High-precision data processing method for GNSS data quality in complex environments

Abstract

SOLUTION: Provided is a complex environment GNSS data quality-oriented high-precision data processing method which includes the steps of: preprocessing complex environment GNSS data (S1); searching for large errors in the complex environment GNSS data (S2); searching for the complex environment GNSS data (S21); identifying large errors in GNSS data (S22); processing large errors in the complex environment GNSS data (S3); removing the large errors in the complex environment GNSS data (S31); processing the GNSS data by an interpolation method (S32); analyzing and calculating ambiguities (S4); evaluating the quality of the complex environment GNSS data (S5); fusing multi-source data (S6); performing adaptive adjustment of parameters (S7); handling abnormal situations (S8); and performing real-time monitoring and correction of the complex environment GNSS data (S9).EFFECT: When processing GNSS observation values of complex environments such as a narrow-space environment, strong reflections, and multi-frequency multi-systems, the combination of the mean-shift and dispersion-broadening concepts represents a significant improvement in large-error search and data processing.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to the technical field of data processing, in particular to a high-precision data processing method for complex environment GNSS data quality. [Background technology]

[0002] Global Navigation Satellite Systems (GNSS) include global and regional satellite navigation systems and augmentation systems, such as the US GPS, China's Beidou BDS, Russia's GLONASS, Europe's Galileo, and Japan's Quasi-Zenith System (QZSS). These systems provide positioning, navigation, and time signal services within global or regional scopes and are widely used in many fields. BDS is a satellite navigation system independently developed, operated, and managed by China. It has undergone three development stages: BDS-1, BDS-2, and BDS-3. BDS-3 has been implemented and will provide navigation and positioning services to users worldwide in 2020. This system has improved signal frequency points, the number of visible satellite signals, the quality of available observation data, and satellite geometry, further improving positioning accuracy. The global service level and altitude positioning accuracy can reach 10 meters and 5 meters in the Asia-Pacific region, and the system service availability is above 95%.

[0003] The application of GNSS systems is often affected by complex environments, such as large buildings, trees, and vehicles in urban areas that cause signal reflection and refraction, resulting in signal interference, interruption, and a reduction in the number of visible satellites, affecting satellite distribution. This generates a large number of large errors, which interfere with parameter estimation and affect the accuracy and reliability of GNSS positioning. As user needs for real-time, precision, and reliability of positioning continue to increase, how to effectively deal with large errors in complex conditions such as multi-frequency, multi-system, and urban areas has become a key issue.

[0004] Currently, two main methods are used to deal with gross errors: the mean-shift-based method and the variance-expansion-based method. The mean-shift-based method focuses on detecting outliers in the data and determines gross errors by identifying significant deviations between data points and the mean value. The variance-expansion-based method places more emphasis on data robustness and detects possible gross errors by estimating the variance of data points. The choice of method depends on the needs and environmental conditions of specific applications. For example, the mean-shift-based method may be more appropriate in some cases, while the variance-expansion-based method may be more effective in others. These methods aim to address the problem of gross errors in GNSS data processing and ensure the accuracy and reliability of positioning results. Summary of the Invention [Problem to be solved by the invention]

[0005] The technical problem to be solved by the present invention is to provide a high-precision data processing method for GNSS data quality in complex environments. Conventional methods have poor accuracy when processing complex data. The present invention combines the concepts of mean shift and variance expansion to achieve significant improvements over conventional methods in large error detection and data processing. Compared with other high-precision data processing algorithms, the method of the present invention has higher robustness and versatility in complex conditions, effectively solving the problem of complex data quality, especially when there are large errors in the observed values. This innovation enables more accurate and reliable positioning results to be obtained in various fields, satisfying the needs for high-precision data. [Means for solving the problem]

[0006] In order to solve the above technical problems, the technical means adopted in the present invention are: S1: Preprocessing of GNSS data; S2: Search for large errors in GNSS data, S21: GNSS data exploration, S22: Identification of large errors in GNSS data, S3: Processing of large GNSS data errors, S31: Removal of large errors in GNSS data, S32: GNSS data processing by interpolation method, S4: analytical calculation of ambiguity, S5: GNSS data quality assessment, S6: Multi-source data fusion, S7: Adaptive adjustment of parameters, S8: Handling abnormal situations, S9: A high-precision data processing method for complex environment GNSS data quality, which includes steps such as real-time monitoring and correction of GNSS data.

