Satellite positioning error optimization method based on big data analysis

Abstract

The invention discloses a satellite positioning error optimization method based on big data analysis, and the method comprises the steps: S1, obtaining multi-source historical data, and forming a unified data set; s2, performing exception elimination, interpolation completion and standardization on the data, and constructing a multi-dimensional feature vector; s3, dividing the positioning error into four channels of an ionosphere, a troposphere, a multipath and an equipment drift error, and respectively establishing improved autoregressive mobile diffusion models; s4, performing parameter estimation by using batch training and a robust loss function; s5, constructing a geographic diffusion diagram of space grid nodes, determining an edge weight according to the similarity between the nodes, and taking the edge weight as a regular constraint; s6, inputting real-time satellite observation and environment characteristics, calculating error prediction values of all channels, and performing weighted fusion to obtain total error correction and uncertainty; and S7, outputting an optimized positioning coordinate with a confidence interval. According to the method, the satellite positioning precision, stability and anti-interference performance are remarkably improved.

Description

Technical Field

[0001] This invention relates to the fields of big data analysis and satellite navigation and positioning technology, and in particular to a satellite positioning error optimization method based on big data analysis. Background Technology

[0002] With the rapid development of applications such as high-precision positioning, autonomous driving, geographic information acquisition, and disaster monitoring, the requirements for positioning accuracy and reliability of satellite navigation systems are constantly increasing. To eliminate errors generated during satellite signal propagation and reception, existing research generally employs methods such as differential positioning, model correction, or empirical compensation to model and correct major error sources such as ionospheric delay, tropospheric delay, multipath effects, and receiver drift. However, in complex real-world environments, satellite signal propagation paths are influenced by multiple factors, including meteorology, ionospheric activity, topography, and building obstruction, resulting in various errors exhibiting strong nonlinearity, time-varying characteristics, and spatial correlation. Traditional single-channel models or independent error correction methods are insufficient to accurately characterize these complex spatiotemporal coupling properties.

[0003] Most existing satellite positioning error optimization methods are based on empirical parameters or static models, such as estimating delay using dual-frequency ionospheric observations, correcting delay based on empirical tropospheric formulas, and reducing multipath interference through geometric path analysis. These methods typically assume that the error terms are independent and time-stable, neglecting the dynamic correlation between multi-source environmental features. This leads to a significant decrease in error correction effectiveness in areas with complex terrain or frequent electromagnetic disturbances. Furthermore, existing models generally rely on a small number of ground reference stations or single-type observation data, making it difficult to effectively integrate meteorological monitoring, ionospheric detection, and geographic environmental information, resulting in limited model input dimensions and insufficient spatial resolution.

[0004] In terms of data processing, traditional error modeling processes mainly employ single-source time series analysis or simple statistical smoothing methods, lacking a unified management and spatiotemporal alignment mechanism for multi-source heterogeneous data. Different sampling frequencies, observation periods, and time reference differences across sensing systems often lead to data misalignment, inconsistent features, and missing samples. Existing compensation strategies for these problems mostly employ linear interpolation or mean filling, which struggles to maintain the dynamic continuity and abrupt changes of the error signal, thus affecting the stability and accuracy of subsequent modeling. Furthermore, during model training, conventional least-squares regression or mean squared error optimization functions are highly sensitive to anomalous noise samples, easily leading to parameter shifts or overfitting.

[0005] Regarding the handling of spatial correlation, most existing error correction models only perform spatial smoothing based on point-to-point distance or regional averages, failing to consider the non-uniform propagation effects caused by geographical environment, building occlusion, and terrain differences. They also lack spatial constraint mechanisms during the training phase, leading to unstable error propagation and poor spatial consistency in boundary areas or with non-uniform node distributions. Overall, existing satellite positioning error correction methods generally suffer from poor data synchronization, weak model adaptability, insufficient spatial constraints, and a lack of uncertainty assessment, making it difficult to achieve high-precision, stable real-time positioning optimization in complex multi-source environments.

[0006] Therefore, how to provide a satellite positioning error optimization method based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose a satellite positioning error optimization method based on big data analysis. This invention constructs an improved autoregressive motion diffusion model and introduces spatial regularization constraints to achieve unified modeling of multi-source satellite observations and environmental characteristics. It utilizes a robust loss function and adaptive training mechanism to improve the stability of parameter estimation, and combines geographic diffusion maps to achieve spatial smoothing and error consistency correction. Finally, it outputs optimized positioning results with confidence intervals, effectively improving the accuracy and reliability of satellite positioning in complex environments.

