物理信息神经网络在地质与矿山工程中的进展与挑战综述

An Overview of Advances and Challenges in Physical Information Neural Networks for Geology and Mining Engineering

摘要: 物理信息神经网络通过将偏微分方程等物理定律嵌入深度学习框架，为地质与矿山工程中普遍存在的小样本正反问题求解开辟了新路径。本文系统综述了PINNs在地下水渗流与水文地质参数反演、岩石裂纹扩展与裂隙渗流传热、煤矿煤体动力灾害预测、隧道围岩变形与采动覆岩智能感知等方向的研究进展，涵盖了近三年的代表性工作。重点分析了硬约束改进算法、多域物理约束策略、小样本条件下的参数反演能力以及与传统有限元方法的互补关系。最后，总结了PINNs在地质与矿山工程应用中面临的计算效率瓶颈、非线性训练困难及矿山尺度外推不足等核心挑战，并对混合架构和现场数据同化等未来方向进行了展望。

Abstract: Physics-informed neural networks (PINNs) open a new pathway for solving forward and inverse problems with small samples, which are prevalent in geological and mining engineering, by embedding physical laws such as partial differential equations into deep learning frameworks. This paper systematically reviews recent research progress of PINNs in groundwater seepage and hydrogeological parameter inversion, rock crack propagation and fracture seepage heat transfer, dynamic disaster prediction in coal mines, tunnel surrounding rock deformation, and intelligent perception of mining-induced overburden. Representative works from the past three years are covered. Key aspects are analyzed, including hard-constraint improved algorithms, multi-domain physics constraint strategies, parameter inversion capability under few-shot conditions, and the complementary relationship with traditional finite element methods. Finally, the core challenges faced by PINNs in geological and mining engineering applications are summarized, such as computational efficiency bottlenecks, nonlinear training difficulties, and insufficient extrapolation at the mine scale, and future directions including hybrid architectures and in-situ data assimilation are prospected.