Beidou-based space-time reference dynamic correction method

Abstract

The invention relates to the technical field of Beidou reference correction, in particular to a Beidou-based space-time reference dynamic correction method. The method comprises the following steps: firstly, collecting Beidou multi-frequency-point observation data, and preprocessing by adopting a Kalman filtering algorithm; then, based on the preprocessed data, a base station network time synchronization mechanism is established by using Beidou satellite signal propagation time difference, spatial reference data of different base stations are integrated through a multi-sensor fusion algorithm and a least square method, and smoothing processing is performed by using a spatial interpolation algorithm; thirdly, selecting an optimal baseline combination based on a space-time synchronization result, constructing a difference observation equation of a Beidou multi-frequency carrier phase and a pseudo range, fixing integer ambiguity by adopting an LAMBDA algorithm, and establishing a unified space-time reference model; and finally, resolving the clock error of the high-precision receiver by utilizing the double-difference integer ambiguity constraint, generating a correction parameter, and dynamically adjusting and updating the correction parameter by adopting a model algorithm. According to the invention, high-precision dynamic correction of the space-time reference is realized.

Description

Technical Field

[0001] This invention relates to the field of BeiDou reference calibration technology, specifically to a dynamic calibration method for spatiotemporal references based on BeiDou. Background Technology

[0002] The existing BeiDou high-precision positioning service has the following shortcomings in spatiotemporal reference processing: In BeiDou data processing, spatial and temporal references are processed independently, lacking a unified spatiotemporal reference modeling method. Station elevation changes are highly correlated with parameters such as tropospheric delay, orbital radial error and vertical coordinate components, clock errors and atmospheric delay. Existing methods cannot achieve unified modeling and effective integration of spatiotemporal references, resulting in insufficient accuracy and reliability of BeiDou correction parameters.

[0003] To address this, a dynamic correction method for spatiotemporal reference based on BeiDou is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a method for dynamic correction of spatiotemporal reference based on BeiDou.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for dynamic correction of spatiotemporal reference based on BeiDou, comprising:

[0007] Collect BeiDou multi-frequency observation data and use the Kalman filter algorithm to preprocess the multi-frequency observation data;

[0008] Based on the preprocessed multi-frequency observation data, a time synchronization mechanism for the reference station network is established by utilizing the difference in the propagation time of BeiDou satellite signals. A multi-sensor fusion algorithm combined with the least squares method is used to integrate the spatial reference data of different reference stations, and a spatial interpolation algorithm is used to smooth the integration results.

[0009] Based on the spatiotemporal synchronization integration results, the optimal baseline combination is selected to establish the differential observation equations for BeiDou multi-frequency carrier phase and pseudorange. The LAMBDA algorithm is used for search and fixation, and a unified spatiotemporal reference model is established to achieve rapid convergence and integrated solution of spatiotemporal reference parameters.

[0010] Based on the solution results of the spatiotemporal reference model, high-precision receiver clock bias is obtained by using double-difference integer ambiguity constraints. BeiDou satellite clock bias, ionospheric grid and tropospheric delay correction parameters are generated. The correction parameters are automatically adjusted and dynamically updated by model algorithm according to satellite signal quality, geometric distribution and atmospheric delay changes.

[0011] Furthermore, the preprocessing of multi-frequency observation data includes:

[0012] Gross errors were detected in the pseudorange observations at each frequency point, and outlier observations exceeding 3 times the standard deviation were removed.

[0013] Carrier phase smoothing pseudorange technology is used to reduce the impact of multipath effects, and low-quality observation data is filtered out by signal-to-noise ratio threshold screening and satellite elevation angle mask setting;

[0014] An adaptive recursive processing model based on observation error modeling and system noise estimation is established, and a recursive filtering algorithm is used to perform quality control and noise filtering on multi-frequency observation data.

[0015] Furthermore, the process of establishing a time synchronization mechanism for the reference station network by utilizing the propagation time differences of BeiDou satellite signals, and integrating spatial reference data from different reference stations using a multi-sensor fusion algorithm combined with the least squares method, includes:

[0016] Using pseudorange observation data of the same BeiDou satellite from various reference stations, the influence of satellite clock bias is eliminated through inter-station single difference processing, and the inter-station time deviation is calculated in combination with the known precise coordinates of the reference stations;

[0017] Establish a time synchronization model for the reference station network, take the reference station with the best time stability as a reference, calculate the time correction of other reference stations relative to the reference station, and achieve time unification of the reference station network.

[0018] Based on the time-unified observation data, Kalman filtering and weighted least squares algorithm are used to determine the weights by combining the positioning accuracy and stability of each reference station, and the spatial reference data of different reference stations are integrated.

