往往受到許多限制,且需耗費相當多的時間、精力在分析比對的工作上。然而隨著類神經網路的應用範圍越來越廣,將之應用於污染源追蹤的研究也越來越多,但是在戴奧辛污染源追蹤的部分仍闕如。
本研究便應用類神經網路中的自組織特徵映射網路(SOM),作為戴奧辛指紋比對的模式,以期更有效率地追蹤到污染來源;並進一步探討戴奧辛之衰變、代謝所造成的指紋改變,及多個排放源共同污染對指紋比對結果的影響。 研究結果顯示,SOM戴奧辛指紋比對模式的主要優點為:(1)在短時間、有限資訊下,便能有效率地找到污染源;(2)因使用U-matrix呈現拓樸結果,故不需要其他方法輔助,便能很快地辨識出分析結果;(3)無須定出指標元素,亦不需要假設樣本內的變數為獨立線性組合所構成,故較無部分資訊遺失的問題;(4)具有『鄰近區域』的觀念,更能呈現樣本空間訊息。然而亦有待改進之處,包括當有兩個以上且貢獻度相當的污染源時,便無法確切得知污染源究竟有哪些,此問題仍待後續研究解決之。 在戴奧辛衰變方面,發現戴奧辛於大氣、植物和土壤中的衰變,並不會對其指紋造成明顯的改變,但可進一步用來判別最初追蹤到的污染源是否為正確或誤判;而於水中及生物體內(本研究使用戴奧辛在人體血液中的半衰期數據)的衰變、代謝,則對戴奧辛指紋有很大的影響,進而造成污染源追蹤的誤判,故當受體為水及生物,或是這些介質為戴奧辛傳輸至受體過程中重要的一環,則於污染源追蹤工作時,考量戴奧辛在水與生物中的衰變是必須的。 此外,亦發現戴奧辛衰變配合於某介質停留的時間,會對戴奧辛指紋改變與否有很大的影響。故若能取得更可靠且在地性的半衰期數據,加上對戴奧辛的傳輸途徑有完整的認知,以及擁有於各介質存留時間的相關可靠資訊,將能更確切瞭解戴奧辛衰變對其指紋的影響,且對於戴奧辛指紋比對結果的準確性將有很大助益。 而將本研究建立之SOM戴奧辛指紋比對模式,配合模式測試所得的法則,應用在94年度高雄市區的戴奧辛實際檢測資料,結果確實能找到幾處採樣點的戴奧辛主要來源,且與階層集群分析所追蹤到的污染源雷同,但本研究所使用之方法的輸出結果較容易辨識,且更能觀察出考量衰變因子前後的指紋比對結果之異同,進而能夠剔除污染源追蹤之誤判情形。 為能盡快找到污染事件的責任歸屬,進而有效遏止戴奧辛污染事件的發生,尋求更有效率、準確度佳的污染源追蹤方法是必須的。除此之外,預防重於治療,故有賴政府各部會、民間各企業組織及社會大眾等共同努力,才能確實有效地減少污染事件的發生。 Nowadays, with fast development of industry in Taiwan, dioxin pollution events take place more often. The methods usually used for identifying pollution sources are bounded, and take a lot of time to analyze the results. To find out pollution sources effectively, a dioxin fingerprint matching model based on SOM Toolbox is built in this study. The half-lives of dioxin and the scenario of multiple pollution are also considered to investigate the influence of them on dioxin fingerprints. The main advantages of SOM dioxin fingerprint matching model are as follows: a) it is able to find out the dioxin pollution sources in the short time and from the limited information; b) by using U-matrix to visualize the SOM results, no other means are needed to help analyzing the topologies; c) there is no need to transform the variables to some linear functions, so that no information would be missed; and d) because of the concept of ‘neighborhood’, the spatial information of samples are displayed well. However, when there are more than two pollution sources, and the contribution of these sources are similar, SOM can’t identify exactly which the sources are. Moreover, the decay of dioxin in air, plants and soil doesn’t have any obvious influence on fingerprints, but it is the basis for judging whether there are only one or more than two pollution sources. On the other hand, the half-lives of dioxin in water and human blood are influential on fingerprints and may cause error of identifying pollution sources. As deduced from here, dioxin decay data and residence time in media both play very important roles in affecting the changes of dioxin fingerprints. As a result, the more local and complete decay data is gained, and the more specific the transportion of dioxin is known, the more accurate the dioxin fingerprint matching will be. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/29627 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 環境工程學研究所 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-96-1.pdf Restricted Access | 2.47 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.