[0007] Preferably, in step S21, a global test is adopted, and T q Construct statistics,

number

number

number

number

number

[0008] Preferably, the features identified in step S22 are as follows: The test statistic w is calculated using the standardized residuals. i Build

number

number

number

[0009] After completing the removal and readjustment, the algorithm performs a global test using the statistics it initially constructed to assess whether other gross errors still exist in the data set. If the test results indicate that the data still contains gross errors, the algorithm repeats the identification and removal steps. This process is continuous, constantly iterating until the global test reveals that no gross errors exist in the data set, ensuring high accuracy and quality of data processing.

[0010] Preferably, step S31 includes the following steps: First, we analyze the signal trend term by introducing the first derivative, The threshold is calculated by Minimax law to separate the abnormal derivative value, and the empirical equation is used to achieve better results. In terms of threshold setting, the threshold calculation formula is as follows:

number

[0011] Preferably, step S32 includes the following steps: W in unit area e has r removal points and x r ∈[x e-1 ,x e ] and the expansion area W ′ e has s nodes in x s ∈[x′ e-1 ,x′ e ] is satisfied, In the extended domain, the approximate function y e Construct (x),

number

Number

[0012] Preferably, the step S4 includes the following steps. The CLAMBDA algorithm. Strengthen the GNSS model using a known baseline length and impose constraints. Here, l is the known baseline length, and the standard GNSS model is extended as follows.

Number

number

number

number

number

number

number

number

number

number

number

[0013] Preferably, step S5 includes: Root Mean Square Error (RMSE): Root Mean Square Error is the square root of the variance between the original signal and the signal after noise reduction. The smaller the value, the better the noise reduction quality.

number

number

number

number

number

[0014] Preferably, step S6 includes the following steps: Data alignment and calibration, A coordinate transformation is a process that aims to ensure that different data sources are in the same spatial reference system. This typically involves converting a geographic coordinate system, such as WGS-84, to a projected coordinate system, or converting between different geographic coordinate systems. The transformation formula can be determined according to the specific coordinate system transformation needs, and can be, for example, a transformation formula from geographic coordinates, latitude φ, longitude λ, to a Cartesian coordinate system (x,y,z). Time stamp alignment is the process of ensuring that data is consistent in time, and may involve converting the time stamps of the data to a uniform time standard, such as Universal Time (UT) or Coordinated Universal Time (UTC). Converting the time stamps can be accomplished by simple mathematical operations, such as adding or subtracting a specific time offset. It is a data fusion algorithm that can fuse two data sources and fuse the values ​​using the weighted average method. The calculation formula is as follows: Y=αX1+(1-α)X2 where X1 and X2 are the values ​​of the two data sources and α is a weighting coefficient.

[0015] Preferably, step S7 is specifically as follows: The parameter set of the algorithm is θ={θ1,θ2,…,θn} and the performance index is P, the goal of adaptive parameter adjustment is expressed as follows: maxP(θ|D,E) where D is the input data, E is the environmental conditions, The gradient descent method is used as a method for adaptively adjusting parameters, which is used to adjust the parameters to minimize the loss function. In the adaptive parameter adjustment, the parameters are updated according to the gradient of the loss function with respect to the parameters. θ new =θ old -α∇ θ J(θ) where α is the learning rate, J(θ) is the loss coefficient, and θ old represents the current parameter values, i.e., the parameter set used in the algorithm before parameter update, and θ new represents the updated parameter value.