[0019] Furthermore, the process of smoothing the integrated result using a spatial interpolation algorithm includes:

[0020] Based on the integrated base station spatial reference data, a regular grid spatial reference data is generated using an inverse distance weighted interpolation algorithm.

[0021] Establish an interpolation quality control mechanism, calculate interpolation accuracy and residual distribution through leave-one-out cross-validation, and identify abnormal interpolation regions;

[0022] For the abnormal regions identified by interpolation quality control, a moving average filtering algorithm is used for local smoothing.

[0023] Furthermore, the process of selecting the optimal baseline combination includes:

[0024] The comprehensive evaluation coefficient of each baseline is calculated based on the distance between reference stations, the satellite geometric distribution intensity factor (PDOP) value, and the observation data quality evaluation index.

[0025] By combining the comprehensive evaluation coefficient and the baseline length, the optimal baseline combination is selected, and the differential observation equations for BeiDou multi-frequency carrier phase and pseudorange are established based on the optimal baseline combination.

[0026] Furthermore, the process of establishing the spatiotemporal reference model includes:

[0027] A joint observation equation set for multi-frequency carrier phase and pseudorange is established based on the differential observation equation;

[0028] The LAMBDA algorithm is used to search and fix the integer ambiguity in the carrier phase observation equation;

[0029] By using fixed integer ambiguities as equality constraints, a spatiotemporal reference model is established to achieve rapid convergence and integrated solution of spatiotemporal reference parameters.

[0030] Furthermore, the generation process of BeiDou satellite clock bias, ionospheric grid, and tropospheric delay correction parameters includes:

[0031] Using fixed double-difference integer ambiguity as a strong constraint, the constrained least squares method is adopted to simultaneously calculate the high-precision receiver clock error and coordinates of the reference station.

[0032] Based on the calculated high-precision receiver clock bias, BeiDou satellite clock bias correction parameters are generated through clock bias fitting and prediction models.

[0033] Regional ionospheric grids and tropospheric delay correction parameters are established and generated by using carrier phase geometry-independent combinations and tropospheric delay estimates, respectively.

[0034] Furthermore, the process of automatically adjusting and dynamically updating the correction parameters using model algorithms includes:

[0035] Establish a multi-parameter monitoring system to collect multi-dimensional monitoring data that affect the correction parameters in real time;

[0036] A neural network algorithm is used to train deep learning on historical monitoring data and correction parameter change data to establish a nonlinear mapping relationship between monitoring data and correction parameters;

[0037] The trained neural network model is used to predict the optimal update timing and numerical adjustment range of various correction parameters. When the prediction results show that the parameters deviate from the threshold, the corresponding correction parameters are automatically updated.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] 1. By eliminating satellite clock bias through inter-station single-difference processing, establishing a time synchronization model, and integrating the data using Kalman filtering combined with the least squares algorithm, time unification of the reference station network and effective integration of spatial data are achieved. This enhances the stability and consistency of positioning services and reduces accuracy loss caused by independent processing. Inverse distance weighted interpolation is used to generate grid data, leave-one-out cross-validation is employed for quality control, and moving average filtering is used to process outlier areas. This improves the smoothness and interpolation accuracy of the integrated results. Addressing the bottleneck of insufficient accuracy in the static reference frame in the background, dynamic influences such as geophysical deformation are compensated for, ensuring the continuity and accuracy of spatial reference data.

[0040] 2. By calculating a comprehensive evaluation coefficient based on the distance between reference stations, PDOP value, and observation quality, the optimal baseline is selected to establish the differential observation equation. This optimizes the baseline configuration, improves the geometric strength and data quality of differential observations, enhances the solution efficiency and reliability of the spatiotemporal reference model, and reduces error amplification caused by suboptimal combinations. Integrated solution is achieved through joint observation equations, the LAMBDA algorithm to fix integer ambiguities, and equality constraints. This promotes rapid convergence of spatiotemporal reference parameters, enables unified modeling of spatial and temporal parameters, improves the overall system integration and accuracy, and enhances the ability to handle complex errors.

[0041] 3. By simultaneously solving clock errors and coordinates using double-difference integer ambiguity constraints and constrained least squares method, and generating regional correction parameters through fitting models and geometrically independent combinations, the accuracy of clock error and atmospheric delay correction is improved, errors caused by parameter correlation are reduced, and the reliability and real-time performance of positioning services are ensured. By establishing a multi-parameter monitoring system, using neural networks to train nonlinear mappings, and predicting update timing, dynamic adaptation and automatic optimization of correction parameters are achieved, improving the system's robustness and long-term accuracy, overcoming the limitations of traditional static models, and ensuring the continuous high-precision maintenance of the spatiotemporal reference. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating a BeiDou-based dynamic correction method for spatiotemporal reference according to the present invention.

