Determination of an infill well placement using a data-driven multi-modal convolutional neural network

Abstract

This study determines the optimal placement for a vertical infill well using a multi-modal convolutional neural network (CNN). 3D arrays composed of static and dynamic reservoir properties near a candidate infill well are inputted to the convolution stage of CNN. Multi-modal learning is applied to CNN for feature extraction of inputs. The features are compressed via fully connected layers for evaluating the productivity of every candidate infill scenario. The proposed CNN is applied to a channelized oil reservoir, and its performance is compared to that of a feedforward neural network. Dataset for the neural networks is obtained by running full-physics simulations for selected scenarios. CNN outperforms the feedforward neural network for the test scenarios of single- and dual-modal cases. Both neural networks yield comparable predictability for a quad-modal case. Results of the quad-modal CNN are in agreement with reservoir simulation results at cheaper computational costs. The results highlight the potential of data-driven machine learning in expediting the optimal well placement by partially replacing expensive simulation runs.