Selection of a vertical infill well placement using a multi-modal convolutional neural network trained using reservoir simulation
Abstract
This study selects an optimal placement for a vertical infill well at an oil reservoir using a multi-modal convolutional neural network (CNN). Multi-modality is implemented for importing a multi-dimensional input array that consists of reservoir properties near a candidate infill well (e.g., permeability and saturation) to the convolution layer, which is the first layer of the CNN to extract features from the input array. The features are obtained by repeating convolution, pooling, and activation processes, and then delivered as a one-dimensional array to a fully connected layer of the CNN. Ancillary 1D input data such as well location are added to the fully connected layer for improving the training performance of the CNN. The fully connected layer is, via hidden layers, linked to the output layer that evaluates the productivity at the infill-well placement. The proposed data-driven CNN is designed using Google's TensorFlowTM. The performance of the CNN is tested with application to a benchmark oil reservoir model SPE10 and compared to the performance of a conventional neural network. Training and test data for the neural networks are obtained by running full-physics simulations for selected infill well scenarios. The trained CNN is used to evaluate productivities at candidate well placements. The accuracy of the CNN results is comparable with full-physics simulation results at cheaper computational costs. The results highlight the potential of the array-based data-driven machine-learning application to expedite the selection of the optimal well location with partially replacing computationally expensive full-physics reservoir simulation.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFM.H23M2143M
- Keywords:
-
- 1816 Estimation and forecasting;
- HYDROLOGYDE: 1846 Model calibration;
- HYDROLOGYDE: 1847 Modeling;
- HYDROLOGYDE: 1873 Uncertainty assessment;
- HYDROLOGY