Dynamic scheduling with fuzzy clustering based Q-learning

For dynamic multi-machine scheduling problem, a multi-agent dynamic scheduling system was proposed. The system was based on an improved contract net mechanism, in which the jobs were invited to bid for the available time of the equipment. In order to ensure that the equipment agents can select the most appropriate bidder according to the current system transient state, an adaptive bid selection strategy was proposed. Considering the state space was too large in dynamic scheduling environment, the strategy reduced the dimension of system state space through the extraction of state feature and fuzzy clustering firstly, and then the Q-learning procedure of equipment agent was implemented based on clustering system states. The simulation results showed that the proposed fuzzy clustering Q-learning based bid selection strategy was superior to single bid selection rule, and can improve the adaptability of the scheduling system in dynamic scheduling environment.