[0016] Preferably, step S8 includes the following steps: The definition and scope of abnormal situations are clarified, and abnormal data are all data that do not meet the pre-defined criteria. Incorrect data format is the data item x i does not match format F, then x i can be defined as abnormal and expressed by the following formula:

number

number

[0017] This will be explained using formulas and algorithms. For numerical anomalies: Set thresholds α and β (α<β). Anomaly detection formula: anomaly (x i )=x i <α or x i<β. Abnormal processing: Adjustment (x i )=max(min(x i ,β),α). Consideration of diversity of abnormal situations: In case of data loss, an interpolation policy is adopted. For example, if the mean value μ is used as the interpolation formula,

number

[0018] The present invention provides a high-precision data processing method for complex environment GNSS data quality, and has the following beneficial effects: 1. High-precision data processing algorithm: This invention proposes a high-precision processing algorithm aimed at improving the quality of GNSS data in complex environments such as high mountain ranges. Through steps such as data preprocessing, large error detection, large error processing, and real-time monitoring, the accuracy and quality of complex data are significantly improved. 2. Robustness and versatility: The algorithm of the present invention has higher robustness and versatility under complex conditions, can solve various data quality problems, and is particularly effective when high-precision data is required. 3. Combining the principles of mean shift and variance expansion: The present invention combines the concepts of mean shift and variance expansion in the aspects of gross error search and data processing, which has better performance when dealing with complex data quality problems compared with traditional methods. 4. Gross error detection method: The present invention uses global testing, statistical construction, chi-square testing and other means to perform gross error detection, thereby improving the accuracy of gross error identification. 5. Large error processing policy: The present invention uses interpolation to process large errors, including large error removal and large error processing by interpolation, which effectively solves the problems of data interruption and missing data. 6. Real-time monitoring and correction: The present invention achieves real-time monitoring and correction by periodically checking the statistical properties of the data and comparing it with known true values ​​or reference station data, ensuring the accuracy and reliability of the data. [Brief explanation of the drawings]

[0019] [Figure 1] 1 is an overall flowchart of the present invention. [Figure 2] 1 is a flowchart of gross error processing according to the present invention. [Figure 3] FIG. 1 is a diagram of the expansion region and corresponding function values ​​of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0020] The high-precision data processing method for complex environment GNSS data quality, as shown in Figure 1, includes the following steps:

[0021] Step 1: Preprocessing of GNSS data. First, in data sampling, an appropriate sampling frequency must be determined to ensure that the density of data points is high enough to meet the needs of subsequent data processing. The selection of the sampling frequency must take into account factors such as the system bandwidth and computing resources.

[0022] Second, data correction aims to deal with common problems in the data, such as clock errors, ionospheric delays, etc. These problems can lead to data deviations and need to be corrected to ensure data accuracy.

[0023] Finally, data filtering aims to remove high-frequency noise and instability in the data to obtain smoother and more stable data. The choice of filtering algorithm should be determined based on the characteristics of the data and your needs.

[0024] Step 1 above is completed by computer processing.

[0025] Step 2: Search for gross errors in GNSS data. The core of the present invention is the application of a gross error search algorithm, which, based on the principles of mean shift and variance expansion, can identify potential gross error points from the data, i.e., data points that are obviously inconsistent with the surrounding data points.

[0026] In a specific implementation, potential gross error points can be determined by first calculating the mean and variance of each data point and its surrounding data points, and then comparing the differences between each data point and its surrounding data points. Data points whose differences exceed a preset threshold are marked as potential gross error points.

[0027] The specific steps are as follows: 1) Exploration The overall test was adopted, and T q Construct statistics,

number

number

[0028] 2)Identification Using the standardized residuals, w i Test statistic w i Build

number

number

number

[0029] Step 2 above is completed by computer processing.

[0030] Step 3: Processing large GNSS data errors. Data Removal: The steps of the method for removing gross errors are shown in FIG. (1) First, the first derivative is introduced to analyze the signal trend term. After calculating the first derivative of the signal, the wavelet coefficients similar to the high scale are obtained. Based on the idea of ​​wavelet thresholding, traditionally, the wavelet thresholding function filters out the low amplitude wavelet coefficients in the signal. Currently, the first derivative points greater than the threshold are considered to be large error points. (2) The threshold is determined by the minimax law of wavelet threshold to separate the abnormal derivative value, where an empirical equation is used. In order to achieve better results, it can be expanded in terms of threshold setting, which is the same as the design of wavelet threshold, and marks the abnormal points, and most of the abnormal points are removed here. The threshold calculation formula is as follows:

number

[0031] Large error processing by interpolation: After the gross errors are removed, signal interruptions and data gaps will occur in the initial signal, and in order to make the monitoring data continuous, an interpolation operation is required. In this paper, a generalized extension interpolation method is used to complete the data. The generalized extension incorporates the advantages and characteristics of existing interpolation and fitting methods, adopts a piecewise smoothing approach, and uses the extension region to construct a unit region fitting function, locking the unit region boundary nodes to achieve the best approximation effect.

