Deep Learning Reservoir Parameter Derivation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining reservoir properties in subsurface reservoirs struggle to effectively integrate seismic and flow-related data, leading to inefficiencies in hydrocarbon production due to the lack of high-resolution parameter derivation that honors both data types.
Innovation Solution
A method combining stochastic inversion and deep learning using conditional generative adversarial networks (cGAN) to generate high-resolution reservoir parameters, which involves Bayesian models for joint AVA and production data inversion, and the use of transfer-learning techniques to accelerate the workflow and improve history matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional geostatistical inversion and statistical rock physics are used to transform coarse-scale geophysical parameters to fine-scale permeability models, then the workflow can produce initial models, but the process requires considerable time and computing resources and requires history matching to production data to close the loop
Solution Approach 1:
The patent replaces traditional geostatistical inversion and statistical rock physics methods with a machine learning-based approach. A neural network model is trained to directly map coarse-scale geophysical parameters to fine-scale reservoir parameters, eliminating the need for iterative history matching and significantly reducing computational time while maintaining high resolution.
Solution Approach 2:
The patent performs preliminary training of the neural network model using synthetic data generated from reservoir simulations. This pre-training phase allows the model to learn the complex relationships between geophysical and reservoir parameters before actual field application, enabling direct high-resolution derivation without requiring subsequent iterative adjustments or history matching.
2Reliability
If the loop is closed by cycling between AVA and production models until a common model is found that fits both seismic and production data, then both data types are honored, but the process requires considerable time and computing resources
Solution Approach 1:
The patent substitutes the iterative cycling process with a single-pass neural network inference. The model is trained to simultaneously honor both seismic (AVA) and production data by learning their joint statistical relationships, allowing direct generation of a unified high-resolution model without repeated cycling between different data constraints.
Solution Approach 2:
The patent transforms the problem from iterative parameter adjustment to direct parameter prediction. By changing the approach from cyclic refinement to a learned mapping function, the system achieves the same reliability of fitting both data types while dramatically improving workflow efficiency through the neural network's ability to perform the transformation in a single operation.
3Productivity
If machine learning methods are used to link coarse-scale geophysical parameters to fine-scale permeability models, then the process can be accelerated, but the lack of real training data limits the application
Solution Approach 1:
The patent creates synthetic training data by copying and adapting results from reservoir simulation studies. Synthetic seismic data and corresponding reservoir models are generated through forward modeling, providing sufficient training examples for the neural network without requiring extensive real field data. This synthetic data approach enables the ML model to achieve high productivity while overcoming the scarcity of real training data.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method is described for deriving high-resolution reservoir properties for a subsurface reservoir. The method may include receiving a seismic dataset; inverting the seismic dataset to generate an ensemble of coarse-scale seismic parameters, wherein the inverting may use one of Bayesian models with Markov Chain Monte Carlo (MCMC) sampling, simulated annealing, partial swarm, or analytic Bayes formulations; receiving fine-scale lithotype models; developing deep learning neural networks based on transfer learning using the fine-scale lithotype models to generate a conditional probability distribution of high-resolution reservoir parameters; generating an ensemble of high-resolution reservoir parameters using the deep learning neural network to condition the ensemble of coarse-scale seismic parameters; and displaying, on a user interface, the ensemble of high-resolution reservoir parameters. The method is executed by a computer system.