| Intelligent factory is the main carrier of intelligent manufacturing,and intelligent production scheduling is the core of intelligent factory operation.In the environment of intelligent factory,because there are many uncertain and interfering factors,the production scheduling system is required to be dynamic,real-time and self-adaptive.At the same time,smart factory production scheduling has the characteristics of big data driven,distributed cooperation,so it is difficult for traditional production scheduling methods to meet the requirements of smart factory production scheduling.Aiming at the intelligent production scheduling requirements of intelligent factories,this paper adopts the method of Multi-Agent and reinforcement learning integration to study the production scheduling problem of intelligent factories,and develops its prototype system.The specific research contents are as follows:(1)Design of double-layer scheduling structure of intelligent workshop based on MAS.Firstly,the architecture and scheduling process of smart factories are analyzed,and the characteristics of smart factory production scheduling are distributed and agile.Based on this combination of multi-agent technology,a hierarchical distributed workshop scheduling structure for intelligent factories is established,in which the functions and structures of agents are discussed in detail.(2)Research on distributed operation workshop scheduling for smart factories.Under the duallayer multi-agent scheduling architecture,the distributed workshop scheduling problem of the intelligent factory is analyzed firstly,and the double-layer scheduling mechanism based on multi-agent is secondly,the upper layer proposes a dynamic scheduling method integrating Q reinforcement learning and contract network to achieve reasonable allocation of tasks to the processing workshop,and the lower layer adopts heuristic scheduling rules to realize the local scheduling decision of the workshop.Finally,a study is constructed to verify the effectiveness of the scheduling mechanism under the two-layer architecture.(3)Research on adaptive scheduling of flexible operation workshop based on multi-agent.A distributed scheduling architecture based on multiple agents is proposed to solve the problem of dynamic flexible job shop scheduling,focusing on achieving two practical goals of minimizing late penalties and reducing the total load of machines.Firstly,the scheduling problem is described by mathematical modeling,and then the self-organization negotiation mechanism of multi-agent is described,in which the workpiece agent and the device agent call their own scheduling decision module to complete task allocation and buffer job ordering.The near-end policy optimization(PPO)algorithm is used to train the scheduling decision module to realize the self-learning process of the agent.Finally,the effectiveness of the proposed scheduling method is verified by scheduling example.(4)Workshop scheduling prototype system design and development.The Java-based JADE framework develops the scheduling system,in which the corresponding functional modules and related system software architecture are designed,and the related database system is also designed.Finally,the core functions of the scheduling system are introduced.The research results show that under the proposed two-layer distributed scheduling mechanism,the use of integrated reinforcement learning and multi-agent technology can solve the scheduling problem of smart factories,which is conducive to improving the rapid response and adaptive ability of enterprises to deal with dynamic factors. |
