An enhanced data-driven framework for subsurface fluids compositional equilibrium distribution modeling

In this study, we introduce an advanced data-driven model aimed at accurately characterizing the three-dimensional distribution of subsurface fluid’s compositions, crucial for ultra-deep, high-temperature, and high-pressure volatile oil reservoirs. Existing models often overlook thermal diffusion and convection, leading to inaccurate simulations. We propose a combined algorithm integrating three-dimensional modeling, reservoir simulation, and history matching inversion to improve accuracy. The process begins with individual well profiles, establishing temperature and pressure fields across the reservoir, followed by calculating the molar fractions of oil compositions in each grid. Acknowledging uncertainties in interpolation modeling and thermal diffusion, we generated 200 prior models and performed additional calibration for another well. The models were optimized using the Particle Swarm Optimization algorithm, generating 20 posterior models that significantly reduced uncertainties. This framework can be applied to other reservoirs with compositional gradients, offering a new approach to initializing compositional models. Our research presents a data-driven framework that enhances reservoir simulation accuracy and provides insights into fluid distribution complexities in volatile oil reservoirs, holding potential for improving oil recovery strategies and optimizing reservoir management.