Uncertainty assessment and risk analysis of steam flooding by proxy models, a case study

Abstract

Many uncertainties exist in development of oil fields. Developing proxy models as substitutes for reservoir simulators is a fast method for uncertainty analysis. Artificial neural network (ANN) and polynomial regression models (PR) were used as proxy models in a sector of an Iranian heavy oil reservoir under steam flooding scenario. Screening analysis was used to find influential uncertain parameters. Different experimental designs have been applied to construct proxy models. A combination of Box-Behnken and inscribed central composite designs was most informative design. Comparison between proxy models results and simulator outputs shows that ANN and PR models can accurately predict the simulator outputs. However, the deviation of ANN from actual results is less than quadratic polynomials. The constructed proxy models were used in Monte Carlo simulation to obtain probabilistic production forecasts. The combination of experimental design and proxy models is a fast and accurate tool for risk analysis and prediction.