[0032] First, we define the set segments to be removed as unit areas W e and extend it to a predetermined length on both sides to form an expanded region W' e Let's say.

[0033] W in unit area e has r removal points and x r ∈[x e-1 ,x e ] and the extension domain W′ e has s nodes in xs ∈ [x′ e-1 , x′ e is satisfied and as shown in Figure 3, Construct an approximation function y e (x) in the extended region,

Mathematics

[0034] Construct a generalized extended approximation interpolation model,

Mathematics

[0035] In the unit region W e , data is missing due to removing large errors, and it is necessary to perform fitting on the entire extended region W′ e using the data outside the unit region to record the numerical values of each missing point within the unit region. Based on this, ensure that all points outside the unit region are determined, and all points in the extended region are repeatedly iterated by the generalized extension model until the stable value of the missing point is obtained.

[0036] The above step 3 is completed by computer processing.

[0037] Step 4: Analysis and calculation of ambiguity. It is the principle of the CLAMBDA algorithm, Augment and constrain the GNSS model using a known baseline length, where l is the known baseline length. Thus, the standard GNSS model is extended as follows:

number

number

number

[0038] 2. Search method In an actual on-board attitude determination experiment, the baseline length is measured in advance, so the following formula is used:

number

number

number

number

[0039] Step 4 above is completed by computer processing.

[0040] Step 5: Assessing complex environment GNSS data quality. Root mean square error: Root mean square error is the square root of the variance between the original signal and the signal after noise reduction, and the smaller the value, the better the noise reduction quality.

number

[0041] Signal-to-noise ratio: The signal-to-noise ratio is the ratio between the signal power and the noise power. The larger the signal-to-noise ratio, the less noise is mixed into the signal, and the better the noise reduction quality.

number

[0042] Correlation coefficient: Correlation coefficient is a quantity that studies the degree of linear correlation between variables. It is a statistical indicator that reflects the closeness of the correlation between variables. It refers to the similarity between the original signal including the noise-containing signal and the signal after noise reduction. The larger the value, the better the noise reduction quality. It is calculated by the quotient of the covariance and standard deviation of the original signal and the signal after noise reduction.

number

[0043] Smoothness: Smoothness is the ratio of the cumulative variance of the first-order differential of the noise-reduced signal to the first-order differential of the original signal. The smaller this value is, the better the noise reduction effect is.

number

[0044] Traditional noise reduction quality evaluation indices are based on statistical principles and have significant limitations. They can only accurately judge noise reduction quality when the true value is known. However, in actual deformation monitoring data processing, the true value is generally unknown. Therefore, any single evaluation index cannot meet the needs of quality assessment, and different indices may even contradict each other during the evaluation process. Therefore, new evaluation indices are needed to guide noise reduction. To this end, scholars have proposed the composite evaluation index T and the denoise signal-to-noise ratio (dnSNR). The former comprehensively considers signal details and approximation information, and combines two anti-correlated indices, root mean square error and smoothness, with different weights to form T, which is theoretically rigorous. The latter takes into account the fact that the true value is unknown in actual deformation monitoring, making it difficult to calculate the signal-to-noise ratio.

[0045] Composite index evaluation: When selecting a single index to evaluate noise reduction quality, the noise reduction effect evaluation using various indexes may not match, and may even be contradictory. The comprehensive evaluation index T comprehensively considers two indexes, root mean square error and smoothness, and accumulates them by weighting them with the coefficient of variation. When T has a minimum value, the noise reduction quality is the best. Here, W RMSE and W r is the variable weight. T=W RMSE ×RMSE+W r ×r

[0046] Noise reduction error ratio: In actual deformation monitoring, both the useful components of the signal and the noise are unknown, and the signal-to-noise ratio cannot be calculated. Therefore, the noise reduction error ratio is introduced and defined as the ratio of the noise-containing signal power to the noise-removed signal power.

number

[0047] Step 5 above is completed by computer processing.

