Terrain-Based Genetic Algorithm (TBGA): Modeling Parameter Space as Terrain

The Terrain-Based Genetic Algorithm (TBGA) is a self-tuning version of the traditional Cellular Genetic Algorithm (CGA). In a TBGA, various combinations of parameter values appear in different physical locations of the population, forming a sort of terrain in which individual solutions evolve. We compare the performance of the TBGA against that of the CGA on a known suite of problems. Our results indicate that the TBGA performs better than the CGA on the test suite, with less parameter tuning, when the CGA is set to parameter values thought in prior studies to be good. While we had hoped that good solutions would cluster around the best parameter settings, this was not observed. However, we were able to use the TBGA to automatically determine better parameter settings for the CGA. The resulting CGA produced even better results than were achieved by the TBGA which found those parameter settings.