New Parameterization Methods Applied to Data Assimilation for Carbonate Reservoirs

• Grant number: 25/05377-2
• Support Opportunities: Scholarships in Brazil - Doctorate
• Start date: August 01, 2025
• End date: July 31, 2029
• Field of knowledge: Engineering - Mechanical Engineering - Transport Phenomena
• Principal Investigator: Marcio Augusto Sampaio Pinto
• Grantee: Jahanzeb Tariq
• Host Institution: Escola Politécnica (EP). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Company: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Mecânica (FEM)
• Associated research grant: 17/15736-3 - Engineering Research Centre in Reservoir and Production Management, AP.PCPE

Abstract

Nowadays, ensemble-based methods have become the state-of-the-art technique to perform history matching in large-scale reservoirs (Lorentzen et al., 2020). Among the different types of ensemble-based formulations, ensemble smoother methods have become the main approach in the petroleum literature, mainly due to their ease of application in comparison with other techniques (for example, the sequential Ensemble Kalman Filter). One of the main smoother methods is the Ensemble Smoother with Multiple Data Assimilations - ESMDA (Emerick and Reynolds, 2013). One of the main drawbacks in this kind of method occurs due to the Gaussian assumption during its formulation, making its application prohibitive when the uncertain parameters do not follow a Gaussian behavior. Regarding the many non-Gaussian parameters usually found in the petroleum reservoirs (e.g., categorical variables), the main method to make possible the assimilation with ensemble-based methods is the class of techniques called parameterization, which maps the original parameter field N^m to a Gaussian domain N^z: N^m¿N^z. Despite the several parameterization methods in the literature, the application of generative adversarial networks (GAN) has gained attention in the reservoir community (Canchumuni et al., 2021; Bao et al., 2022) as a potential solution to the parameterization problem. However, its application in more realistic/complex reservoirs is at an incipient stage. In this context, this project aims to address this gap by developing a robust workflow that incorporates current techniques in ensemble-based history matching. The goal of this proposal is to create a novel methodology using the parameterization techniques for large-scale carbonate reservoirs. One promising application is in the use of advanced data augmentation techniques specifically designed for GANs (Karras et al., 2020; Zhao et al., 2020; Tran et al., 2021). One of the primary challenges of using GANs for large-scale reservoirs is the significant dataset size required to train larger network structures. By implementing effective data augmentation techniques tailored to reservoir problems, we may bridge this gap and make the application of 3D-GANs more practical. In this way, this research will assess the effectiveness of these innovative techniques and develop a suite of robust algorithms as part of the deliverables. Furthermore, this project also presents opportunities for the development of additional parameterization techniques. As part of the research plan, we will apply these methodologies and evaluate their performance using the UNISIM-III benchmark (Correia et al., 2020), which is a large-scale carbonate reservoir model containing karst trends. (AU)