| Under the rapid development of informatization and networking in nowadays,significant changes have been taken place in various industries centering on big data.Positioning information like transportation,personal travel,enterprise equipment management has become increasingly important.In outdoors,the general solution for vehicles to obtain location information is to use a combination system of global navigation satellite system(GNSS)and Inertial Navigation System(INS).GNSS is a space-based radio navigation and positioning system,which is sensitive to satellite signal.The precision of INS is limited to its cost and volume.Therefore,when GNSS is in locklose state,the vehicle will be difficult to locate.Existing domestic and international research suffers from problems such as data accuracy,hardware volume,high-cost,complex system and insufficient generalization.In this paper,we propose a data-driven and learning model-based positioning algorithm that utilizes multiple motion sensor data of smartphones to jointly train models by establishing a federated localization framework under the condition that the data set is retained locally by the participants.In this framework,it improves model accuracy while ensuring data privacy.In this paper,the edge nodes can collect data and store data.At the same time,the edge nodes deploy learning units.During the training process,each edge node cooperates with the central node in parameter exchange to build the whole model.While at the central node,different federated fusion algorithms are used.In the actual environment,two field scenes that satisfy--Independently Identically Distribution(IID)and Non-Independently Identically Distribution(Non-IID)are modeled separately.Finally,we teste and compare systems with two existing typical optimization algorithms.The experimental results show that the position error is only about 12 m in about 1.8km with every 60 s to acquire GNSS signals.However,on even ground,the existing algorithms exceeded 100 m,which fully proves the feasibility of the algorithm in this paper and gives the best positioning algorithm for different scenarios.The main work and innovations of this paper can be seen in the following:1.Aiming at the shortcomings of the current vehicle GNSS/INS combined system in some special scenarios,a joint positioning algorithm based on data-driven and learning models is studied and proposed.Under GNSS low-sampling conditions,the feasibility of the algorithm in this paper is verified in two typical practical scenarios,and the advantages and disadvantages of the algorithm are analyzed.2.Based on the low-precision sensor data of the smartphone,the existing vehicle positioning system is optimized,which greatly reduces the update frequency of the GNSS signal required by the positioning system,and reduces the cost and sensor volume;3.Based on the federated learning positioning framework proposed in this paper,it solves the dilemma that each participant does not have a large-scale training set locally,and avoids the leakage of the position information of the participants in the joint construction of the positioning system. |
