Research On A Class Of Nonlinear Robust Filter Algorithms For Integrated Navigation System

In the modern military and civilian fields,the high-altitude solar unmanned aerial vehicle(UAV)owns the advantages of long duration and wide application prospects so the UAV has become a research hotspot of many countries.During the flight,the high-precision navigation system could provide reliable performance guarantee for the UAV flight mission.However,the characteristic of high-altitude environment is unknown and navigation components may suffer failure which would decrease the accuracy and reliability of calculated results in the navigation system.In this way,the aerial flight mission will be affected.Around the above demand background,this paper studies the nonlinear robust filter algorithms with high performance for the integrated navigation system of high-altitude solar UAV.In addition,the multi-sensor based integrated navigation system vehicle experiment is designed to verify the performance of the proposed algorithms.The main work is shown as follows:(1)In order to improve the reliability and accuracy of UAV navigation system,the Strapdown inertial navigation system/ Global positioning system/ Beidou satellite navigation system(SINS/GPS/BDS)in-motion alignment model and Strapdown inertial navigation system/ Global positioning system/ Beidou satellite navigation system/ Celestial navigation system(SINS/GPS/BDS/CNS)integrated navigation system filter model are proposed based on the federal fusion structure.The navigation calculated value of SINS is utilized to constitute the local filter subsystems with the measurement output from other subsystems,respectively.The filter estimated value of each subsystem is obtained by the filter algorithm.The state estimated value and covariance of each subsystem are input into the fault detection module,the corresponding fault detection value is calculated as well.Then,by means of the allocation function,the weight of each subsystem is calculated and the global state estimation is completed in the main filter.The global state estimated value is fed back to each filter subsystem and could compensate for the SINS that will decrease the SINS calculated error and obtain the accurate navigation information.The federal fusion structure could help reduce the influence of fault information on the navigation accuracy and improve system reliability.(2)To meet the performance requirements of alignment accuracy and computation real-time capability during the UAV flight process,a robust H infinity cubaure Kalman/Kalman hybrid filter based aerial in-motion alignment algorithm is proposed.Based on the idea of dual filters algorithm,the filter model of in-motion alignment system is decomposed into two parts including the nonlinear state part and linear state part that the parallel state estimation process could be conducted by the CKF and KF algorithms,respectively.In this way,the number of nonlinear sampling points is reduced while the availability and efficiency of sampling points could be enhanced effectively.Combined with the H infinity robust filter technology,the influence of external inference and system uncertainty on the filter process is restrained,which improves the accuracy and anti-interference ability of in-motion alignment system.The stochastic stability of the proposed algorithm is proved by the discrete Lyapunov function.The simulations demonstrate that the proposed algorithm could effectively reduce the computation load and improve the accuracy in the in-motion alignment process.In addition,the influence of external disturbance is decreased as well.(3)As the high accuracy nonlinear filter algorithm would bring about a large computational burden and the external outlier as well as the non-Gaussian distributed measurement noise could decrease the filter accuracy,a measure of nonlinearity(Mo NL)based robust adaptive cubature Kalman filter based integrated navigation algorithm is proposed to solve the problem.According to the integral characteristic of cubature rule,the high degree and low degree cubature rules can be obtained through the combination of spherical rule and radial rule under different accuracy levels.By means of nonlinearity evaluation function which is based on the local measurement method,the Mo NL can be calculated in the system state equation and measurement equation,respectively.Based on the comparison result with threshold,the appropriate cubature rule is selected to conduct the filter process.In this way,the high accuracy filter result is achieved with lower computational burden.The outliers and non-Gaussian distributed measurement noise can be fused into measurement update process by the introduction of M-estimation method.Then,the measurement noise covariance could be rebuilt to enhance the robustness and adjust the varying curve of Mo NL which is helpful to the reasonable selection of cubature rules.Thus,the calculated accuracy and computation real-time capability could be improved in the abnormal condition.The stochastic stability of the proposed algorithmis is proved by the discrete Lyapunov function.The simulations demonstrate that the proposed integrated navigation algorithm could obtain the higher accuracy with lower computational burden when the outliers occur or the measurement noise does not satisfy the Gaussian distribution.(4)In the uncertain environment,the priori model uncertainty and measurement noise with unknown characteristic would decrease the calculated accuracy in the integrated navigation system,a robust cubature smooth variable structure filter based integrated navigation algorithm is proposed to solve the problem.The proposed algorithm utilizes the idea of smooth variable structure filter as well as the cubature rule and square root filter framework.In this way,the filter linearization error is reduced while the positivity and symmetry of computation process are improved as well.Meanwhile,the smooth subspace could be adaptively adjusted based on the real-time state estimated error,which could adjust the filter gain and restrain the influence of model uncertainty.Combined with the variational Bayesian strategy and M-estimation method,the real-time unknown measurement noise characteristic is estimated accurately.The influence of outliers on the filter progress is decreased as well.The simulations demonstrate that the proposed integrated navigation algorithm would effectively improve the calculated accuracy in condition of the complex application environment.(5)In order to verify the validity of the proposed algorithms in this paper,the ground vehicle experiment is designed to test the performance of the in-motion alignment algorithm and integrated navigation system filter algorithm.Based on the existing experiment equipments in the laboratory,multi navigation sensors are utilized to build the vehicle experiment data acquisition platform for the SINS/GPS/BDS in-motion alignment system and SINS/GPS/BDS/CNS integrated navigation system.The SPAN-CPT is taken as the navigation benchmark unit that the error characteristic of calculated results could be analyzed precisely.Through the vehicle experiment,each navigation system data is collected and the online calculation is conducted.The result is transmitted to the upper monitor to show up and saved to the memory.Finally,the calculated result is compared with the data of benchmark unit,which could testify the practical application performance of the proposed algorithms in this paper.