Artificial Neural Networks for Estimating Arsenic Variation in the Regional Ground Water
Resource
臺灣水利,57(1),1-12
Journal
臺灣水利
Journal Volume
57
Journal Issue
1
Pages
1-12
Date Issued
2009-03
Date
2009-03
Author(s)
Abstract
致癌毒物－砷對環境與人類健康之影響係為社會所關切的議題，本研究架構砷濃度類神經網路推估模式，對區域內砷濃度進行推估工作，提供區域內砷濃度變化，以維護當地民眾使用地下水之安全。台灣西南沿海地區有嚴重砷污染地下水情形，水利署於民國81年至94年間於雲林沿海地區，共設置28口水質監測井，蒐集此區域水質資料作為架構模式之用，在過程中發現本區域雖有長期水質監測資料，惟資料時常發生零星遺漏情形，為解決此問題，本研究以雲林沿海地區為研究範例，架構類神經網路模式推估各水質站內遺漏砷濃度資料，並對沿海地區進行區域砷濃度推估工作。另一方面，由於本區域內水質站之資料不足，造成模式之穩定性不佳，無法建構可靠補遺推估模式，為改善此問題，首先採用交叉驗證以確認模式架構後，並採修正型目標函數搜尋模式中參數，其結果證明此兩種方法可改善模式不穩定性及過度訓練情況，擴大類神經網路之用範圍及應用領域，提供可靠砷濃度空間推估結果，對於了解此區域內砷濃度在時間與空間變化有很大助益，同時依據此結果可減少居民誤引用高濃度砷地下水之危險，達地下水有效管理及利用之目的。
With the great concern for the potential effects of aresenic on human health and the environment, there is a growing need for efficiently determining and modeling the presence and amount of aresenic in the ecohydrogeological systems. In this study,we construct the ANNs model to complete the lost aresenic data according to the relationship of the aresenic concentration of the monitoring wells in the region. The results offer the realization for the aresenic variation and keep the safe from the resident ingesting and usage of the groundwater. Blackfoot disease was once common the southwestern coast of Taiwan, especially in the alluvial fan of Chou-Shui River. In order to monitoring the aresenic concentration, the Water Resource Agency has setup twenty-eight water quality wells which distribute in coasltal area in Yun-Lin county from 1992 to 2005. However, some water quality data were lost or not record because of the man-made amd money factors. The lost data affect the development of the arsenic associate research. Therefore, we choose this region as the example to construct ANNs model to estimate the lost aresenic data.
Due to sparse and lost data, there are not steady and over-fitting problem in the process of constructing the ANNs model. To solve the problems, we first apply cross-validation to assure the architecture of the model amd adopt the modify-objective function to search the optimal weights. It is proven that these two methods can reduce the unsteady and over-fitting problems. The results have been apparently improved the realization of the spatial-temporal distribution of aresenic. Based on the results, the risk of ingesting the high aresenic groundwater can be decreased to reach the goal of efficiently controlling and usage of the groundwater.
Subjects
砷
地下水
人工智慧
類神經網路
修正型目標函數
Arsenic
Groundwater
Artificial intelligence
A.I.
Artificial neural networks
ANNs
Modify-objective function
SDGs
Type
journal article
File(s)![Thumbnail Image]()
Loading...
Name
類神經網路於區域地下水砷濃度的推估.pdf
Size
2.08 MB
Format
Adobe PDF
Checksum
(MD5):f63bfe2fda7c93faa2735954c4eb60bc