Research On Content Driven Computing Task Unloading And Resource Allocation Algorithm In Internet Of Vehicle

Video information provides rich information for vehicle networking,and the information contained in video data needs to be obtained through content understanding and analysis,making the understanding and analysis of vehicle video content gradually become a trend,and improving the ac-curacy of video content understanding has become a major challenge to promote vehicle networking.At the same time,the development of moving edge computing provides a lot of computing resources for vehicles,which makes up for the lack of vehicle computing ability.Therefore,how to use moving edge computing to improve the accuracy of video content under-standing has become an important issue.Mobile Edge Computation(MEC)in the Internet of Vehicles scenario focuses on providing cloud computing power and information technology services at the Edge of Mobile networks,and is considered a key technol-ogy to drive the development of intelligent connected vehicles.Mobile edge computing in the scenario of Internet of Vehicles is mainly divided into two aspects:computing unloading decision and resource allocation.The performance of video content analysis is affected by how computing offloads and resources are allocated.At present about the calculated un-loading decision-making and resource allocation has been around for a lot of research,such as based on the quality of service(quality of service,QoS)or based on the quality of experience(quality of experience,QoE)resource allocation scheme,however,these methods on the one hand,unable to make full use of resources to meet the vehicle for the accuracy of video content understanding,on the other hand,the existing high algorithm com-plexity and the ability to adapt to the car network dynamic change of envi-ronment.In view of the above problems,in order to improve the performance of video content analysis,this thesis considers the content-based calcula-tion unloading and resource allocation scheme,preprocesses the vehicle video information collected,and guides the calculation unloading and re-source allocation by building the importance model of video content.The main research contents are as follows:On the one hand,aiming at the problem of computing task unloading decision of video content understanding task in Internet of vehicles,a con-tent driven algorithm for computing task unloading decision is proposed.Firstly,the relationship between the video content understanding task and the state of the edge server is analyzed,and the utility of the computational offloading decision system is quantified.Then,the corresponding delay energy consumption model is constructed.Finally,the computational of-floading decision problem of the video content understanding task is trans-formed into the mixed integer linear programming problem(MINLP).Sec-ondly,in order to solve the problem of the optimal solution can involve huge amounts of state decision,MCTS based on reinforcement learning algorithm is proposed in this thesis,by improving the confidence limit al-gorithm to help choose the motion of the state,and in order to improve the calculated unloading the convergence and reduces the complexity of the decision-making algorithm,by using the method of training within DNN state transition probability for the algorithm is improved.Finally,the sim-ulation results show that the proposed algorithm reduces the algorithm complexity and convergence speed,and improves the effectiveness of the system and the accuracy of the video content understanding task.On the other hand,aiming at the problem of computing task unloading and resource allocation in the process of video content understanding task uploading and processing in the Internet of vehicles(IOT),a content driven joint optimization algorithm of computing task unloading and resource al-location is proposed.Firstly,the optimization model between the decision of computing unload and the accuracy of resource allocation and video content understanding task is established.The utility of unloading decision and resource allocation under video content understanding task is quanti-fied according to the optimization model.A multi-objective optimization problem is constructed.According to the utility of the decision-making of computing task unloading,communication resources and computing re-source allocation,the optimization model is constructed to maximize In order to improve the total efficiency of computing task offloading and re-source allocation system in the Internet of vehicles(IOT),it is used to im-prove the effectiveness of computing task offloading and resource alloca-tion.Then,the joint optimization problem of vehicle computing task un-loading decision and resource allocation is constructed as Markov decision process(MDP).In order to solve this MDP,considering the huge state space and action space in the joint optimization problem mentioned above,this thesis uses the deep reinforcement learning method based on Actor-Critic algorithm to deal with it.Simulation results show that this method has faster convergence speed and better learning performance compared with other DRL methods.Compared with DQN,which require a large number of steps to reach near-optimal strategy,this method achieves higher system utility with fewer learning steps,and effectively utilizes communication resources and computing resources at the edge.