Good solutions will emerge without a global objective function: applying organizational-learning oriented classifier system to printed circuit board design

Abstract

This paper describes a novel evolutionary computational model: organizational-learning oriented classifier system (OCS), and its application to printed circuit boards (PCBs) design problems. The idea of OCS comes from the theory of organizational learning, in organizational sciences. OCS is an extended multiagent version of a conventional learning classifier system to learn adaptive rules in a given environment. OCS adaptively learns good knowledge for problem solving via interaction among the agents without explicit control mechanisms for a global optimization function. To validate the effectiveness of OCS, we have conducted intensive experiments on a real scale PCB design problem for electric appliances. The experimental results have suggested that (1) OCS has found feasible solutions with the same quality of the ones by human experts; (2) the solutions are not only locally optimal, but also globally better than the ones by human experts with regard to the total wiring length; and (3) the solutions are more preferable than the ones from the conventional computer aided design (CAD) systems.