Research On Filtering Technology In GFMINS/GNSS Integrated Navigation For Range Rocket

The long-range rocket projectile,called the "approximate missile",contrast with short-range missile,which has advantages of light weight,small size,low cost and so on.Especially,the speed of the rocket projectile is very fast(about 3 or 6 Mach),and so this make the rocket have the very strong defense penetration ability and almost can not be intercepted.So the long-range rocket projectile has been the important part of the modern artillery currently.Because of the launch large overload high-speed rotation of the projectile,the conventional INS can not be used for the rocket projectile system.However,the Gyro-Free micro inertial navigation system(GFMINS)based on the MEMS accelerometer can bear the large overload and impact,and also the manufacturing technology of MEMS accelerometer is more mature than the MEMS gyro.Therefore,in this dissertation,the integrated navigation system GFMINS/GNSS is proposed,which is constituted by the GFMIS and the global navigation satellite system(GNSS)and used for the position and attitude determination of the rocket projectile.The main task of dissertation is researching on this GFMINS/GNSS integrated navigation system and the mainly content includes the accelerometer configuration?integrated navigation system analysis?adaptive UKF filtering?nonlinear H? filtering and so on.Firstly,the navigation principle of GFMINS is introduced and the common accelerometer configurations and angular velocity solutions methods are analyzed.The analysis results show that the angular velocity errors of square root method?differential method and logarithmic method can be bounded but the absolute errors of the three solutions are significantly larger and can not meet the engineering requirements.In addition to the three solution methods above,the integral method is another important angular velocity solution method,although its error accumulates fast because of the component error,the absolute error is satisfactory in a short time.For example,the angular velocity error is equivalent to gyro drift of 1(°/h)in 10s.A nine-accelerometer configuration is designed and a scheme of random error detection and compensation for MEMS accelerometer is proposed,and this scheme work before the launch of the rocket projectile in any environment.Then,the redundancy of this nine-accelerometer configuration is analyzed and the result shows that there are three redundancy schemes that respond the accelerometer fault in the nine-accelerometer configuration.Secondly,on the base of the analysis of a variety of satellite navigation systems,the combination mode of GFMINS/GNSS is determined.On the base of deriving error equation of nine-accelerometer GFMINS,the integrated navigation system model of GFMINS/GNSS is built,which is corrected by the system output with the linear Kalman filer.Next,with the analysis method of Piece Wise Constant System(PWCS)and on the base of singular value decomposition,the observability of three states of motion are analyzed and the system feedback correction amount are been identified with the analysis results.A linear Kalman filter is built to simulate these analysis methods.For the problem of poor observability of angular velocity error of GFMINS/GNSS system,a H? filter is designed and which is compared with the Kalman filter.The simulation results show that the precision of Kalman filter is better than the H? filter in short time but the H? filter is more robustness and stability.Next,the initial alignment of GFMINS system is analyzed and the GFMINS/GNSS integrated system is nonlinear,then the nonlinear model of GFMINS/GNSS integrated system is derived in the case of large azimuth misalignment angle.At the same time,on the base of study of existing nonlinear filtering methods,the EKF and UKF methods are used in this nonlinear system and are compared by computer simulation.For the high dynamic and real-time requirements of long-distance rocket projectile,a Rao-Blackwellized Unscented Kalman Filter with Minimal Skew Simples Sampling(MS-RBUKF)is proposed and the simulation result shows that its processing time reduce about 33%compared to traditional UKF while ensuring system estimation accuracy.Thirdly,on the base of analysis of noise model of integrated system GFMINS/GNSS,design a nonlinear "Sage-Husa" maximum posterior noise estimator and propose a two parallel BP neural network controller to approximate the noise estimator above to solve the problem that the noise estimator can not estimate the system noise Qk and the observation noise Rk simultaneously,and then a adaptive UKF algorithm based on two parallel neural network is proposed base on this.Next,the integrated system is simulated in the case of uncertain noise and the simulation result is analyzed.This result shows that the data saturation will make the filter divergent and so a fading UKF algorithm based on variance inflation factor is proposed to solve this problem.At last,the two filtering algorithms above are combined used with the conversion judge,which is simulated with the better simulation results.Last,on the base of equivalent conversion relationship between H? norm of transfer function and Riccati equation,the nonlinear H? filtering algorithm is studied and an extended H? filtering algorithm is proposed for the strongly nonlinear system.This algorithm can only make the system model linear and didn't change the equivalent relationship between H?.norm and Riccati equation.Then,the problem of model drift of GFMINS/GNSS integrated system is analyzed and an extendedH?filtering algorithm based on gradually disappear memory is proposed to aim at this problem,which do not expand the model drift to the H? norm definition and so ensure the integrity of the H? filtering theory.At last,combined with the filter convergence criteria,the selecting method of gradually disappear factor is given,which is simulated in the case of random disturbance and also has been verified effectively.