Date of Award

1-1-1995

Degree Type

Thesis

Degree Name

Master of Science

Faculty

Faculty of Science and Technology

First Advisor

Associate Professor James Cross

Abstract

Simulated annealing (Kirkpatrick et al, 1983) is used to estimate the parameters of a mathematical model that predicts the water yield from a catchment. The calibration problem involves finding the global minimum of a multivariate function that has many extraneous local minima, a situation in which conventional optimisation methods are ineffective. The objective function which quantifies discrepancies between the computed and observed streamflows must be carefully selected to satisfy the least square assumptions. Several published simulated annealing algorithms have been implemented, tested and evaluated using standard test functions. Appropriate cooling schedules are found for each algorithm and test function investigated. The number of function evaluations required to find the minimum is compared to published results for the test functions using either simulated annealing and other global optimisation methods. A new simulated annealing algorithm based on the Hooke and Jeeves (1961) pattern search method is developed and compared with existing algorithms from the literature.

Share

 
COinS