Research On Job Shop Scheduling Problem Based On Machine Learning And Rule-based Scheduling

Job-shop scheduling problem is an important part of the enterprise production management and optimization,and also the foundation and key to realize advanced manufacturing and improve production efficiency.The job shop is a typical NP hard combination optimization problem.In dealing with the dynamic job shop scheduling problem,the rule-based scheduling method is a very common method.However,the dispatching rules applicable to a particular scenario are difficult to meet the scheduling target under the new manufacturing conditions.Therefore,for the scheduling problem of dynamic job shop,this thesis studies the job scheduling method which integrates DQN algorithm and rule-based scheduling method.The aim is to select the scheduling rules that satisfy the multi-objective scheduling target from the set of candidate dispatching rules designed under the new manufacturing conditions,using appropriate machine learning methods.The specific research contents of this thesis are as follows.This thesis summarizes the domestic and foreign research status of rule scheduling methods and machine learning methods in job shop scheduling problems,summarized the advantages and disadvantages of various methods and analyzed the feasibility of machine learning methods in solving job shop scheduling problems.In order to design a scheduling strategy that integrates the DQN algorithm with the rule-based scheduling method,this thesis studies the problem description of the job shop scheduling problem.By analyzing various dynamic events in the dynamic job shop,this thesis established a dynamic job shop mathematics model under the DQN algorithm.According to the requirements of the multiobjective optimization problems,the commonly used evaluation indicators are selected,and the basic method of the multi-objective optimization problem is studied.Taking into account the characteristics of the rule-based scheduling method,this thesis uses a multiple-dispatchingrules strategy(MDR)to sequence jobs,and design a dispatching rules library that meets the requirements of the DQN algorithm.In order to the state description of the job shop problem under the DQN algorithm,this thesis introduces the Markov decision process and studies the theoretical basis of the DQN algorithm.In order to select appropriate dispatching rules in different states of the job shop,this thesis first establishes the state description of the job shop problem under the DQN algorithm and establishes the update strategy of the job shop state under the DQN algorithm;secondly,the dispatching rules commonly used in the job shop are selected,and an action set containing 59 different dispatching rules forms is established;finally,for the multiple scheduling objectives of the job shop,corresponding rewards are established mechanism.Through simulation verification,the selection strategy of scheduling rules based on DQN algorithm proposed in this thesis can optimize and generate corresponding dispatching rule combinations under different manufacturing conditions,and construct a scheduling sequence that meets the job shop scheduling objective.The results show that the method proposed in this thesis has a good effect on optimizing the multi-objective scheduling problem of dynamic job shop.