| Smart manufacturing is the main direction of the strategy of "manufacturing power" in China,and it is the key to the high-quality development of manufacturing industry.As one of the core technologies of smart manufacturing,Industrial Internet of Things(IIoT)combining Big Data,Internet of Things(IoT),Artificial Intelligence,Cloud Computing,Edge Computing and other key technologies.has effectively promoted the development process of industrial digitalization and intelligence.With the comprehensive promotion of smart manufacturing,the number of industrial sensor devices is increasing rapidly,and massive industrial data will be collected and processed in the IIoT.These data which cover many aspects such as production,quality,environment,are diverse and random.On the one hand,in view of the heterogeneity and large scale of various industrial data,IIoT platforms are faced with problems such as difficulty in real-time data analysis,task concurrent processing,and system intelligent decision-making.On the other hand,due to the random generation of industrial data,the computing resource requirements of edge servers also change dynamically,and the imbalance between task volume and computing resources will lead to system latency and energy consumption.Aiming at the above problems.this thesis carries out research work on the design and construction of IIoT platform for cloud-edge-device collaboration,and edge-server activation and task offloading strategy in IIoT scenarios,respectively,as follows:(1)Design and construction of IIoT platform based on cloud-edge-device collaboration.Firstly,for the key application scenario of smart manufacturing:the quality detection of air conditioner external appearance,this thesis designs and sets up an IIoT experimental platform based on cloud-edge-device collaboration.The platform is composed of the Internet of things layer,the edge layer and the cloud center layer,which realizes the functions of data perception,image analysis and production line control.Secondly,in order to solve the problem of concurrent accumulation of tasks generated by the platform,this study established a system model of cloud-edge collaboration based on the average processing latency and transmission rate of the device,and proposed a task offloading strategy based on discrete particle swarm optimization algorithm to realize the latency optimization of the "single-cloud and single-edge" system.The advantages and applicability of the proposed model and strategy are proved by comparing the simulation results with the measured data from the platform.Finally,in the face of the limited computing resources of single edge server,the "single-cloud and single-edge" system model was extended to "single-cloud and multi-edges".By comprehensively considering communication and computing resources,the bandwidth allocation and task offloading strategy based on particle swarm optimization algorithm was proposed which significantly reduces the latency of system.(2)Research on edge server activation and task offloading strategies in IIoT scenarios.Considering the multiple task types and the dynamic randomness of task generation in the IIoT,this thesis introduced the "active-dormant" mechanism of the edge server in the"single cloud and multi-edges" system,combined with the latency and energy consumption indicators of the system in a long time,and proposed a system cost optimization method based on deep reinforcement learning and multiple knapsack algorithm.This strategy uses deep neural networks to generate the number of edge server activations and the number of tasks to be offloaded,and uses a multiple backpack algorithm to determine the destination of task offloading while effectively utilizing computational resources,thus completing the periodic adjustment of the number of edge server activations and task offloading policies,significantly reducing the overall system cost. |
