Physics-informed neural networks (PINNs) have emerged as a novel framework that integrates physical laws with deep learning, demonstrating significant potential in recent geological reservoir studies. This approach offers an innovative pathway for addressing challenges in conventional reservoir research, such as high computational costs and heavy reliance on extensive data. The application progress of PINNs in various domains was systematically reviewed, including reservoir numerical simulation, parameter inversion, multiphase flow simulation, geothermal development, and CO2 geological storage. By incorporating partial differential equations into the loss function of the neural network, PINNs enable high-precision predictions under data-sparse conditions while ensuring that solutions adhere to physical constraints. Results indicate that PINNs exhibit computational efficiency and mesh-free flexibility in areas such as surrogate model construction, history matching, and simulation of complex media. However, challenges remain, including high training costs and limited generalization capabilities. In the future, through algorithmic improvements, multi-source data integration, and hybridization with conventional methods, PINNs are expected to advance reservoir research toward more intelligent and efficient development.
ZHANG Xue-min, WAN Huan, HU Yun-ting.
Review of Application of Physics-informed Neural Networks in Geological Reservoir Research[J].
Science Technology and Engineering, 2026, 26(19): 8085-8096 DOI:10.12404/j.issn.1671-1815.2505326
除了常规油气藏,PINNs在非常规能源和环境地质领域也有所应用。例如,在地热储层中,温度场和渗流场的耦合模拟对资源评价至关重要。文献[28]将PINN用于地热井的热恢复过程模拟,通过同时约束热传导和流体流动方程,预测注入停止后井周温度随时间的回升。结果表明,PINN模型能够较好地拟合温度观测曲线,为地热井的生产调度提供参考。在CO2地质封存(carbon capture, utilization and storage,CCUS)方面,PINNs被用于模拟CO2在盐水层中的运移和相变。例如,文献[22]构建PINN模型求解CO2-盐水两相流的组分模型,通过自动微分计算CO2溶解度和密度变化对流动的影响。PINN模型在训练中结合了实验室测得的CO2物性参数,确保模拟结果符合物理规律。与传统模拟器相比,PINN在给定封存场地少量监测数据的情况下,即可快速预测CO2羽流的扩散范围和压力变化[29],为封存安全性评估提供支持[30]。这些应用表明,PINNs能够有效处理地热和CCUS等涉及多物理场耦合的复杂系统,为非常规能源开发和环境工程提供新的建模工具。
在地质CO2封存中,PINNs结合傅里叶神经算子(Fourier neural operator,FNO)可预测CO2羽流迁移和压力场演化。例如,Yan等[30]的深度学习工作流通过分离注入期和封存期模型,将预测速度提升251倍,且CO2饱和度预测均方根误差(root mean squared error,RMSE)低至0.025(图7)。
(1)训练效率与计算成本:PINN的训练通常涉及大量配点的偏导数计算,优化高维、复杂问题时计算成本巨大。对于三维大尺度油藏模型,PINN可能需要非常深广的网络和海量训练数据,导致训练时间过长甚至不可行[46]。此外,PINN常采用自适应矩估计算法(adaptive moment estimation,Adam)结合有限内存布罗伊登-弗莱彻-戈德法布-香诺算法(limited-memory Broyden-Fletcher-Goldfarb-Shanno,L-BFGS)等混合优化策略,优化过程复杂且难以收敛到全局最优[47]。 如何提高PINN训练的效率和稳定性,是当前研究的重点之一。
HasanN, FentonT, PereraM S A, et al. Physics-informed neural networks with periodic activation functions for solute transport in heterogeneous porous media[J/OL]. arXiv Preprint, arXiv: 2212. 08965, 2022.
[2]
WenG, LiX, WangG. Improving deep learning performance for predicting large-scale heterogeneous reservoir models with a feature coarsening technique[J]. Scientific Reports, 2022, 12(1): 20704.
ShiBowei, ZhengHeng, LiJunliang, et al. Fracturing parameter optimization and yield prediction based on machine learning algorithms[J]. Science Technology and Engineering, 2025, 25(25): 10663-10670.