[0008] A satellite positioning error optimization method based on big data analysis according to an embodiment of the present invention includes the following steps:

[0009] S1. Acquire historical data from multiple sources and perform time synchronization and spatial mapping to form a unified dataset;

[0010] S2. Perform anomaly removal, interpolation completion, and feature standardization on the unified dataset to construct a multidimensional feature vector;

[0011] S3. The positioning error is divided into four channels: ionospheric error, tropospheric error, multipath error and equipment drift error. An improved autoregressive motion-diffusion model is established for each channel. The model includes an autoregressive term, a moving average term, a drift term and a diffusion term.

[0012] S4. Parameter estimation is performed using batch training and a robust loss function to obtain the parameter set for each error channel;

[0013] S5. Construct a geographic diffusion map containing spatial grid nodes, determine edge weights based on the similarity of building occlusion and terrain between nodes, and use the spatial association relationship of the geographic diffusion map as a regularization constraint;

[0014] S6. Input real-time satellite observations and environmental characteristics, call the improved autoregressive mobile diffusion model to calculate the error prediction values ​​of each channel and perform weighted fusion to obtain the overall error correction amount and uncertainty;

[0015] S7. Based on the error correction amount and uncertainty, output the optimized positioning coordinates with confidence interval.

[0016] Optionally, step S1 includes:

[0017] S11. Collect multi-source historical data, including satellite observation residual data, pseudorange observation data, carrier phase observation data, satellite azimuth and elevation data, receiver latitude and longitude and elevation data, meteorological parameter data, ionospheric electron content data, and terrain and building obstruction data.

[0018] S12. Determine the spatial position coordinates of each satellite at the observation time based on the satellite ephemeris information, calculate the geometric distance data between the satellite and the receiver, and establish the spatial association between the satellite and the receiver;

[0019] S13. Perform time synchronization processing on data from satellite receivers, meteorological monitoring systems, ionospheric monitoring systems and geographic information databases, and align data from different sampling frequencies to the same time series using satellite observation time as a unified benchmark.

[0020] S14. Perform spatial mapping processing on the meteorological parameter data, total ionospheric electron content data, terrain and building occlusion data, project them onto a unified geographic reference coordinate system, and calculate the environmental characteristic parameters of the corresponding location based on the receiver's latitude, longitude and elevation.

[0021] S15. The time-synchronized satellite observation data and the spatially mapped environmental feature data are jointly registered according to the time index and spatial coordinates to form a unified dataset containing time, location, observation features and environmental features.

[0022] Optionally, step S2 includes:

[0023] S21. Perform integrity screening on the unified dataset and remove data records with a data missing rate exceeding a preset threshold.

[0024] S22. Establish independent data channels for different types of feature data in the unified dataset, construct time index sequences in each data channel, calculate the time interval and rate of change of adjacent observations, determine the location of abnormal observation points based on the time series continuity detection results, and use the sliding window interpolation method to complete the data of the abnormal points.

[0025] S23. Perform statistical smoothing on the interpolated data sequence to eliminate short-term fluctuations introduced by observation noise;

[0026] S24. Standardize the characteristic data of meteorological parameters, ionospheric electron content, satellite azimuth, elevation, receiver latitude and longitude and elevation of different dimensions, and map each feature to a unified scale of zero mean and unit variance.

[0027] S25. In the standardized feature space, verify the consistency of data from each data channel at the same observation time according to the unified time index, and delete data records that do not meet the time alignment requirements.

[0028] S26. After anomaly removal, interpolation completion, smoothing and standardization, and passing the consistency check, the data are renumbered and multi-dimensional feature vectors with observation features and environmental features as components are generated in time series order.

[0029] Optionally, step S3 includes:

[0030] S31. The multidimensional feature vector is divided into ionospheric feature set, tropospheric feature set, multipath feature set and equipment drift feature set according to the error source, which correspond to ionospheric error channel, tropospheric error channel, multipath error channel and equipment drift error channel respectively;

[0031] S32. Select the corresponding time series features in each error channel as input variables to establish an improved autoregressive moving-diffusion model, which includes an autoregressive term, a moving average term, a drift term, and a diffusion term.

