Abstract In this study, we investigate the use of deep learning-based and kernel-based proxy models in nonlinearly constrained production optimization compare their performances with directly using high-fidelity simulators (HFS) for such terms computational cost optimal results obtained. One is embed to control observe (E2CO), a model, other model proxy, least-squares support-vector regression (LS-SVR). Both proxies have capability predicting well outputs. The sequential quadratic...

Summary Life-cycle production optimization is a crucial component of closed-loop reservoir management, referring to optimizing production-driven objective function via varying well controls during reservoir's lifetime. When nonlinear-state constraints (e.g., field liquid rate and gas rate) at each control step need be honored, solving large-scale problem, particularly in geological uncertainty, becomes significantly challenging. This study presents stochastic gradient-based framework...

Abstract Population-based (gradient-free) methods have become attractive for solving optimization problems since they been known to locate "optimal" (best) positions within a search space. In this study, we investigate the use of three different gradient-free namely: Iterative Latin Hypercube Sampling (ILHS), Particle Swarm Optimization (PSO), and Genetic Algorithm (GA) well placement controls problem CO2 underground storage in 2D saline aquifer with bound constraints on design variables. We...

Abstract This study presents an efficient gradient-based production optimization method that uses a deep-learning-based proxy model for the prediction of state variables (such as pressures and saturations) well outputs bottomhole injection rates) to solve nonlinearly constrained with geological uncertainty. The surrogate is Embed-to-control Observe (E2CO) deep-learning model, consisting four blocks neural networks: encoder, transition, transition output, decoder. use output block in E2CO...

Abstract This paper presents new semilog-straight line and temperature-derivative methods (similar to pressure-derivative method commonly used in pressure transient analysis) for interpreting analyzing temperature data from constant-rate drawdown buildup tests conducted infinite-acting reservoirs containing slightly compressible fluid of constant compressibility viscosity. The procedures are based on the analytical solutions accounting Joule-Thomson (J-T) heating/cooling, adiabatic...

Summary In this work, we investigate the efficient estimation of optimal design variables that maximize net present value (NPV) for life-cycle production optimization during a single-well carbon dioxide (CO2) huff-n-puff (HnP) process in unconventional oil reservoirs. A synthetic reservoir model based on Bakken Formation composition is used. The accounts natural fracture and geomechanical effects. Both deterministic (based single model) robust an ensemble models) strategies are considered....

Abstract The objective of this study is to develop a computationally efficient methodology for the prediction oil rate, water and injection bottomhole pressure (BHP), history matching such well outputs estimate important rock fluid parameters that have significant impact on reservoir conformance after in situ polymer gel treatment. Two different machine learning (ML) proxy methods are investigated performing output data as production and/or BHP may be acquired before One ML used...

Summary In this study, we present a framework for efficient estimation of the optimal carbon dioxide (CO2)-water-alternating-gas (WAG) parameters robust production-optimization problems by replacing high-fidelity model with least-squares support-vector regression (LS-SVR) model. We provide insight and information on proper selection feature space training samples LS-SVR proxy CO2-WAG life cycle production optimization problem. Given set points generated from model-based simulation results,...

Summary This study focuses on the optimization of net present value (NPV) with respect to well controls, such as injection rates and producing bottomhole pressures (BHPs), subject nonlinear constraints field liquid production rate (FLPR) water for producers (WWPR). The core challenge is maximize NPV while adhering both linear constraints, crucial effective reservoir management. research presents a detailed comparison three sequential quadratic programming (SQP)-based optimizers—the Tulsa...

Summary We present a computationally efficient methodology based on the use of machine learning (ML) for predicting and history matching oil rate, water injection bottomhole pressure (BHP) data recorded during an in-situ polymer gel treatment. Using this methodology, we can estimate rock/fluid parameters affecting process that impact reservoir conformance. The two ML methods is investigated purpose. One used least-squares support vector regression (LS-SVR), kernel-based method, other long...

Abstract Optimizing well controls such as producing bottom-hole pressures (BHPs) and injection rates becomes more challenging when geological uncertainty nonlinear constraints field liquid water production rate are presented. Hence, the main objective of this study is to present an efficient optimization tool handle nonlinear-state for well-controlled waterflooding problems under uncertainty. In study, proposed uses our improved physics-based data-driven interwell simulator (INSIM-2P) that...

Abstract This study introduces an efficient deep learning-based reduced-order modeling (ROM) approach for reservoir history matching (HM) applications. The builds on the existing Embed-to-Control and Observe (E2CO) framework, which integrates autoencoder projecting state variables from a high-dimensional space to low-dimensional space, linear transition network predicting evolution of in latent output extending predictions well over time. E2CO framework is implemented using Proper Orthogonal...

Summary This paper presents a carbon dioxide (CO2) storage application of data-driven approach in forecasting reservoir performance (via the prediction state variables and well outputs) compositional fluid system multi-objective optimization under geological uncertainty using deep-learning-based surrogate model. The model used this work is based on multi-model Embed-to-control Observe (E2CO) architecture, consisting four blocks neural networks: encoder, transition, transition output,...

Abstract A physics-based data-driven model is an appealing alternative in situations where full-scale reservoir simulation becomes cost-prohibitive, and empirical correlations are inadequate for addressing the specific case at hand. In most recent publication related to Interwell Simulator (INSIM), which referred as INSIM-BHP, we addressed issue of pressure accuracy during waterflooding. this study, improve INSIM formulation by incorporating saturation information into equation prove that...

Abstract In the past decade, data-driven physics-based interwell numerical simulator (INSIM) has shown to be an attractive alternative high-fidelity simulators when developing full-scale reservoir models is not feasible for practical applications. However, INSIM-based are currently constrained water flooding simulations. this study, we present a novel simulator, as referred INSIM-3P, which considers three-phase flow (oil, water, and gas) in porous media based on approach. With given space...

Abstract Water alternating gas (WAG) injection has been a popular method for commercial projects worldwide. The of water and alternatively offers better mobility control hence, improves the volumetric sweep efficiency. Although WAG process is conceptually sound, its field incremental recovery disappointing as it rarely exceeds 5 to 10 % OOIP. Apart from operational problems, mechanism suffers inherent challenges such blocking, gravity segregation, in high viscosity oil, decreased oil...