Prediction of Fracturing Pressure and Parameter Evaluations at Field Practical Scales.

Hydraulic fracturing pressure is one of the key criteria in defining the pumping schedule and in dimensioning surface plants for fracturing operations. However, it is difficult to predict at a practical field scale due to the complex determining factors, involving geological, hydrodynamic and rock mechanical parameters. This study proposes a synthetic data-driven workflow, integrating both hydrodynamic and rock mechanical models, to predict the pressure based on fracturing experience in neighboring wells. The data quality and performance of the geological, hydrodynamic and rock mechanical features are evaluated based on a backward elimination strategy and control variate method, using error evolutions as criteria. Relatively small errors (root mean square error 5.1 ~ 5.7 MPa and mean absolute error 5.6 ~ 8.3%) are returned for wells/cases within the same region as the training wells, demonstrating the superior performance of the workflow. The prediction errors increase significantly with increasing distance between training and testing wells – defining the range of applicability of the workflow. The rock mechanical feature (represented by the brittleness index) provides a larger contribution to the prediction than that of the hydrodynamic feature (represented by the proppant accumulation) in most of the testing cases. Young's Modulus exhibits a higher performance (induces the smallest errors in pressure prediction) to characterize the rock mechanical feature of the formation, compared with the brittle mineral ratio and Poisson's Ratio. The data quality of geological stresses and Poisson's Ratio may require improvements based on the irregular error evolutions. The remaining errors and proximity/regional limitations of the data-driven workflow are critically discussed. This new method provides a platform for the prediction of pressures at field-practical scales, which may be significant for both hydraulic fracturing and geological storage of CO2 and H2 in depleted reservoirs with sufficient historical hydraulic fracturing records. Highlights: A data-driven workflow is built to predict fracturing pressure at field scales. Errors increase with the increasing distance between training and testing cases. Contribution of rock mechanical feature is larger than hydrodynamic feature. Young's Modulus performs better in characterizing rock mechanical feature. Lower data qualities of geological stresses and Poisson's Ratio are revealed. [ABSTRACT FROM AUTHOR]