Evolving Dispatching Rules with Greater Understandability for Dynamic Job Shop Scheduling

Heuristic dispatching rules are one of the most popular and widely used methods of scheduling in dynamic job shop environments. The manual development of such dispatching rules is time consuming and requires substantial knowledge of the domain. There have been numerous works into using genetic programming (GP) as a framework for the automated generation of dispatching rules for job shop scheduling environments. However most existing works do not take into account the interpretability and semantic validity of the evolved dispatching rules. In this paper we propose the use of a grammar, implemented through strongly typedGP, to restrict the GP programs to only contain semantically meaningful expressions. Experimental results show that the dispatching rules evolved in the semantically constrained search space do not have (on average) performance that is as good as unconstrained. However the interpretability of evolved rules is substantially improved. This is the firstwork usingGP for the automatic discovery of dispatching rules that has explored their interpretability in depth and considered it as an important trait of an effective dispatching rule.