Idea
A generative inversion platform for real-time subsurface flow modeling and uncertainty quantification benefiting geoscientists and energy companies
Research Paper
Core Innovation
This paper introduces SURGIN, which uniquely integrates a U-Net enhanced Fourier Neural Operator surrogate with a score-based generative model to enable zero-shot conditional generation for inverse modeling. Unlike prior methods requiring retraining for new data, SURGIN performs posterior sampling guided by a differentiable surrogate, allowing efficient and real-time assimilation of unseen observations. This approach unifies generative learning with surrogate-guided Bayesian inference in parametric functional spaces.
Market Size (TAM)
$2–10B TAM for subsurface modeling and simulation software; $1–2B SAM from oil and gas, environmental monitoring sectors. Driven by increasing demand for real-time reservoir management and regulatory compliance.
Potential Customers & Pain Points
- Oil and Gas Companies Needing Accurate Reservoir Characterization
- Environmental Agencies Monitoring Groundwater Contamination
- Geoscientists Requiring Fast Data Assimilation for Subsurface Models
- Energy Firms Seeking Efficient Multiphase Flow Predictions
Business Model
Subscription-based SaaS platform with tiered pricing for different data volumes and support levels; enterprise licensing for large energy firms; consulting services for custom integration
Competitive Landscape
- Schlumberger DELFI
- CMG
- Kongsberg Digital
Implementation Challenges
- Integration with existing workflows
- High computational resource requirements
- Adoption resistance due to model complexity
Validation Strategy
- Pilot deployment with partner oil and gas company
- Benchmarking against traditional inversion methods on real datasets
- User feedback collection and iterative model refinement
Research Paper Overview
SURGIN: SURrogate-guided Generative INversion for subsurface multiphase flow with quantified uncertainty
Summary
We present SURGIN, a direct inverse modeling framework combining a U-Net enhanced Fourier Neural Operator surrogate with a score-based generative model to enable zero-shot conditional generation for subsurface multiphase flow data assimilation. SURGIN pretrains an unconditional generative model to capture geological priors and performs posterior sampling guided by a differentiable surrogate, allowing real-time assimilation of unseen monitoring data without retraining. Extensive experiments show its ability to infer heterogeneous geological fields and predict spatiotemporal flow dynamics with quantified uncertainty across diverse measurement settings.