[0048] Step 6: Multi-source data fusion. Data alignment and calibration: Coordinate transformation: Coordinate transformation aims to ensure that different data sources are in the same spatial reference system. This generally involves converting a geographic coordinate system, such as WGS-84, to a projected coordinate system, or converting between different geographic coordinate systems. Transformation formulas can be determined based on specific coordinate system transformation needs, such as the transformation from geographic coordinates, latitude φ, longitude λ, to a Cartesian coordinate system (x,y,z).

[0049] Time Stamp Alignment: Time stamp alignment is the process of ensuring that data is consistent in time, which may involve converting the timestamps of the data to a uniform time standard, such as Universal Time (UT) or Coordinated Universal Time (UTC). Converting the timestamps can be accomplished by simple mathematical operations, such as adding or subtracting a specific time deviation.

[0050] Data fusion algorithm: This is the core of multi-source data fusion, and involves various data fusion techniques, such as weighted average, Kalman filtering, neural network, etc. For example, when fusing two data sources, the values ​​can be fused using the weighted average method, and the calculation formula is as follows: Y=αX1+(1-α)X2 where X1 and X2 are the values ​​of the two data sources and α is a weighting coefficient.

[0051] Step 6 above is completed by computer processing.

[0052] Step 7: Adaptively adjust parameters. The core idea of ​​adaptive parameter adjustment is that an algorithm can automatically adjust its parameters due to changes in input data or changes in external environmental conditions, with the goal of achieving better performance in specific conditions, for example, reducing error and improving accuracy or response speed.

[0053] The parameter set of the algorithm is θ={θ1,θ2,…,θ n} and the performance index is P, the goal of adaptive parameter adjustment can be expressed as: maxP(θ|D,E) where D is the input data and E is the environmental condition.

[0054] Methods for adaptive parameter adjustment: Gradient descent is a commonly used optimization algorithm that is used to adjust parameters to minimize a loss function, and adaptive parameter adjustment can update the parameters according to the gradient of the loss function with respect to the parameters. θ new =θ old -α∇ θ J(θ) where α is the learning rate and J(θ) is the loss function.

[0055] Step 7 above is completed by computer processing.

[0056] Step 8: Handling abnormal situations. Clarify the definition and scope of abnormal conditions. Abnormal data are all data that do not meet the pre-defined criteria, and incorrect data formatting is a data item x i does not match format F, then x i can be defined as abnormal and expressed by the following formula:

number

[0057] The processing mechanism is described in detail. For malformed data, the processing procedure may include detection, recording, automatic correction or marking. The expression is

number

[0058] This will be explained using formulas and algorithms. For numerical anomalies: Set thresholds α and β (α<β). Anomaly detection formula: anomaly (x i )=x i <α or x i <β. Abnormal processing: Adjustment (x i )=max(min(x i ,β),α)). Consideration of diversity of abnormal situations: In case of data loss, an interpolation policy is adopted.

[0059] For example, if the mean value μ is used as the interpolation formula,

number

[0060] Step 8 above is completed by computer processing.

[0061] Step 9: Real-time monitoring and correction of complex environment GNSS data. Monitoring can be achieved by periodic examination of the statistical properties of the data and comparison with known true values ​​or reference station data. If anomalies are discovered, corresponding corrective measures can be taken, such as data re-correction or gross error search threshold adjustment.

[0062] Step 9 above is completed by computer processing.

[0063] The above embodiments are merely preferred technical embodiments of the present invention and should not be considered as limiting the present invention, and the scope of protection of the present invention should be the technical embodiments set forth in the claims and the equivalent replacement modes of the technical features in the technical embodiments set forth in the claims. That is, the equivalent replacement and improvement within this scope is also included in the scope of protection of the present invention.

