HR: 08:45h
AN: H11B-04 [PDF]
TI: A New Function Approximation Optimization Algorithm for Automatic Calibration of Complex Nonlinear
Water Resource Models
AU: * Shoemaker, C A
EM: cas12@cornell.edu
AF: Civil and Environmental Engineering, Cornell University, Ithaca, NY 14853 United States
AU: Regis, R
EM: rregis@orie.cornell.edu
AF: Operations Research, Cornell University, Ithaca, NY 14853
AB:
Site-specific pollution transport models are based on field data and the calibration of those parameter values that cannot be
directly measured. This combination of field data and calibrated model is a powerful tool in environmental analysis.
However, one of the difficulties with complex nonlinear models is that the calibration process can be time consuming for the
modeler and computationally very demanding if it is done thoroughly. Automatic calibration methods employing optimization
methods can reduce both the human and computer time required to do a thorough calibration.
This paper will discuss the use of function approximation optimization for automatic calibration of complex models arising in
environmental and water resource management. We will describe a new algorithm developed by the authors for optimization of
nonlinear, non-convex problems, including those arising in water resources. The procedure is designed for use with "costly"
functions, i.e. those that take a substantial amount of CPU for each evaluation of the water resources simulation model. The
method uses function approximation of the objective function to guide the search procedure and to reduce the number of times
the objective function must be evaluated.
A comparison will be presented of the performance of different algorithms (including the new function approximation
algorithm) on test functions that have multiple local optima. The new function approximation method performs better than
earlier function approximation optimization methods on most of the test problems. This analysis will also be related to two
other papers in this session that apply the new function approximation algorithm to optimal calibration to field-data based
models including a) a transport model of enhanced bioremediation of chlorinated ethenes (Mugunthan et al.) at a DOD site
and b) a watershed model in NY state (Tolson et al.).
DE: 1831 Groundwater quality
DE: 1832 Groundwater transport
DE: 1860 Runoff and streamflow
DE: 3230 Numerical solutions
DE: 3260 Inverse theory
SC: Hydrology [H]
MN: 2003 Fall Meeting