A parallel genetic-neuro scheduler for job-shop scheduling problems

Abstract Despite relentless efforts on developing new approaches, there are still large gaps between schedules generated through various planning systems, and schedules actually used in the shop floor environment. An effective schedule generation is a knowledge intensive activity requiring a comprehensive model of a factory and its environment at all times. There are four main difficulties that need to be addressed. First, job shop scheduling belongs to a class of NP-hard problems. Second, it is a highly constrained problem that changes from shop to shop. Third, scheduling decisions depend upon other decisions which are not isolated from other functions. Thus, it is subject to random events and finally scheduling problems usually tend to embrace multiple schedule objectives to be optimized. These difficulties stimulate the need to develop more robust and effective approaches to scheduling problems. In this paper, genetic algorithms and artificial neural networks to the solution of scheduling problems are discussed. After a brief review of the classical method of schedule generation the basics of evolutionary programming and artificial neural networks are introduced and their possible use in the schedule generation process are examined. The reviews are followed by a real-life example. It demonstrates the combined use of genetic algorithms and neural networks and points out the benefits to be gained through parallel implementation of the genetic neuro-scheduler on a neuro-computer.