Reservoir Facies and Petrophysical Models with Quantile Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional facies and petrophysical property models in reservoir development rely on limited direct measurements from sparse wells, often excluding valuable secondary data, leading to inaccurate volume and flow performance predictions.
Innovation Solution
Incorporate multiple secondary datasets using a combination of machine learning algorithms and geostatistical programs, employing a quantile-trained machine learning model to optimize information extraction and provide uncertainty ranges, followed by geostatistical probability field simulations to build three-dimensional models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional geostatistical toolboxes are used with limited direct measurements from sparse wells, then the modeling process is simple and straightforward, but the accuracy of reservoir facies and petrophysical property models is insufficient
Solution Approach 1:
The patent segments the modeling process into distinct modules: machine learning model training, quantile prediction, and geostatistical simulation. This allows complex secondary data to be processed through specialized algorithms while maintaining overall system manageability and improving accuracy through focused computational approaches.
Solution Approach 2:
The patent introduces machine learning models as intermediary components between raw secondary data and final geostatistical models. These intermediaries process and interpret complex relationships in the data, enabling more accurate reservoir characterization without directly complicating the core geostatistical modeling workflow.
2Reliability
If multiple secondary datasets are incorporated into reservoir modeling, then the accuracy and reliability of volume and flow performance predictions improve, but the device complexity and difficulty of data processing increase
Solution Approach 1:
The patent employs a universal machine learning framework that can process multiple types of secondary data (seismic attributes, well logs, geological data) through a single integrated approach. This multi-functional system handles diverse data formats and sources while maintaining consistent processing standards, thereby improving prediction reliability without proportionally increasing complexity.
Solution Approach 2:
The patent transforms the modeling approach by changing key parameters from traditional single-dataset methods to multi-dataset integration with machine learning. This parameter change enables the system to handle multiple secondary datasets systematically, improving reliability through comprehensive data utilization while managing complexity through standardized processing protocols.
3Loss of information
If traditional methods using only one type of secondary data are used, then the modeling process remains simple and manageable, but valuable data from multiple sources is excluded
Solution Approach 1:
The patent merges multiple secondary datasets into a unified machine learning model that processes all available data types simultaneously. This combining approach ensures that no valuable information is lost or excluded, as the integrated model utilizes seismic attributes, well logs, and geological data together to maximize the utilization of available information while managing complexity through cohesive processing.
Data Source
AI summary
Some implementations relate to a method for generating, at least in part by a quantile-trained learning machine, a model of a formation property across one or more subsurface formations of a reservoir using a plurality of external data sources different than the formation property.