YangLi, ZhouZixi, WangTingting, et al. Single well production forecasting method based on CNN-GRU-LightGBM model[J]. Science Technology and Engineering, 2024, 24(18): 7606-7614.
ZengRui, ShaoYanlin, HuangYu, et al. Intelligent identification of logging lithology based on deep learning[J]. Science Technology and Engineering, 2020, 25(30): 12804-12812.
[9]
WangN, ChenY, ZhangD. A comprehensive review of physics-informed deep learning and its applications in geoenergy development[J]. The Innovation Energy, 2025(2): 100087.
[10]
BehlM, TyagiM. Development of CRM for reservoir simulations using PINNs[EB/OL]. (2023-03-22)[2024-10-26]. https//www.mathworks.com/videos/mathworks-energy-conference-2022-development-of-crm-for-reservoir-simulations-using-pinns-1679461536392.html.
ZhangZitong, ShiChenyi, FuYingkun, et al. Production forecast of shale gas wells based on long short-term memory and bayesian optimization[J]. Science Technology and Engineering, 2025, 25(27): 11569-11580.
[13]
KarniadakisG E, KevrekidisI G, LuL, et al. From PINNs to PIKANs: recent advances in physics-informed machine learning[J/OL]. arXiv Preprint, arXiv: 2410. 13228, 2024.
[14]
RaissiM, PerdikarisP, KarniadakisG E. Physics informed neural networks: a deep learning framework for solving forward and inverse problems involving non linear partial differential equations[J]. Journal of Computational Physics, 2019, 378: 686-707.
[15]
IshitsukaY, LinY. Modeling unobserved geothermal structures using a physics-informed neural network with transfer learning of prior knowledge[J]. Geothermal Energy, 2024, 12(1): 38.
[16]
LiuY, ZhangD. On acoustic fields of complex scatters based on physics-informed neural networks[J]. Ultrasonics, 2022, 123: 106721.
[17]
ZhaoW, HuangG, WangK, et al. Dynamic weight strategy of physics-informed neural networks for solving the navier-stokes equations[J]. Proceedings of the National Academy of Sciences, India Section A: Physical Sciences, 2022, 92(3): 545-555.
[18]
LiuY, WangJ, ZhangD. Physics-informed graph neural network for predicting fluid flow in subsurface porous media[J]. Petroleum Science, 2025, 22(6): 1-15.
LiuXueqi, WangZhiyuan, WeiQiang, et al. A wellbore temperature field solution method driven by physics-informed neural networks[J]. Acta Petrolei Sinica, 2025, 46(2): 413-425.
PanXiaotian. Application research on dynamic prediction of water-flooding reservoirs based on physics-informed neural networks[D]. Xi’an: Xi’an Shiyou University, 2024.
[23]
HanJ X, XueL, WeiY S, et al. Physics-informed neural network-based petroleum reservoir simulation with sparse data using domain decomposition[J]. Petroleum Science, 2023, 20(6): 3450-3460.
ZhangSheng, LanPeng, SuJingjing, et al. Solution of groundwater seepage model and parameter inversion based on PINNs algorithm[J]. Chinese Journal of Geotechnical Engineering, 2023, 45(2): 376-383, 443-444.
[26]
TartakovskyA M, PeregoM, KoustovV, et al. Physics-informed deep neural networks for learning parameters and constitutive relationships in subsurface flow problems[J]. Water Resources Research, 2020, 56(5): e2019WR026731.
[27]
QinZ, JiangA, FaulderD, et al. Physics-guided deep learning for prediction of energy production from geothermal reservoirs[J]. Geothermics, 2024, 116: 102824.
[28]
ZhangJ, ZhangD. Physics-informed neural networks for multiphase flow in porous media[J]. Energy & Fuels, 2024, 38(18): 15019-15030.
[29]
ZhangJ, ZhangD. Physics-informed neural networks for multi-phase flow in porous media[J/OL]. arXiv Prepint, arXiv: 2404. 12502, 2024.
ZhangCheng, LiXue, YeMao, et al. Application of physics-informed neural networks in two-phase flow[J]. CIESC Journal, 2024, 75(11): 3835-3856.
