Geophysical Model Inversion Using Bayesian Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In geophysical applications, inversion methods fail to fully utilize geophysical datasets, leading to incomplete representation of subterranean region physical properties, which in turn affects the accuracy of hydrocarbon reservoir identification.
Innovation Solution
A method involving preprocessing of observed geophysical datasets, formation of training datasets, iterative determination of simulated datasets, and use of machine learning networks to predict geophysical models, updating the current model based on data and model loss functions, enhances the representation of subterranean region properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional inversion methods are used to determine geophysical models, then the process is computationally simpler, but the geophysical model fails to fully utilize the geophysical dataset and does not robustly represent physical properties
Solution Approach 1:
A machine learning network is introduced as an intermediary between the geophysical dataset and the geophysical model. The network is trained on training datasets to learn the complex mapping relationships, thereby fully utilizing the geophysical data while automating the inversion process and improving robustness without requiring manual intervention in the complex inversion calculations
Solution Approach 2:
Training datasets are generated in advance through preliminary forward modeling and synthetic data creation. This preliminary action prepares the machine learning network with comprehensive examples of geophysical data-model relationships before actual inversion is performed, enabling the network to robustly represent physical properties during the actual inversion process
2Measurement precision
If traditional inversion methods are used, then the computational process is faster, but the geophysical model does not accurately identify hydrocarbon reservoirs
Solution Approach 1:
The traditional mechanical inversion process (iterative numerical optimization) is replaced with a machine learning-based approach. The machine learning network, once trained, can rapidly predict geophysical models with high accuracy for hydrocarbon reservoir identification, achieving both precision and computational efficiency by substituting the traditional computational mechanics with learned patterns
3Reliability
If machine learning networks are trained on comprehensive training datasets to improve model accuracy, then the geophysical model representation improves, but the computational time and resources increase
Solution Approach 1:
The machine learning network is trained in advance on comprehensive training datasets that capture the full range of geophysical data variations. This preliminary training action transfers the computational burden to the training phase, allowing the network to perform rapid, accurate predictions during actual inversion operations without requiring extensive computation time when processing real geophysical data
Data Source
AI summary
A system and methods for determining an updated geophysical model of a subterranean region of interest are disclosed. The method includes obtaining a preprocessed observed geophysical dataset based, at least in part, on an observed geophysical dataset of the subterranean region of interest, and forming a training dataset composed of a plurality of geophysical training models and corresponding simulated geophysical training datasets. The method further includes iteratively determining a simulated geophysical dataset from a current geophysical model, determining a data loss function between the preprocessed observed geophysical dataset and the simulated geophysical dataset, training a machine learning (ML) network, using the training dataset, to predict a predicted geophysical model and determining a model loss function between the current and predicted geophysical models. The method still further includes updating the current geophysical model based on an inversion using the data loss and model loss functions.