Friday, November 20, 2026
Science
No Result
View All Result
  • Login
  • HOME
  • SCIENCE NEWS
  • CONTACT US
  • HOME
  • SCIENCE NEWS
  • CONTACT US
No Result
View All Result
Scienmag
No Result
View All Result
Home Science News

Implicit Physics Informed Neural Network Simulates Complex Two Phase Reservoir Flow

November 20, 2026
in Science News
Cassandra Pierce
By Cassandra Pierce Scienmag Editorial Profile - Systems Neuroscience
Reading Time: 6 mins read
0
Implicit Physics Informed Neural Network Simulates Complex Two Phase Reservoir Flow

Implicit Physics Informed Neural Network Simulates Complex Two Phase Reservoir Flow

65
SHARES
587
VIEWS
Share on FacebookShare on Twitter
ADVERTISEMENT

Simulating multiphase fluid displacement in subsurface formations represents an enduring computational challenge, particularly when fluids pass through tightly packed, low-permeability geological formations where conventional flow assumptions break down. In these challenging subsurface environments, fluid behavior deviates from standard linear Darcy flow, exhibiting low-velocity non-Darcy mechanics alongside prominent capillary forces, intricate fracture pathways, and pervasive spatial heterogeneity. These physical mechanisms create tightly coupled, highly nonlinear governing equations for fluid pressure and phase saturation that traditional numerical schemes struggle to solve efficiently without immense computational overhead. Addressing this computational barrier, researchers have established a standalone simulation framework known as the Fully Implicit Scheme-based Enriched Physics-Informed Neural Network, or EPINN-FIS. Designed specifically to resolve strongly coupled oil and water systems without relying on precomputed, labeled simulation datasets, this optimization-driven neural network architecture provides an alternative approach for subsurface modeling by evaluating the governing nonlinear conservation laws across consecutive temporal increments.

Traditional petroleum reservoir engineering relies heavily on numerical modeling to evaluate water-based displacement processes, assess ultimate recovery efficiency, and map trapped hydrocarbon distributions within geological formations. Conventional numerical modeling frameworks typically deploy finite volume discretizations combined with classical iterative nonlinear solvers, such as Newton-Raphson iterations, to resolve system variables. In a standard Newton-based fully implicit formulation, the discretized nonlinear residual equations are linearized at each step, necessitating the repeated computation, assembly, and inversion of large Jacobian matrices. This repeated linear system assembly and inversion often dominates overall runtime and can suffer from convergence bottlenecks when faced with extreme mathematical nonlinearities. Alternative physics-informed machine learning models have previously been introduced to bypass some of these limitations, but many earlier formulations adopted implicit-pressure-explicit-saturation formulations or relied exclusively on automatic differentiation. In tight, low-permeability rock formations, explicit saturation tracking encounters severe numerical stability constraints, while automatic differentiation often exhibits numerical instability across abrupt spatial property contrasts and localized source or sink well boundaries.

To overcome the operational vulnerabilities inherent in explicit saturation updates and automatic differentiation, EPINN-FIS directly discretizes the governing immiscible mass conservation equations using the finite volume method. By applying the divergence theorem across discrete grid elements, the framework inherently enforces local mass flux continuity between neighboring grid blocks while honoring physical closed boundary conditions. Fluid compressibility and rock matrix compressibility are explicitly incorporated through exponential constitutive relations that link density and rock porosity to fluid pressure. Within the low-permeability matrix, fluid velocity obeys an empirical low-velocity non-Darcy flow relation governed by laboratory-derived nonlinear coefficients that suppress flow velocity under shallow pressure gradients. Rather than decoupling the equations or updating saturation explicitly, the new formulation evaluates fluid pressure, phase saturation, relative mobilities, non-Darcy flow terms, and capillary pressure relationships simultaneously at the advanced time level, establishing a truly fully implicit physics-informed neural solver.

