A high-speed scheduling method by parallel and asymptotic taboo searches for large-scale job shops problems with group constraints

The target of this research is a job shop problem of about 2000 jobs with group constraints where jobs are grouped and processed. Our research group has proposed a scheduling method whose initial solution is improved by applying several rules concerning evaluation elements by using a taboo search. However, the scheduling method has the following problems. The first is that the evaluation value of the best solution has large variance because of random factors. Another is that the search often falls into a local minimum after rapid increases in one of the evaluation elements, the number of group changes. This research proposes a technique of applying parallel taboo searches by plural computers to one solution, sending the best obtained solutions to the other computers, and repeating the parallel search. To avoid duplication of search areas, a taboo list that prevents repetition of searches is sent along with the best result to the other computers. The proposed method has been applied to an actual large-scale job shop problem. The method can generate solutions that are better than 95% of experts solutions in less than one hour using three 2.4GHz Pentium JV PCs.