[0043] Figure 2 This is a flowchart illustrating the adaptive multi-frequency weight optimization mechanism of the present invention.

[0044] Figure 3 This is a schematic diagram of the process for integrating spatial reference data according to the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see Figures 1 to 3 This invention provides a method for dynamic correction of spatiotemporal reference based on BeiDou, the technical solution of which is as follows:

[0047] Example 1:

[0048] Collect BeiDou multi-frequency observation data and use the Kalman filter algorithm to preprocess the multi-frequency observation data;

[0049] Furthermore, the regional observation network is distributed with multiple BeiDou reference stations, collecting observation data from multiple frequency points such as BeiDou B1I (1561.098MHz), B3I (1268.52MHz), B1C (1575.42MHz), and B2a (1176.45MHz), with a sampling interval of 1 second; the observation data includes: carrier phase observations, pseudorange observations, Doppler frequency shift, signal-to-noise ratio, etc.

[0050] Furthermore, the preprocessing of multi-frequency observation data includes:

[0051] Gross errors were detected in the pseudorange observations at each frequency point, and outlier observations exceeding 3 times the standard deviation were removed.

[0052] Carrier phase smoothing pseudorange technology is used to reduce the impact of multipath effects, and low-quality observation data is filtered out by signal-to-noise ratio threshold screening and satellite elevation angle mask setting;

[0053] An adaptive recursive processing model based on observation error modeling and system noise estimation is established, and a recursive filtering algorithm is used to perform quality control and noise filtering on multi-frequency observation data.

[0054] Furthermore, the mean and standard deviation of pseudorange observations at each frequency point are calculated; the monitoring threshold is set to 3 times the standard deviation. When the observation deviates from the mean by more than 3 times the standard deviation, it is judged as a gross error, and the abnormal observations judged as gross errors are removed. At the same time, the removal rate is recorded; when the data removal rate exceeds 20%, the entire observation arc is marked as unusable.

[0055] Furthermore, the carrier phase smoothing pseudorange technique employs a Hatch filter, which can utilize continuously tracked carrier phase information to correct errors in pseudorange measurement;

[0056] Furthermore, carrier phase smoothing pseudorange technology is employed to reduce the impact of multipath effects. An adaptive multi-frequency point weight optimization mechanism is introduced during the filtering of low-quality observation data through signal-to-noise ratio threshold screening and satellite elevation angle mask settings. The process is as follows: Figure 2 As shown, specifically: Real-time evaluation of the signal-to-noise ratio, multipath effect, and carrier phase continuity of BeiDou multi-frequency observation data is performed. Simultaneously, a quality score for each frequency point is generated by combining the ionospheric TEC change rate and regional meteorological data. An exponential normalization method is used to dynamically allocate weights to each frequency point based on the quality score, and these weights are applied to a Hatch filter for carrier phase smoothing pseudorange, optimizing the smoothing effect of pseudorange observations. Based on the current quality score, the signal-to-noise ratio threshold and satellite elevation angle mask are dynamically adjusted. The impact of environmental changes on the threshold is predicted using a machine learning model (such as random forest), generating an adaptive screening threshold.

[0057] By introducing an adaptive multi-frequency weight optimization mechanism, the weights are dynamically adjusted according to the real-time signal quality and environmental characteristics of each frequency point, thereby further improving the robustness and accuracy of the preprocessed data.

[0058] Furthermore, an adaptive recursive processing model based on observation error modeling and system noise estimation is established. The process of using a recursive filtering algorithm to perform quality control and noise filtering on multi-frequency observation data includes: classifying and modeling various error sources in the BeiDou multi-frequency observation data, establishing a system noise model to distinguish between white noise and colored noise components, and tracking their variance changes in real time; using a prediction-update recursive filtering framework to process time series data, in the prediction stage, the parameter values ​​and covariance at the current time are predicted based on the parameter state at the previous time and the established dynamic model, and in the update stage, the prediction results are corrected using the current observation data; the processing weights are dynamically adjusted according to the real-time estimation error and noise level. When the estimation error of a certain observation is large, its weight is automatically reduced, and when the system noise level increases, the filtering gain is adjusted accordingly; the weight adjustment is not a one-time event, but is recalculated in each processing cycle.

[0059] By employing gross error detection, carrier phase smoothing pseudorange technology, and an adaptive recursive processing model, quality control and noise filtering are performed on the observation data. This helps reduce multipath effects, low-quality data interference, and observation errors, directly improving the utilization efficiency of multi-frequency data and providing high-quality input for subsequent spatiotemporal benchmark modeling.

