GAN Generator Fills Missing MWD Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In measuring or logging while drilling (MWD) or logging while drilling (LWD), data may be missing due to various reasons such as non-capture, transmission errors, or tool design limitations, leading to incomplete data sets that can negatively impact drilling plans.
Innovation Solution
A method using a generative adversarial network (GAN) is employed to fill in missing data by training a discriminator to differentiate between original and modified images with artificial subsurface data, and a generator to replace missing data with artificial subsurface data, ensuring the discriminator cannot distinguish between original and modified images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If multiple passes are made along the well to obtain multiple readings, then the completeness of data is improved, but the time required for data collection increases
Solution Approach 1:
The system performs preliminary actions by making multiple passes along the well in advance to collect redundant readings. These pre-collected multiple readings are then used to fill in missing data portions, ensuring data completeness is achieved before the drilling plan execution begins, thereby avoiding time delays during the actual drilling operation.
2Measurement precision
If artificial subsurface data is generated to fill missing data, then the accuracy of drilling plans is improved, but the complexity of data processing increases
Solution Approach 1:
The system creates artificial subsurface data by copying and generating fake readings that replicate the characteristics of real MWD/LWD data. These artificial data copies are designed to be indistinguishable from actual measurements, allowing them to seamlessly fill missing portions while maintaining the same statistical properties and geological context, thus improving accuracy without requiring complex verification processes.
Solution Approach 2:
The system employs a feedback mechanism where the discriminator continuously evaluates the generated artificial data against the original real data distribution. This feedback loop adjusts the generator's output to ensure the artificial readings closely match the characteristics of actual MWD/LWD measurements, thereby achieving high measurement precision through iterative refinement rather than complex one-time processing.
Data Source
AI summary
A discriminator of a training model is trained to discriminate between original training images without artificial subsurface data and modified training images with artificial subsurface data. A generator of the training model is trained to: replace portions of original training images with the artificial subsurface data to form the modified training images, and prevent the discriminator from discriminating between the original training images and the modified training images.