應用因子分析與人工神經網路建立硝酸鹽氮污染潛勢的預測模型

NCU Institutional Repository > 地球科學學院 > 應用地質研究所 > 博碩士論文 > Item 987654321/86914

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/86914

題名:	應用因子分析與人工神經網路建立硝酸鹽氮污染潛勢的預測模型
作者:	林采蓁;Lin, Tsai-Chen
貢獻者:	應用地質研究所
關鍵詞:	硝酸鹽氮;因子分析;人工神經網路
日期:	2021-08-30
上傳時間:	2021-12-07 13:25:29 (UTC+8)
出版者:	國立中央大學
摘要:	近年來臺灣用水需求大增，水資源的品質與供應量成為重要議題。臺灣雖降雨充沛但地表水資源保留不易，地下水成為重要水資源來源。不適當之土地使用產生的有害化學物質滲透到非飽和土壤中，最終會到達下面的含水層系統，因此，淺層含水層的地下水水質受到土地利用和非飽和土壤的顯著影響。本研究的目的為應用因子分析及人工神經網路的方法建立土地利用與硝酸鹽氮污染潛勢區的預測模型，首先應用因子分析的方法，分析結果指出因子三中硝酸鹽氮、果樹與礫石層呈現正相關，根據此結果將因子三之得分作為人工神經網路的輸入參數，以硝酸鹽氮濃度作為輸出結果，而後根據前人研究加入其他水質參數，包含鈣、鎂、鈉、鉀、砷、氯鹽、硫酸鹽、鐵、錳、總有機碳、碳酸氫根做為人工神經網路的輸入參數。人工神經網路的結果顯示在2層隱藏層且每層神經元個數為12個的模式下有最好的預測效果，R^2達到0.664，證實土地利用、地質材料、水質與硝酸鹽氮的污染相關性。得出此架構為最佳的預測硝酸鹽氮濃度模型後，本研究將所有的123筆資料以此模型架構進行最後的預測，得出R^2為0.95，達到很好的預測效能，為預測地下水硝酸鹽氮的污染潛勢提供了一個很好的預測模式。政府可依據結果採取土地利用管理措施更好地預防或控制地下水污染，規劃完善的水質保護計畫以維護臺灣居民使用地下水的安全。;In recent years, the demand for water greatly increases with Taiwan’s population growing cause the water quality and supply become an important issue. Although Taiwan has abundant rainfall, which is not easy to retain surface water, and cause the groundwater become an important water resource. The harmful chemicals coming from inappropriate land use permeate through unsaturated soil and ultimately reach the underlying aquifer system. Groundwater quality in shallow aquifer is thus significantly affected by the land use and unsaturated soil. This study’s target is to develop a predictive model for occurrence of nitrate pollution in Taiwan coupled with artificial neural network. First, the method of factor analysis is applied. According to the results of factor analysis, it is indicated that the nitrate, fruit trees, and gravel layer are positively correlated. Based on the result, the scores for factor 3 used as the input data of the artificial neural network, and the nitrate concentration is the output result. Then, based on previous studies, other parameters were added, including Cl^-, SO_4^(2-), TOC, Fe^(3+), Mn^(2+), Ca^(2+), Mg^(2+), Na^(2+), K^+, HCO_3^- and As. The result of artificial neural network show that the model with 2 hidden layers and 12 neurons in each hidden layers has the best predictive effect. The determination coefficient is 0.664. This framework as the best model for predicting the concentration of nitrate. The study uses the final model to predict, R^2 can reach 0.95, which achieves good prediction performance, and provides a good prediction model for predicting the groundwater nitrate vulnerability. The government can strength land use management or take a better plan to prevent or control groundwater pollution and plan a comprehensive water quality protection plan to maintain the safety of Taiwan residents using groundwater.
顯示於類別:	[應用地質研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	125	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....