Research On Edge Computing Task Offloading And Migration Strategies For RTK Positioning

Location-based services have been widely used in people’s transportation,logistics management and engineering surveying.The positioning system based on real-time dynamic measurement(RTK)can realize centimeter-level positioning accuracy under ideal conditions,which meets the various positioning requirements of people outdoors.However,in mobile cloud computing,IoT devices that need to perform RTK positioning send the data collected by the rover and reference station to the cloud server or data center for processing.The massive growth of devices will cause the access network to be congested,making it difficult to perform real-time computing.At the same time,for the positioning system,it is difficult to manage massive devices.In mobile edge computing,migrating computing tasks to edge servers close to base stations not only reduces task transmission delay,but also relieves network bandwidth pressure.This thesis firstly proposes a task offloading algorithm based on device power sensing,which respectively considers the delays and energy consumption of local computing and edge computing in the RTK edge computing task offloading model.In addition,the remaining power factor of the device is introduced to balance the time weight and energy weight,in order to reasonably represent the overall system cost.The offloading model is transformed into an optimization problem with multiple constraints and solved by using the Double deep Q-network(DDQN)in reinforcement learning.Simulation results show that the proposed task offloading algorithm has lower overall system cost and lower energy consumption under the constraint of maximum delay.In the RTK edge computing architecture,the mobility of IoT devices can cause task migration when devices enter other cells,which results in additional cost.This thesis proposes a task migration algorithm based on movement prediction,and introduces a short-term movement prediction algorithm based on Kalman filter to predict the position at the next moment in order to perform task migration in advance.After a comprehensive analysis about connection-close strategy and the simultaneous connection strategy,a task migration strategy with pre-assigned dual connections is proposed.It is verified by simulation that the task migration strategy based on equipment movement prediction has lower system cost,lower latency and lower energy consumption than other strategies when the average residence time is very short.The proposed strategy can maintain a stable connection when devices frequently enter and leave the cell.Finally,this thesis uses the open-source RTKLIB tool to develop an RTK positioning system based on edge computing.The system can perform RTK positioning,task offloading,task migration and real-time viewing of positioning results.The system architecture includes four parts:RTK-Cloud,RTK-Server,RTK-Client and RTK-Map.The result of stress test show that the upper limit of the number of tasks that the system supports at the same time is 5K,and the bottleneck is the frequency of CPU.