[0032] S33. In the autoregressive term, historical observation sequences are extracted based on the input time series features, the difference between observation error values ​​between adjacent time steps is calculated, and a weighted linear combination of historical observation values ​​is generated according to a preset lag order to form a time prediction term for the current observation error.

[0033] S34. In the moving average term, calculate the difference between the observed residual and the predicted residual at each time step, and perform a weighted average of the difference with the mean of the residuals at the most recent several time steps to generate a noise correction term; the observed residual is the actual observed value minus the model predicted value; the predicted residual is the current predicted value minus the predicted value at the previous time step.

[0034] S35. In the drift term, a time variation function is constructed based on the environmental characteristics input, and meteorological parameters, ionospheric electron content and building shading angle are mapped as time driving factors, and their dynamic offset to the prediction error is calculated.

[0035] S36. In the diffusion term, the geographical distance and spatial correlation coefficient are calculated based on the spatial coordinates of adjacent receiver nodes, and the error values ​​in the spatial neighborhood are weighted and averaged according to the diffusion kernel weight to generate the spatial propagation term.

[0036] S37. The time prediction term, noise correction term, environmental offset and spatial propagation term calculated by the autoregressive term, moving average term, drift term and diffusion term are jointly optimized to form an improved autoregressive moving diffusion model structure for each error channel.

[0037] Optionally, step S4 includes:

[0038] S41. Divide the multidimensional feature vectors into training and validation sets according to time order, and assign them to the corresponding improved autoregressive moving diffusion models according to channels.

[0039] S42. Load the training set data using batch training, input each batch of samples into the model, and calculate the residual sequence between the predicted output and the observed values.

[0040] S43. Define a robust loss function based on the residual sequence. The robust loss function adopts the Huber form. When the absolute value of the residual is less than the threshold δ, it is calculated as squared error. When the absolute value of the residual is greater than the threshold δ, it is calculated as linear error. The threshold δ is adaptively set according to the quantiles of the residual distribution of the training data.

[0041] S44. With the goal of minimizing the robust loss function, gradient updates are performed on the autoregressive coefficients, moving average coefficients, drift function parameters, and diffusion function parameters, respectively. The gradient update adopts an adaptive learning rate mechanism.

[0042] S45. After each round of training, the model performance is evaluated based on the mean square error of the validation set residuals. When the performance improvement is lower than a preset threshold in several consecutive iterations, the training process is terminated.

[0043] S46. After training is completed, extract the final parameter set of each error channel. The parameter set includes the autoregressive coefficient vector, the moving average coefficient vector, the drift function time factor parameter, and the diffusion kernel space parameter.

[0044] Optionally, step S5 includes:

[0045] S51. Based on the latitude, longitude and elevation data of the receiver, determine the spatial distribution range of the receiver in the target area, divide the target geographical area into a set of regular spatial grid nodes, record the geographical coordinates, average elevation and local terrain feature parameters of each node, and form a set of nodes covering the target geographical area.

[0046] S52. Calculate the geographical distance between any two spatial grid nodes, and determine the spatial association weight based on the similarity of building occlusion and terrain between nodes. The spatial association weight is calculated based on the Gaussian radial function.

[0047] S53. Construct a geographic diffusion graph based on the node set and the corresponding spatial association weights. In the geographic diffusion graph, nodes are used as vertices and spatial association weights are used as edge weights, forming a weighted undirected graph structure that reflects geographic proximity and environmental similarity.

[0048] S54. Convert the node connection relationships in the geographic diffusion map into matrix form to obtain the node connection relationship matrix. Element corresponding node With nodes Spatial correlation weights;

[0049] S55. Introduce the node connection relationship matrix into the constraints of model training, and perform constraint correction in each round of parameter update according to the weights corresponding to the matrix elements. Calculate the squared difference term of the prediction error between adjacent nodes and accumulate it into a space regularization term.

[0050] S56. After training, output the node spatial relationship and weight matrix data, and generate a structured result file containing node index, spatial distance and associated weight.

[0051] Optionally, step S6 includes:

[0052] S61. Acquire real-time satellite observation data and environmental feature data. The real-time satellite observation data includes pseudorange observations, carrier phase observations, satellite azimuth and elevation angles. The environmental feature data includes meteorological parameters, ionospheric electron content, terrain and building obstruction angles.

