Training Well Selection for Accurate Target Well Log Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is no systematic and quantitative approach to optimize the training data set for machine learning to achieve high prediction accuracy in hydrocarbon reservoir properties, particularly for target wells or zones.
Innovation Solution
A systematic workflow is developed to select a training set of relevant wells by filtering a multi-well database based on location proximity, stratigraphic zonation, and operational settings, using neural networks to calculate relevancy levels, and applying coverage analysis to ensure the training set's adequacy, followed by training an AI network with complex algorithms to predict reservoir properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large number of training wells are used in machine learning, then prediction accuracy may improve, but computational resources and processing time increase
Solution Approach 1:
The patent extracts and removes irrelevant training wells from the dataset using filtering criteria (geological similarity, operational parameters, data quality). This reduces the training dataset to only the most relevant wells, decreasing computational resources while maintaining prediction accuracy by eliminating unnecessary data processing.
Solution Approach 2:
The patent applies different selection criteria to different subsets of training wells based on their relevance to the target well. By evaluating each training well's specific characteristics (geological formation, operational settings, data completeness) and assigning different relevancy scores, the system optimizes the training set composition to improve prediction accuracy while reducing overall computational burden.
2Quantity of substance
If all available training wells are used for machine learning, then data coverage is maximized, but data quality and relevance deteriorate
Solution Approach 1:
The patent changes the selection parameters from simple quantity-based inclusion to multi-criteria evaluation including geological similarity, operational parameter matching, and data quality assessment. By transforming the selection criteria from broad coverage to targeted relevance based on multiple parameters, the system maintains adequate data coverage while significantly improving data relevance for prediction accuracy.
3Measurement precision
If extensive filtering and selection processes are applied to training wells, then prediction accuracy improves, but processing time and operational complexity increase
Solution Approach 1:
The patent performs preliminary filtering and preprocessing of training wells before the main machine learning training process. By pre-evaluating and selecting relevant training wells based on geological and operational criteria, and pre-processing their data to ensure quality and consistency, the system reduces the workload for subsequent training operations, thereby improving prediction accuracy while minimizing additional processing time.
Data Source
AI summary
Systems, methods, and apparatus including computer-readable mediums for managing training wells for target wells in machine learning are provided. In one aspect, a method includes: for each training well of a plurality of training wells, building a training network for the training well based on well log data of the training well, predicting a target well log of a target well using the training network built for the training well, determining a relevancy level between the training well and the target well based on the predicted target well log of the target well and a measured target well log of the target well, and selecting relevant training wells among the plurality of training wells based on the relevancy levels associated with the plurality of training wells.