Multi-view 3D seismic facies classifier
| Autor: | Mauro Roisenberg, Elton Alves Trindade |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Network architecture
Artificial neural network Heuristic Computer science 020207 software engineering 02 engineering and technology computer.software_genre Convolutional neural network Synthetic data 020204 information systems 0202 electrical engineering electronic engineering information engineering Reservoir modeling Benchmark (computing) Data mining Classifier (UML) computer |
| Zdroj: | SAC |
| DOI: | 10.1145/3412841.3441976 |
| Popis: | Technological advances in oil and gas reservoir characterization, such as 3D seismic and seismic attributes, enriched the subsurface's description made by specialists. Nevertheless, the analysis of this now huge volume of data became a complex task. In order to more efficiently manage big seismic data, this work explores a computationally cheaper network with the use of 2D orthogonal planes convolutional neural networks for 3D seismic cube facies classification and lithostratigrafic groups, supported by an heuristic based on geological principles, which is one of the steps of reservoir characterization and oil and gas exploration. We proposed a 2D to 3D transfer learning in which we split the training samples of our 3D data as 3 orthogonal slices and convert the trained parameters to a 3D counterpart of the network, with each direction of the training planes conveniently converted to a 3D convolution, based on geological insights. Through a sampling method that captures spacial information of seismic data, the proposed model was applied in the synthetic data of the Stanford VI-E reservoir and the real seismic data of F3-block dataset. Compared to other models in the same benchmark, the proposed AH-Net Ensemble classifier, with a geological heuristic obtained better results than other architectures of literature with a very feasible computational cost in both datasets, which suggests that such approach is a promising one and of easy replication, since it can be seamlessly applied to any network architecture. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|