Robust start for population-based algorithms solving job-shop scheduling problems

Abstract

Most of the methods to solve job-shop scheduling problem (JSSP) are population-based and one of the strategies to reduce the time to reach the optimal solution is to produce an initial population that firstly has suitable distribution on space solution, secondly some of its points settle nearby to the optimal solution and lastly generate it in the shortest possible time. But since JSSP is one of the most difficult NP-complete problems and its space solution is complex, most of the previous researchers have preferred to utilize random methods or priority rules for producing initial population. In this paper, by mapping each schedule to a unique sequence of jobs on machines matrix (SJM), we have proposed the novel concept of plates, and have redefined and adapted concepts of tail and head path and have designed evaluator functions between SJM matrix and its corresponding schedule aiming at identifying gaps in the obtained schedule, we have proposed three novel initialization procedures. The proposed procedures have been run on 73 benchmark datasets and their results have been compared with some existing initialization procedures and even some approximation algorithms for solving JSSP. Based on this comparison, we have seen the proposed procedures have the significant advantage both in the quality-generated points and in the time producing them. The more interesting point in the implementation of proposed procedures on some datasets is that we see the best known solution in the produced initial population.