[32]
YanB, WangS, SunS. Physics-informed neural network simulation of two-phase flow in heterogeneous and fractured porous media[J]. SPE Journal, 2024, 29(3): 1348-1366.
[33]
AlmajidM M, Abu-Al-SaudM. Physics-informed neural network simulation of two-phase flow in porous media[J]. Journal of Petroleum Science and Engineering, 2024, 238: 111719.
YanXia, HuaTianying, ZhangKai, et al. A physics-informed neural network numerical simulation method suitable for fractured shale gas reservoirs: CN119962260B[P]. 2025-06-13.
YanXia, QiuJunchen, ZhangKai, et al. Physics-informed neural network numerical simulation method suitable for low-permeability water-flooding reservoirs: CN119720800B[P]. 2025-05-27.
[38]
SuryanarayanaS, PutraH, AzisM. The application of physics-informed neural networks(PINNS) for geothermal well thermal recovery post-injection[J]. IOP Conference Series: Earth and Environmental Science, 2025, 1456: 012006.
MaXianlin, LiuZhengzhi, ZhanJie, et al. CO2 plume distribution prediction method based on physics-informed neural networks[J]. Petroleum Drilling Techniques, 2024, 52(5): 69-75.
[41]
YanB C, ChenB L, JiaW, et al. A robust deep learning workflow to predict multiphase flow behavior during geological CO2 sequestration injection and post-injection periods[J]. Journal of Hydrology, 2022, 607: 127542.
ZouJingbo, LiuCai, ZhaoPengfei. Research progress on physics-informed neural network algorithms for seismic wave forward modeling problems[J]. Progress in Geophysics, 2023, 38(1): 430-448.
[44]
KarimpouliS, TahmasebiP. Physics-informed neural networks(PINNs) for wave propagation and full-waveform inversion[J]. Journal of Geophysical Research: Solid Earth, 2022, 127(5): e2021JB023120.
[45]
Rasht-BeheshtM, HuberC, ShuklaK, et al. Physics-informed neural networks (PINNs) for wave propagation and full waveform inversions[J]. Journal of Geophysical Research: Solid Earth, 2022(5): 12035.
[46]
EmeribeV, MaJ. Time-lapse seismic history matching with an iterative ensemble smoother and a generative adversarial network[J]. Geophysics, 2020, 85(4): 115-129.
[47]
CuomoS, Di ColaV S, GiampaoloF, et al. Understanding physics-informed neural networks[J]. Entropy, 2023, 25(8): 1218.
ZhuY, ZabarasN, KoutsourelakisP S, et al. Uncertainty quantification in scientific machine learning: methods, metrics, and comparisons[J]. Journal of Computational Physics, 2023, 477: 111909.
[50]
LiuA, LiJ, BiJ, et al. A novel reservoir simulation model based on physics informed neural networks[J]. Physics of Fluids, 2024, 36(11): 116617.
[51]
KarniadakisG E, KevrekidisI G, LuL, et al. Physics-informed machine learning[J]. Nature Reviews Physics, 2021, 3(6): 422-440.
[52]
ZhangZ. A physics-informed deep convolutional neural network for simulating and predicting transient darcy flows in heterogeneous reservoirs without labeled data[J]. Journal of Petroleum Science and Engineering, 2022, 211: 110179.
[53]
RaissiM, PerdikarisP, KarniadakisG E. Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations[J]. Journal of Computational Physics, 2019, 378: 686-707.
[54]
JagtapA D, KharazmiE, KarniadakisG E. Physics-informed neural networks for solving transient unconfined groundwater flow[J]. Computers & Geosciences, 2023, 179: 105422.
[55]
YaqoobM, AnsariM Y, IshaqM, et al. FluidNet-Lite: lightweight convolutional neural network for pore-scale modeling of multiphase flow in heterogeneous porous media[J]. Advances in Water Resources, 2025, 197: 104952.
[56]
ZhangJ J, Braga-NetoU, GildinE. Physics-informed neural networks for multiphase flow in porous media considering dual shocks and interphase solubility[J]. Energy & Fuels, 2024, 38(18): 17781-17795.
