跳到主要內容 :::

年度	111	專案性質	實驗性質
專案類別	研究專案	研究主題	調查
申請機構	國立中央大學	申請系所	應用地質研究所
專案主持人	陳瑞昇	職等/職稱	教授
專案中文名稱	利用人工智慧結合政府跨部會多元資料建立地下水硝酸鹽氮污染潛勢預測模型
中文關鍵字	硝酸鹽氮，人工智慧，人工神經網路，多元資料，預測變量，反應變數
專案英文名稱	Combining Artificial Intelligence and inter-ministerial and inter-agency multivariate data to establish a predictive model for groundwater nitrate pollution vulnerability
英文關鍵字	Nitrate,Artificial intelligence(AI),Artificial neural network,Multivariate data,Explanatory variable,Response variable
執行金額	610,000元
執行期間	2022/7/11 至 2023/6/30
計畫中文摘要	地下水為我國淡水資源供給的重要來源，高硝酸鹽氮濃度地下水會威脅人體健康且是地下水質保護關注的議題，預測高濃度硝酸鹽氮地下水發生位置的工作，對於水資源達到保護與發展減緩衝擊的策略。以數據驅動化的機器學習為基礎的人工智慧方法在近年已成功應用於準確地預測地下水硝酸鹽氮污染情勢。本計畫目標為發展結合機器學習與跨部會政府多元資料建立地下水硝酸鹽氮污染潛勢預測模式，機器學習所需的預測變量資料庫包括來自政府跨部會土地利用、水文、土壤型式、水文地質與硝酸鹽氮輸入等資料。考慮的4個機器學習方法包括線性迴歸、隨機森林、極限梯度提昇與人工神經網路。4個機器學習模型的功能評估藉由15折交叉驗證進行，利用Shapley 值法來篩選主要影響的預測變量，結果顯示果樹土地利用為最重要的影響變數，接著為未飽和層坋土厚度與未飽和層礫石厚度，此結果建議應該教育果農應減少化肥或雞糞的使用，推廣有機果樹種植以有效減少地下水硝酸鹽氮污染。
計畫英文摘要	Groundwater is a vital source of the fresh water supplies in Taiwan. Presence of high nitrate nitrogen concentration can pose threat to human health and is already a major concern for groundwater quality protection. Efforts for predicting the locations of high nitrate nitrogen groundwater are required for the protection of water resources and development of mitigation strategies. Most recently, artificial intelligence (AI) techniques like data-driven machine learning (ML) approaches have proven their worth for successively and accurately characterizing nitrate nitrogen groundwater pollution occurrence. In this project, we aim to develop groundwater nitrate nitrogen pollution vulnerability prediction model by combining machine learning and inter-ministerial and inter-agency multivariate data. The data set for predictor variables used for machine learning includes land use, hydrological conditions, soil type, hydrogeology and nitrate concentration collected from inter-ministerial and inter-agency. Four m machine learning methods including linear regression (LR), random forest (RF), extreme gradient boosting method (XGBboost) and artificial neural network (ANN) are used to develop the AI prediction model. Fifteen-fold cross-validation is performed to evaluate the prediction performance. Ultimately, dominant predictors that significantly influence the nitrate nitrogen concentration are examined using Shapley value evaluation method. Land use type of fruit farming is the most important factors, follows by silt thickness of unsaturated soil and gravel thickness of unsaturated soil. This finding implies that fruit farmers should educate the farmer to reduce the use of chemical fertilizers and chicken manure and promote the organic fruit farming for effectively minimizing nitrate nitrogen groundwater contamination.