利用機器學習預測濁水溪沖積扇區域之地下水砷汙染;Using Machine Learning to Predict Groundwater Arsenic Pollution in Choushui Alluvial Fan Area

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/89813

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/89813

題名:	利用機器學習預測濁水溪沖積扇區域之地下水砷汙染;Using Machine Learning to Predict Groundwater Arsenic Pollution in Choushui Alluvial Fan Area
作者:	時于凱;Shih, Yu-Kai
貢獻者:	資訊工程學系
關鍵詞:	砷;地下水;地下水水質;濁水溪沖積扇;機器學習;神經網路;隨機森林;Arsenic;Groundwater;Groundwater quality;Choushui Alluvial Fan;Machine Learning;Artificial Neural Network;Random Forest
日期:	2022-07-27
上傳時間:	2022-10-04 12:00:49 (UTC+8)
出版者:	國立中央大學
摘要:	台灣對地下水資源之高度依賴，使得水體品質(簡稱水質)之優劣更為重要。近年來由於台灣濁水溪沖積扇區域嚴重超抽地下水於農業灌溉與養殖漁業，造成水中砷濃度上升，進而影響地下水的用水安全、作物生長甚至對人體健康造成危害，因此，探究並預測地下水砷濃度之變化將有助於強化水資源之使用與管理。目前與地下水砷汙染相關之研究缺乏關注人為抽水之影響，同時，過去研究已指出地下水抽取量可利用抽水馬達用電量推估。因此，本研究使用濁水溪沖積扇區域抽水井之用電資料、觀測井不同含水層水位資料及氣象測站雨量資料作為特徵，同時，採用隨機森林(Random Forest)和人工神經網路(Artificial Neural Network)建構濁水溪沖積扇區域地下水砷濃度預測模型，並探究抽用地下水造成砷濃度變動之關鍵特徵。其中，濁水溪沖積扇砷濃度最高之兩口監測井，彰化縣東興國小與雲林縣台西國小，採用人工神經網路建構之預測模型判定係數(Coefficient of Determination, R2)分別達到 0.723 和 0.705，相關係數(Correlation Coefficient, COR)達 0.999 和 0.989。此外，特徵重要性分析結果顯示，監測井東半邊抽水井之抽水活動，對水中的砷濃度會產生較大的影響，同時，地下水位觀測井之第二含水層與第四含水層之水位變動對於地下水砷濃度預測模型有重要影響。因此，本研究結果顯示利用人工神經網路並使用濁水溪沖積扇區域抽水井之用電資料、觀測井不同含水層水位資料及氣象測站雨量資料作為特徵可有效預測地下水砷濃度。;Taiwan is highly dependent on groundwater resources, making water quality important. In recent years, due to the severe over-pumping of groundwater in the alluvial fan area of Taiwan′s Choushui River for agricultural irrigation and aquaculture, the concentration of arsenic in the water has increased, which in turn affects the water safety of groundwater, the growth of crops and even harms human health. Therefore, investigating and predicting changes in groundwater arsenic concentrations will help strengthen the use and management of water resources. At present, studies related to arsenic pollution of groundwater have not paid attention to the impact of artificial water pumping. At the same time, previous studies have pointed out that groundwater pumping amount can be estimated by using the electricity consumption of pumping motors. This study uses two different machine learning algorithms, Random Forest and Artificial Neural Network, Therefore, this study uses the electricity consumption data of the pumping wells in the Choushui alluvial fan area, the water level data of the observation wells in different aquifers and the rainfall data of the meteorological station as the characteristics. At the same time, two machine learning algorithms, Random Forest and Artificial Neural Network, were used to construct a prediction model of arsenic concentration in groundwater in the Choushui alluvial fan area, and the key characteristics of arsenic concentration changes caused by groundwater pumping were explored. The two monitoring wells with the highest arsenic concentration in the alluvial fan of Choushui is Dongxing Elementary School in Changhua County and Taixi Elementary School in Yunlin County, the Coefficient of Determination (R2) of the prediction model constructed by artificial neural network reached 0.723 and 0.723. And the Correlation Coefficient (COR) reached 0.999 and 0.989. In addition, the results of feature importance analysis show that the pumping activities of the pumping wells in the east half of the monitoring wells will have a greater impact on the arsenic concentration in the water. At the same time, the fluctuation of the water level of the second aquifer and the fourth aquifer in the groundwater level observation well has an important influence on the prediction model of groundwater arsenic concentration. Therefore, the results of this study show that the artificial neural network can effectively predict the arsenic concentration in groundwater using the power consumption data of the pumping wells in the Choushui alluvial fan area, the water level data of the observation wells in different aquifers and the rainfall data of the meteorological station.
顯示於類別:	[資訊工程研究所] 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	36	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.....