Efficient workflow for assisted history matching and Brownfield Design of experiments for the Tengiz field

The main goal of history matching is to calibrate a reservoir model to production data and provide a model on which field development decisions can be made. Most of the history matching effort conducted in the industry is based on a single deterministic simulation model. With recent developments in probabilistic history matching techniques, faster computers, and efficient numerical algorithms, the opportunity to exploit these resources for efficient history matching workflows becomes available. In this paper we present a combined approach of assisted history matching followed by Brownfield Design of Experiment (DoE or ED) for model development for the Tengiz field. The Tengiz oilfield, located along the northeastern margin of the Caspian Sea, is the world's deepest producing supergiant oil field. The underlying simulation model of the field uses a dual-porosity, dual-permeability compositional formulation. The first step in the workflow was performing sensitivity study using one variable at a time analysis (OVAT) to reduce the number of uncertainties. This was followed by several Design of Experiments cycles to minimize the model misfit in static bottomhole pressures and MDT pressures and to modify the uncertainty ranges for the uncertain variables impacting the history match. The history match for the Tengiz field was achieved without any local modifications. The next step in the workflow was simulating the Brownfield DoE to create probabilistic reservoir models. The objective of this step was to develop multiple history matched models that generate a range of prediction outcomes. Two sets of prediction scenarios were considered: a base case which included all business plan work and a future development case. 203 DoE cases were run for each of the development scenarios. All DoE cases were used to generate proxy equations for the pressure mismatch functions (relative error and L2-norm) and cumulative oil production for the base case and the future development case. Multiple proxies were used to select probabilistic reservoir models while maintaining a good history match. The resulting probabilistic models will be used for reserves estimations, production optimization from existing infrastructure, design of future projects (ongoing drilling and expansion of an existing miscible flood), and to assess any future opportunities.