[0064] (Addendum) (Appendix 1) S1: Preprocessing of GNSS data; S2: Search for large errors in GNSS data, S21: GNSS data exploration, S22: Identification of large errors in GNSS data, S3: Processing of large GNSS data errors, S31: Removal of large errors in GNSS data, S32: GNSS data processing by interpolation method, S4: analytical calculation of ambiguity, S5: GNSS data quality assessment, S6: Multi-source data fusion, S7: Adaptive adjustment of parameters, S8: Handling abnormal situations, S9: Includes steps such as real-time monitoring and correction of GNSS data. In step S21, a global test is adopted, and T q Construct statistics,

number

number

number

number

number

number

number

number

[0065] (Appendix 2) The step S31 includes the following steps: First, we analyze the signal trend term by introducing the first derivative, The threshold is calculated by Minimax law to separate the abnormal derivative value, and the empirical equation is used to achieve better results. In terms of threshold setting, the threshold calculation formula is as follows:

number

[0066] (Appendix 3) The step S32 includes the following steps: W in unit area e has r removal points and x r ∈[x e-1 ,x e ] and the expansion area W ′ e has s nodes in x s ∈[x′ e-1 ,x′ e ] is satisfied, In the extended domain, the approximate function y e Construct (x),

number

Number

[0067] (Supplementary Note 4) The step S**4** includes the following steps, the CLAMBDA algorithm, strengthens the GNSS model using the known baseline length and applies constraints. Here, l is the known baseline length, and the standard GNSS model is extended as follows,

Number

Number

number

number

number

number

number

number

number

number

number

[0068] (Appendix 5) Step S5 includes the following: Root mean square error RMSE:

number

number

number

number

number

[0069] (Appendix 6) Step S6 Data alignment and calibration; coordinate transformations aimed at ensuring that different data sources are in the same spatial reference system; timestamp alignment, which is the process of ensuring that data is consistent in time; The method includes the steps of: fusing two data sources, fusing the values ​​of the two data sources by a weighted average method, and using a data fusion algorithm whose calculation formula is as follows: Y=αX1+(1-α)X2 10. The high-precision data processing method for complex environment GNSS data quality according to claim 1, wherein X1 and X2 are values ​​of two data sources, and α is a weighting coefficient.

[0070] (Appendix 7) Specifically, step S7 is as follows: The parameter set of the algorithm is θ={θ1,θ2,…,θ n} and the performance index is P, the goal of adaptive parameter adjustment is expressed as follows: maxP(θ|D,E) where D is the input data, E is the environmental conditions, The gradient descent method is used as a method for adaptively adjusting parameters, which is used to adjust the parameters to minimize the loss function. In the adaptive parameter adjustment, the parameters are updated according to the gradient of the loss function with respect to the parameters. θ new =θ old -α∇ θ J(θ) where α is the learning rate, J(θ) is the loss function, and θ old represents the current parameter values, i.e., the parameter set used in the algorithm before parameter update, and θ new represents the updated parameter value.

[0071] (Appendix 8) The step S8 includes the following steps: The definition and scope of abnormal situations are clarified, and abnormal data are all data that do not meet the pre-defined criteria. Incorrect data format is the data item x i does not match format F, then x i can be defined as an abnormality, and is expressed by the following formula:

number

number

Claims

1. S1: Preprocessing of GNSS data; S2: Search for large errors in GNSS data; S21, a sub-step of S2: Searching for GNSS data; S22, a sub-step of S2: Identification of large errors in GNSS data; S3: Processing of large errors in GNSS data; S31, a sub-step of S3: Removal of large errors in GNSS data; S32, a sub-step of S3: GNSS data processing using a generalized extended interpolation method; S4: analytical calculation of ambiguity, S5: Evaluation of GNSS data quality; S6: Fusion of multi-source data, which is GNSS data from different data sources; S7: Adaptive adjustment of parameters used in the algorithm; S8: Handling abnormal situations including erroneous data and data format errors; S9: Real-time monitoring of GNSS data and data re-correction or gross error search threshold adjustment steps; In step S21, a hypothesis test is adopted, and T ｑ Construct statistics, [Equation 1] however, [Equation 2] is the residual of the observation, which is the difference between the actual observation value l and the GNSS observation model predicted value, where the actual observation value l is the GNSS data, [Equation 3] and [Equation 4] is a vector [Equation 5] represents the transformation or processing of the matrix A, q is the number of redundant observations for the actual observation l, and Q ｕ －１ is the variance matrix of the observations, and the statistic T ｑ A hypothesis test was performed using the chi-square test method, and the significance level was α １ Set T ｑ >χ α１ ２ In the case of T q ≦χ α1 2 (q, 0), the hypothesis that the observed value contains a large error is correct, and the process proceeds to S22. In the case of T q ≦χ α1 2 (q, 0), the null hypothesis that the observed value does not contain a large error is correct, and the GNSS data search is stopped and the process proceeds to S3, where χ α１ ２ (q, 0) is for determining gross errors in the GNSS data, In step S22, the test statistic w i is used to identify whether the i-th observation is a large error as follows: Using the standardized residuals, the test statistic w ｉ Build [Equation 6] However, c ｉ = [0, ..., 0, 1, 0, ..., 0] Ｔ is an m × 1 observation vector, where m is the total number of observations, [Equation 7] represents the residual cofactor matrix, [Equation 8] where the test statistic is a standard normal distribution, i.e., w ｉ ~N(0,1), and |w ｉ |>U （１／２）α１ In the case of (0,1), U （１／２）α１ A high-precision data processing method for complex environment GNSS data quality, characterized in that (0, 1) is determined to be a threshold value of a standard normal distribution having a significance level.