The neural network architecture of EPINN-FIS is structured to reflect the fundamental coupled nature of fluid pressure and water saturation while minimizing computational complexity. The framework takes the spatial distribution of oil-phase pressure and water-phase saturation at the current time level as input and maps them directly to the corresponding field distributions at the subsequent time level. Rather than deploying two isolated networks or forcing a single network to output both properties without structural separation, the architecture incorporates shared initial hidden layers that split into two independent downstream branches dedicated to pressure and saturation. Grid connectivity derived from the underlying spatial discretization is explicitly integrated into the network through an adjacency-location anchoring matrix, which links contiguous spatial elements. Gradient flow is stabilized across the intermediate layers using batch normalization, while adaptive rectified linear unit activation functions incorporate learnable scaling parameters that accelerate optimization convergence compared to conventional static activation functions.

To guarantee robust stability across long temporal simulations, the network incorporates skip connections and gated updating mechanisms at the terminal output layers. The spatial pressure and saturation distributions from the previous time step are routed directly across skip connections to combine linearly with the transformed representations emerging from the dedicated hidden branches, governed by learnable weight parameters. Network training is executed by minimizing a composite loss function defined by the mean squared finite-volume mass balance residuals of both the oil and water phases across all computational cells. Because the initial conditions are directly provided at the first temporal step and closed boundary conditions are naturally satisfied by the finite volume flux formulation, the loss formulation does not require artificial penalty weights for boundary enforcement. Furthermore, to accelerate convergence across sequential time intervals, the model leverages transfer learning, reusing optimized network parameters from preceding time steps to initialize training for subsequent temporal increments.

The researchers conducted systematic architecture screening to identify the optimal configuration of neural layers for balancing prediction accuracy against training time. Testing four distinct configurations on a two-dimensional waterflooding problem with a twenty-five by twenty-five grid system, the investigators evaluated setups with varying numbers of shared and dedicated hidden layers. The experimental results demonstrated that an architecture containing two shared coupling hidden layers followed by one dedicated hidden layer for pressure and one dedicated hidden layer for saturation achieved an optimal balance. This configuration achieved Euclidean norm relative errors below 0.0001 for both oil-phase pressure and water saturation during the initial time step while requiring markedly less training duration than deeper or more partitioned alternative designs. Consequently, this screened hybrid layout served as the core architecture for subsequent numerical benchmark investigations.

The performance of the framework was rigorously evaluated across two-dimensional heterogeneous reservoir domains under varying physical conditions, comparing neural predictions against high-fidelity numerical solutions generated by an established in-house fully implicit finite volume simulator. Over a simulated waterflooding span of one thousand days, the neural solver demonstrated remarkable fidelity, maintaining pressure relative errors below one percent and saturation absolute errors below 0.01. The comparative analyses highlighted the profound influence of tight reservoir physics on displacement dynamics. When low-velocity non-Darcy flow and capillary pressure were fully incorporated, the required pressure differential between injection and production wells increased noticeably to overcome low-speed flow resistance. Simultaneously, spontaneous capillary imbibition within the water-wet tight matrix promoted water entry into smaller pores, expanding the waterflood swept area and visibly delaying water breakthrough at the production well relative to non-capillary scenarios.

Further validation examined scenarios featuring different intensities of non-Darcy flow resistance as well as varying capillary pressure magnitudes. When non-Darcy flow effects were amplified, the heightened flow resistance severely impeded pressure propagation toward the displacement front, leading to elevated bottom-hole pressure at injection wells, decreased pressure at production wells, and reduced waterflood sweep efficiency. Conversely, higher capillary pressure in the water-wet system provided an additional driving force that enhanced water propagation across poorly permeable zones, slowing the increase in the production well water cut. Across all variations of non-Darcy severity and capillary strength, the predictions generated by EPINN-FIS matched the reference numerical solutions with high precision, consistently restricting pressure relative errors to less than one percent and saturation absolute errors to under 0.01 across the spatial domain.