[0060] Based on the preprocessed multi-frequency observation data, a time synchronization mechanism for the reference station network is established by utilizing the difference in the propagation time of BeiDou satellite signals. A multi-sensor fusion algorithm combined with the least squares method is used to integrate the spatial reference data of different reference stations, and a spatial interpolation algorithm is used to smooth the integration results.

[0061] Furthermore, a time synchronization mechanism for the reference station network is established by utilizing the propagation time differences of BeiDou satellite signals. The process of integrating spatial reference data from different reference stations using a multi-sensor fusion algorithm combined with the least squares method is as follows: Figure 3 As shown, it includes:

[0062] Using pseudorange observation data of the same BeiDou satellite from various reference stations, the influence of satellite clock bias is eliminated through inter-station single difference processing, and the inter-station time deviation is calculated in combination with the known precise coordinates of the reference stations;

[0063] Establish a time synchronization model for the reference station network, take the reference station with the best time stability as a reference, calculate the time correction of other reference stations relative to the reference station, and achieve time unification of the reference station network.

[0064] Based on the time-unified observation data, Kalman filtering and weighted least squares algorithm are used to determine the weights by combining the positioning accuracy and stability of each reference station, and the spatial reference data of different reference stations are integrated.

[0065] Furthermore, the difference in the inter-station single-difference processing is the difference in the geometric distance from each station to the satellite, plus the difference in the clock bias of the station's receiver, multiplied by the speed of light. Since the precise coordinates of the base station are known, the geometric distance difference can be accurately calculated, thereby deducing the clock bias difference between the two stations.

[0066] Furthermore, based on the time-unified observation data, the time series observation data of each reference station are filtered using Kalman filtering; an integrated observation equation is established based on the Kalman-filtered observation values ​​and corresponding covariance information of each reference station, and then the integrated observation equation is solved using weighted least squares to integrate the spatial reference data.

[0067] Furthermore, the observation equations are integrated into a set of equations describing the relationship between multi-sensor observations and the parameters to be estimated; wherein, the observation vector contains the processed observations of each reference station, and the parameter vector contains regional spatial reference parameters, such as reference station coordinates, regional coordinate reference point positions, etc.; a matrix is ​​designed to describe the geometric relationship between the processed observations and the regional reference parameters.

[0068] Furthermore, a weighted least squares method is employed to solve the integrated observation equations. A system of normal equations is constructed, where the coefficient matrix equals the transpose of the design matrix multiplied by the weight matrix, and then multiplied by the design matrix; the right-hand side vector equals the transpose of the design matrix multiplied by the weight matrix, and then multiplied by the observation vector. Solving this system of linear equations yields parameter estimates. The weight matrix employs a block-diagonal structure, with each diagonal block corresponding to a weight sub-matrix for a single reference station, including covariance information obtained from Kalman filtering. This process realizes the transformation from local observation information from each reference station to unified spatial reference parameters for the region.

[0069] By eliminating satellite clock bias through inter-station single-difference processing, establishing a time synchronization model, and integrating it using Kalman filtering combined with the least squares algorithm, this achieves time unification of the reference station network and effective integration of spatial data, promotes coupled modeling of spatiotemporal parameters, enhances the stability and consistency of positioning services, and reduces accuracy loss caused by independent processing.

[0070] Furthermore, the process of smoothing the integrated result using a spatial interpolation algorithm includes:

[0071] Based on the integrated base station spatial reference data, a regular grid spatial reference data is generated using an inverse distance weighted interpolation algorithm.

[0072] Establish an interpolation quality control mechanism, calculate interpolation accuracy and residual distribution through leave-one-out cross-validation, and identify abnormal interpolation regions;

[0073] For the abnormal regions identified by interpolation quality control, a moving average filtering algorithm is used for local smoothing.

[0074] Furthermore, the residual distribution is calculated by taking the mean, standard deviation, and maximum value of the interpolation residuals; when the interpolation residual is greater than twice the standard deviation, it is marked as an outlier.

[0075] Furthermore, the filter window size of the moving average filtering algorithm can be set to 3×3 or 5×5 grid points, and the weight function adopts the Gaussian weight function.

[0076] By employing inverse distance weighted interpolation to generate grid data, leave-one-out cross-validation for quality control, and moving average filtering to handle outlier areas, the smoothness and interpolation accuracy of the integrated results are improved, compensating for dynamic effects such as geophysical deformation, and ensuring the continuity and accuracy of spatial reference data.

[0077] Based on the spatiotemporal synchronization integration results, the optimal baseline combination is selected to establish the differential observation equations for BeiDou multi-frequency carrier phase and pseudorange. The LAMBDA algorithm is used for search and fixation, and a unified spatiotemporal reference model is established to achieve rapid convergence and integrated solution of spatiotemporal reference parameters.

