Research On Flexible Job Shop Scheduling Of Power Plant Valves

Scheduling plays a crucial role in improving the production efficiency of manufacturing industry.The appropriate and reasonable scheduling method is used to schedule the manufacturing system,so that the enterprise has the advantages of shortening the construction period,reducing inventory,energy saving and emission reduction.In this paper,the production process of valve workshop is taken as the research object.The static scheduling of the target workshop is carried out to obtain the pre-scheduling results of the workshop,and the dynamic events that may be encountered in the actual production are scheduled in real time.The main research contents are as follows:Firstly,from the current overview of the valve workshop,the process flow of the workpiece and the scheduling process,it is concluded that the management efficiency,information transmission efficiency and robustness to dynamic events of the valve workshop need to be improved.It is concluded that the valve workshop needs an intelligent and real-time scheduling system.The valve shop scheduling problem is abstracted into a flexible job shop scheduling problem,and the mathematical model of the valve workshop scheduling problem is built.Secondly,analysis of the static scheduling problem of the valve workshop and establishment of a multi-objective static scheduling model for the valve workshop.Proposed a Q-learning improved non-dominated sequencing genetic algorithm(NSGAII)and used for the pre-scheduling of workshop.Regarding the selection of hyperparameters for NSGA-II,the standard method is to take fixed values.This approach may cause the disadvantages of insufficient algorithm diversity,easy prematureness,and local convergence.In response to this problem,this chapter proposes a Q-learning algorithm that accomplishes hyperparametric search based on Q-learning rewards such that the crossover probability,variation probability,and population segmentation probability take different values for each cycle.The performance of the Q-learning improved NSGA-II algorithm is verified using the Brandimarte standard arithmetic example.Then,real-time scheduling for dynamic events that may occur in actual production.Analysis of dynamic events in the valve workshop,selection of the unexpected arrival of the artifact as the dynamic event under study,and establishment dynamic scheduling model of the valve workshop.Reinforcement learning of the agent’s state,action,and reward settings is performed by analyzing the real-time state of workpieces and machines in the production environment.This chapter performed the set of the state,action,and reward of the agent.This chapter proposed a distributed approach to train two types of agents,machine selection agents,and workpiece selection agents,using improved DDQN algorithm.Choose from 12 different machining environments to simulate the algorithm.The algorithm reduces the total delay time in a machining environment where workpieces arrive randomly.Finally,develop a production scheduling system for valve workshop.The demand analysis of production scheduling system,scheduling system architecture design and interface design,and design the database form and system module.Built the scheduling system using the MFC development framework in C# and verified the effectiveness of the scheduling system using actual production examples in the valve workshop.