Improving reservoir characterization using the adjoint method in history matching

With globally rising demand for fossil fuels and the continuous maturation of oil and gas fields worldwide, reservoir characterization is becoming increasingly important in reservoir management studies for the successful development of realistic reservoir models. Most reservoir models constructed from cores, logs, seismic data etc. are only satisfactory replica of subsurface reservoirs in best case scenarios. This is so because data acquired at various scales are only sufficient to estimate reservoir size, structure, massive faults and petrophysical properties at well locations. Information on inter-well permeability/porosity are unknown thus leading to the omission of inter-well flow barriers and channels. In other cases, oil and gas companies build reservoir models using sparse data either due to time or budget constraints. The consequences of building subpar reservoir models is that they never reproduce measured data even after spending considerable efforts in history matching. State-of-the-art methods available today for improving reservoir characterization in reservoir models are mostly data-driven methods that use only measured injection and production data to infer hidden reservoir features like faults and channels. These methods suffer from major limitations such as the lack of support for simulation models with constant injection rates, poor support for simulation models undergoing significant changes in flow patterns over time (e.g. introducing an infill well), satisfactory support for aquifer-dominated reservoirs, limitations on well architectures supported and the inability to actually reveal hidden reservoir features in reservoir models. This dissertation introduces a proposed method for improving reservoir characterization using a history matching method that utilizes the adjoint method which is a powerful method for computing sensitivities. These sensitivities are needed to fit reservoir simulation models to measured data through unconstrained history matching (UHM) for the purpose of revealing hidden reservoir features like faults, channels etc. The llowed to vary from zero to their default maximum value or possibly higher. Subsequently, developed detection algorithms are applied directly on best-case permeability and/or porosity arrays obtained from UHM in order to reveal the location, shape and other essential properties of hidden reservoir features. Finally, classical history matching (CHM) is performed on the improved model (new base model containing revealed reservoir features) with the original or geologically constrained permeability and/or porosity distribution respected and other model parameters included. The proposed method is evaluated with a handful of synthetic and real field reservoir simulation models in order to ascertain its effectiveness in revealing hidden reservoir features under various practical reservoir settings. Results obtained show that the proposed method outperforms other methods existing in literature. In addition, the proposed method does not suffer from aforementioned limitations. It is also discovered that production noise greater than ±15% as well as numerical noise can have a detrimental impact on the performance of the proposed method. Finally, the capability of the proposed method in revealing certain types of reservoir features during UHM is heavily dependent on the model parameters defined.