Extension Research On UKF Algorithm And Data Fusion Technology For Integrated Navigation

Modern vehicles put forward increasingly high request on precision and reliability for navigation systems, and any single navigation mode has no longer satisfied this requirement. Consequently, integrated navigation becomes the inevitable trend for navigation technology. An integrated navigation system is essentially a muti-sensor information fusion system. As the information fusion technology is employed to design an integrated navigation system, the filtering algorithm and muti-sensor data fusion are the two fundamental issues need to be solved. This thesis provides in-depth research on the aforementioned two issues.There are two parts in this thesis. Based on the platform of INS/GNSS integrated system, the first half of the thesis has done some exploratory works on the performance improvement of UKF in integrated navigation. Specifically, the latter part of the thesis has been focus on the muti-sensor data fusion technology in INS/GNSS/CNS integration.The main contributions and innovations of the thesis are as follows:(1) A nonlinear measurement equation is constructed for the tightly coupled INS/GNSS integrated system by truncating the Taylor expansion of pseudorange measurements to the second-order term. To eliminate the standard UKF's drawback of redundant computation due to the linear system state equation of the tightly coupled INS/GNSS integrated system, a derivative UKF is further presented. In the time update process, the proposed filter avoids the redundant computations involved in the UKF due to the use of unscented transformation; and in the measerment update process, it inherits the excellent properties of the UKF in dealing with nonlinear filtering problems. The proposed filter can effectively reduce the compulational complexity of the standard UKF when using in the tightly coupled INS/GNSS integration, without the loss of filtering accuracy.(2) The stochastic stability of the proposed derivative UKF is studied. It is proved that the estimation error of the derivative UKF remains bounded in mean square if some certain conditions are satisfied. Based on the impact analysis of the measurement noise covariance matrix on the error behavior, it is demonstrated that by adding small positive definite matrix to the measurement noise covariance, the stability of the derivative UKF can be achieved with large initial error. However, a very large value for the extra additive matrix will make its filtering accuracy decrease.(3) A windowing and random weighting based adaptive UKF is presented to address the performance degradation and divergence of the standard UKF due to unknown or inaccurate statistical characteristics of system noise. The proposed filter makes full use of the information obtained in the filtering process to dynamically estimate and adjust system noise statistics. It can resist the disturbance on the filtering performance due to the uncertainty involved in system noise statistics. The effectiveness of the proposed filter was verified through numerical simulations as well as the application examples in direct filtering approach for INS/GNSS integration.(4) A modified strong tracking UKF is presented for state estimation in nonlinear dynamic systems with process model uncertainty. The proposed filter adopts the hypothesis testing method to identify process model uncertainty. In the occurrence of process model uncertainty, a defined suboptimal fading factor is introduced into the prediction covariance to to decrease the weight of the prior knowledge on filtering solution; in the time segments without process model uncertainty, it runs optimal with the basic formulae of the standard UKF, avoiding the loss of precision. Numerical simulations and application examples in direct filtering approach for INS/GNSS integration verify the robustness of the proposed filter against process model uncertainty.(5) A modified federated Kalman filter (FKF) is designed to improve the FKF's computational efficiency involved in the master filter for INS/GNSS/CNS integration. During the master filtering process, the modified FKF firstly decomposes the global state vector of INS/GNSS/CNS integration into three sub-states. Subsequently, it fuses the sub-state estimations from INS/GNSS and INS/CNS subsystems with the corresponding ones from the time-update solution of the master filter, respectively. Eventually, the fused sub-state estimations are recombined to yield the global fusion state estimation. The modified FKF has the capability for distributed and parallel data processing. It can effectively reduce the computational load involved in the master filtering process of the FKF with slight loss of precision.(6) A matrix weighted multi-sensor data fusion methodology with two-level structure is designed to address the information fusion problem in INS/GNSS/CNS integration. On the first level, GNSS and CNS are integrated with INS by two local filters to constitute the INS/GNSS and INS/CNS subsystems and obtain the local optimal state estimations. On the second level, two different matrix weighted data fusion algorithms, one based on generic weighting matrices and the other based on diagonal weighting matrices, are presented to fuse the local state estimations for generating the global optimal state estimation. The designed multi-sensor data fusion methodology overcomes the limitation of the FKF in no-reset mode by improving the data fusion accuracy when it is applied in INS/GNSS/CNS integration. Compared with the data fusion algorithm based on generic weighting matrices, the computational load of the algorithm based on diagonal weighting matrices is significantly reduced even thouth its accuracy is slightly lower, making it is more suitable for real-time applications. Besides, since it is not required to reset the local filters by the global state estimation, the designed multi-sensor data fusion methodology has the capability of strong fault tolerance.