Document Type

Thesis - University Access Only

Award Date


Degree Name

Master of Science (MS)

Department / School

Civil and Environmental Engineering

First Advisor

Thomas Van Lent


This thesis applies artificial neural network theory and the simulated annealing algorithm to the problem of selecting candidate gages for removal from a rain gage network. A network is defined using a modified Hopfield neural network. The Hopfield architecture is modified by allowing node activations to assume the discrete values of either 1 (gage present) or 0 (gage removed). Hebbian learning is accomplished by decomposing and backsubstituting a set of simultaneous equations that equate the global variance increase of rainfall measurement with the node multiplication of the Hopfield defined rain gage network. System energy is minimized through simulated annealing and computed using the Hopfield energy equation. The configuration with the lowest energy value is selected as the solution. The model is computationally efficient and correctly selects the lowest energy network configuration. Application to large problems can be improved by using a better method for solving overconstrained systems of equations.

Library of Congress Subject Headings

Rain and rainfall -- Measurement -- Mathematical models
Neural networks (Computer science)
Rain gauges




South Dakota State University



Rights Statement

In Copyright