Workflow to Enable Effective Uncertaimty Propagation and Decreasing Bias on Predictive Models Used For Field Development Decisions

Open Access
Author:
Moreno, Juan Carlos
Graduate Program:
Petroleum and Mineral Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
September 07, 2016
Committee Members:
  • Turgay Ertekin, Dissertation Advisor
  • Turgay Ertekin, Committee Chair
  • Eugene C Morgan, Committee Member
  • Zuleima T Karpyn, Committee Member
  • Elizabeth Ann Hajek, Outside Member
Keywords:
  • Uncertainty
  • field development
  • predictive models
  • risk on field development
  • history match
  • dynamic conditioning
  • ensemble
Abstract:
The aim of this research is to enhance the overall reliability of subsurface models used for production forecasting with the purpose of reserves classification and field development decisions in brown fields. Given the inaccurate and sparse nature of subsurface measurements used to build subsurface models, it is deemed essential to address quantification of the impact of uncertainty on production forecasting and assessment of remaining hydrocarbon potential. To achieve these objectives, traditional workflows for deterministic and stochastic subsurface modeling with significant production history were reviewed and fundamental key steps believed to inhibit the ability to propagate uncertainty from geological modeling to production forecasting were addressed. In general three main aspects of industry-wide established workflows were revised - namely, model upscaling, history matching, and probabilistic forecasting. In the context of this research, upscaling referred to the process of homogenizing reservoir properties inferred from well logs (approximately 1ft vertically) to accommodate a model grid allowing to perform numerical simulation. There is another level of homogenization related to observations and measurements done on cores and how they relate to log properties. Such upscaling was not addressed in this research mainly due to the –more often than not – condition of not having enough and representative core coverage. In the past, upscaling has been governed by decision timelines, computer power limitations, and overall dynamic model runtimes. To address the upscaling process, recent advances in parallel computing, load balancing, and scalable linear solvers were successfully implemented avoiding the need for model coarsening. The second aspect refers to the calibration of subsurface models with production information. The use of diagnostic plots as objective function was implemented as a way to address non-uniqueness. Diagnostic plots have proved to be a valuable tool to identify production behavior associated with particular geological characters. It also allows segregating some of the non-reservoir observed behavior. Diagnostic plots were used as a guide to pre-condition static models prior to numerical simulation, and as an objective function to optimize and reduce the difference between simulated and observed behavior stochastically. Having achieved a geologically consistent dynamic calibration stochastically, probabilistic production forecasting and identification of bypassed hydrocarbon zones were performed with the aim of stochastic reserves classification and mapping of bypassed zones making the selection of production enhancement initiatives and infill prospects more effective. The workflow was tested using a conceptual model of known properties where it was possible to track the propagation of uncertainty in terms of property error and the Root Mean Square Error (RMSE) between observed data and simulated results. As a result, the workflow provided a seamless platform to integrate geological and dynamic uncertainty and allows one to perform dynamic conditioning and history match using parameters governing propagation of geological properties, thereby reducing the need for artificial modifiers. Results showed a consistent reduction of error in terms of properties after dynamic conditioning and history match, and also a significant reduction in terms of RMSE on the ensemble of models. Having a robust ensemble within reasonable ranges of RMSE, the set of models was used for reserves assessment and classification showing that the true model reserves estimate fell well within 1P and 3P estimates and very close to 2P estimates as calculated probabilistically from the ensemble. In addition, when inspecting remaining prospective areas using remaining mobile oil maps, there was a good agreement of potential areas between the true model areas, and the average of the potential zones from all realizations on the ensemble.