2. The step S31 includes the following steps: First, we introduce the first derivative of the GNSS observation signal with respect to time to analyze the signal trend term, which is the signal and noise term; The threshold value λ for determining large errors is obtained by the Minimax method, and abnormal derivative values ​​greater than the threshold value λ are separated from the first-order time derivative of the GNSS observation signal. The threshold value λ is calculated using an empirical equation. To achieve better results, a logarithmic term of N is added. The calculation formula for the threshold value λ, which is the empirical equation, is as follows: [Equation 9] where σ represents the noise level of the GNSS data, N represents the total number of observations in the GNSS data large error removal or the number of samples in the GNSS data set in the GNSS data large error removal, After removing the large errors, the GNSS data is subjected to 3σ removal once. The high-precision data processing method for complex environment GNSS data quality as described in claim 1, characterized in that finally, large errors are removed from the GNSS data using a wavelet analysis means, a low-frequency trend term is obtained by wavelet decomposition, and a residual is obtained by calculating the difference between the signal before the large errors are removed and the low-frequency trend term.

3. In step S31, after completing the removal of the large errors, a hypothesis test is performed to evaluate whether there are other large errors in the set of GNSS data, and if the test result shows that the GNSS data contains large errors, the removal is repeated; The step S32 includes the following steps: Let the continuous data interval having r removal points be the unit volume W e , and let x ｒ ∈[x ｅ－１ , x ｅ ], and has s nodes. The effective data section obtained by extending the unit area W e is defined as an extended area W ′ e . ｓ ∈[x′ ｅ－１ , x′ ｅ ] is satisfied, In the extended domain, the approximate function y ｅ Construct (x), [Equation 10] However, x ｅ－１ is the set of GNSS data from which large errors have been removed in S31 before the e-1th removal operation, and x ｅ is the set of GNSS data after the e-th removal operation, and x′ ｅ－１ In the above, x' represents the GNSS data that has undergone data preprocessing in S1, e-1 represents the GNSS data after the e-1th iteration and operation, and x' ｅ represents the GNSS data after the e-th iteration and operation, a ｊ are undetermined coefficients, j = 1, 2, ..., v, and g ｊ Is W ｅ where j=1, 2, ..., v, v is the number of terms in the approximate function, and satisfies r<v<s. When solving for the undetermined coefficients, the value of x is set to [x' ｅ－１ , x′ ｅ ] and construct a generalized extended approximate interpolation model. [0011] where I(a １ , a ２ , ..., a ｊ 2. The high-precision data processing method for complex environment GNSS data quality according to claim 1, wherein the error between the approximation function and the prior value is the error between the approximation function and the prior value.

