Task Scheduling And Resource Allocation Under Cloud-Edge Collaboration In Intelligent Internet Of Things

The technology of the Internet of Things(Io T)aims to realize the Internet of Everything,so terminal devices,including sensors and intelligent devices,emerge in endlessly.However,due to the low computing power,low storage capacity and limited battery capacity of terminal equipment,it is not realistic to analyze data on it.In such a big data environment of the Io T,the problem of data processing can be solved by introducing cloud computing technology.Through cloud computing as a link,it is expected to realize the landing of Io T terminal to intelligent application.However,cloud platform resources are not inexhaustible,and the massive data transmission between Io T terminal and cloud platform brings pressure to communication link and network environment.In order to balance the contradiction between the limited capabilities of Io T terminals and the efficiency of network resource management,it is of great significance to carry out in-depth research on task scheduling and resource allocation strategy in the Io T.To this end,this thesis intends to carry out research on task scheduling and resource allocation strategy based on the background of Io T AI application,focusing on the efficient task scheduling and resource allocation strategies under cloud-edge collaboration to adapt to the dynamic context of the Io T,thereby improving the efficiency of network resource management and the quality of user service experience.The existing researches on Io T AI applications rely on offline data processing and analysis modes.Considering that the deployments of Io T AI applications in cloud computing are limited by the actual environment,including communication methods,data processing and analysis methods,as well as the participation of professionals in the field.The edge computing system for artificial intelligence application of Internet of Things is studied in this thesis.Firstly,the AI application cases in smart city and smart healthcare are introduced,and the framework of edge AI system is proposed.And the problems that need to be considered in access management,network management,security and privacy,and application management of edge AI system are also discussed.Then a case of edge skin disease recognition system is given,including skin database construction,skin disease classification model training,and system delay testing.Finally,the edge AI system model is abstracted.The system utility objective in the scenario of terminal-edge cloud-cloud task scheduling is designed,and the impacts of key objective variables on the system utility are evaluated in the experiment.By distributing and deploying cloud computing applications to the network edge,the task computing results can be quickly sent back to the Io T terminal.However,due to the heterogeneity of cloud and edge cloud computing resources,and the complexity of computing and communication processes in multi-edge cloud environment,the existing solutions of task placement are difficult to apply.A task placement strategy for maximizing the revenue of supply and demand in cloud-edge collaborative computing is studied in this thesis.Firstly,a deep reinforcement learning controller based cloud-edge collaborative computing framework is proposed.Then,by analyzing the task model in cloud-edge collaborative computing environment,the system Qo S model is constructed from the perspective of user benefits and service provider benefits.By using Q-table and deep Q-network to analyze the target problem,a deep reinforcement learning based collaborative task placement algorithm is proposed.Finally,experiments show that the proposed method has a good learning ability for the computing cost of cloud and edge cloud and the communication cost between multiple edge clouds.Compared with Q-table learning,random computing and cloud computing,the system utility of the proposed method is improved by 10%.The existing computing service modes can not meet the requirements of dynamic Io T context environment,including the diversity of terminal tasks,and the randomness of link qualities and network resources,which brings challenges to the design of new computing paradigm.A distributed competitive and cooperative task offloading strategy in cloud-edge communication system is studied in this thesis.Firstly,a distributed reinforcement learning based offloading computing framework is proposed.Then,the user utility objective of task computing is established from the perspective of service delay and terminal energy consumption.By transforming the optimization of user utility into a partially observable reinforcement learning problem of distributed offloading computing,a distributed reinforcement learning based offloading computing algorithm is proposed.Finally,experiments show that the proposed method has a good learning ability for the interactive information between Io T terminals,and between Io T terminals and cloud servers.Compared with other distributed offloading computing methods,higher system utility of the proposed method is obtained.At the same time,the computing time of the proposed method is reduced by 60% compared with the centralized offloading computing method.The access scheme for Io T terminals that meets the quality of service(Qo S)requirements is also a key issue in the intelligent Io T.With the randomness of wireless channel state information,the Qo S requirements of users bring difficulties to multi-terminal wireless access.A Qo S-aware resource allocation strategy in multi-task communication system is studied in this thesis.Firstly,the data interaction process between Io T terminal and cloud is given.Then,based on the fairness criterion,the mathematical expression of the minimum transmission rate of the user pair is derived,and the sum transmission rate optimization model with the constraints of user Qo S and system resource capacity is established.Considering that the available space of time slot is limited,the objective problem is transformed into a convex optimization problem and solved by KKT condition and greedy strategy.Finally,simulation experiments are carried out in Rayleigh fading and Rician fading channels.The experimental results show that compared with the three methods of single energy resource allocation,single time resource allocation and single user pair,the sum transmission rate of the proposed multi-user pair dynamic resource allocation method is significantly improved.