DST Matching and Interpretation through Numerical Well Testing on the Johan Sverdrup Field

Abstract

AbstractThe Johan Sverdrup oil field was discovered in 2010 and is one of the largest in the North Sea having exceptional reservoir quality with matrix permeabilities of more than 50 Darcy. However, even with more than 30 appraisal wells, 8 drillstem tests (DST) and extensive core sampling, some uncertainties remain large. Conventional pressure transient analysis (PTA) indicates ambiguity in for example kh and barriers, so in order to support the selection of development concepts and optimize the drainage strategy, there was an opportunity to reduce this subsurface uncertainty by combining several DSTs into one single reservoir simulation model. This would take advantage of overlapping radii of investigation (ROI) from the large drainage areas of each well.A fully integrated computer assisted history matching (CAHM) approach was used to consolidate DSTs into one single simulation model by varying both static and dynamic parameters simultaneously. Static parameters such as structure and sand thickness were varied together with permeability, fault transmissibility and relative permeability. Simulation grids were optimized before implementation in the full field model to better understand and minimize numerical artifacts. Characteristics of the derivative were transformed to numerical values through an objective function and calculation of derivatives was automated for inclusion in the integrated workflow. For each well test, pressure build-up derivatives were divided into early, mid and late time periods and used as partial objective elements in the global objective function (J). Absolute values, shape and time until an event/feature were matched and/or monitored for derivative diagnostics. Finally, several alternatives were investigated to how the DSTs could condition the final integrated static and dynamic uncertainty assessment for probabilistic forecasting.The identification of uncertainties was a multidisciplinary approach, as was the definition of ranges for adjusting the uncertainties. This team approach ensured consistent parameter changes and realistic models.Several thousand models were run to better understand and calibrate the reservoir models to DST data. Large amounts of output from reservoir simulations were analyzed and acceptance ranges for each well was defined. Considerable insight was achieved in how each uncertainty impacts different periods of the pressure derivatives, and how the match in one well correlated with another well. As an example, ranges of uncertainties and range of alternative models suggested by analytical PTA were discarded when tested in the full field model using DST data combined into a single model. Results were integrated into an uncertainty assessment workflow to condition all reservoir models to DST data. Acceptable models were used to calculate probabilistic production profiles (e.g. Mean, P10-P50-P90), and a few discrete models were selected for further analysis and studies (e.g. Low-Mid-High). Based on this study, the uncertainty in initial oil in place volume (STOIIP) was reduced, the placement and number of development wells was improved and thus the drainage strategy was optimized. The resulting Mean production profiles were used together with discrete models and sensitivities for field development planning, resource management and investment decisions. The workflow was robust and repeatability is high as the work can be redone when new data is available, and also deployed in other projects, with little effort. It may be implemented on simple desktop computers, but works best on larger computer clusters with sufficient calculation power.This paper describes the fully integrated uncertainty assessment workflow for probabilistic forecasting, and focuses on automating pressure transient derivative diagnostics and presenting some alternatives of using, and integrating, numerical well testing with uncertainty assessment, history matching and probabilistic forecasting.