Researches On Resources Scheduling Optimization In Cloud Manufacturing Based On Deep Reinforcement Learning

With the development of high technology such as cloud computing,artificial intelligence,and big data,cloud manufacturing has become a popular research direction in academia and industry.Cloud manufacturing is a new service-oriented networked manufacturing paradigm.Manufacturing resources can be provided to users in a low-cost,resource-sharing,highly efficient and coordinated manner.However,because the manufacturing resources of different companies are usually highly heterogeneous,the geographical location of the manufacturing resources may also vary greatly,and it is very challenging to arrange large-scale manufacturing projects in such a paradigm.How to solve the constraint problems in the process of service composition and to design the composition path as optimally as possible is the difficult point in the implementation of the cloud manufacturing paradigm.In view of multi-objective optimization problem in cloud manufacturing,further improvements have been made on the basis of existing researches.First,in the aspect of cloud manufacturing resource scheduling model design,a cloud manufacturing service optimization model based on the logistics topological relationship between tasks is proposed.This model not only considers the multi-objective optimization problems,but also combines the logistics problem with the timing constraint of tasks on the evaluation metrics,and further satisfies the cloud manufacturing realistic needs.Aiming at the proposed cloud manufacturing model,a new genetic algorithm DRL-MGA(Deep Reinforcement Learning based Multi-objective Genetic Algorithm)is designed based on the NSGA-Ⅱ(Nondominated Sorting Genetic Algorithms Ⅱ)and deep reinforcement learning algorithm.The algorithm can dynamically adjust the parameters in the genetic algorithm iteration process through self-learning manner,so as to obtain a better Pareto front and improve the effect of cloud manufacturing resource scheduling.Finally,two sets of experiments are designed to verify the cloud manufacturing service optimization model and DRL-MGA algorithm.First,by introducing the DTLZ series of benchmark problems to compare the algorithm DRL-MGA with the algorithm NSGA-Ⅱ,it is verified that the proposed algorithm has a significant improvement in the IGD(Inverted Generational Distance)metric compared with NSGA-Ⅱ.Secondly,through simulation,the proposed algorithm is applied to the designed model,and then the analysis is carried out by means of diagrams.The experimental results effectively demonstrate the rationality of the proposed model.