Genetic network programming with exception control

As an graph based evolutionary technique, Genetic Network Programming (GNP) has manifested impressive performances on complex optimization problems. However, GNP also exposes several inherent problems, which causes nonlogical exceptions, and is baffling the efficiency of the evolution. This paper thereby proposes a new method named Genetic Network Programming with Exception Control (GNP-EC) in order to resolve this situation. GNP-EC places three modifications on the original GNP, covering its structure and genetic operations. The individuals created by GNP-EC have a much larger chance to make valid decisions than before. In the experimental studies, GNP-EC is compared with the original GNP on the Tileworld testbed, and the result verifies its effectiveness in both training and testing phases.