[0078] Furthermore, the process of selecting the optimal baseline combination includes:

[0079] The comprehensive evaluation coefficient of each baseline is calculated based on the distance between reference stations, the satellite geometric distribution intensity factor (PDOP) value, and the observation data quality evaluation index.

[0080] By combining the comprehensive evaluation coefficient and the baseline length, the optimal baseline combination is selected, and the differential observation equations for BeiDou multi-frequency carrier phase and pseudorange are established based on the optimal baseline combination.

[0081] Furthermore, the comprehensive evaluation coefficient is expressed as the product of length weight, geometric weight, and quality weight; the length weight is determined by the maximum distance between the base station and neighboring stations; the geometric weight is determined by the range of PDOP values; and the quality weight is determined by the range of values ​​of the multipath RMS index.

[0082] Furthermore, in order to further eliminate receiver clock bias, the differential observation equation adopts double-difference processing; double-difference is based on single-difference, which is the difference between the single difference of one reference satellite and the single difference of another satellite; among them, the satellite with a higher elevation angle and better signal quality is selected as the reference satellite;

[0083] Furthermore, the differential observation equations for BeiDou multi-frequency carrier phase and pseudorange include the double-difference carrier phase observation equation and the double-difference pseudorange observation equation.

[0084] Furthermore, in the expression of the double-difference carrier phase observation equation, the left side is the double-difference carrier phase observation value, and the right side includes the double-difference geometric distance, the double-difference integer ambiguity multiplied by the wavelength, and the observation noise; the double-difference geometric distance is equal to the difference in the geometric distance between the two reference stations and the two satellites; the double-difference pseudorange observation equation can be expressed as: the double-difference pseudorange observation value is equal to the double-difference geometric distance plus the observation noise;

[0085] Furthermore, the process of selecting the optimal baseline combination is dynamically optimized. Specifically, after calculating the comprehensive evaluation coefficient of each baseline, a genetic algorithm is used to perform selection, crossover, and mutation operations within a fixed time window (e.g., 10 seconds) through a fitness function based on the comprehensive evaluation coefficient. After iterative optimization, the optimal baseline combination is selected, prioritizing combinations with high comprehensive evaluation coefficients and moderate baseline lengths. During this process, if the signal interruption rate of a certain reference station exceeds 10%, the relevant baselines are automatically removed. Based on the newly selected baseline combination, the double-difference carrier phase and pseudorange observation equations are updated.

[0086] By introducing a dynamic optimization mechanism in the process of selecting the optimal baseline combination, it is possible to adapt to the dynamic changes in satellite geometry or the real-time status of the reference station (such as equipment failure or signal blockage), thereby enhancing the solution efficiency and stability of the spatiotemporal reference model.

[0087] By calculating a comprehensive evaluation coefficient based on the distance between reference stations, PDOP value, and observation quality, the optimal baseline is selected to establish the differential observation equation. This optimizes the baseline configuration, improves the geometric strength and data quality of differential observations, enhances the solution efficiency and reliability of the spatiotemporal reference model, and reduces error amplification caused by suboptimal combinations.

[0088] Furthermore, the process of establishing the spatiotemporal reference model includes:

[0089] A joint observation equation set for multi-frequency carrier phase and pseudorange is established based on the differential observation equation;

[0090] The LAMBDA algorithm is used to search and fix the integer ambiguity in the carrier phase observation equation;

[0091] By using fixed integer ambiguity as an equality constraint, a spatiotemporal reference model is established to achieve rapid convergence and integrated solution of spatiotemporal reference parameters.

[0092] Furthermore, the process of searching and fixing the integer ambiguity in the carrier phase observation equation using the LAMBDA algorithm includes: obtaining the floating-point estimate of the integer ambiguity and its covariance matrix based on the floating-point solution of the joint observation equation set; reducing the correlation between ambiguity parameters through Z-transform to generate the decorrelated ambiguity vector; and searching for the optimal integer solution in the decorrelated ambiguity space using the integer least squares method.

[0093] Furthermore, based on the joint observation equations, the fixed integer ambiguity is used as an equality constraint to update the equations. After fixing the ambiguity, the high-precision constraint of carrier phase observation reduces the uncertainty of parameter estimation and accelerates the convergence of spatiotemporal reference parameters.

[0094] By using a joint observation equation set, the LAMBDA algorithm to fix integer ambiguities, and equality constraints to achieve integrated solution, the spatiotemporal reference parameters converge rapidly, enabling unified modeling of spatial and temporal parameters, improving the overall system's integration and accuracy, and enhancing its ability to handle complex errors.

[0095] Based on the solution results of the spatiotemporal reference model, high-precision receiver clock bias is obtained by using double-difference integer ambiguity constraints. BeiDou satellite clock bias, ionospheric grid and tropospheric delay correction parameters are generated. The correction parameters are automatically adjusted and dynamically updated by model algorithm according to satellite signal quality, geometric distribution and atmospheric delay changes.