To assess performance in structurally complex geological environments, the framework was deployed on a two-dimensional domain embedded with eleven discrete fractures under constant bottom-hole pressure control. Capillary action and non-Darcy effects were neglected within the high-conductivity fracture network, while Darcy flow governed fluid movement along the fractures. Despite severe local permeability contrasts and non-uniform displacement fronts caused by preferential fluid channeling through the fracture network, EPINN-FIS accurately captured the complex saturation evolution, maintaining pressure relative errors below one percent and saturation absolute errors below 0.02. The framework also replicated the time-dependent injection and production flow rates, which fluctuated dynamically despite constant pressure boundaries due to nonlinear saturation shifts, relative permeability variations, and matrix capillary interactions.

The methodology was ultimately tested on a three-dimensional heterogeneous reservoir model discretized via corner-point geometry comprising 963 grid cells, four production wells, and five injection wells operating under constant pressure constraints. Simulating waterflooding over an extended period of 3600 days with thirty-day time steps, EPINN-FIS maintained pressure relative errors below one percent and capped absolute saturation errors below 0.004, capturing the irregular spatial sweep patterns dictated by the heterogeneous permeability field. While the framework demonstrated exceptional numerical accuracy and eliminated the need for explicit Jacobian matrix construction, the authors noted that its current training speed remains slower than highly optimized conventional Newton solvers due to the necessity of multi-epoch gradient optimization at every time step. Nevertheless, the successful implementation confirms that EPINN-FIS provides a robust, physically consistent, Jacobian-free solution framework for complex multiphase subsurface flow, with future efforts focusing on scaling the architecture to larger reservoir domains and inverse parameter estimation.

Subject of Research: Two-phase fluid flow simulation in low-permeability heterogeneous reservoirs using physics-informed neural networks

Article Title: Two-phase flow simulation in low-permeability heterogeneous reservoirs with a fully implicit scheme-based enriched physics-informed neural network

Article References: Li, Y., Shen, X., Li, D., Liu, Y., Huang, W., Qiu, J., Yan, X., & Zhang, K. (2026). Two-phase flow simulation in low-permeability heterogeneous reservoirs with a fully implicit scheme-based enriched physics-informed neural network. Journal of Petroleum Exploration and Production Technology, 16(11), Article 108. https://doi.org/10.1007/s13202-026-02216-7

Image Credits: AI Generated

DOI: 10.1007/s13202-026-02216-7

Keywords: Physics-Informed Neural Networks, Two-Phase Flow, Reservoir Simulation, Low-Permeability Reservoirs, Finite Volume Method, Fully Implicit Scheme, Non-Darcy Flow, Capillary Pressure, Corner-Point Grid, Petroleum Engineering, Fluid Dynamics, Heterogeneous Media, Embedded Discrete Fracture Model, Computational Fluid Dynamics, Scientific Machine Learning, Nonlinear Solver

Cite Scienmag News

Cassandra Pierce. (November 20, 2026). Implicit Physics Informed Neural Network Simulates Complex Two Phase Reservoir Flow. Scienmag. https://scienmag.com/implicit-physics-informed-neural-network-simulates-complex-two-phase-reservoir-flow/

Cassandra Pierce. "Implicit Physics Informed Neural Network Simulates Complex Two Phase Reservoir Flow." Scienmag, 20 November 2026, https://scienmag.com/implicit-physics-informed-neural-network-simulates-complex-two-phase-reservoir-flow/. Accessed 20 November 2026.

Cassandra Pierce. "Implicit Physics Informed Neural Network Simulates Complex Two Phase Reservoir Flow." Scienmag. November 20, 2026. https://scienmag.com/implicit-physics-informed-neural-network-simulates-complex-two-phase-reservoir-flow/

Tags: capillary pressurecomputational fluid dynamicsCorner-Point GridEmbedded Discrete Fracture Modelfinite volume methodfluid dynamicsFully Implicit SchemeHeterogeneous MediaLow-Permeability ReservoirsNon-Darcy FlowNonlinear Solverpetroleum engineeringphysics-informed neural networksreservoir simulationscientific machine learningtwo-phase flow
Share26Tweet16
Previous Post

