GAとNNを用いたジョブショップ・スケジューリングにおける突然変異率操作の改良法 Improvement of Mutation Manipulation in GA・NN Job-shop Scheduling
In a job-shop scheduling method proposed in a previous paper using a three-layered neural network optimized by a genetic algorithm, an improved method where the mutation probability is manipulated in a simulated annealing-like way is introduced to escape from a local optimum solution. Its effect on the variety of networks is investigated from the viewpoint of the structure of gene arrangement, and the efficiency of the present improvement is made clear. As a result, easy escape from a local optimum of 981 hours is realiged. Moreover, some excellent gene blocks (blocks of connection weights) for superior schedules are discovered.
化学工学論文集 23(5), 726-729, 1997-09-10
The Society of Chemical Engineers, Japan