[57]
YanX, LinJ Q, WangS, et al. Physics-informed neural network simulation of two-phase flow in heterogeneous and fractured porous media[J]. Advances in Water Resources, 2024, 193: 104731.
[58]
KrishnapriyanA, GholamiA, ZheS, et al. Characterizing possible failure modes in physics-informed neural networks[J]. Advances in Neural Information Processing Systems, 2021, 34: 26548-26560.
[59]
WangN, ZhangD. A critical review of physics-informed machine learning applications in geoenergy and geosciences[J]. Geoenergy Science and Engineering, 2024, 240: 213228.
[60]
AhmedA M A. Physics-informed heterogeneous spatiotemporal graph neural network for reservoir simulation[R]. Baton Rouge: Louisiana State University, 2024.
[61]
BadawiD, GildinE. Physics-informed neural network for the transient diffusivity equation in reservoir engineering[J/OL]. arXiv Preprint, arXiv: 2309. 17345, 2023.
[62]
JinT, XiaY, JiangH L. A physics-informed neural network approach for surrogating a numerical simulation of fractured horizontal well production prediction[J]. Energies, 2023, 16(24): 7948.
ShuWei, JiangJianguo, WuJichun. Physics-informed neural networks with hard constraints for hydraulic conductivity field inversion[J]. Earth Science Frontiers, 2026, 33(1): 500-510.
BonfantiA, SantanaR, ElleroM, et al. On the generalization of PINNs outside the training domain and the hyperparameters influencing it[J]. Neural Computing and Applications, 2024, 36(36): 22677-22696.
[67]
ZhangZ, YanX, LiuP Y, et al. A physics-informed convolutional neural network for the simulation and prediction of two-phase darcy flows in heterogeneous porous media[J]. Journal of Computational Physics, 2023, 477: 111919.
[68]
ElfetniS, KamachaliR D. PINNs-MPF: a physics-informed neural network framework for multi-phase-field simulation of interface dynamics[J]. Engineering Analysis with Boundary Elements, 2025, 176: 106200.
[69]
ChenJ, GildinE, KilloughJ E. Transfer learning-based physics-informed convolutional neural network for simulating flow in porous media with time-varying controls[J]. Mathematics, 2024, 12(20): 3281.
[70]
LüX C, QinX, VoskovD, et al. Self-adaptive physics-informed neural networks for CO2 immiscible flooding in low-permeability oil reservoirs[C]// SPE Reservoir Simulation Conference. Galveston: SPE, 2025: 223838.
[71]
CuomoS, Di ColaV S, GiampaoloF, et al. Scientific machine learning through physics-informed neural networks: where we are and what’s next[J]. Journal of Scientific Computing, 2022, 92(3): 88.
[72]
WangS F, TengY J, PerdikarisP. Understanding and mitigating gradient flow pathologies in physics-informed neural networks[J]. SIAM Journal on Scientific Computing, 2021, 43(5): 3055-3081.
[73]
BiJ, LiJ, WuK, et al. A physics-informed spatial-temporal neural network for reservoir simulation and uncertainty quantification[J]. SPE Journal, 2024, 29(4): 18.
[74]
LiS X, TianW C, YanH X, et al. Modeling transient mixed flows in sewer systems with data fusion via physics-informed machine learning[J]. Water Research, 2024, 25: 100266.
[75]
WangN Z, ChangH B, ZhangD X. Theory-guided auto-encoder for surrogate construction and inverse modeling[J]. Computer Methods in Applied Mechanics and Engineering, 2021, 385: 114037.
[76]
LiuZ, ZhangJ H, ChenY R, et al. From physics constraints to trustworthy Bayesian reasoning: synergies of PINN, iPINN, and iBPINN in prestack AVO inversion[J]. IEEE Transactions on Geoscience and Remote Sensing, 2025, 63: 5926424.
[77]
SinhaU, DindorukB. Review of physics-informed machine learning (PIML) methods applications in reservoir engineering[J]. Geoenergy Science and Engineering, 2025, 244: 213713.
[78]
PsarosA F, MengX H, ZouZ R, et al. Uncertainty quantification in scientific machine learning: methods, metrics, and comparisons[J]. Journal of Computational Physics, 2023, 477: 111902.