| Global Navigation Satellite System(GNSS)is a kind of spatial-temporal reference service technology with the characteristics of real-time,all-weather and high-precision,which is widely used in a variety of research related to geosciences.The deformation monitoring of mining areas and structures generally adopts a short baseline relative positioning.The double-difference observation model can eliminate receiver-and satellite-related errors,and basically eliminate distance-related errors,but multipath error cannot be weakened by double-differenced technique.Therefore,multipath becomes the main error source that limits the high-precision separation of deformation information.Many researches have been carried out on GPS multipath processing.However,at the present stage,researches on the constellation heterogeneous BDS and the multi-GNSS multipath mitigation are still relatively limited.Moreover,GPS multipath can be mitigated in the observation-domain based on spatial-temporal correlation,but the large number of multi-GNSS observed satellites greatly limits the processing efficiency of multipath in the observation-domain.Therefore,this paper focus on the space-domain correlation of BDS multipath in the observation-domain and the non-linear characteristic of multi-GNSS multipath in the coordinate-domain,and carries out researches on the multipath real-time mitigation,dynamic deformation extraction model of mining area,and real-time deformation information separation model of structure based on deep learning method.The main work and results of this paper are as follows:1)A single-difference multipath hemispherical map(SD-MHM)is proposed,which is based on the spatial correlation of multipath errors in the BDS-2/BDS-3 observation-domain,mitigates the BDS multipath and improves the BDS short baseline positioning accuracy.In addition,some BDS-3 MEO satellites have inconsistencies in the observation duration of the two-phase orbital repetition period,which will affect the processing accuracy of the sidereal filtering method,and SD-MHM can effectively reduce the effect of this phenomenon on the positioning accuracy.This model reduces the RMS of positioning errors in the 9-day average E,N,and U directions by 56.4%,63.9%and 67.4%.Moreover,the model can significantly mitigate the multipath error of observation data with a span of 7 days.We provide a universal BDS multipath processing strategy for conventional receivers.2)Combining empirical wavelet transform and independent component analysis with reference signal.Using a priori multi-GNSS multipath error as the reference signal,the EWT-ICA-R method is constructed to mitigate the multipath error in the GNSS monitoring series and accurately extract the deformation information.Experiments show that the EWT-ICA-R method has a very high correlation between the settlement deformation information separated in the E,N,and U directions with the simulated settlement deformation for adding 60mm settlement deformation every day,and the separation accuracy of the deformation information reaches the millimeter level.Compared to the daily sedimentation volume,it is tiny.In addition,the algorithm is also effective for vibration deformation.3)A CNN-LSTM deep learning method based on the fusion of convolutional neural network and long-short-term memory network is proposed for real-time prediction and correction of multipath error.Analyzed the feasibility of the CNN-LSTM method to mitigate the multi-GNSS multipath error in real-time,and verified the stability and reliability of the method from the statistical results of the 13-day multi-GNSS observation data;for GPS,the CNN-LSTM method is used with sidereal filtering method is comparatively studied;the universality of CNN-LSTM method to reduce the multipath error of combined data between different systems is analyzed.Experimental results show that this method can weaken the multi-GNSS multipath error in real-time,and improve the combined observation accuracy of different systems to varying degrees.In addition,the CNN-LSTM method is not affected by the time span of the 13-day GPS coordinate sequence,and the improvement degree of each day is better than the sidereal filtering method.4)A joint time-frequency masking and convolutional neural network method(TMF-CNN)is proposed for extracting distortion information aliased in multi-GNSS data.Through the convolutional neural network,the massive simulation deformation information and the time-frequency masking characteristics of the prior multipath error are learned,and then the aliasing deformation in the multi-GNSS raw data is separated in real time in the decoding stage.Firstly,the feasibility of separating deformation information of TMF-CNN is analyzed from the two perspectives of time domain and frequency domain;then,generalization experiments are carried out on different training samples,different observation scenarios,and different system combinations.Experiments show that TMF-CNN can accurately extract the deformation information in the GNSS monitoring sequence.In addition,different parameter models do not affect the accuracy of the deformation information extraction,and only use specific orientation training data(such as the north direction)to train the parameter model.The observation data in the migration scene still has excellent extraction accuracy.Figure[47];Table[15];Reference[134]... |
PDF Full Text Request