Research On Adaptive Multi-Source Information Fusion Methods For Polar Integrated Navigation

Precise navigation method for the polar regions is a core technology that supports scientific investigations,military activities,and commercial activities in these regions.Affected by the limitation of traditional coordinate system,current inertial navigation means present worse heading angle error,latitude error,and longitude error when it is used in the polar regions.Current polar satellite navigation means face low altitude angle problem for visible satellites,and its availability is limited in the polar regions.To deal with the polar application limitations of current single navigation methods,this dissertation adopts multiple navigation sources including inertial navigation,satellite navigation,and other sensors to establish an integrated navigation system,and the corresponding research issues are as follows: error formation mechanism and correction models;design of optimal state estimation adaptive filters;multi-sources information fusion methods based on unified coordinate systems and temporal alignment.From aspects of problem analysis,theory derivation,algorithm verification,this dissertation provides new ideas for establishing much more precise polar navigation systems and spatio-temporal information service systems.Based on the analysis of polar application environment,this dissertation chooses inertial navigation system,satellite navigation system,celestial navigation system,and Doppler velocity log to form a polar vehicle integrated navigation system and corresponding information fusion scheme.And ships sailing in the Arctic region are selected as the main analysis and simulation background.On the basis of decentralized information fusion scheme,the system architecture designed by this dissertation is composed of sensors level design,two sensors fusion level design,and multi-sensors fusion level design.Thus,research issues concerned with the three levels constitute the main research contents of this dissertation.At sensors level,this dissertation proposes refined tropospheric delay correction models for satellite navigation signal suffering low altitude angle.From the aspect of mitigating sensor error and improving sensor availability in Polar,this dissertation aims the problem that tropospheric delay for satellite navigation is the main error source because of the low altitude angle of visible satellites,and researches on refining tropospheric delay correction models for polar visible satellites under low altitude angle condition,which is helpful to lower the requirement of cutoff angle of satellite receivers.To improve traditional models which depend on local meteorological parameters,this dissertation proposes a tropospheric delay representation method using spatio-temporal parameters based on a nonlinear hypothesis.And then the global zenith hydrostatic delay,zenith wet delay,and zenith total delay are modeled as black boxes;thus,a global navigation satellite tropospheric zenith delay correction model is built on a grid.Validation results based on real reference sites data prove the effectiveness of the proposed model,and a detail analysis concerned with latitude shows that the zenith delay correction accuracy is better than state of the art models.Aiming at correcting zenith delay for arbitrary coordinate,arbitrary height,and arbitrary time,corresponding interpolation methods and refined models are proposed.Combining the proposed models with Vienna mapping function,slant delay models are obtained which are available for low satellite altitude angle scenarios in the polar regions.Validations where the satellite altitude angle is lower than 10° prove the advancement of the proposed slant delay model.The proposed methods provide theories and model validations for improving the availability of current satellite navigation systems in the polar regions.At a two sensors fusion level,this dissertation proposes an information fusion algorithm based on interacting dual models and adaptive filters.When fusing information for two sensors,this dissertation aims at solving inertial navigation problems in Polar that it is hard to find the North direction and the longitude converges.This dissertation analyzes the feasibility of transversal coordinate system in polar coordinate unification and information fusion,and proposes a universal two sensors information fusion scheme and algorithm based on adaptive filters.To deal with potential non-stationary and non-Gaussian measurement noise during sensors observation processes,and the change of measurement noise when executing coordinate system transformation,this dissertation proposes a dynamic estimation method for measurement noise covariance matrices.The proposed adaptive filter,which is based on this covariance matrix estimation method and an interacting dual model structure,is able to provide interacting fusion output on the basis of dynamically updated dual models probabilities.Under several atypical measurement noise conditions,simulation results show that the proposed information fusion algorithm performs better than conventional algorithms based on filters,and it can provide higher navigation accuracy,presenting good adaptivity for dynamically changed measurement noise environment.At a multi-sensors fusion level,this dissertation proposes an asynchronous information fusion algorithm based on performance degradation indicators and a temporal alignment method.From the aspect of information fusion for multiple sensors,this dissertation fuses the subsystems fusion results again to find the global optimal fusion result and improve the robustness of the whole system.To deal with the negative effects of subsystems performance degradation on fusion center,the dissertation proposes a synchronous multi-sensors information fusion algorithm based on so-called degradation indicators,where an information sharing factor allocation strategy is used,and the confidence levels of the subsystems can be timely estimated to change the weights participating in the fusion center.To deal with the asynchronous sensors whose data update rates are different,the dissertation proposes an asynchronous multi-sensors fusion algorithm based on a temporal alignment method considering the confidence levels;once if the confidence levels are beforehand obtained and stored as a digital map,a concept of confidence field and the corresponding asynchronous multi-sensors fusion algorithm are proposed.Performance degradation and temporary outage of subsystems are introduced in the simulation scenarios,and the simulation results validate that the proposed three synchronous and asynchronous fusion algorithms are effective and robust.Finally,this dissertation explore a multi-sensors fusion network based on ensemble learning,which is the first trial in the field of navigation to the best of the author's knowledge.By learning the relationship between reconstructed figures of covariance matrices and the fusion optimality,the fusion network is able to mitigate the performance fluctuation when facing sudden faults.