Research On Nonlinear Filtering Algorithm For INS/GNSS Integrated Train Positioning

With the rapid development of science and technology,the investment in railway transportation has maintained a strong growth.Train mileage increases year-to-year,trains travel faster and more frequent,stringent requirements for accuracy and reliability of train positioning systems are necessary.At present,positioning methods with individual sensor are difficult to satisfy the requirements both from the accuracy and the stability of the positioning system.Occasional accidents are more likely to affect the driving safety of the train.To solve this kind of problem,some positioning methods with complementary advantages are organically combined to form integrated positioning system for trains,which can improve the train positioning accuracy and reliability greatly.The INS/GNSS,as the most widely used integrated positioning system,provides efficient and stable positioning services for trains.At the same time,the odometer system was used in INS/GNSS,and INS/GNSS/ODO integrated positioning system was established,which improved the accuracy and reliability of the positioning system.Some new non-linear filtering methods and data fusion algorithms were put forward for the train integrated positioning system.Application of nonlinear filtering algorithm and new data fusion method in train integrated positioning were discussed.The main contributions and innovations:(1)When the INS/GNSS integrated positioning system is used in high-speed train,system state equations which describe system dynamics and measurement need to be established.Firstly,the coordinate system and the transformation between various coordinate systems was introduced,conversion matrices used in INS/GNSS integrated positioning system from different coordinate systems to the train navigation coordinate system were listed;Secondly,the system equations for direct filtering in INS/GNSS train were given;At last,the state equation and the measurement equation of INS/GNSS/ODO integrated positioning system were established.(2)The complicated railway environment and the vibration and swing of the train will cause interference to the inertial components,which may get errors to system model.An adaptive UKF with noise statistic estimator was proposed to overcome the limitation of the standard UKF for the using in train integrated positioning system of INS/GNSS.The algorithm based on covariance matching technique,and the innovation and residual sequences were used to determine the covariance matrices of the process and measurement noise,and feeded back into the standard UKF to compensate for prior knowledge of INS and GNSS noise statistics online.This algorithm improved the adaptive ability of the standard UKF in dynamic state and parameter estimation system.The random stability of the CMAUKF algorithm was analyzed.The proposed CMAUKF was applied to the train integrated positioning system,and the validity of the algorithm is verified.(3)Consider the non-Gaussian noise that may occurs in the positioning system when the train is running,the characteristics of particle filtering were studied.Particle filter importance density function is difficult to select,new unscented particle filter algorithm based on Gaussian process regression was proposed,which can alleviate the problem of PF particle degradation.GPR is a self-learning regression method.It estimates and adjusts the covariance of process noise and measurement noise online.The regression model of the system can be obtained to improve the filtering accuracy.The proposed filter estimated and adjusted system noise statistics dynamically,improved the accuracy of train integrated positioning system.(4)In order to ensure the positioning accuracy and stability of the train in the mountainous tunnel and some aera where satellite signals are susceptible to interference,the INS/GNSS/ODO train integrated positioning system was established,and a multi-sensor optimal data fusion method based on UKF was designed for this integrated system.First,a data fusion method with a two-level fusion structure is proposed,a sub-filter was formed by INS and GNSS,another was made by INS and ODO to obtain the local optimal state estimates.UKF is used instead of KF for the filter,and it can be used to deal with non-linear problem of the INS/GNSS/ODO integrated system;second,a matrix weighted data fusion algorithm is derived based on linear minimum variance to fuse the local state estimates,and the global optimal state estimate is obtained.The upper bound technique is not needed to eliminate the correlation between local states,and the proposed data fusion algorithm reduces the conservation derived from the upper bound technique.The positioning accuracy achieved by the proposed method is improved significantly.(5)To improve the adaptive ability of the INS/GNSS/ODO train integrated positioning system,The adaptive filter is introduced into the federated Kalman filter to improve the accuracy of data fusion.First,CMAUKF is used as a local filter in the federated Kalman filter for the non-linear system of INS/GNSS/ODO integrated system;At the same time,the calculation method of allocation factor is construct based on the adaptive factor,matrix weighting method for fusion filtering is also used for comparison.The improved algorithm overcomed the influence of external information for train dynamics model effectively.