Field Realization of Adaptive Reservoir Simulation for Steam Injection Forecasting

Abstract

Globally, steam injection for heavy and high-viscous oil recovery is increasing, including carbonate reservoirs. Lack of full understanding such reservoir heating and limited information about production and injection rates of individual wells require to forecast steam injection not only deterministic and simple liquid displacement characteristic modeling types, but also the data-driven one, which covers the adaptive modeling. The implementation and validation of the adaptive system is presented in this paper by one of the world's largest carbonate reservoirs with heavy and high-viscous oil of the Usinsk field. Steam injection forecasting in such reservoirs is complicated by the unstable well interactions and relatively low additional oil production. In the adaptive geological model, vertical dimensions of cells are similar to gross thicknesses of stratigraphic layers. Geological parameters of cells with drilled wells do not necessarily match actual parameters of those wells since the cells include information of neighboring wells. During the adaptive hydrodynamic modeling, a reservoir pressure is reproduced by cumulative production and injection allocation among the 3D grid cells. Steam injection forecasting is firstly based on the liquid displacement characteristics, which are later modified considering well interactions. To estimate actual oil production of steamflooding using the reservoir adaptive geological and hydrodynamic models, dimensionless interaction coefficients of injection and production wells were first calculated. Then, fuzzy logic functions were created to evaluate the base oil production of reacting wells. For most of those wells, actual oil production was 25 – 30 % higher than the base case. Oil production of steamflooding for the next three-year period was carried out by modeling two options of the reservoir further development - with and without steam injection. Generally, forecasted oil production of the option with steam injection was about 5 % higher. The forecasting effectiveness of cyclic steam stimulations of production wells was done using the cross-section method, when the test sample was divided into two groups - the best and the worst, for which the average forecasted oil rates after the stimulations were respectively higher or lower than the average actual oil rate after the stimulations for the entire sample. The difference between the average actual oil rates after the stimulations of the best and the worst groups was 32 %, i.e. this is in how much the actual oil production could have increased if only the best group of the sample had been treated.