Techniques for effective simulation, optimization, and uncertainty quantification of the in-situ upgrading process

Strongly temperature-dependent compositional flow/transport, chemical reactions, delivery of energy into the subsurface through downhole heaters, and complex natural fracture architecture render the dynamic modeling of in-situ upgrading process (IUP) a computationally challenging endeavor for carbonate extra-heavy-oil resources. Economic-performance indicators for IUP can be considerably enhanced via pattern optimization. IUP is endowed with uncertain subsurface parameters as in the case of other recovery mechanisms. Simulation results must reflect the impacts of these uncertainties; hence they should always deliver “expected-value production functions” and their attached uncertainty ranges, in short, the “error bars”. Both the optimization and uncertainty quantification workflows require (typically multiple) multi-scenario simulations, and are therefore very compute intensive. cribe our recent developments in simulation techniques, optimization algorithms, tool capabilities, and high-performance computing protocols that in unison form a massively parallel simulation/optimization/uncertainty–quantification workflow, in which it is almost equally easy to produce recovery time-functions with an attached uncertainty range, as it is to run a single simulation. Our simulation platform supports various optimization and uncertainty quantification methods, such as conventional as well as robust optimization using a novel simultaneous perturbation and multivariate interpolation technique, experimental design, and Monte Carlo simulation, that can be linked together through a unified script-based interface, to carry out optimization in the presence of subsurface uncertainties and to quantify the impact of these uncertainties on simulation results. Application of our massively parallel dynamic modeling workflow is illustrated on a proprietary IUP recovery method for a complex naturally fractured extra-heavy oil (bitumen) reservoir as example. After briefly explaining these recovery processes and the modeling approach, we show the techniques (including their accompanying application results) that notably accelerate the (single-model) simulation process; effectively identify the predominant subsurface uncertainties; rapidly optimize heater-producer patterns under the influence of predominant subsurface uncertainties; and efficiently compute expected-value production functions with error bars.