Abstract:
This thesis describes a computer program for optimizing the layout geometry of looped water pipe networks. The program uses production rules that it learns through autonomous experimentation with sample problems. The program consists of three components, an agitator, a rule formulator and a production system. The agitator and rule formulator enable the program to derive new rules thereby extending the ability of the production system to improve solutions. The thesis also describes an evolution based method for solving the same problem. Comparison of the two methods demonstrates the advantages of the learning program's ability to utilize previous experience.
