Well Data Augmentation for AI Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing well data acquisition methods often produce inaccurate measurements due to unusual geological and drilling conditions, which can lead to incomplete training datasets for models used in subsurface formation analysis.
Innovation Solution
The method involves generating augmented well data using geological and drilling factors, combined with machine-learning epochs to create a model that replicates and adjusts well data, thereby expanding the training dataset to account for various scenarios encountered during well data acquisition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If well data is acquired using logging tools, then subsurface formation properties can be measured, but measurement accuracy deteriorates due to unusual geological and drilling conditions
Solution Approach 1:
The patent creates synthetic copies of well data by generating augmented well data that replicates the patterns and characteristics of real well data. These synthetic copies are used to train machine learning models, allowing the system to learn from multiple similar scenarios without being limited by the accuracy of individual real measurements under unusual conditions.
Solution Approach 2:
The patent transforms real well data into augmented well data by applying various parameter changes and transformations. This process modifies the data to create variations that represent different geological and drilling conditions, enabling the model to handle a broader range of scenarios while maintaining measurement accuracy.
2Adaptability or versatility
If machine learning models are trained using limited real well data, then model development is simpler, but the model's ability to capture diverse geological scenarios is insufficient
Solution Approach 1:
The patent generates synthetic copies of well data through data augmentation techniques. These copies replicate the essential patterns and features of real well data while introducing variations that represent different geological scenarios. This allows the training dataset to be expanded significantly without requiring additional real measurements.
Solution Approach 2:
The patent performs preliminary data augmentation before model training to create a comprehensive synthetic training dataset. By preparing augmented data in advance, the system ensures that the model receives diverse training examples that cover various geological conditions, improving the model's adaptability without delaying the training process.
3Measurement precision
If data augmentation is applied to expand training datasets, then model accuracy improves, but the complexity of data processing increases
Solution Approach 1:
The patent applies parameter transformations and data augmentations that modify well data in systematic ways. These parameter changes create varied training examples while following established transformation rules, which manages processing complexity through structured approaches rather than random complex operations.
Solution Approach 2:
The patent replaces manual data processing and analysis with automated machine learning models that consume augmented data. By substituting mechanical data processing steps with automated AI algorithms, the system handles the complexity of data augmentation and model training through computational processes rather than manual intervention.
Data Source
AI summary
A method may include obtaining first acquired well data. The method may further include generating augmented well data based on the first acquired well data. The augmented well data may be generated using a geological factor and a drilling factor. The method may further include generating a model using various machine-learning epochs and the first acquired well data and the augmented well data. The model may be trained by replicating a portion of the first acquired well data and the augmented well data during a machine-learning epoch among the machine-learning epochs. The method may further include generating adjusted well data for a region of interest using the model and second acquired well data.