Design And Verification Of Intelligent Scheduling System For Computation Resource In Industrial Internet Of Things

In recent years,industrial Internet of things(IIo T)has emerged with the rapid evolution of network and information technology,and became the research hotspot among academia and industry.In the IIo T system,the communication network composed by sensors,communication nodes,controllers and other intelligent devices can achieve efficient and convenient data interaction between human and machines or between machines and machines,which provides the important infrastructure and technical support for industrial big data analysis and intelligent production.However,in the current IIo T system,industrial devices generally have the problem of low computing energy efficiency,and the collected industrial data also has high security risks in the process of data transmission and processing.Besides,due to the huge volume and scale of IIo T devices group,the lack of reasonable resource allocation leads to the excessive waste of the system computation resource,which is also a prominent problem in the current IIo T system.In view of the above problems,we firstly analyze the characteristics of IIo T architecture,and according to the hybrid computing model theory of cloud computing and mobile edge computing(MEC),propose a data computation offloading system architecture for the IIo T,then achieve the optimal offloading strategy based on reinforcement learning.Further,by researching the principles and characteristics of blockchain technology,we propose an IIo T system architecture that integrating MEC and blockchain,then iteratively solve the optimal computation resource allocation strategy based on deep reinforcement learning(DRL).Finally,based on the proposed IIo T system architecture and resource allocation optimization method,we design and build an intelligent dynamic scheduling platform of computation resource for industrial Smart Park.In the platform,the functions of the DRL model training for the simulated park scene,the DRL model execution for the practical park environment,as well as the real-time interface display of computation resource allocation actions,periodic optimization effects and optimization summary,are preliminarily implemented and verified.The specific work is as follows:(1)Research on computation offloading strategy of the IIo T based on reinforcement learningAiming at the problem of data computation offloading strategy of devices in the IIo T system,and according to the technical characteristics of MEC and cloud computing as well as the architecture characteristics of the IIo T,we propose a data computation offloading system architecture for the IIo T,then achieve the optimal computation offloading strategy based on reinforcement learning,to effectively reduce the overall consumption caused by computation offloading in the system.(2)Intelligent scheduling strategy design of computation resource in the IIo T system that integrating MEC and blockchainBased on the system architecture of data computation offloading for the IIo T,we further focus on the computing energy efficiency and data security of the devices in the system.And then relying on the technical advantages of MEC and blockchain,we propose an IIo T system architecture integrating MEC and blockchain.In the above framework,we iteratively solve the optimal strategy of computation resource scheduling in the system,to achieve the effective optimization of the devices energy consumption and system computation overhead.(3)Design and verification of intelligent dynamic scheduling platform of computation resource for industrial Smart ParkBased on the proposed IIo T system architecture and resource scheduling optimization method,we design and build the intelligent dynamic scheduling platform of computation resource for industrial Smart Park consisted of training module and execution module.By simulating the scene of park controller system,the training module trains the DRL model for intelligent scheduling of computation resource,which effectively saves the model training time.And the execution module applies the trained DRL model to the practical park controller system.The computation resource scheduling strategy optimized by DRL model can effectively reduce the devices energy consumption and computation overhead in the system.Meanwhile,the computation resource scheduling actions,periodic optimization effects and optimization summary can be displayed on the interface in real time.The effectiveness and feasibility of the proposed architecture and optimization algorithm are verified by the platform.