Methodology for robust production optimization for the management and redevelopment of petroleum reservoirs using machine-learning techniques

• Grant number: 22/13501-7
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Effective date (Start): February 01, 2023
• Effective date (End): April 21, 2023
• Field of knowledge: Engineering - Mechanical Engineering
• Principal Investigator: Denis José Schiozer
• Grantee: Isabela Magalhães de Oliveira
• Supervisor: Eduardo Gildin
• Host Institution: Centro de Estudos de Energia e Petróleo (CEPETRO). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Research place: Texas A&M University, United States
• Associated to the scholarship: 20/14092-8 - Methodology for robust production optimization for the management and redevelopment of petroleum reservoirs using machine-learning techniques, BP.DR

Abstract

The management of complex reservoirs, such as the naturally fractured typical of Brazilian pre-salt, demands high computational cost and time, requiring practical and efficient methodologies. The Closed-Loop Reservoir Development and Management (CLRDM) proposed by Schiozer et al (2019) is a promising methodology, as it describes 12 steps to assist the model-based decision-making process. Despite being designed for practical applications, this methodology bears challenges in complex cases, such as (i) the necessity of thousands of evaluations during robust optimization and (ii) a high number of uncertainties that increases even further the number of simulations runs and the time of the process. Machine learning (ML) techniques are being used by the literature during optimization procedures to speed up the life-cycle process and to increase the precision of the production forecast in the short term. The main objective of this work is to develop a methodology that improves the management of the field through the study of CLRDM cycles in a carbonate reservoir with WAG-CO2 injection utilizing reservoir simulation and ML techniques. The focus is on the execution of step 10 of the CLRDM process, which is responsible for robust production optimization. The PhD methodology is divided into four approaches: (i) life cycle decision-making based on model-based numerical simulation, (ii) life cycle decision-making based on ML techniques, (iii) short-term decision-making for a cycle of the management phase (1 year) utilizing model-based numerical simulation, and (iv) short term decision-making for a cycle of the management phase (1 year) utilizing ML techniques. Finally, the feasibility of integrating numerical simulation with ML techniques during the management phase is analyzed. Regarding the variables to be optimized in this work, they are called control variables (G2), to establish the operational specifications of equipment over time, and revitalization variables (G3), to evaluate the necessity of the second wave of wells. The optimizer is the Iterative Discrete Latin Hypercube (IDLHC), and the case study is the public access benchmark UNISIM-IV-2022, a carbonate reservoir model with characteristics of the Brazilian pre-salt created by the UNISIM research group at UNICAMP. This benchmark also has a reference model, UNISIM-IV-R, also created by the UNISIM group, and it is considered the real reservoir of the UNISIM-IV-2022 case study. (AU)