Scholarlink

<p>CH<sub>4</sub>, the primary constituent of shale gas, typically exists in a supercritical state within shale pores. Investigating its adsorption and displacement behaviors under confinement is pivotal for the accurate assessment of shale gas resources. Coarse-grained (CG) molecular dynamics simulation serves as a powerful tool for elucidating the microscopic mechanisms governing supercritical fluids (SCFs) behavior. However, the physical properties of SCFs vary drastically under different pressures. Conventional fixed empirical potential parameters become inadequate, thereby compromising the reliability of CG models. In this work, machine learning (ML) is employed to obtain interaction potential parameters of supercritical CH<sub>4</sub> and N<sub>2</sub> under different pressures, and corresponding Lennard-Jones (LJ)-type CG models are constructed. The results show that SCFs exhibit distinct mechanical sensitivities to potential parameters, leading to intrinsic limitations of conventional LJ-type CG models under high temperature and high pressure. Based on the ML-driven potential parameters, the dynamic process of supercritical N<sub>2</sub> displacing CH<sub>4</sub> was investigated, revealing adsorption competition, density reconstruction, and diffusion differences governed by molecular mechanical interactions. Compared with all-atom simulations, the CG model exhibits a deviation of 6% in displacement efficiency. This study provides mechanistic insights into SCF displacement in nanoconfined pores, with implications for enhanced unconventional gas recovery, CO<sub>2</sub> sequestration, and gas transport in nanomaterials.</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

Machine learning-driven coarse-grained molecular dynamics of supercritical fluid displacement in nanoconfined pores

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Mingzhu Xu,
  • Qunazi Yuan,
  • Xianfu Huang

摘要

CH4, the primary constituent of shale gas, typically exists in a supercritical state within shale pores. Investigating its adsorption and displacement behaviors under confinement is pivotal for the accurate assessment of shale gas resources. Coarse-grained (CG) molecular dynamics simulation serves as a powerful tool for elucidating the microscopic mechanisms governing supercritical fluids (SCFs) behavior. However, the physical properties of SCFs vary drastically under different pressures. Conventional fixed empirical potential parameters become inadequate, thereby compromising the reliability of CG models. In this work, machine learning (ML) is employed to obtain interaction potential parameters of supercritical CH4 and N2 under different pressures, and corresponding Lennard-Jones (LJ)-type CG models are constructed. The results show that SCFs exhibit distinct mechanical sensitivities to potential parameters, leading to intrinsic limitations of conventional LJ-type CG models under high temperature and high pressure. Based on the ML-driven potential parameters, the dynamic process of supercritical N2 displacing CH4 was investigated, revealing adsorption competition, density reconstruction, and diffusion differences governed by molecular mechanical interactions. Compared with all-atom simulations, the CG model exhibits a deviation of 6% in displacement efficiency. This study provides mechanistic insights into SCF displacement in nanoconfined pores, with implications for enhanced unconventional gas recovery, CO2 sequestration, and gas transport in nanomaterials.