Research On Dual-System Knowledge Modeling And Scheduling Method Oriented To Flexible Job Shop Scheduling

The processing of aerospace structural parts has the characteristics of multiple varieties,small batches,and complex processes.During the production process,batch production tasks and research and development（R＆D）production tasks coexist.R＆D tasks lead to frequent emergence of urgent orders,which affects the production process and efficiency of batch production tasks.This paper takes flexible job shop scheduling as the research object,and uses a perception-cognitive dual system to combine job proc ess knowledge with scheduling methods.A multi-level knowledge modeling method is proposed,and a dual-loop scheduling method based on deep reinforcement learning is studied.The research content includes three aspects:（1）During the scheduling process of aerospace structural parts processing workshop,it is difficult to organize and manage knowledge modeling and scheduling strategy generation.A dual system method of perception-cognition oriented to flexible job shop scheduling is proposed.The knowledge graph constructed by the perception system provides real-time data.Combine these data with historical data provided by the experience memory module of the cogniti on system.The dual system realizes the fast response of dynamic scheduling.（2）The hierarchical and structural relationships of knowledge in the flexible job shop scheduling process are complex and changeable.The ontology knowledge modeling method of shop scheduling element set based on knowledge graph is proposed.First,a four-dimensional disjunction graph representation model for scheduling is given,and a multi-layer knowledge graph is used to classify scheduling data.Then the ontology modeling method is used to divide the scheduling data into three levels,which are mai nly divided into three levels:equipment resources,process flow and processing tasks.This method establishes the hierarchical relationship of static data fusion and the inter-layer relationship of dynamic data fusion.Finally,the proposed multi-dimensional information matrix mapping rule is used to generate the multi-dimensional information matrix corresponding to the scheduling method to provide parameters.This module realizes the management and integration of scheduling knowledge,and ensures the accuracy and timeliness of scheduling parameters.（3）Aiming at the optimization problem of flexible job shop scheduling strategy,a dual-loop scheduling method is proposed.A dual-loop deep Q network（DL-DQN）learning method based on resource allocation agent and process sequencing agent is proposed,the optimization goals and reward functions are designed.Verified by standard calculation examples,the proposed method has superior solution performance compared with traditional scheduling methods.Finally,take the rocket engine structural parts processing workshop of an aerospace institute as an example.Apply the knowledge modeling and dual-loop scheduling methods mentioned in this paper to workshop scheduling.Cases verify and analyze the effects of knowledg e modeling fusion,transformation and scheduling application.The method has certain practical value for improving the data management and scheduling utilization of the flexible workshop scheduling process.