4. Step S4 includes the following steps: Step S4 employs the CLAMBDA algorithm; Augmenting and constraining the GNSS observation model with a known baseline length, where l is the known baseline length, by extending the GNSS observation model as follows: [0012] where E(y) is the expectation function, which represents the expected value of the random variable y; D(y) is the variance function, which represents the variance of the random variable y; and Q ｙｙ is the covariance matrix, and Q ｙｙ describes the linear relationship between the random variable y and itself, a is the unknown parameter vector, which needs to be estimated by GNSS data, and Z ｎ is the integer part of the unknown parameter vector a, b represents the constant term vector in the equation, and R ｐ represents the weight matrix, The minimization problem is converted into a quadratically constrained integer least squares problem, and the solution obtained by orthogonal decomposition is as follows: [0013] however, [0014] represents the estimate or solution for the parameters a, i.e., the optimal estimate of the unknown parameter vector a obtained by solving a least squares problem, which estimate is obtained by minimizing the sum of squares of the differences between the observed values ​​and the model predictions; [Equation 15] represents an estimate of the constant term b, and in the least squares problem, [0016] is the parameter vector [Equation 17] Together, we construct an optimal solution for the GNSS observation model, The objective function can be expressed as follows: [Equation 18] The search method is as follows: [Equation 19] The solution to the problem is as follows: [Equation 20] where: [Equation 21] and The solution to the approximate constrained integer least squares ambiguity is [Equation 22] The high-precision data processing method for complex environment GNSS data quality according to claim 1,

5. Step S5 includes six evaluations of GNSS data quality, including root mean square error (RMSE), signal-to-noise ratio (SNR), correlation coefficient (R), smoothness (r), composite index evaluation (T), and noise reduction error ratio (dnSNR); Root mean square error RMSE: [Equation 23] where x(t) represents the initial noise-containing signal preprocessed in S1, i.e., the GNSS data at time t, and contains noise components; x'(t) represents the signal obtained by performing noise reduction processing on the GNSS data interpolated in S32 and then performing ambiguity analysis calculation in S4; that is, the signal at time t has already had some noise removed by the noise reduction algorithm; Signal-to-noise ratio SNR: [0000] Correlation coefficient R: [Equation 25] Smoothness r: [Equation 26] Composite index evaluation T: T=W ＲＭＳＥ ×RMSE+W ｒ ×r Noise reduction error ratio dnSNR: [0000] Here, p ｓ＋ｎ is the noise-containing signal power, and p ｍ is the noise-removed signal power, where x'(t) is the signal obtained by performing noise reduction processing on the GNSS data interpolated in S32 and then performing ambiguity analysis calculation in S4, and x(t) is the initial noise-containing signal preprocessed in S1, as described in claim 1.

6. Step S6 GNSS data alignment and calibration; a coordinate transformation for converting GNSS data sources that exist in different coordinate systems into a uniform spatial reference system; timestamp alignment, which is the process of aligning the time base of GNSS data; The method includes the steps of: fusing two GNSS data sources, and fusing the values ​​by a weighted average method with a GNSS data fusion algorithm, the calculation formula of which is as follows: Y=αX １ + (1-a) X ２ Here, X １ and X ２ The high-precision data processing method for complex environment GNSS data quality as claimed in claim 1, characterized in that: α is the value of two GNSS data sources, and α is a weighting coefficient.

7. Specifically, step S7 is as follows: The parameter set of the algorithm is θ = {θ １ , θ ２ , …, θ ｎ } and the performance index is P, the goal of adaptively adjusting the parameters that affect the accuracy of processing GNSS data is expressed as follows: maxP(θ|D,E) where D is the input data, which is the GNSS data preprocessed in S1, and E is the environmental condition. The gradient descent method is used as a method for adaptively adjusting parameters, which is used to adjust the parameters to minimize the loss function. In the adaptive parameter adjustment, the parameters are updated according to the gradient of the loss function with respect to the parameters. i ｎｅｗ =θ ｏｌｄ -α∇ θ J (the) where α is the learning rate, J(θ) is the loss function, and θ ｏｌｄ represents the current parameter values, i.e., the parameter set used in the algorithm before parameter update, and θ ｎｅｗ The high-precision data processing method for complex environment GNSS data quality according to claim 1, characterized in that: represents the updated parameter value.

8. Step S8 includes the following steps: The abnormal situations are defined as abnormal data, data format errors, and data loss. The abnormal data is all GNSS data that does not meet the preset criteria. For data format errors, the data item x ｉ x if does not match format F ｉ can be defined as an abnormality due to an error in the data format, and the abnormality due to an error in the data format can be expressed by the following formula: [0000] The high-precision data processing method for complex environment GNSS data quality as claimed in claim 1, characterized in that for GNSS data with an incorrect data format, the procedure for processing abnormal data includes detecting, recording, automatically correcting or marking the abnormal data.

Images