Transfer-Learning Validation Improves Hydraulic Fracture Design

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper IPTC 24850, “Physics-Informed Machine Learning for Hydraulic Fracturing—Part III: The Transfer-Learning Validation,” by Abdul Muqtadir Khan, SPE, Esteban Ugarte, and Stevanus Kurniadi, SPE, SLB, et al. The paper has not been peer reviewed. Copyright 2025 International Petroleum Technology Conference. _ The growth of machine learning (ML) approaches has sparked innovations in many applications including hydraulic fracturing design. The crucial drawback in these models is the subjectivity and expertise of the design engineers. In this study, the authors conduct a transfer-learning validation with another robust real-field data set by focusing on predicting different fracture-design parameters by investigating the advantages of transfer learning using a pretrained backbone with physics-based data and tailoring it toward real-field data. Results from this study showed the advantages of using the transfer-learning approach. Introduction By integrating physical laws and constraints, especially in complex, multiphysics systems, ML models can be enhanced with hybrid solutions. Physics-informed machine-learning (PIML) techniques and physics-informed neural networks enable the model to function in a middle ground between fully data-driven models and fully physics-based models. The complete paper is the third in a series of papers devoted to PIML applications. PIML has emerged as a hybrid technique that combines the strengths of physics-based models with data-driven ML approaches. Its key innovation is its ability to encode the underlying physics of hydraulic fracturing processes into the ML model by leveraging synthetically generated data sets, which simulate real-world conditions. These synthetic data provide a comprehensive and robust foundation that captures the critical relationships between input parameters—such as reservoir properties, fluid efficiency, and proppant parameters—and the resulting production outcomes. Basin Overview and Challenges The data for this study was taken from Middle East fracturing experience of the past 15 years. The A and B fields consist of three main formations—X, Y, and Z. Sandstone thicknesses of 550 m have been penetrated within the basin; however, the sandstone trends much thinner toward the north (50 to 100 m). The X formation is the most prolific among the three formations, with average permeability of approximately 0.1 md and approximately 6% porosity. The Y formation is well-developed in the basin with a maximum thickness of over 350 m. The Y is of lower petrophysical quality when compared with the X but features thicker continuous gross height. The Z formation is basal conglomeratic sandstones that are believed to represent alluvial fan complexes. These Z sandstones possess very low permeability, with values between 0.001 and 0.1 md. Despite the low permeability, the formation has significant gross thickness, up to 120 m. Following an extensive exploration and appraisal program, it was determined that hydraulic fracturing should be used to achieve the economical rate of the well in A and B Field development. To understand the full cycle of technical planning, deployment, and evaluation, an integrated surveillance road map was created for the field-development plan. The data set is termed high-quality because integrated surveillance was deployed for various fracture geometries and related measurements, including fracture height, fracture length, fracture conductivity, and well testing. These surveillance methods are detailed in the complete paper.