Developing a self-learning adaptive genetic algorithm

Introduces an approach to developing an adaptive real code genetic algorithm (ARGA). In developing the algorithm, we first use the ordinal optimisation concept to soften the goals, and then quick factorial design experiments are run to identify "important" and "sensitive" parameters. These "important" and "sensitive" parameters are dynamically changed during the search process by efficient computing budget allocation. At the end of the search process, not only the optimum of the original problem is found, but also the adaptive changing pattern of the GA parameters is captured. This algorithm was successfully used to solve some benchmark problems. The results show that ARGA outperforms simple GAs and other adaptive GAs. Moreover, ARGA is able to find the optimum for some difficult problems while the simple GAs with best parameter combination can only reach the local optimum.