[0096] Furthermore, the generation process of BeiDou satellite clock bias, ionospheric grid, and tropospheric delay correction parameters includes:

[0097] Using fixed double-difference integer ambiguity as a strong constraint, the constrained least squares method is adopted to simultaneously calculate the high-precision receiver clock error and coordinates of the reference station.

[0098] Based on the calculated high-precision receiver clock bias, BeiDou satellite clock bias correction parameters are generated through clock bias fitting and prediction models.

[0099] Regional ionospheric grids and tropospheric delay correction parameters are established and generated by using carrier phase geometry-independent combinations and tropospheric delay estimates, respectively.

[0100] Furthermore, the time series of receiver clock bias is fitted using a polynomial or physical model, and combined with the observation data of the reference station network, the BeiDou satellite clock bias is deduced; a time series analysis model (such as ARIMA) is used to predict short-term clock bias changes and generate satellite clock bias correction parameters.

[0101] Furthermore, geometrically independent combinations are constructed using multi-frequency carrier phase observations (such as B1I and B3I); ionospheric delay is extracted through geometrically independent combinations, and combined with observation data from the reference station network to generate a regional ionospheric grid; inverse distance weighting (IDW) or kriging interpolation algorithms are used to interpolate discrete ionospheric delay observations into a continuous grid.

[0102] Furthermore, the tropospheric delay is estimated using the Hopfield model. Based on the observation data of the reference station network, the regional tropospheric delay parameters are estimated by least squares or Kalman filtering, and gridded correction parameters are generated.

[0103] By utilizing double-difference integer ambiguity constraints and constrained least squares to simultaneously solve clock errors and coordinates, and generating regional correction parameters through fitting models and geometrically independent combinations, the accuracy of clock error and atmospheric delay correction is improved, errors caused by parameter correlation are reduced, and the reliability and real-time performance of positioning are ensured.

[0104] Furthermore, the process of automatically adjusting and dynamically updating the correction parameters using model algorithms includes:

[0105] Establish a multi-parameter monitoring system to collect multi-dimensional monitoring data that affect the correction parameters in real time;

[0106] A neural network algorithm is used to train deep learning on historical monitoring data and correction parameter change data to establish a nonlinear mapping relationship between monitoring data and correction parameters;

[0107] The trained neural network model is used to predict the optimal update timing and numerical adjustment range of various correction parameters. When the prediction results show that the parameters deviate from the threshold, the update of the corresponding correction parameters is automatically triggered.

[0108] Furthermore, monitoring data include, for example: signal-to-noise ratio, multipath effect, satellite elevation angle, PDOP geometric precision factor, ionospheric TEC rate of change, tropospheric wet delay variation, reference station receiver performance, regional meteorological data, etc.

[0109] Furthermore, the neural network model uses a long short-term memory network, containing 3-5 hidden layers, each with 128-256 neurons. The specific number should be selected according to the actual situation.

[0110] Furthermore, the model takes multidimensional monitoring data and correction parameters as input and outputs predicted values ​​of the correction parameters. The training process is as follows: historical monitoring data and historical correction parameters are used as training and validation test sets; the input data is normalized to handle the time dependence of time series data; the Adam optimizer is used with an initial learning rate of 0.001; the training objective is set to minimize the mean square error between the predicted value and the actual correction parameter.

[0111] Furthermore, when the predicted correction parameter deviates from the current value by more than a set threshold, an update is triggered; based on the difference between the predicted value and the current value, an adjustment amount is calculated, which is applied to the correction parameter to generate new satellite clock bias, ionospheric grid, and tropospheric delay parameters;

[0112] Furthermore, the process of using a multi-timescale and multi-dimensional monitoring data fusion method to predict the timing and magnitude of correction parameter updates in the neural network model includes: real-time acquisition and normalization of multi-dimensional monitoring data to construct input feature sequences at multiple time scales (e.g., 1 second, 10 seconds, 60 seconds); employing a multi-scale LSTM model, where the input features at each time scale are processed through independent LSTM sub-modules, each sub-module containing an input gate, a forget gate, an output gate, and cell states to preserve the long-term dependencies of the time series; and assigning weights to the outputs at different time scales through the attention mechanism of the multi-scale feature fusion layer, with the weights dynamically adjusted based on the contribution of each scale feature to the correction parameters.

[0113] By introducing multi-scale time series processing and multi-scale feature fusion on the basis of neural network model, the adaptability of the model to rapid changes (such as ionospheric scintillation) and slow changes (such as tropospheric diurnal variation) can be enhanced, and it has higher prediction accuracy and robustness compared with single-scale LSTM.

