Parameters selection of fitness scaling in genetic algorithm and its application

Fitness scaling is an important element affecting the evolutionary performance of genetic algorithm. The scaling transformation parameters decide the efficiency. Firstly, two conditions for efficient fitness scaling are proposed. The first condition is that the domination relationship should be kept after the transformation; the second condition is that fitness should be different after transformation. Based on the two conditions, the formulation with roulette wheel selection is given. Secondly, the scopes of parameters of three kind of fitness scaling are deduced. At last, based on the two conditions, the fitness scaling based on logarithm function and triangle function are given. The above study of fitness scaling enriches the theory of genetic algorithm.