Machine Learning Earth Model for Real-Time Wellbore Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional earth modeling methods are cumbersome, require complex software and expertise, and are slow to update, leading to inaccuracies and inefficiencies in drilling operations due to limitations in seismic inversion, machine learning model instability, and reliance on single algorithms.
Innovation Solution
Implementing a machine learning model that carries over weighting values between instances, automatically selects inputs, and uses ensembles to generate real-time earth models, allowing for dynamic updates and improved accuracy by integrating multiple algorithms and data types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional seismic inversion processes are used to create earth models, then the models can be generated with detailed geological information, but the process becomes time-consuming and expensive
Solution Approach 1:
The patent replaces the traditional mechanical seismic inversion process with a machine learning-based system. The machine learning model is trained on seismic data and well log data to directly predict earth model parameters, eliminating the need for complex iterative inversion calculations. This substitution dramatically reduces computation time while maintaining or improving accuracy.
Solution Approach 2:
The patent implements preliminary action by pre-training the machine learning model offline using extensive training datasets comprising seismic data and corresponding well log data. The pre-trained model captures complex relationships between seismic attributes and geological parameters, enabling rapid real-time or near-real-time earth model generation during drilling operations without requiring complex processing at that stage.
2Productivity
If machine learning models are used for earth modeling, then update frequency can be increased, but model instability occurs when using random weighting values
Solution Approach 1:
The patent implements feedback mechanisms where the machine learning model continuously receives new data from drilling operations and updates the earth model accordingly. The model compares predictions with actual measurements and adjusts its parameters to minimize errors, ensuring stability while enabling frequent updates. This feedback loop allows the model to adapt to new information while maintaining consistency with previously learned patterns.
Solution Approach 2:
The patent changes the parameters of the machine learning model dynamically based on incoming data. Instead of using fixed random weighting values, the model adjusts its parameters (weights and biases) through continuous training on new seismic and drilling data. This parameter adaptation enables frequent model updates while maintaining stability through systematic learning rather than random initialization.
3Measurement precision
If seismic data is converted to depth domain to relate with log data, then accurate tying is achieved, but the conversion process is slow
Solution Approach 1:
The patent changes the dimensional approach by working with seismic data in the time domain rather than converting to depth domain. The machine learning model learns to map relationships between time-domain seismic attributes and depth-domain well log data directly, effectively bridging the time-depth relationship through the learned model rather than through explicit domain conversion. This approach maintains tying accuracy while eliminating the slow conversion step.
4Device complexity
If manually selected fixed input variables are used in machine learning models, then model structure is simple, but the model becomes biased and over-relies on redundant variables
Solution Approach 1:
The patent implements self-service by enabling the machine learning model to automatically select and weight input variables based on their predictive power. The model performs feature selection and importance weighting autonomously during training, identifying the most relevant seismic attributes and well log parameters without human intervention. This self-determined variable selection eliminates bias in manual selection and reduces over-reliance on redundant variables.
Solution Approach 2:
The patent introduces dynamics into the model structure by allowing input variable weights and selections to change adaptively based on the specific drilling scenario and data quality. Rather than using fixed manually selected variables, the model dynamically adjusts which variables are most important for prediction in each context, improving objectivity and reducing bias while maintaining reasonable complexity.
Data Source
AI summary
Aspects of the present disclosure relate to earth modeling using machine learning. A method includes receiving detected data at a first depth point along a wellbore, providing at least a first subset of the detected data as first input values to a machine learning model, and receiving first output values from the machine learning model based on the first input values. The method includes receiving additional detected data at a second depth point along the wellbore, providing at least a second subset of the additional detected data as second input values to the machine learning model, and receiving second output values from the machine learning model based on the second input values. The method includes combining the first output values at the first depth point and the second output values at the second depth point to generate an updated model of the wellbore, the updated model comprising an earth model.