[論文レビュー] A Critical Review of Physics-Informed Machine Learning Applications in Subsurface Energy Systems

Abdeldjalil Latrach, Mohamed Lamine Malki+3 more|arXiv (Cornell University)|Aug 6, 2023

Seismic Imaging and Inversion Techniques参考文献 159被引用数 8

ひとこと要約

A comprehensive review of physics-informed machine learning (PIML) for subsurface energy systems, detailing modes of integration, PINNs theory, and applications in geoscience, drilling, reservoirs, and production forecasting.

ABSTRACT

Machine learning has emerged as a powerful tool in various fields, including computer vision, natural language processing, and speech recognition. It can unravel hidden patterns within large data sets and reveal unparalleled insights, revolutionizing many industries and disciplines. However, machine and deep learning models lack interpretability and limited domain-specific knowledge, especially in applications such as physics and engineering. Alternatively, physics-informed machine learning (PIML) techniques integrate physics principles into data-driven models. By combining deep learning with domain knowledge, PIML improves the generalization of the model, abidance by the governing physical laws, and interpretability. This paper comprehensively reviews PIML applications related to subsurface energy systems, mainly in the oil and gas industry. The review highlights the successful utilization of PIML for tasks such as seismic applications, reservoir simulation, hydrocarbons production forecasting, and intelligent decision-making in the exploration and production stages. Additionally, it demonstrates PIML's capabilities to revolutionize the oil and gas industry and other emerging areas of interest, such as carbon and hydrogen storage; and geothermal systems by providing more accurate and reliable predictions for resource management and operational efficiency.

研究の動機と目的

• Identify limitations of standard ML in scientific domains and motivate physics-informed approaches.
• Systematically categorize modes of integrating physics into data-driven models for subsurface applications.
• Present the theoretical foundation and practical variants of physics-informed neural networks (PINNs).
• Survey current PIML applications across geology/geophysics, drilling, reservoir engineering, and production forecasting.
• Highlight best practices, knowledge gaps, and future research directions in PIML for subsurface energy systems.

提案手法

• Explain the spectrum of physics integration modes including data/feature engineering, postprocessing, initialization, optimizer design, architecture design, loss function, and hybrid models.
• Present the PINN framework and objective function combining IB, PDE, and data losses to enforce physical laws.
• Discuss neural solvers and neural operators as neural simulators for ODEs/PDEs/SDEs and their comparative attributes.
• Theoretically formulate PDEs with initial and boundary conditions and describe how PINNs enforce these through automatic differentiation.
• Review and synthesize PIML applications in geoscience, drilling, reservoir engineering, production forecasting, and CO2 storage.

実験結果

リサーチクエスチョン

• RQ1 What are the dominant modes of incorporating physics into data-driven subsurface models and when are they appropriate?
• RQ2 How do PINNs operate to solve and discover governing equations in subsurface applications?
• RQ3 What is the current state of PIML applications across geoscience, drilling, reservoir engineering, production forecasting, and CO2 storage?

主な発見

• PIML enhances model generalization and adherence to physical laws in data-scarce regimes.
• PINNs can solve and discover PDEs by enforcing their residuals in the loss function, enabling physics-consistent predictions and potential extrapolation beyond training data.
• Hybrid models that couple physics-based losses with physics-aware architectures yield improved adherence to governing equations and physical plausibility.
• Geology and geophysics benefit from PINNs in seismic wave modeling, inversion workflows, and high-frequency wavefield predictions with methods addressing spectral bias.
• Neural operators like Fourier Neural Operators offer mesh-free alternatives with potential data requirements differing from PINNs, useful for seismic problems.
• The review underscores practical best practices and identifies gaps such as optimizer design and broader adoption in industry.