Research On Large-scale And Dynamic Cloud Resources Scheduling Based On Platform Intelligent Coordination

With the development of the Internet,the Internet of Things,and the Deep Learning technologies,the manufacturing industry is rapidly transforming towards digitalization,networking,and intelligence.The scope of enterprise's manufacturing activities has gradually expanded from the workshop environment to a cloud environment.Demander and provider of the manufacturing resources can collaborate efficiently through cloud platforms to complete manufacturing tasks.One of the key issues in the cloud environment is how to reasonably schedule massive manufacturing resources that distributed in different regions to determine the assignment of complex and changeable manufacturing tasks among manufacturing resources for processing.However,the networked manufacturing model supported by the cloud platform has a wider supply-demand time-space scale.It has complex features such as real-time change of manufacturing resource availability,multiple manufacturing tasks that are constrained by the processing precedence,and a large feasible scheduling space,which will enhance the difficulty in solving this NP-Hard scheduling problem.Therefore,in order to coordinate providers and demanders for the processing of manufacturing tasks,this thesis studies the large-scale and dynamic cloud resources scheduling based on platform intelligent coordination.While efficiently scheduling large-scale manufacturing resources,the scheduling approach will ensure the constraints of task processing precedence,single resource capacity limitation,and the time consumption of logistics.In solving the scheduling problem in the mentioned high-dynamic,large-scale,complex and variable manufacturing environment,the research work in this thesis include:(1)Study the service model based on cloud platform coordination for manufacturing.Establish a service model according to the characteristics of registered manufacturing resources and the processing flow of manufacturing tasks.With the help of the proposed service model,the solving technical framework for the scheduling problem in cloud environment is proposed,which aims at minimizing service cost,minimizing manufacturing make-span,and maximizing service quality.The proposed scheduling method is named as ANNRL.(2)Study the intelligent manufacturing resources allocation method at the platform level.According to the procedure of coordinating the demander of manufacturing tasks with the provider of manufacturing resources,a feasible schedule generation scheme that supports parallel processing is formulated.The manufacturing tasks without conflicts in processing precedence are assigned to candidate manufacturing resources for processing under the feasible schedule generation scheme.Through the Artificial Neural Network model,the make-span of any task obtained after being processed on each candidate manufacturing resource is estimated to intelligently allocate the manufacturing resources.In addition,with the forecast of recent task demand,the accuracy of the trained make-span estimation model has been improved.(3)Research on intelligent manufacturing task arrangement method at the resource provider level.The resource provider will put the assigned manufacturing tasks into the pending queue,and then change the priority values in the pending queue according to the scheduling rule that is adaptively suitable for current resource workload status,with the help of the priority rule updating model based on the Reinforcement Learning provided by the platform.(4)Evaluate the proposed intelligent coordination scheduling approach via the simulation environment of cloud resource scheduling.By transforming the international standard case library and the trained generator,the simulation datasets are generated.A scheduling method based on evolutionary algorithms is designed as the reference in the experiments to evaluate the proposed scheduling method that integrates the resource allocation and the task arrangement.Results from the experiments show that: 1)The scheduling method ANNRL has high response performance.The proposed scheduling method is suitable for highly dynamic environment because the average scheduling solution determination time in any task-scale project can be controlled within 40 s,which is only about 4.4% of the referring algorithm,and the average decision time for the scheduling of any task can be controlled within 50ms;2)For any task-scale project,the scheduling results obtained using ANNRL are superior to the referring algorithm in each dimension of the scheduling optimization objectives,which indicates that ANNRL has the potential in the application for any cloud based manufacturing environment.(5)Demonstrate the process of coordinated scheduling for cloud resources based on the cloud platform prototype.Based on the original platform of our research group,a cloud platform prototype that can support the proposed intelligent coordinated scheduling method is designed.This prototype can demonstrate manufacturing activities of coordination between the demander and the provider,and it describes the way to apply the proposed scheduling method ANNRL.The problem of massive resource scheduling in the dynamic environment studied in this thesis is a common problem of many other cloud-based networked manufacturing models.Thus,the proposed intelligent coordinated scheduling method ANNRL can be applied to the more general networked manufacturing environments.