The Research Of Scheduling Rules Decision-Making Based On Production Data Of Dynamic Job Shop

Nowadays,the study of data-driven production management methods has become an important research point in the trend of intelligent manufacturing.There are often abnormal production disturbances in the production process of the general job shop of the manufacturing enterprise,making the production scheduling scheme need to be adjusted dynamically and timely.It is the dynamic job shop scheduling problem.In solving the dynamic job shop scheduling problem,the heuristic priority scheduling rule method has the advantages of rapid scheduling and flexible implementation compared with search optimization algorithms,integer programming and etc.Therefore,the scheduling rule-based scheduling method has been widely used in dynamic job shop scenarios of real enterprises to respond timely to scheduling requirements.However,due to the different types of scheduling rules with different applicable conditions,how to select scheduling rules involves the decision-making of scheduling rules.At present,in the actual dynamic job shop scheduling the decision process of the selection of the scheduling rules mainly depends on the artificial experience,which makes the decision-making for selecting right scheduling rules to be biased,and it is difficult for managers with artificial experience to determine the scheduling rule that meets the current production conditions best in real time.Therefore,this thesis combines the data-driven decision-making idea to study the production data-driven scheduling rule decision-making method for dynamic job shop scheduling.Firstly,on the basis of analyzing the general scheduling optimization mechanism of production data-driven dynamic scheduling for the dynamic job shop,the production system attribute data is determined as the input data of the scheduling rule decision-making model.The corresponding scheduling rule obtained from the decision-making model is the output target of the model.The input data and output target constitute a scheduling sample.The scheduling rule decision-making model is built by the machine learning method.In order to obtain the optimized scheduling sample data for training the decision-making model,the optimized scheduling sample generation method based on Multi-pass simulation mechanism is studied,and the production scheduling simulation platform for scheduling problem instances simulation optimization is built.Then,a scheduling feature selection method based on the improved binary firefly algorithm(EDSBFA)is proposed for scheduling feature selection from production system attributes of the scheduling samples that contain the redundancy or noise attributes affecting the accuracy of the decision-making model.The feature selection optimization ability of EDSBFA was tested on UCI data set and scheduling sample data set.Afterwards,a wrapped scheduling rule decision-making model construction method based on EDSBFA for scheduling feature selection and based on extreme learning machine(ELM)for scheduling sample learning is proposed.The experimental tests show that the performance of EDSBFA-ELM-based scheduling rule decision-making method outperforms the traditional data-driven scheduling rule decision-making methods proposed in literatures.In addition,two dynamic scheduling mechanisms with human-machine coordination are proposed for applying the scheduling rule decision-making models to dynamic job shop scheduling task.Finally,the feasibility and effectiveness of the proposed scheduling rule decision-making method and dynamic scheduling mechanisms are verified by a self-made job shop scheduling problem instance with the co-simulation approach,and a design idea of the application system of the scheduling rule decision-making method is given.