Evaluation of proppant injection based on a data-driven approach integrating numerical and ensemble learning models
Lei Hou, Derek Elsworth, Fengshou Zhang, Zhiyuan Wang, Jianbo Zhang
Abstract
Injecting proppant to prop open fluid-driven fractures in subsurface reservoirs is one of the key missions of hydraulic fracturing. However, quantitative evaluation of the distribution of successfully propped fractures is limited due to the infeasibility of direct measurement. This work defines an indexing parameter for field practice to estimate the proportion of proppant-filled fractures in the reservoir – the proppant filling index (PFI). A new data-driven workflow, combining numerical models and an ensemble learning algorithm, is proposed and trained on field records of both screen-out and near screen-out cases and is then applied to predict PFIs for regular cases. The algorithm performance is promoted via variable importance measure (VIM) analyses and a backward elimination strategy. Four screen-out and twelve regular cases are presented to demonstrate the predicted PFI and highlight its potential utilizations. The new PFI and workflow evaluate the proppant injection quantitatively and reveal any mismatch between proppant injection and underground fractures, which may be essential for post-fracturing analyses and reservoir characterization to improve both oil & gas recovery, the sequestration of CO2, storage then recovery of H2 and the recovery of deep geothermal fluids as important components in enabling the energy transition.