Research On Vehicle Integrated Navigation Algorithm Based On Federated Kalman Filter

The Vehicle Navigation System(VNS)can provide a variety of navigation parameters for car driving.It is widely used in civil and military fields such as driverless,car navigation,robot navigation,etc.,and due to the rapid development of the Internet and the Internet of Things,vehicle navigation systems will play an increasingly important role in the future.At present,accuracy and reliability are important performance indicators for in-vehicle integrated navigation systems.In the car navigation system,the strap-down inertial navigation system,global positioning system and odometer are commonly used navigation sensors.How to combine them with effective multi-sensor and improve the traditional filtering algorithm to improve the system's anti-interference ability and high-precision navigation parameter output ability is the focus of this paper.This paper focuses on the principle and error analysis of sensors,the selection of multi-sensor filter structure,the design of information distribution coefficients and the construction of adaptive robust federated filtering algorithms.The main research work includes the following:Firstly,it introduces the working principle of strap-down inertial navigation system,global positioning system and odometer,and deduces their respective error models.At the same time,it compares the advantages and disadvantages of these three sensors in practical applications.From a theoretical point of view,the shortcomings of the three sensors alone and the necessity of combining them are explained.Secondly,the application background,mathematical formula and algorithm flow of standard kalman filter are introduced.When multiple sensors are used in combination,two filtering methods,centralized filtering and federated filtering,are introduced.The working principle and mathematical model are described respectively,and the advantages and disadvantages of the two are compared.The mathematical model of the car navigation system based on the federated filter structure is modeled,and from the perspective of improving the accuracy,an optimal distribution scheme of federal filtering information is designed.Then,the problem of error model in the process of filtering and solving the car navigation system is studied,including the problem of the abnormality of the observation model and the uncertainty of the statistical characteristics of the system state noise.A cascaded adaptive robust federated filtering algorithm is proposed.According to the real-time online matching idea of covariance matrix,the improved adaptive innovation filtering method is used to eliminate the influence of the observed model anomaly on the filtering accuracy.At the same time,the simplified Sage-Husa adaptive filtering method is used to solve the uncertain system state noise problem,and the system noise update also helps to correct the abnormal model.Finally,the proposed algorithm is verified from two aspects of simulation and sports car experiments.The results show that the proposed cascaded adaptive robust federated filtering algorithm can effectively control the influence of model uncertainty and make car navigation system has better robustness and higher accuracy.