Petrophysical Inversion with Machine Learning Geologic Priors

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing geophysical prospecting methods face challenges in accurately imaging subsurface structures and identifying hydrocarbon-bearing formations due to non-uniqueness in geophysical inversion problems and limitations in data resolution at sub-seismic scales.

Innovation Solution

The method employs a machine learning-based approach by training a network to predict rock type probabilities using petrophysical parameters, incorporating geological priors, and performing petrophysical inversion to generate updated models, thereby enhancing the accuracy and resolution of subsurface modeling.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional geophysical inversion methods are used, then the inversion process can be performed with available data, but the results suffer from non-uniqueness and lack of resolution at sub-seismic scales

Engineering Contradiction:

Improveresolution at sub-seismic scalesVSAvoidnon-uniqueness in inversion results

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The patent transforms elastic parameters (seismic velocity, density) to petrophysical parameters (porosity, volume of clay) through petrophysical inversion. This parameter transformation enables the model to capture subsurface properties at multiple scales including sub-seismic scales, resolving the limitation of traditional seismic inversion that only provides elastic parameters with limited resolution

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent introduces machine learning-based geologic priors as an intermediary between seismic data and inversion results. These priors, trained on well log data and geological knowledge, guide the inversion process to produce geologically plausible models that reduce non-uniqueness while enhancing resolution at sub-seismic scales

Inventive Principle:

Principle #24Intermediary (Mediator)

2Manufacturing precision

If seismic data with limited frequency band is used, then data acquisition is practical, but the resulting models are bandlimited and cannot resolve sharp layer boundaries

Engineering Contradiction:

Improvesharpness of layer boundariesVSAvoidbandlimited nature of seismic data

Core Design Contradiction:

Manufacturing precisionVSLoss of information

Solution Approach 1:

The patent performs petrophysical inversion that transforms bandlimited elastic parameters into petrophysical parameters. This transformation process, guided by machine learning priors, recovers sharp layer boundaries and high-frequency geological features that are not directly present in the bandlimited seismic data, effectively overcoming the information loss

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent applies machine learning-based geologic priors before the inversion process to establish expected geological patterns and structures. This preliminary action guides the inversion to recover sharp boundaries and high-frequency features that would otherwise be lost, preparing the inversion process to overcome bandlimitation effects

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If uniform discretization with many voxels is used, then the model can match simulated data to observed seismic data, but the model complexity increases significantly

Engineering Contradiction:

Improvefidelity of data matchingVSAvoidnumber of volume elements in model

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent changes the parameterization from uniform elastic-wave velocity voxels to petrophysical parameters (porosity, volume of clay) that have direct geological meaning. This parameter transformation allows the model to achieve high fidelity in matching seismic data while using fewer, more meaningful parameters that reduce model complexity and improve interpretability

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent inverts the traditional approach by first transforming seismic elastic parameters to petrophysical parameters through petrophysical inversion, rather than directly modeling elastic parameters. This inverted approach reduces model complexity by working with fewer, more constrained petrophysical parameters that naturally capture subsurface variability

Inventive Principle:

Principle #13The other way round (Inversion)

Data Source

PatentEP4010740B1Petrophysical inversion with machine learning-based geologic priors

Publication Date: 2025.02.19 EXXONMOBIL TECHNOLOGY & ENGINEERING CO

Figure 1
Figure 2
Figure 3A~3B

AI summary

A method and system for modeling a subsurface region include applying a trained machine learning network to an initial petrophysical parameter estimate to predict a geologic prior model; and performing a petrophysical inversion with the geologic prior model, geophysical data, and geophysical parameters to generate a rock type probability model and an updated petrophysical parameter estimate. Embodiments include managing hydrocarbons with the rock type probability model. Embodiments include checking for convergence of the updated petrophysical parameter estimate; and iteratively: applying the trained machine learning network to the updated petrophysical parameter estimate of a preceding iteration to predict an updated rock type probability model and another geologic prior model; performing a petrophysical inversion with the updated geologic prior model, geophysical seismic data, and geophysical elastic parameters to generate another rock type probability model and another updated petrophysical parameter estimate; and checking for convergence of the updated petrophysical parameter estimate.