Scholarlink

<p>The enrichment and migration of geogenic high-arsenic contaminated groundwater are controlled by complex hydrological conditions and hydrogeochemical processes. Conventional reactive transport models encounter dual challenges, including poor convergence and low computational efficiency, when simulating multi-pathway reaction networks. To address these limitations, this study develops an improved physics-informed neural network (PH-PINNs) by integrating physical constraints with the PHREEQC geochemical module. The proposed model explicitly accounts for microbially mediated reaction networks involving iron-sulfur-carbon-nitrogen cycles, which drive arsenic enrichment and mobilization in groundwater. Leveraging field monitoring data from the Shanyin experimental site in the Datong Basin, Shanxi Province, a two-dimensional numerical model was constructed to simulate site-scale arsenic reactive transport. Results indicate that the PH-PINNs model significantly outperforms traditional PINNs in capturing nonlinear and non-stationary dynamics, reducing the root mean square error (RMSE) by over 50%. The model exhibits superior predictive accuracy, numerical stability, and adaptability for complex, multi-component reaction networks. This framework provides a robust tool for advancing both theoretical research and practical management of arsenic-contaminated groundwater systems.</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

Improved physics-informed neural network for reactive transport modeling of groundwater arsenic enrichment

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • Fang Deng,
  • Jichun Wu,
  • Yun Yang,
  • Junxia Li,
  • Xianjun Xie,
  • Jianguo Jiang,
  • Xiaobin Zhu

摘要

The enrichment and migration of geogenic high-arsenic contaminated groundwater are controlled by complex hydrological conditions and hydrogeochemical processes. Conventional reactive transport models encounter dual challenges, including poor convergence and low computational efficiency, when simulating multi-pathway reaction networks. To address these limitations, this study develops an improved physics-informed neural network (PH-PINNs) by integrating physical constraints with the PHREEQC geochemical module. The proposed model explicitly accounts for microbially mediated reaction networks involving iron-sulfur-carbon-nitrogen cycles, which drive arsenic enrichment and mobilization in groundwater. Leveraging field monitoring data from the Shanyin experimental site in the Datong Basin, Shanxi Province, a two-dimensional numerical model was constructed to simulate site-scale arsenic reactive transport. Results indicate that the PH-PINNs model significantly outperforms traditional PINNs in capturing nonlinear and non-stationary dynamics, reducing the root mean square error (RMSE) by over 50%. The model exhibits superior predictive accuracy, numerical stability, and adaptability for complex, multi-component reaction networks. This framework provides a robust tool for advancing both theoretical research and practical management of arsenic-contaminated groundwater systems.