Stochastic Engine: Direct Incorporation of Measurements Into Predictive Simulations
We are creating a new method of combining disparate types of geologic observations and process simulations. Using Bayesian inferencing and an efficient search algorithm, we obtain a consolidated body of knowledge in the form of multiple configurations and parameter values of the system that are consistent with our existing data and process models. In so doing, we effectively estimate the distributions of both individual parameters and system-wide states, and their likelihood of occurrence. This is in contrast with conventional inversion methods, which produce a single deterministic understanding lacking quantitative information about the distribution of uncertainty. We call this combination of probabilistic evaluation and deterministic process simulators the stochastic engine. Our approach allows the investigators to rapidly improve their understanding of system progress, making it particularly valuable for active processes like injection. The Bayesian inferencing is driven by forward process models that predict data values, such as temperature or electrical voltage, for direct comparison to measured field values. We stage the stochastic searches of possible configurations and run the simplest models, such as lithology estimators, at the lower stages. The majority of possible configurations are eliminated from further consideration by the higher stages' more complex models, such as electrical resistance models for geophysical imaging, or flow and transport models for fluid movement. The approach allows for the continuous augmentation of existing data with newly available information to enhance our understanding and reduce the number of high likelihood configurations. This effectively creates a tool capable of dynamically finding models of underground geological systems that are consistent with all available data. The stochastic engine approach will dramatically increase our understanding of large-scale complex systems and the accuracy of predicting their future behavior under natural or man-made conditions.
- Research Organization:
- Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)
- Sponsoring Organization:
- US Department of Energy (US)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 15004639
- Report Number(s):
- UCRL-JC-145116; TRN: US200320%%263
- Resource Relation:
- Conference: International Ground Water Symposium, Berkeley, CA (US), 03/25/2002--03/28/2002; Other Information: PDF-FILE: 17 ; SIZE: 1 MBYTES; PBD: 2 Aug 2001
- Country of Publication:
- United States
- Language:
- English
Similar Records
Stochastic Engine Final Report: Applying Markov Chain Monte Carlo Methods with Importance Sampling to Large-Scale Data-Driven Simulation
Source Inversion for contaminant plume dispersion in urban environments using building-resolving simulations