skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Combining Models and Data for Predicting, Parameter Estimation, and UQ

ORCiD logo [1]
  1. Los Alamos National Laboratory
Publication Date:
Research Org.:
Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
Sponsoring Org.:
USDOE Office of Science (SC)
OSTI Identifier:
Report Number(s):
DOE Contract Number:
Resource Type:
Resource Relation:
Conference: Workshop of Astrophysical Population Emulation and Uncertainty Quantification ; 2017-04-03 - 2017-04-07 ; Research Triangle Park, North Carolina, United States
Country of Publication:
United States
Atomic and Nuclear Physics; Mathematics

Citation Formats

Lawrence, Earl Christopher. Combining Models and Data for Predicting, Parameter Estimation, and UQ. United States: N. p., 2017. Web.
Lawrence, Earl Christopher. Combining Models and Data for Predicting, Parameter Estimation, and UQ. United States.
Lawrence, Earl Christopher. Mon . "Combining Models and Data for Predicting, Parameter Estimation, and UQ". United States. doi:.
title = {Combining Models and Data for Predicting, Parameter Estimation, and UQ},
author = {Lawrence, Earl Christopher},
abstractNote = {},
doi = {},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Mon Apr 03 00:00:00 EDT 2017},
month = {Mon Apr 03 00:00:00 EDT 2017}

Other availability
Please see Document Availability for additional information on obtaining the full-text document. Library patrons may search WorldCat to identify libraries that hold this conference proceeding.

Save / Share:
  • Stochastic modeling techniques are being used increasingly to describe the geometries of fluvial channel reservoirs as a basis for field development decisions such as optimal well configuration. One of the main problems associated with the application of such tools is the inference of the necessary input parameters to describe the models. It is difficult to estimate the lateral continuity and stacking pattern of the sandstone reservoirs. This paper will focus on the application of outcrop data sets for inferring realistic input parameters. Input to two different stochastic models is described: one simple model (a binary truncated Gaussian field), and onemore » more complex model (a hierarchical marked-point model with spatial interaction functions). Parameters have been estimate from outcrops in Spain and the United States and the models have been tested on data from a North Sea oil field. Indicator variograms estimated from outcrops can be used to define the spatial correlation structure of a truncated Gaussian field and control the continuity of sandstones simulated using such a model. Although indicator variograms are only two-point statistics, they provide a reasonable description of sandstone geometries and correlation lengths reflect bed thicknesses and widths. The hierarchical marked-point model comprises a population of [open quotes]points[close quotes] (lines in 3-D models), which describe the positions of channel belts. The spatial arrangement of the lines is partly controlled by interaction functions, which are used to define the (non-Poisson) distribution of the channel belts. Each channel belt is associated with a series of [open quotes]marks,[close quotes] which define the channel dimensions, number of channels, and channel stacking. Parameter estimation in this model is more complex, but typical values of parameters such as channel dimension, channel stacking patterns, and channel-belt stacking patterns have been estimated from outcrop data.« less
  • The single-well chemical tracer method has been used in over 140 tests since 1968 to determine residual oil saturation after waterflooding. Interpretation of the test involves matching tracer data from the field using computer simulation programs. As the complexity of the simulator model increases, the number of trial runs that are needed to satisfactorily fit field data increases rapidly. The conventional fitting procedure can thus become very cumbersome and can involve prohibitive computer costs. A new method has been devised to model tests in heterogeneous formations. These rigorous simulators have been replaced by simple response functions capable of being usedmore » in a standard parameter estimation routine. Matching the tracer data now involves function evaluations rather than full simulator runs, thus resulting in a large reduction in computing time. An algorithm has been developed to find best fit values for all unknown test variables, such as the flow distribution into and out of the layers, the dispersive mixing lengths and the residual oil saturations in the different layers. The method bas been verified by using it to explain both simulated and actual field data from tracer tests conducted in multi-layer reservoirs.« less
  • Development of optimal control strategies for complex heating, ventilating, and air-conditioning (HVAC) systems often requires a model of the system. Models of conventional heating or cooling plant equipment are nonlinear with multiple input and output variables. Model parameter values (e.g., an overall heat transfer coefficient, UA, in a heat exchanger model) need to be assigned so that the model adequately predicts the performance of the actual system. Parameters for a model of an operating system can be determined with regression techniques using measurements of model variables. The work presented in this paper considers the multiresponse nonlinear parameter estimation problem associatedmore » with a system of interconnected components in a chilled-water plant. Methods studied for solving the estimation problem include ordinary least squares, weighted least squares, and a determinant criterion derived from Bayesian estimation theory. Potential pitfalls in multivariate regression, such as linear dependencies among responses, are identified. The application and analysis of the parameter estimation methods are directed toward building a predictive model for use in optimal supervisory control strategies. Various solution methods are reviewed and applied to a simulated chilled-water plant for comparison and analysis. The parameter estimation techniques are then used to create a predictive model of an operational chilled-water plant. The plant has both electric and steam-driven chillers and cooling towers with multispeed fans. The goal of the predictive model is to predict variables associated with operational costs such as electric motor power and steam consumption. The relative merits of the different regression techniques are judged (compared) according to how well the model can predict these cost-related variables.« less
  • Abstract not provided.
  • Abstract not provided.