Solving The Green Flexible Job Shop Scheduling Problem Based On Deep Reinforcement Learnin

Green manufacturing is an inevitable requirement for China to cope with climate change and transition to high quality development.As one of the most widely distributed types of workshop in the current manufacturing industry,the research on green flexible job-shop scheduling has become a hot spot in the field of combinatorial optimization.Green scheduling problem is a more complex NP-hard problem.At present,the solution of this problem is mainly composed of meta-heuristic algorithm and mixed algorithm after adding other algorithms.Such method can obtain a better solution within a certain time,but it lacks stability,generalization and timeliness of response scheduling system,which leads to more unnecessary time and labor cost.Deep reinforcement learning algorithm has powerful decision-making ability,and once a mature network model is trained,it can solve homogeneous problems quickly and solve green scheduling problems with lower computational cost and higher computational efficiency,which to some extent makes up for the shortcomings of existing algorithms.Therefore,this paper adopts deep reinforcement learning algorithm to solve green flexible job-shop scheduling problems.The main work of this paper is as follows:Firstly,the characteristics of energy consumption in the running process of the machine in the shop were analyzed,and the energy consumption of the machine on/off,processing and idling were considered as the green indicators.At the same time,the workpiece completion time was taken as the economic indicators,and the mathematical modeling of the green flexible job shop scheduling problem was carried out.In order to obtain a satisfactory solution that takes into account both the completion time and the total energy consumption,the paper transforms the multi-objective into a single objective by means of weighted normalization,and assigns different weights to the two objectives to obtain the corresponding green scheduling arrangement.Secondly,DQN algorithm is used to solve the green flexible job shop scheduling problem.According to the characteristics of the model,the green scheduling problem is transformed into a Markov decision process.Seven universal state features are defined for it to indicate the production state of the scheduling problem.Sixteen scheduling rules,one heuristic scheduling rule and two compound scheduling rules are set as action candidate sets to provide the basis for the selection of the workpiece and machine in the scheduling node.A reasonable reward function is defined according to the optimization objective,so that the agent can obtain the satisfactory green scheduling result according to the maximum reward value.Finally,the effectiveness of the problem transformation design is proved by example test and application.Finally,the DQN algorithm is improved and trained to obtain a universal network,and the data set is used to verify the solving performance of the network,so as to verify the generalization of the proposed algorithm.On this basis,the network models under three different production modes are retrained,and the satisfactory solutions under different production preferences can be obtained in a very short time by directly using the network model to solve the green flexible job shop scheduling problem,which makes up for the shortcomings of the traditional solution methods.