Implementation of Assisted History Matching Under Uncertainty in Integrated Reservoir Modeling

Abstract

Abstract This paper presents an application of probabilistic, ensemble-based computer-Assisted History Matching (AHM) with uncertainty to Integrated Reservoir Model (IRM) of a Middle Eastern reservoir. The paper outlines the most important characteristics of the AHM workflows for rigorous quantification of model uncertainty, optimization of history matching parameters and execution of large-scale reservoir simulations using Massive Parallel Processing technology. The AHM approach integrates probabilistic Bayesian inference using Ensemble Smoother with Multiple Data Assimilation (ES-MDA), which simultaneously assimilates the data and generates maximum a-posteriori updates of reservoir model parameters in a variance-minimizing update scheme. Variability and sensitivity analyses are conducted to identify the most dominant reservoir parameters and a large number of geo-cellular model realizations is generated to rigorously capture the uncertainty ranges. The AHM workflow was applied to a synthetic Dual-Porosity Dual-Permeability (DPDP) oil reservoir model with approximately (~) 34 million grid-cells. The simulation model span ~50 years of production with flank water injection. The optimization objective was to minimize the joint misfit of watercut, oil-rate and static well pressure in ~50 producing wells and improve well-level history match. An enhancement of AHM workflow is proposed to improve the simulation model connectivity as well as the accuracy of the history match by implementing the streamline-based approach to update fracture network through drainage volume analysis of injector-producer pairs. While the computational performance of the used ES-MDA algorithm was found very robust and fairly independent of the geological and engineering complexity of studied simulation cases, the overall complexity of IRMs can raise memory-allocation, computation and information technology (IT) communication challenges. The paper discusses these challenges and proposes measures to alleviate them for successful deployment of AHM workflows to large-scale models.