Optimal Placement of Infill Wells at a Petroleum Reservoir based on Time-series Forecasting using Multi-modal Convolutional Neural Network
Abstract
This study develops a multi-modal convolutional neural network coupled with long short-term memory (M-CNN-LSTM) to determine the optimal placement of infill wells at a petroleum reservoir. The developed algorithm identifies the optimal placement maximizing the cumulative oil production. The M-CNN-LSTM correlates static (e.g., permeability and porosity) and dynamic (e.g., pressure and saturation) properties near each well as inputs with time-series cumulative oil production as outputs. Multi-modal learning is employed to import static and dynamic properties simultaneously for feature extraction in M-CNN. LSTM handles the time-dependence of reservoir properties in the algorithm. The performance of M-CNN-LSTM is tested with application to a benchmark channelized oil reservoir. The results of this study highlight the efficacy of coupling CNN and LSTM for handling the time-dependence of a petroleum reservoir in a well-placement optimization problem, revealing accuracy and savings at the computational cost.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H15O1220K