We present a methodology, called Constraint Partition and Coordinated Reaction (CP&CR), for collective, evolutionary problem solving. Problem solving is viewed as an emergent functionality from the evolving process of a group of diverse, interacting, and well-coordinated reactive agents. Cheap and effective search knowledge is extracted from local interactions and embedded in the coordination mechanism. Our domain of problem solving is constraint satisfaction problems. We have applied the methodology to job shop scheduling, an NP-complete constraint satisfaction problem. Experimental results on a benchmark suite of problems show that CP&CR outperformed three other state-of-the-art direct search scheduling techniques, in both efficiency and number of problems solved. In addition, CP&CR was experimentally tested on problems of larger sizes and showed favorable scaling-up characteristics
Relation:
International Conference on Evolutionary Computation , pp. 575-578