A method for assessing the performance of genetic algorithm optimization for water distribution design

Abstract

The paper proposes a new methodology for assessing the effectiveness of GA parameters in consistently finding similar low cost solutions over a broad range of different starting random number seeds. The method involves testing the parameters of probabilities of crossover (for various crossover types) and probabilities of bitwise mutation with 1000 different random number seeds. In addition, this methodology allows different varieties of GA to be compared so that the methodology with the best performance can be determined. The proposed methodology has been verified for its effectiveness on the previously published benchmark network New York Tunnels Problem.