Novel Wheelchair Armrest Device Encourages Self-Directed Upper Limb Practice After Stroke

Next Post

Chemometric Integration of Spectroscopy and Chromatography Enables Accurate Hardwood Classification

Related Posts

Telemedicine Guides Rapid Oxytocin Stabilization of Severe Postpartum Hemorrhage in Sow
Medicine

Telemedicine Guides Rapid Oxytocin Stabilization of Severe Postpartum Hemorrhage in Sow

November 20, 2026
Cancer Survivors Support Tailoring Mindfulness Relapse Prevention for Smoking and Drinking
Psychology & Psychiatry

Cancer Survivors Support Tailoring Mindfulness Relapse Prevention for Smoking and Drinking

November 20, 2026
Educational Video Game Boosts Early Engineering Design Skills and Student Confidence
Science Education

Educational Video Game Boosts Early Engineering Design Skills and Student Confidence

November 20, 2026
Biogas Digestate Yields Nitrogen-Rich Artificial Humic Acid Through Hydrothermal Humification
Chemistry

Biogas Digestate Yields Nitrogen-Rich Artificial Humic Acid Through Hydrothermal Humification

November 20, 2026
Porous Mini-Channels Enhance Photovoltaic Cooling While Machine Learning Predicts Performance Numerically
Technology and Engineering

Porous Mini-Channels Enhance Photovoltaic Cooling While Machine Learning Predicts Performance Numerically

November 20, 2026
Extremely Preterm Infants Show Motor and Cognitive Gains by Age Three
Medicine

Extremely Preterm Infants Show Motor and Cognitive Gains by Age Three

November 20, 2026
Next Post
Chemometric Integration of Spectroscopy and Chromatography Enables Accurate Hardwood Classification

Chemometric Integration of Spectroscopy and Chromatography Enables Accurate Hardwood Classification

  • Mothers who receive childcare support from maternal grandparents show more optimized

    Mothers who receive childcare support from maternal grandparents show more parental warmth, finds NTU Singapore study

    27656 shares
    Share 11059 Tweet 6912
  • University of Seville Breaks 120-Year-Old Mystery, Revises a Key Einstein Concept

    1061 shares
    Share 424 Tweet 265
  • Bee body mass, pathogens and local climate influence heat tolerance

    682 shares
    Share 273 Tweet 171
  • Researchers record first-ever images and data of a shark experiencing a boat strike

    546 shares
    Share 218 Tweet 137
  • Groundbreaking Clinical Trial Reveals Lubiprostone Enhances Kidney Function

    531 shares
    Share 212 Tweet 133
Science

Embark on a thrilling journey of discovery with Scienmag.com—your ultimate source for cutting-edge breakthroughs. Immerse yourself in a world where curiosity knows no limits and tomorrow’s possibilities become today’s reality!

RECENT NEWS

  • Telemedicine Guides Rapid Oxytocin Stabilization of Severe Postpartum Hemorrhage in Sow
  • Cancer Survivors Support Tailoring Mindfulness Relapse Prevention for Smoking and Drinking
  • Educational Video Game Boosts Early Engineering Design Skills and Student Confidence
  • Biogas Digestate Yields Nitrogen-Rich Artificial Humic Acid Through Hydrothermal Humification

Categories

  • Agriculture
  • Anthropology
  • Archaeology
  • Athmospheric
  • Biology
  • Biotechnology
  • Blog
  • Bussines
  • Cancer
  • Chemistry
  • Climate
  • Earth Science
  • Editorial Policy
  • Marine
  • Mathematics
  • Medicine
  • Pediatry
  • Policy
  • Psychology & Psychiatry
  • Science Education
  • Science News
  • Social Science
  • Space
  • Technology and Engineering

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 5,151 other subscribers

© 2026 Scienmag - Science Magazine

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • HOME
  • SCIENCE NEWS
  • CONTACT US

© 2026 Scienmag - Science Magazine

Discover more from Science

Subscribe now to keep reading and get access to the full archive.

Continue reading