| With the rapid development of mobile communication technology and the Internet,autonomous driving and intelligent transportation are important development directions for the transformation of the automobile industry.People’s demand for positioning services is increasing,and the requirements for positioning accuracy are also getting higher and higher.Satellite navigation and positioning technology,especially Beidou and GPS,is widely used in traffic management,electronic information,forest fire prevention,transportation control and other industries.However,due to the influence of the troposphere,ionosphere and multipath effects,there may be some regional errors in GPS positioning.Some vehicles with high-precision positioning function sensors,such as radio frequency identification,lidar,and cameras,can improve the positioning accuracy of their own vehicles through the perception information of the sensors.But for some vehicles without high-precision positioning function sensors,how to optimize their positioning is the focus of this thesis.Aiming at the problem of protecting user privacy and improving vehicle positioning accuracy,this thesis studies a set of mobile vehicle positioning optimization techniques.This thesis starts with the relevant technical theory,introduces the theoretical basis of deep neural network,and focuses on the multi-layer perceptron model in deep neural network.The characteristics of edge computing,federated learning and other technologies are briefly analyzed.A vehicle localization optimization method based on distributed machine learning is studied.With the goal of improving the positioning accuracy of mobile vehicles,the multi-sensor vehicles in VANET that can be accurately positioned,use the sensor data obtained by their sensors into a distributed machine learning framework for training,while protecting the privacy of the user’s vehicle location,Make the neural network have the error properties of this region.In this way,it can help some ordinary vehicles without high-precision positioning function sensors to achieve positioning optimization.Aiming at the problem that the neural network model has few training samples and the model is difficult to converge,this thesis further studies a reliability-based vehicle distributed co-location optimization method.For ordinary vehicles without highprecision positioning function sensors,the surrounding multi-sensor vehicles assist in positioning,correct their own GPS positioning deviation,and participate in the training of neural networks.Considering that the accuracy of the co-located vehicle is affected by the number and distance of the vehicles participating in the co-location,this thesis proposes the concept of reliability.According to the number and distance of the vehicles participating in the co-location,the reliability of the co-located vehicle is calculated to affect the vehicle.Participate in the weight occupied by the distributed machine learning model training to avoid the phenomenon of non-convergence of the system model.Finally,the feasibility of the positioning optimization method proposed in this thesis is verified by simulation experiments,which meets the expected design goals.The research content of this thesis provides a set of reliable optimization methods for improving the positioning accuracy of mobile vehicles,and provides important support for the future development of vehicles such as autonomous driving and smart transportation. |
