改良2-opt法を組み込んだ遺伝的アルゴリズムによるQAPの解探索

抄録

We propose a new 2-opt base method as a local search approach used with Genetic Algorithms (GAs) in Memetic Algorithm. We got a hint from the fast 2-opt method and devised the new 2-opt method. The main different point is such that our method exchanges genes by using histories of contributions to fitness value improvement. The contribution level is represented by the value `Priority’.
In computer experiment, Quadratic Assignment Problem (QAP) instances are solved by GA with the 2-opt method(First Admissible Move Strategy, the Best Admissible Move Strategy), the fast 2-opt, and our proposed method for comparative evaluation. The results showed that our improved method obtained better solutions at ealier generation of the GA and our method required less computation time than the others at some upper bound value of appropriate `Priority’ setting values. Specially, at the average elapsed time of the fast 2-opt method’s 1000th generation, the exact solution findings of ours is more than the others.
In further experiment, we observe that the searching capability depends on the number of levels of `Priority’. The ratio between two different Priority level sets becomes 1.59 in computation time in solving problem instance “char25a". This characteristic is shown to be statistically significant in ten instances among eleven.