Efficient Simulation of Multispecies Contaminant Transport via Analytical Solutions and Machine Learning Surrogate Models

NCU Institutional Repository > 地球科學學院 > 應用地質研究所 > 博碩士論文 > Item 987654321/99471

請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/99471

題名:	Efficient Simulation of Multispecies Contaminant Transport via Analytical Solutions and Machine Learning Surrogate Models
作者:	阮秋媛;UYEN, NGUYEN THI THU
貢獻者:	應用地質研究所
關鍵詞:	One keyword per line;One keywordmultispecies contaminant per line;Decay on sorbed phase;mobile-immobile
日期:	2026-01-21
上傳時間:	2026-03-06 19:04:40 (UTC+8)
出版者:	國立中央大學
摘要:	地下水中的多組分污染會危害人類健康並破壞周邊生態系統。準確預測這些污染物在地下水中的遷移與歸趨過程，對於污染修復與風險評估至關重要，尤其是在地質條件複雜的含水層中。本文建構並評估了三種不同的模型，用以模擬多組分污染物在多孔介質中經歷序列降解過程的運移行為。首先，本文提出一種二維解析模型，用於描述多組分污染物在速率受限吸附條件下的運移過程，其中一階降解反應可同時發生於溶解相與吸附相中。該模型綜合考慮對流–彌散、動力學吸附及序列降解機制，並適用於不同的入口邊界條件。其半解析解與相應的數值模型結果顯示出極佳的一致性。將該模型應用於放射性核素衰變鏈的模擬結果表明，與僅考慮水相降解的常見假設相比，同時考慮溶解相與吸附相中的降解過程會導致顯著較高的溶解相濃度，突顯了雙相降解對長期濃度預測與風險評估的重要影響。其次，本文建立了一個雙域移動–非移動（mobile–immobile，MIM）模型，用以模擬具有流動水區與滯留水區共存之多孔介質中多組分污染物的運移行為。該模型同時考慮各域內的一階降解反應、水相與固相之間的速率受限吸附過程，以及移動域與非移動域之間的一階質量交換。將該模型應用於典型的氯代溶劑降解鏈顯示，所提出的 MIM 架構能夠再現多種污染物明顯的濃度拖尾現象，從而為長期污染物持久性提供一種物理上一致的描述。第三，本文提出一種基於人工神經網路（ANN）的機器學習替代模型，用於快速預測二維域內多組分污染物的濃度分布。該 ANN 模型以解析解所生成的合成資料進行訓練，能夠準確重現母體污染物及其降解產物在時空上的濃度分布特徵。模型完成訓練後，在維持高預測精度的同時，其計算效率相較於解析模型提升約 100 至 700 倍，具體取決於污染物種類與模擬情境。總結而言，本文提出的三種模型為理解地下水中多組分污染物的運移行為提供了一個系統且全面的研究架構，並為大尺度地下水污染評估提供高效且可靠的分析工具。;Multispecies contamination in groundwater can compromise human health and degrade surrounding ecosystems. Accurately predicting the fate and transport of these contaminants is essential for effective remediation and risk assessment, particularly in geologically complex aquifers. This study develops and evaluates three distinct models to simulate the transport of multispecies contaminants undergoing sequential degradation in porous media. First, a two-dimensional analytical model is developed for rate-limited sorption of multispecies contaminants, in which first-order decay is allowed to occur in both the dissolved and sorbed phases. The model describes advection–dispersion, kinetic sorption, and sequential degradation under different inlet boundary conditions, and its semi-analytical solutions show excellent agreement with a corresponding numerical model. Applied to a radionuclide decay chain, the results demonstrate that including decay in both the dissolved and sorbed phases can lead to substantially higher dissolved-phase concentrations compared with the commonly used assumption of decay occurring only in the aqueous phase, highlighting the strong impact of decay in both phases on long-term concentration predictions and risk assessment. Second, a dual-domain mobile–immobile framework (MIM) is formulated to simulate multispecies transport in porous media with coexisting flowing and stagnant water zones. The model incorporates first-order degradation in each domain, rate-limited sorption between the aqueous and solid phases, and first-order mobile-immobile mass transfer. Application of the model to a realistic chlorinated-solvent decay chain demonstrates that the new MIM framework can reproduce pronounced concentration tailing for multiple species, providing a physically consistent description of long-term contaminant persistence. Third, a machine-learning surrogate model based on an artificial neural network (ANN) is proposed to rapidly predict multispecies concentrations in two-dimensional domains. The ANN is trained on synthetic data generated from the analytical solution and accurately reproduces the spatial and temporal concentration patterns of both parent and daughter species. Once trained, the surrogate reduces computation time by approximately 100 to 700 times compared with the analytical model, depending on the species and scenario, while maintaining high predictive accuracy. These three models together offer a comprehensive framework for understanding multispecies transport and provide efficient tools for large-scale groundwater contamination assessments.
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	25	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.....