Research On Resource Allocation Mechanism For Internet Of Vehicle Based On Machine Learning

In recent years,the popularity of automobiles has promoted the efficiency of people's travel and expanded the range of people's daily activities.However,it also brings a lot of traffic problems.Hence,the research on the technology of smart and connected vehicles is necessary.As we all know,it is of great importance to reduce traffic accidents and improve the road safety through the transmission of safety messages in Internet of Vehicle(IoV).With the increase of the number of vehicles,the demands of communication between vehicles continues to grow.And the demand of resource in IoV also varies with the geographical differences.Therefore,for the limited communication resources,reasonable allocation is of great significance to the research of vehicle interconnections in IoV.To solve these problems,the thesis analyzes the network characteristics of IoV,summarizes the research status of IoV's network architecture and resource allocation,and studies some theories and methods in Machine Learning.Firstly,for the existing architecture of cloudbased IoV and the communication resource,we propose a centralized-distributed architecture of cloud-based IoV which includes vehicular cognitive layer,roadside-microcloud layer and data-center-cloud layer.In the proposed architecture,communication resource is virtualized as resource blocks,and we put forward a resource allocation mechanism for IoV based on Machine Learning.The presented mechanism which works in data-center-cloud layer includes the data file of IoV,the external influence file,the data processing model,Support Vector Regression(SVR)forecasting model and the resource allocation model.Then,we make a description of the proposed network architecture and allocation mechanism in details.Finally,the SVR forecasting model and resource allocation model in the proposed mechanism are mainly analyzed.The SVR forecasting model optimized by Grid Search Algorithm(GSA)and Genetic Algorithm(GA)respectively can make a more accurate prediction to the number of vehicles in different regions which are controlled by roadside-micro-cloud.In the resource allocation model,the strategy combining static and dynamic resource allocation assigns resource for each region(that's roadsidemicro-cloud)in IoV.according to the results of the prediction.Therefore,the resource satisfactions of some regions have been improved,and the whole resource satisfactions of IoV has been balanced in IoV.In the thesis,the performance of SVR forecasting model and resource allocation model is verified,which are parts of the resource allocation mechanism for IoV based on Machine Learning.The results demonstrate that the SVR forecasting model based on GA outperforms the one based on GSA.And it is necessary to update SVR forecasting model with new data so as to ensure the accuracy of the predicted results in reality.Moreover,in the resource allocation model,the strategy combining static and dynamic resource allocation can adjust the resource demands for different regions according to their requirements,and improve the resource satisfaction of some regions.To a certain extent,the resource satisfaction is balanced among different regions.