Machine Learning approach in Geoplat AI
Transfer learning
Neural Networks pre-training
3 stage
2 stage
1 stage
Synthetic data generator
Additional training of neural networks based on
real seismic data and interpreter’s guidance allows for our approach to adapt to any complications.
real seismic data and interpreter’s guidance allows for our approach to adapt to any complications.
ML algorithms have been applied to the synthetic data, gaining ability to recognize faults, noise, geobodies, etc.
~1 mln synthetic data samples with various conditions make our approach universal for any kind of geological environment.
Stage 1
Synthetic data generator
Synthetic data generator has created over 1 mln models for various situations: adding structural and random noise, with complex and simple faults, including horizontal and dipping reflections, considering geobodies, etc. Because of the amount and the automatic creation AI was able to create datasets even for complex geological environments.
Stage 2
Neural networks pre-training
Neural network analyze input data and looking for certain indications that characterise the issue: noise component, fault zone, geobody position, etc. After that it writes down the information into database in a form of weight coefficients (weights).
Result
Ready-made universal solution for searching and highlighting any types of the following features:
Noise component (regular, irregular)
Horizons (for smoothing or increasing resolution)
Fault zones (in the form of a probability cube)
Geobodies (in the form of a probability cube)
Facies (in the form of layer cube)
These universal models are available by default for use on seismic data with any geological structure.
Stage 3
Transfer Learning
- A technique for adapting a neural network model to specific seismic data
- Usually 2-5 marked sections with manual picking are required
- Provides significant improvement in interpretation results, especially in areas with complex tectonic structure
Benefits
Universality ofmodels for all typesof geological assets
Including archive data or data with complex geology
Including archive data or data with complex geology
Quick forecast without any previous tracing
Most of the procedures can give you a result within hours just based on raw seismic dta
Most of the procedures can give you a result within hours just based on raw seismic dta
Low requirements on computing resources
The user needs only a computer with a modern GPU to perform any procedure even on big size data
The user needs only a computer with a modern GPU to perform any procedure even on big size data
Exception of human bias
Allowing the user to focus on the interpretation rather then routine work
Allowing the user to focus on the interpretation rather then routine work
Highspeedof model training
Takes only several days to process data and increase its quality significantly
Takes only several days to process data and increase its quality significantly
Other Features
Horizon interpretation is the core process in understanding the structural features of a geological cross section and conducting reliable dynamic seismic analysis.
Fault interpretation is one of the most difficult tasks within a general structural interpretation workflow.
Liverpool city will open its doors to annual surrealistic show. Tenner Hall / August, 15.
Fault interpretation is one of the most difficult tasks within a general structural interpretation workflow