[0114] By establishing a multi-parameter monitoring system, using neural networks to train nonlinear mappings and predict update timing, dynamic adaptation and automatic optimization of correction parameters are achieved, improving the robustness and long-term accuracy of the system, overcoming the limitations of traditional static models, and ensuring the continuous high-precision maintenance of spatiotemporal references.

[0115] Example 2:

[0116] This embodiment takes a real BeiDou high-precision positioning service network as an example to specifically apply the method proposed in this invention.

[0117] Collect BeiDou multi-frequency observation data and use the Kalman filter algorithm to preprocess the multi-frequency observation data;

[0118] The network consists of 15 BeiDou reference stations distributed within an area of ​​approximately 100 kilometers, and collected observation data from these stations at BeiDou B1I, B3I, and B1C frequencies. In the preprocessing step, a pseudorange gross error detection threshold of 3 times the standard deviation was set and applied to all frequencies. A signal-to-noise ratio threshold of 38 dB-Hz and a satellite elevation angle mask of 10° were also set to ensure that only high-quality observation data was used for subsequent processing.

[0119] Based on the preprocessed multi-frequency observation data, a time synchronization mechanism for the reference station network is established by utilizing the difference in the propagation time of BeiDou satellite signals. A multi-sensor fusion algorithm combined with the least squares method is used to integrate the spatial reference data of different reference stations, and a spatial interpolation algorithm is used to smooth the integration results.

[0120] The reference station A, with the highest time stability in the network, was selected as the reference station, and the time of the other 14 reference stations was aligned with it. During spatial data integration, based on the historical performance of each reference station, stations with high positioning accuracy (better than 3mm) were assigned a weight of 0.7, and stations with good stability (RMS less than 0.5mm) were assigned a weight of 0.6. These weights were then used for integration, and an inverse distance weighted interpolation algorithm was employed to generate spatial reference grid data. In the quality control phase, leave-one-out cross-validation was used to identify anomalous regions near the grid boundaries, and a 5×5 moving average filter window was used to smooth these regions.

[0121] Based on the spatiotemporal synchronization integration results, the optimal baseline combination is selected to establish the differential observation equations for BeiDou multi-frequency carrier phase and pseudorange. The LAMBDA algorithm is used for search and fixation, and a unified spatiotemporal reference model is established to achieve rapid convergence and integrated solution of spatiotemporal reference parameters.

[0122] When selecting the optimal baseline combination, the baseline length (preferably 20-40 km), the satellite geometric distribution intensity factor (PDOP) value (preferably less than 2.5), and data quality evaluation indicators (signal-to-noise ratio, multipath effect, etc.) were comprehensively considered. Ultimately, 10 baseline combinations meeting all optimal conditions were selected. In the spatiotemporal reference model solution, the LAMBDA algorithm was used to fix integer ambiguities. These fixed ambiguities were then used as constraints for integrated solution, achieving rapid convergence of spatiotemporal parameters.

[0123] Based on the solution results of the spatiotemporal reference model, high-precision receiver clock bias is obtained by using double-difference integer ambiguity constraints. BeiDou satellite clock bias, ionospheric grid and tropospheric delay correction parameters are generated. The correction parameters are automatically adjusted and dynamically updated by model algorithm according to satellite signal quality, geometric distribution and atmospheric delay changes.

[0124] Based on the calculation results, BeiDou satellite clock error correction parameters, regional ionospheric grids, and tropospheric delay correction parameters were generated. A multi-parameter monitoring system was used, and a neural network model with three hidden layers was trained on historical data to learn the nonlinear mapping relationship between the monitoring data and the correction parameters. This model can predict the optimal timing for updating the correction parameters. For example, when the predicted residual of the ionospheric grid exceeds 1.5 cm, the system automatically triggers an update, thereby ensuring high positioning accuracy for users within the region.

[0125] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic correction of spatiotemporal reference based on BeiDou, characterized in that, include: Collect BeiDou multi-frequency observation data and use the Kalman filter algorithm to preprocess the multi-frequency observation data; Based on the preprocessed multi-frequency observation data, a time synchronization mechanism for the reference station network is established by utilizing the difference in the propagation time of BeiDou satellite signals. A multi-sensor fusion algorithm combined with the least squares method is used to integrate the spatial reference data of different reference stations, and a spatial interpolation algorithm is used to smooth the integration results. Based on the spatiotemporal synchronization integration results, the optimal baseline combination is selected to establish the differential observation equations for BeiDou multi-frequency carrier phase and pseudorange. The LAMBDA algorithm is used for search and fixation, and a unified spatiotemporal reference model is established to achieve rapid convergence and integrated solution of spatiotemporal reference parameters. Based on the solution results of the spatiotemporal reference model, high-precision receiver clock bias is obtained by using double-difference integer ambiguity constraints. BeiDou satellite clock bias, ionospheric grid and tropospheric delay correction parameters are generated. The correction parameters are automatically adjusted and dynamically updated by model algorithm according to satellite signal quality, geometric distribution and atmospheric delay changes.

