|
Abstract:
|
A dispatching policy can be defined as a set of condition-action (CA) rules in which changing job-shop circumstances correspond to the condition part and the dispatching rules correspond to the action part. Since advanced manufacturing technologies can enable job shops to practice dispatching policies as efficient as dispatching rules, this paper introduces an intelligent scheduling system that can learn dispatching policies depending on the queue lengths of machines by using Pitts approach of genetics-based machine learning (GBML) for dynamic job shops. In our proposed intelligent scheduling system, the Pitts approach of GBML performs a matching between all possible queue lengths of each machine and dispatching policies according to the expected efficiencies. The four objectives, all related to minimization of tardiness, are considered: total tardiness, average tardiness, maximum tardiness, number of tardy jobs. The efficiencies of the dispatching policies are computed by applying the corresponding condition-action (CA) rule-set to the job-shop in a simulation environment. Nine sets of problems were used to compare solutions of our proposed scheduling system with best known nine dispatching rules. The experiments show that dispatching policies learned by the proposed scheduling system outperform the dispatching rules including SPT, EDD, MDD, COVERT, and CR significantly.
|
