Prediction of the Continuous Probability of Sand Screenout Based on a Deep Learning Workflow

Summary Sand screenout is one of the most serious and frequent challenges that threaten efficiency safety hydraulic fracturing. Current low prices oil/gas drive operators to control costs by using lower viscosity lesser volumes fluid for proppant injection—thus reducing sand-carrying capacity in treatment increasing risk screenout. analyses predict as isolated incidents based on interpretation pressure or accumulation. We propose a method continuous evaluation prediction combining data-driven methods with field measurements recovered during shale gas The probability updated, redefined, used label original data. Three determining elements are proposed, which four indicators generated training deep learning model [gated recurrent units (GRU), tuned grid search walk-forward validation]. Training records following manually trimmed force machine algorithm focus prescreenout data, then improves Pearson coefficients analyzed STATA software remove obfuscating parameters from inputs. extracted optimized, via forward selection strategy, their contributions according confusion matrix root mean squared error (RMSE). By optimizing inputs, accurately predicted testing cases, well precursory predictors, evolution prior effect pump rate analyzed, defining U-shaped correlation suggesting safest-fracturing (SFPR) under both low- high-stress conditions. SFPR, together, allow monitoring real time fracturing operations provision appropriate mitigation strategies.