Research On End-edge-cloud Collaborative Computing For Intelligent Railway

The sustainable deepening of intelligent railway technology innovation is an important goal of the development of Chinese railway.However,the current computing capability and intelligence level of railway terminal equipment cannot meet the needs of intelligent railway.At the same time,under the policy of the construction of national new infrastructure,future basic equipment relied on 5G,central cloud and edge cloud will contribute to interconnections and ubiquitous computing for the development of smart applications in various vertical industries,including the rail transit.In order to enhance the intelligent level of railway equipment and support the development of emerging intelligent railway services,the introduction of cloud computing and edge computing as the terminals’ supplementary computing capability as well as the convergence node of analysis of multivariate data will become an inevitable approach for that.In order to enable end-edge-cloud collaborative computing to ensure the real-time requirements of intelligent railway services in the train movement scenario,and to feed back processing results in time for the driving safety of trains,it is necessary to focus on the research of offloading strategies and resource allocation mechanisms of end-edge-cloud collaborative computing to minimize the end-to-end delays of services.However,existing research on end-edge-cloud collaborative computing has not yet considered users’ mobility,and offloading strategies and resource allocation mechanisms cannot be applied to scenarios where the wireless transmission rate is continuously changed due to the train movement during the process of end-edge-cloud collaborative computing.In view of the research deficiencies,this thesis selects two intelligent railway services with different characteristics in the train movement scenario and conducts corresponding research:(1)For the service of train’s life cycle health management,considering the logical relationships between the tasks which cannot be processed in parallel,and the service characteristics of the relationship between the input and output data,this thesis carries out corresponding research.In the single train scenario,this thesis takes the movement of the train into consideration and establishes the service task model,the transmission and calculation time cost model required in each stage of the collaborative computing process,and the optimization problem model of the time cost of this service.This thesis designs a two-layer offloading strategy algorithm to solve this optimization problem by determining the offloading decision of each task.In this algorithm,when the previous task of each task is calculated at the cloud side,the edge side,and the terminal side respectively,the required cost of this task calculated at these three sides is obtained respectively.Then the offloading decision that can obtain a lower cost is reserved for the decision-makings of subsequent tasks.The simulation results show that the offloading strategy proposed in this thesis can save about 40% of the end-to-end delay at most,compared with other offloading strategies.(2)For the service of real-time monitoring of the driving status of train drivers,considering its characteristic that multiple tasks in it can be processed in parallel without considering the data input and output relationship between each task,corresponding research have been carried out.In a multi-train scenario,with the goal of minimizing the sum of the time delay of this service for all trains,this thesis takes the movement of trains into consideration and establishes a time cost model for each stage of end-edge-cloud collaborative computing and an optimization problem.To solve this problem,the thesis divides a wireless coverage cell into several small areas with relatively fixed wireless transmission rate.By solving a convex problem,the data processing volume and computing resource required by each train at the end side,the edge side as well as the cloud side respectively are obtained.By assigning certain weights to the above two results in each small area,the data processing volume and computing resources required by each train at the three sides in the collaborative computing process are solved.The simulation results show that the strategy proposed in this thesis can save about 79% of the end-to-end delay at most,compared with other strategies.