Pressure Transient Analysis in Shale Wells with Heterogeneous Fractures by Using a Deep Learning Based Surrogate Model

Abstract

With hydraulic fracturing technology, manmade fractures can be generated around the shale-gas wells. After hydraulic fracturing at each stage, many wells in shale reservoirs have the "shut-in" process, which providing many precious data for parameter estimation. But, owing to intricate geological and engineering factors, the fractures in reservoirs are asymmetric and heterogeneous, which brings a great challenge for fracture estimation. To improve this situation, coupling the deep learning (DL) approach and field practices, we established a surrogate model for non-uniform fractures at one stage based on deep Bi-directional LSTM model. First, a well testing model containing three distinct flow regions is developed, namely (1) heterogeneous hydraulic fractures, (2) the inner region affected by hydraulic fracturing, and (3) the outer region without stimulation. Laplace transformation method are used for model solutions. Then, with the model solutions, a surrogate model based on deep bidirectional LSTM is built for improve computational efficiency. The results show that the model can effectively reduce the early prediction error of pressure derivative, and the average relative prediction error is 1.67%. Finally, model verification was shown by comparing with the results from traditional well testing model. The results show that the calculation speed of the surrogate model is three orders of magnitude higher than that of the well test model, which helps to efficiently evaluate the fracture parameter in complex fracture system generated by large-scale fracturing treatments in shale reservoirs.