Date of Award

Fall 1993

Document Type

Thesis - Restricted

Degree Name

Master of Science (MS)


Electrical and Computer Engineering

First Advisor

Feng, Xin

Second Advisor

Belfore, Lee A.

Third Advisor

Hock, Jeffrey L.


This thesis presents a new approach for analyzing the solution performance of Genetic Algorithms (GAs). An adaptive filtering algorithm is combined with a predicting algorithm and memory data from previous GA iterations to estimate the GA's "optimal" solution. By normalizing all past iteration points with this optimal prediction, a fuzzy performance measure of the GA's current iteration value is obtained. If the current GA iteration value is above a certain user-defined acceptance level, iteration is stopped and the GA calculates a reliability estimation of the found solution. In summary, the GA analyzes and stops itself when a user-approved level of solution is achieved, which is quite different from the way GAs are conventionally implemented. This new method provides a unique stop criterion for the GA that is based on the dynamic nature of the GA and its past and current performance. Results indicate this new approach is preferable to the traditional GA iteration approach, in terms of cost/performance and in shortening the amount of time the GA searches for acceptable solutions.



Restricted Access Item

Having trouble?