Zobrazit v EDS

Integrating Deep Learning and Distance‐Based Clustering to Optimize the Field Scale In Situ Uranium Leaching System in Heterogeneous Reservoirs.

Uloženo v:
Podrobná bibliografie
Název: Integrating Deep Learning and Distance‐Based Clustering to Optimize the Field Scale In Situ Uranium Leaching System in Heterogeneous Reservoirs.
Autoři: Qiu, Wenjie, Liu, Dianguang, Yang, Yun, Song, Jian, Que, Weimin, Liu, Zhengbang, Weng, Haicheng, Wu, Jianfeng, Wu, Jichun
Zdroj: Water Resources Research; Mar2026, Vol. 62 Issue 3, p1-21, 21p
Témata: URANIUM mining, CLUSTERING algorithms, DEEP learning, VALUE (Economics), MATHEMATICAL optimization, EVOLUTIONARY algorithms
Geografický termín: CHINA
Abstrakt: Utilization of the integrated simulation‐optimization models for supporting decisions of the in situ leaching (ISL) design of uranium (U) mining is often hampered by the physicochemical heterogeneity within the sandstone reservoirs. Nevertheless, the conventional way suffers from a high conceptual uncertainty due to almost ubiquitous simplifying assumptions used in model parameterizations. Additionally, the increasing complexity of process‐based reactive transport simulators results in substantial computational demands, limiting the feasibility of conducting numerous model evaluations. Addressing the optimization challenges posed by geological uncertainty typically involves Monte Carlo‐based population search methods with evolutionary algorithms which are often computationally intensive and suffer from excessive model redundancy. This study presents a novel optimization framework for identifying the optimal well control strategies for a field‐scale neutral ISL of U mining system in the Songliao Basin, China. The proposed approach integrates a deep learning‐based proxy model with distance‐based clustering components. Specifically, a ResNet‐LSTM network is employed to predict dynamic U recovery concentration. A small subset of representative reservoir realizations is selected through clustering analysis, effectively capturing the uncertainty space without relying on the full ensemble. The subset is then embedded into a heuristic evolutionary algorithm with the objective of maximizing economic benefits. The results demonstrate that this integrated framework significantly enhances the decision‐making process in a computationally efficient way. By integrating the proxy model with cluster‐based realization selection, the proposed procedure achieves a 15.2% improvement in net present value compared to unoptimized scenarios. Overall, the framework provides a versatile and powerful tool for robust optimization in heterogeneous reservoirs. Plain Language Summary: Determining the optimal well control strategy is crucial for the design of in situ leaching (ISL) of uranium (U) mining system. In practice, this process typically involves integrating optimization techniques with numerical simulations to identify wellfield operating strategies. However, the high computational cost associated with predictive reactive transport models, coupled with the substantial uncertainty arising from reservoir heterogeneity, poses a significant barrier to field‐scale optimization. Traditional frameworks often address parameter uncertainty using Monte Carlo simulations to ensure robustness and reliability; however, such approaches are computationally intensive and frequently redundant. We thus propose a novel optimization framework that integrates a deep learning‐based proxy model with a cluster‐based subset selection procedure. The selection of representative realizations is guided by a feature space constructed from both static reservoir properties and dynamic U recovery responses. Distance‐based clustering is employed to identify a representative subset that effectively captures underlying geological variability. Robust optimization is subsequently performed on this representative subset using an evolutionary algorithm to identify optimal well control strategies. The results demonstrate that the proposed method effectively balances economic performance with uncertainty considerations and computational efficiency in heterogeneous reservoirs, highlighting its potential to enhance decision‐making for ISL operations regarding geological uncertainty. Key Points: Hybrid convolution‐recurrent proxy model is developed for U recovery prediction and robust optimizationClustering analysis using static reservoir properties and dynamic features allowing for the selection of representative realizationsCluster‐based subset model significantly enhances optimization efficiency and decision robustness [ABSTRACT FROM AUTHOR]
Copyright of Water Resources Research is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Databáze: Biomedical Index
Popis
Abstrakt:Utilization of the integrated simulation‐optimization models for supporting decisions of the in situ leaching (ISL) design of uranium (U) mining is often hampered by the physicochemical heterogeneity within the sandstone reservoirs. Nevertheless, the conventional way suffers from a high conceptual uncertainty due to almost ubiquitous simplifying assumptions used in model parameterizations. Additionally, the increasing complexity of process‐based reactive transport simulators results in substantial computational demands, limiting the feasibility of conducting numerous model evaluations. Addressing the optimization challenges posed by geological uncertainty typically involves Monte Carlo‐based population search methods with evolutionary algorithms which are often computationally intensive and suffer from excessive model redundancy. This study presents a novel optimization framework for identifying the optimal well control strategies for a field‐scale neutral ISL of U mining system in the Songliao Basin, China. The proposed approach integrates a deep learning‐based proxy model with distance‐based clustering components. Specifically, a ResNet‐LSTM network is employed to predict dynamic U recovery concentration. A small subset of representative reservoir realizations is selected through clustering analysis, effectively capturing the uncertainty space without relying on the full ensemble. The subset is then embedded into a heuristic evolutionary algorithm with the objective of maximizing economic benefits. The results demonstrate that this integrated framework significantly enhances the decision‐making process in a computationally efficient way. By integrating the proxy model with cluster‐based realization selection, the proposed procedure achieves a 15.2% improvement in net present value compared to unoptimized scenarios. Overall, the framework provides a versatile and powerful tool for robust optimization in heterogeneous reservoirs. Plain Language Summary: Determining the optimal well control strategy is crucial for the design of in situ leaching (ISL) of uranium (U) mining system. In practice, this process typically involves integrating optimization techniques with numerical simulations to identify wellfield operating strategies. However, the high computational cost associated with predictive reactive transport models, coupled with the substantial uncertainty arising from reservoir heterogeneity, poses a significant barrier to field‐scale optimization. Traditional frameworks often address parameter uncertainty using Monte Carlo simulations to ensure robustness and reliability; however, such approaches are computationally intensive and frequently redundant. We thus propose a novel optimization framework that integrates a deep learning‐based proxy model with a cluster‐based subset selection procedure. The selection of representative realizations is guided by a feature space constructed from both static reservoir properties and dynamic U recovery responses. Distance‐based clustering is employed to identify a representative subset that effectively captures underlying geological variability. Robust optimization is subsequently performed on this representative subset using an evolutionary algorithm to identify optimal well control strategies. The results demonstrate that the proposed method effectively balances economic performance with uncertainty considerations and computational efficiency in heterogeneous reservoirs, highlighting its potential to enhance decision‐making for ISL operations regarding geological uncertainty. Key Points: Hybrid convolution‐recurrent proxy model is developed for U recovery prediction and robust optimizationClustering analysis using static reservoir properties and dynamic features allowing for the selection of representative realizationsCluster‐based subset model significantly enhances optimization efficiency and decision robustness [ABSTRACT FROM AUTHOR]
ISSN:00431397
DOI:10.1029/2025WR041741