Evaluation of rapid performance reservoir models for quantitative risk assessment

Abstract The National Risk Assessment Partnership (NRAP) is a research organization focused on developing methods and tools for long-term quantitative risk assessment for carbon storage. NRAP's approach is to divide the carbon storage system into components—reservoir, wells, seals, groundwater, atmosphere—and to develop reduced order models for each of these components. These rapid performance models are trained and/or validated against full physics reservoir models (e.g., TOUGH2, GEM) so that they reproduce similar results but in a fraction of the time of the reservoir model. The different component models can then be combined in an integrated assessment model that can simulate the full system in a matter of seconds or minutes rather than the days, weeks, or longer that a full physics simulation of the entire system would take. The integrated model can then be run in a Monte Carlo mode to assess the probability of failure of a carbon storage system. In NRAP, part of the focus is on long-term leakage risk, and the rapid performance reservoir models are designed to generate pressures and saturations within the reservoir, and particularly at the reservoir-seal interface, both during injection and for up to 1,000 years post injection. These pressures and saturations can then be used as inputs to wellbore or seal leakage models to predict rates and volumes of leakage of CO 2 and/or in situ fluids. In the past few years, NRAP researchers have developed and applied a number of different reduced order models to saline-, gas-, and oil- bearing storage fields. These models vary significantly in several respects. They range from lookup tables or response surfaces to models such as polynomial chaos expansion to models that rely on data mining and artificial intelligence techniques. Each of these types of rapid performance models has different strengths and weaknesses, depending on the method used, the reservoir type, and the goals. In all cases, the rapid performance models required a geologic model and at least some traditional reservoir simulation runs for training or validation purposes. Also, in all techniques developed, an initial analysis is performed to reduce the number of input parameters and scenarios needed for the final simulations. The number of reservoir simulation runs needed can vary significantly based on the reduced order model used. The time and sophistication that it takes to develop a reduced order model is another major factor that varies among the different types of models. Some models are easily able to handle a significant number of varying spatial inputs, while others are limited in the number of input parameters available. Additionally, while all of the rapid performance models will run much faster than a reservoir model, there run times can vary from fractions of a second to tens of seconds or longer, depending on the situation being modeled. This paper will describe the different types of reduced order reservoir models used within NRAP. It will also provide a critical assessment of these rapid performance models, discuss under what circumstances different rapid performance models would be most effective, and evaluate their utility in the context of quantitative risk assessment.