An adaptive alternation model in genetic algorithms considering landscape complexity

We propose a new adaptive algorithm which can maintain the diversity of population appropriately and accordingly can find more than one local optimum for a given function. The algorithm generates many children by a crossover operator. It estimates a local fitness landscape near the parents using the generated children. The estimated local landscape is utilized as a selection procedure. Furthermore, a mating restriction is implemented for effective search. By applying this algorithm to several benchmark problems, we show that it can converge the population on a global optimum for unimodal functions, and can find plural local optima for multimodal functions