Subsurface Physics Inspired Neural Network to Predict Shale Oil Recovery under the Influence of Rock and Fracture Properties

Abstract

Shale oil reservoir is one of the promising unconventional oil reservoirs with large deposits worldwide. Due to its complex nature and development strategy planning, understanding the behavior and devising a development strategy through numerical simulation is inexpensive and time-consuming. Hence, this study proposes to implement an Artificial neural network to understand the mechanism involved in the shale oil matrix and predict its performance. This study involves, at first, generating a mechanistic model in order to develop a dataset featuring the responses of fracture and rock properties on shale oil recovery. Later, an artificial neural network is trained and tested to predict the oil recovery under the influence of a defined set of fracture and rock properties. The neural network is validated by k-fold cross-validation. Moreover, an optimizer is used to minimize the loss in the trained proxy model. The results show that the generated proxy model efficiently predicts the oil recovery on different fracture and rock properties. Moreover, the model results are found to be consistent with the trend analysis showing the robust nature of the model in predicting oil recovery.