Machine Learning for Inversion Solution Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current inversion schemes for electromagnetic resistivity logging tools are inadequate in distinguishing between good and bad solutions, particularly in ultra-deep measurements, leading to inaccurate formation characterization and reservoir monitoring.
Innovation Solution
The implementation of machine learning methods to identify similarities among inversion solutions and train an information handling system to recognize the best model, reducing the need for randomized initial model calculations and improving the accuracy of formation resistivity modeling by using a multi-step inversion process that combines shallow and deep measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a many-initial-guess approach to inversion is used, then multiple inverted solutions can be calculated, but existing algorithms cannot sufficiently evaluate which solutions are good or bad
Solution Approach 1:
The patent introduces machine learning algorithms as an intermediary evaluation layer between the inversion process and solution selection. The ML model is trained to recognize patterns in inversion solutions and predict their quality, serving as a mediator that guides the selection of good solutions without requiring exhaustive algorithmic evaluation of all possibilities
Solution Approach 2:
The patent replaces traditional deterministic algorithmic evaluation methods with machine learning-based assessment. Instead of relying on complex mechanical algorithmic processes to evaluate solution quality, the system uses trained ML models that have learned to identify good solutions from training data, substituting the evaluation mechanism
2Measurement precision
If randomized initial model calculations are performed to improve inversion accuracy, then better formation profiles can be obtained, but computational time increases significantly
Solution Approach 1:
The patent applies preliminary action by training the machine learning model in advance on a database of inversion solutions with known quality. This pre-trained model can then quickly evaluate new inversion solutions without requiring time-consuming randomized initial model calculations, as the ML system has already learned the patterns of good solutions during the training phase
Solution Approach 2:
The patent changes the evaluation parameter from exhaustive algorithmic computation to machine learning prediction. By transforming the solution evaluation from a computation-intensive process to a pattern-recognition task, the system achieves comparable or better accuracy with significantly reduced computational time
3Measurement precision
If existing inversion algorithms are used for ultra-deep electromagnetic measurements, then formation resistivity can be modeled, but the accuracy is insufficient to distinguish between good and bad solutions
Solution Approach 1:
The patent implements feedback by using the machine learning model to evaluate inversion solutions and provide guidance on which solutions are reliable. The ML system learns from training data what characteristics indicate good solutions and uses this learned knowledge to feedback into the inversion process, improving the reliability of selected solutions
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances the accuracy and efficiency of formation resistivity modeling, allowing for more precise characterization of subterranean formations and improved reservoir monitoring by filtering and sorting inversion results based on similarity and continuity, thereby reducing computational time and improving geological structure visualization.
Implementation Method 1
an electromagnetic transmitter antenna, transmitting electromagnetic fields into a formation
Implementation Method 2
an electromagnetic receiver antenna, measuring the electromagnetic fields
Data Source
AI summary
A method for selecting initial models or inversion solutions during an inversion process from electromagnetic measurements may comprise disposing an electromagnetic well measurement system into a wellbore, transmitting electromagnetic fields into a formation with the electromagnetic transmitter, measuring the electromagnetic fields with the electromagnetic receiver as measurements at a depth in the wellbore, forming initial models for the inversion process based on the measurements and performing an inversion, filtering an inversion solution, forming a solution database from the filtered inversion solutions, building a reference model, calculating a similarity between the reference model and one or more models in the solution database, selecting one or more results from the solution database with the similarity larger than a threshold, and generating a final inversion model image from the one or more results. An electromagnetic well measurement system may comprise an electromagnetic transmitter antenna, an electromagnetic receiver antenna, and an information handling system.