A New Concept for Genetic Algorithms Based on Order Statistics
Daniel Delahaye, Stéphane Puechmorel · 2013
The new genetic algorithm works on individuals representing domains of the state space instead of points, coding each individual in the population as a hypercube for which the exploration and exploitation phases are guided by the probability of including the global optimum. The first part of this chapter discusses important notions of order statistics. The second part describes the principle used to evaluate individuals and the crossover and mutation operators associated with this new algorithm. The final part of the chapter compares the new algorithm with classic genetic algorithms and presents the results of the new algorithm for three functions (Griewank, Rosenbrook and Lennard-Jones). The new algorithm presents clear improvements in the performance of the standard genetic algorithm.