2. The method for dynamic correction of spatiotemporal reference based on BeiDou according to claim 1, characterized in that, The preprocessing of multi-frequency observation data includes: Gross errors were detected in the pseudorange observations at each frequency point, and outlier observations exceeding 3 times the standard deviation were removed. Carrier phase smoothing pseudorange is used to reduce the impact of multipath effects, and low-quality observation data is filtered out by signal-to-noise ratio threshold screening and satellite elevation angle mask setting; An adaptive recursive processing model based on observation error modeling and system noise estimation is established, and a recursive filtering algorithm is used to perform quality control and noise filtering on multi-frequency observation data.

3. The method for dynamic correction of spatiotemporal reference based on BeiDou according to claim 1, characterized in that, The process of establishing a time synchronization mechanism for a reference station network by utilizing the propagation time differences of BeiDou satellite signals, and integrating spatial reference data from different reference stations using a multi-sensor fusion algorithm combined with the least squares method, includes: Using pseudorange observation data of the same BeiDou satellite from various reference stations, the influence of satellite clock bias is eliminated through inter-station single difference processing, and the inter-station time deviation is calculated in combination with the known precise coordinates of the reference stations; Establish a time synchronization model for the reference station network, take the reference station with the best time stability as a reference, calculate the time correction of other reference stations relative to the reference station, and achieve time unification of the reference station network. Based on the time-unified observation data, Kalman filtering and weighted least squares algorithm are used to determine the weights by combining the positioning accuracy and stability of each reference station, and the spatial reference data of different reference stations are integrated.

4. The method for dynamic correction of spatiotemporal reference based on BeiDou according to claim 1, characterized in that, The process of smoothing the integrated result using a spatial interpolation algorithm includes: Based on the integrated base station spatial reference data, a regular grid spatial reference data is generated using an inverse distance weighted interpolation algorithm. Establish an interpolation quality control mechanism, calculate interpolation accuracy and residual distribution through leave-one-out cross-validation, and identify abnormal interpolation regions; For the abnormal regions identified by interpolation quality control, a moving average filtering algorithm is used for local smoothing.

5. The method for dynamic correction of spatiotemporal reference based on BeiDou according to claim 1, characterized in that, The process of selecting the optimal baseline combination includes: The comprehensive evaluation coefficient of each baseline is calculated based on the distance between reference stations, the satellite geometric distribution intensity factor (PDOP) value, and the observation data quality evaluation index. By combining the comprehensive evaluation coefficient and the baseline length, the optimal baseline combination is selected, and the differential observation equations for BeiDou multi-frequency carrier phase and pseudorange are established based on the optimal baseline combination.

6. The method for dynamic correction of spatiotemporal reference based on BeiDou according to claim 1, characterized in that, The process of establishing a spatiotemporal reference model includes: A joint observation equation set for multi-frequency carrier phase and pseudorange is established based on the differential observation equation; The LAMBDA algorithm is used to search and fix the integer ambiguity in the carrier phase observation equation; By using fixed integer ambiguities as equality constraints, a spatiotemporal reference model is established to achieve rapid convergence and integrated solution of spatiotemporal reference parameters.

7. The method for dynamic correction of spatiotemporal reference based on BeiDou according to claim 1, characterized in that, The generation process of BeiDou satellite clock bias, ionospheric grid, and tropospheric delay correction parameters includes: Using fixed double-difference integer ambiguity as a strong constraint, the constrained least squares method is adopted to simultaneously calculate the high-precision receiver clock error and coordinates of the reference station. Based on the calculated high-precision receiver clock bias, BeiDou satellite clock bias correction parameters are generated through clock bias fitting and prediction models. Regional ionospheric grids and tropospheric delay correction parameters are established and generated by using carrier phase geometry-independent combinations and tropospheric delay estimates, respectively.

8. The method for dynamic correction of spatiotemporal reference based on BeiDou according to claim 1, characterized in that, The process of automatically adjusting and dynamically updating correction parameters using model algorithms includes: Establish a multi-parameter monitoring system to collect multi-dimensional monitoring data that affect the correction parameters in real time; A neural network algorithm is used to train deep learning on historical monitoring data and correction parameter change data to establish a nonlinear mapping relationship between monitoring data and correction parameters; The trained neural network model is used to predict the optimal update timing and numerical adjustment range of various correction parameters. When the prediction results show that the parameters deviate from the threshold, the corresponding correction parameters are automatically updated.

Images