Scholarlink

<p>This study presents a novel simulation–optimization framework for comprehensive groundwater contaminant source identification that addresses critical computational limitations in existing methodologies. The framework integrates a high-order numerical scheme, utilizing quintic B-spline functions for spatial discretization and an adaptive Runge–Kutta algorithm for temporal integration, with a robust chaotic differential evolution optimization algorithm. To overcome the computational intensity associated with traditional approaches, two advanced surrogate modeling techniques are implemented and compared: the concentration response matrix approach and support vector regression. These surrogate models substantially reduce computational requirements while maintaining solution accuracy, making the framework feasible for complex real-world applications. A key strength of the proposed methodology is its ability to identify all major source characteristics, including the number of sources, their locations, release histories, and active stress periods, without prior knowledge, even under observational noise levels as high as 20%. Its robustness is demonstrated through two hypothetical aquifer examples. Even under significant observational noise (α = 0.20), the model maintains acceptable accuracy, with a normalized error value of 7.09 while preserving accurate identification of source locations. Quantitative results highlight substantial computational savings. For instance, in the second example, solving the problem requires approximately 90000 simulations, resulting in around 61.25 h of computation. In contrast, the same problem is solved using the response matrix in 0.09 min and support vector regression in 45.94 min. When both the number of sources and active stress periods are unknown, the model must be executed repeatedly across multiple configurations, further increasing the computational load. These findings emphasize the importance of surrogate models in making simulation–optimization approaches computationally feasible and demonstrate the potential of the proposed methodology for practical applications in real-world groundwater contamination scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Quintic B-spline and chaotic differential evolution-based approach for identifying multiple contaminant source characteristics in groundwater systems

  • 期刊论文

  • 发表日期：

    2025年9月12日

查看全文
  • Ersin Bahar,
  • Gurhan Gurarslan

摘要

This study presents a novel simulation–optimization framework for comprehensive groundwater contaminant source identification that addresses critical computational limitations in existing methodologies. The framework integrates a high-order numerical scheme, utilizing quintic B-spline functions for spatial discretization and an adaptive Runge–Kutta algorithm for temporal integration, with a robust chaotic differential evolution optimization algorithm. To overcome the computational intensity associated with traditional approaches, two advanced surrogate modeling techniques are implemented and compared: the concentration response matrix approach and support vector regression. These surrogate models substantially reduce computational requirements while maintaining solution accuracy, making the framework feasible for complex real-world applications. A key strength of the proposed methodology is its ability to identify all major source characteristics, including the number of sources, their locations, release histories, and active stress periods, without prior knowledge, even under observational noise levels as high as 20%. Its robustness is demonstrated through two hypothetical aquifer examples. Even under significant observational noise (α = 0.20), the model maintains acceptable accuracy, with a normalized error value of 7.09 while preserving accurate identification of source locations. Quantitative results highlight substantial computational savings. For instance, in the second example, solving the problem requires approximately 90000 simulations, resulting in around 61.25 h of computation. In contrast, the same problem is solved using the response matrix in 0.09 min and support vector regression in 45.94 min. When both the number of sources and active stress periods are unknown, the model must be executed repeatedly across multiple configurations, further increasing the computational load. These findings emphasize the importance of surrogate models in making simulation–optimization approaches computationally feasible and demonstrate the potential of the proposed methodology for practical applications in real-world groundwater contamination scenarios.