Deep neural network model for estimating montney shale gas production using reservoir, geomechanics, and hydraulic fracture treatment parameters

Abstract

The development of a shale gas project requires an accurate prediction of the production potential owing to the high cost of horizontal well drilling and hydraulic fracturing treatment. Previous research developed models for estimating shale gas production based on the reservoir and hydraulic fracturing data. They reported a high degree of uncertainty in the prediction as a result of the fracture properties' locking information. This research aims to reduce the uncertainty in shale gas production prediction by developing a deep neural network (DNN) model that incorporates reservoir, geomechanics, and fracture treatment parameters to predict monthly Montney shale gas production. The DNN model, optimized using the Keras Tuner optimization tool, employs a unique architecture of eight hidden layers and integrates production time as an input feature. This allows the model to predict production rates at any time point in the reservoir's life cycle. The prediction model was developed using 102,000 simulation data and blind tested using other 18,000 unseen simulation data, which were achieved by coupling a hydraulic fracture simulator with a reservoir simulator. The model revealed an average error of 13.41%, with a promising correlation between predicted and simulated gas production rates (R2 > 0.85), and cumulative gas production (R2 > 0.89). These results demonstrate the model's accuracy in predicting Montney shale gas production and its utility in assessing the economic viability of shale gas projects. In conclusion, the research introduces an important advancement in reducing uncertainty in shale gas production prediction, providing an accurate tool that has wide applications in assessing the viability of shale gas projects and suggesting further exploration in refining the prediction techniques for other geological formations.