当前位置：
X-MOL 学术
›
J. Hydrol.
›
论文详情
Our official English website, www.x-mol.net, welcomes your
feedback! (Note: you will need to create a separate account there.)
A deep-learning-based multiobjective optimization for the design of in-situ uranium leaching system under multiple uncertainties
Journal of Hydrology ( IF 5.9 ) Pub Date : 2024-12-25 , DOI: 10.1016/j.jhydrol.2024.132576 Wenjie Qiu, Yun Yang, Jian Song, Weimin Que, Zhengbang Liu, Haicheng Weng, Jianfeng Wu, Jichun Wu
Journal of Hydrology ( IF 5.9 ) Pub Date : 2024-12-25 , DOI: 10.1016/j.jhydrol.2024.132576 Wenjie Qiu, Yun Yang, Jian Song, Weimin Que, Zhengbang Liu, Haicheng Weng, Jianfeng Wu, Jichun Wu
Optimizing a field-scale in-situ leaching (ISL) of uranium system under uncertainty in sandstone reservoirs by simulation–optimization (SO) models is a challenging problem, often referred to as a noisy optimization problem. The practical utility of classic stochastic optimization methods has been limited, particularly when addressing subsurface spatial heterogeneity and sophisticated reactive transport modeling (RTM). A traditional method of handling uncertainty in optimization is accomplished by the Monte Carlo sampling strategy to discover reliable optimal solutions, but such an optimization process is generally redundant and computationally prohibitive for the ISL design problem. Here, this study intends to incorporate a novel noise-handling strategy into a multiobjective evolutionary algorithm, along with using a deep learning-based proxy model to substitute the time-consuming simulation model. By doing so, the recurrent residual U-Net model is constructed as an alternative to RTM simulator for predicting spatiotemporal uranium recovery concentrations, and a k -Nearest-Neighbor averaging (kNN-averaging) denoising technique is introduced to deal with the uncertainty associated with the objectives in optimization. We then demonstrate the performance of the proposed methodology through the field-scale ISL design problem in northeastern China. Comparative results indicate that the proposed approach is of robust operation and efficient performance in finding optimal solutions to the noisy multiobjective optimization problems in terms of both convergence and diversity. The study provides a novel, computationally efficient approach to assist multiobjective decision-making processes for ISL system design under multiple uncertainties.
更新日期：2024-12-25
京公网安备 11010802027423号