Apply Reduced-Physics Modeling to Accelerate Depletion Planning Optimization Under Subsurface Uncertainty

Abstract

Abstract Robust decision-making regarding reservoir management using model-based strategies requires a large number of evaluations which can be enormously time-consuming if incorporating the full-field simulation. This challenge becomes more acute when dealing with subsurface uncertainty represented by multiple geologic scenarios. Various reduced-physics and reduced-order models such as streamlines and upscaling are commonly applied to accelerate the optimization process by reducing the computational burden of each evaluation. In this paper we propose an innovative integrated workflow that applies Flow Diagnostics, a reduced-physics modelling tool, to accomplish this acceleration and the mesh-adaptive direct search (NOMAD) algorithm to efficiently optimize well location. Flow Diagnostics (FD) is a reduced-physics approach that characterizes key flow behaviors and reservoir heterogeneity by combining a single-phase pressure solution with a time-of-flight estimator based on steady-state flux. The time-of-flight values for each grid cell can subsequently be combined with initial reservoir conditions to estimate saturation-weighted 3-phase production using 1D Buckley-Leverett fractional flow relationships. Our testing of this technique on a 2-phase, water flooding asset showed that, while it does not accurately predict exact volumes, there is a strong rank-order correlation with the full numerical simulator. Thus, this technique is an efficient one to assist in the decision-making and optimization process. To demonstrate this workflow, we applied this strategy to a synthetic base model of a deep water reservoir. Synthetic uncertainty was added to the base case to generate multiple subsurface scenarios. Representative P10, P50 and P90 cases were selected to represent low-, middle-, and high-side cases based on cumulative distribution function (CDF) of NPV. Different well placement optimization studies using high fidelity models and reduced order models were then performed on these cases (optimization of each single scenario on its own and optimization across all representative scenarios simultaneously) to compare solutions and effectiveness. Results indicated that, when the optimized well plans were run on the simulator, optimization under uncertainty provided a solution more robust across the uncertainty space. The results of the full study successfully demonstrate the efficacy of our workflow.