[0053] S62. Perform feature mapping and index matching on the real-time satellite observation data and environmental feature data according to the multi-dimensional feature dimension structure of the model input;

[0054] S63. Input the real-time feature vectors into the improved autoregressive motion diffusion models corresponding to the ionospheric error channel, tropospheric error channel, multipath error channel and equipment drift error channel respectively, and calculate the error prediction values ​​of each channel.

[0055] S64. Determine the weights based on the error variance of each channel model during the verification phase, perform weighted fusion on each error prediction value, and obtain the overall error correction amount;

[0056] S65. Using the node connection matrix and the corresponding spatial association weights, perform spatial smoothing on the fusion result to adjust the consistency of error correction between adjacent nodes;

[0057] S66. Calculate the uncertainty of the overall error correction amount based on the results of weighted fusion and spatial smoothing, and output the overall error correction amount and uncertainty.

[0058] Optionally, step S7 includes:

[0059] S71. Obtain the overall error correction amount and uncertainty, as well as the real-time satellite positioning results;

[0060] S72. Decompose the overall error correction into coordinates and map them to the latitude, longitude and elevation components of the satellite positioning results respectively.

[0061] S73. Calculate the correction value for each coordinate component, and superimpose the correction value onto the real-time positioning result according to the time index to generate the corrected positioning coordinates;

[0062] S74. Calculate the confidence interval range of the corrected positioning coordinates based on the total uncertainty, and determine the upper and lower limits of the coordinates according to the set confidence level;

[0063] S75. Output the optimized positioning coordinates with confidence intervals, and record the positioning timestamp, correction amount, and uncertainty parameters.

[0064] The beneficial effects of this invention are:

[0065] This invention constructs a multi-channel error modeling system that combines an improved autoregressive moving-diffusion model with spatial regularization constraints. Addressing the challenges of strong time-varying characteristics, weak spatial correlation, and non-Gaussian noise interference in complex environments during satellite positioning, including ionospheric, tropospheric, multipath, and equipment drift errors, a unified spatiotemporal reference system is established, and time synchronization and spatial mapping of multi-source data are performed. Combined with anomaly removal, interpolation completion, and feature standardization, multi-dimensional spatiotemporal feature vectors are generated. In the error modeling stage, an improved autoregressive moving-diffusion model incorporating autoregressive, moving average, drift, and diffusion terms is introduced to decompose and model the dynamic evolution of each error channel. Optimization is achieved by combining batch training with a Huber robust loss function. The strategy involves adaptively updating the model parameters using a learning rate, significantly improving the model's convergence stability and robustness under abnormal samples and noisy backgrounds. At the spatial constraint level, a geographic diffusion map incorporating geographical distance, terrain similarity, and building occlusion features is constructed. The spatial relationships between nodes are transformed into a constraint matrix and introduced into the training process. Spatial regularization terms are used to achieve smooth propagation and consistency correction of errors between adjacent nodes. In the prediction phase, real-time satellite observations and environmental features are mapped into real-time feature vectors consistent with the model structure and input into each error channel model. Multi-source error prediction values ​​are obtained and fused using variance weighting from the validation phase. The overall error correction and uncertainty are calculated by combining spatial correlation weights, ultimately outputting optimized positioning coordinates with confidence intervals. Through the above design, this invention achieves multi-channel decomposition modeling, robust parameter estimation, and spatial consistency correction of satellite positioning errors, significantly improving positioning accuracy, reliability, and result confidence in complex environments. Attached Figure Description

[0066] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0067] Figure 1 This is a schematic diagram of the overall process of a satellite positioning error optimization method based on big data analysis proposed in this invention;

[0068] Figure 2 This is a schematic diagram of the structure of the improved autoregressive mobile diffusion model in this invention;

[0069] Figure 3 This is a schematic diagram illustrating the construction of the geographic diffusion map and spatial regularity constraint relationship in this invention. Detailed Implementation

[0070] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0071] refer to Figure 1-3 A satellite positioning error optimization method based on big data analysis includes the following steps:

[0072] S1. Acquire historical data from multiple sources and perform time synchronization and spatial mapping to form a unified dataset;

[0073] S2. Perform anomaly removal, interpolation completion, and feature standardization on the unified dataset to construct a multidimensional feature vector;

[0074] S3. The positioning error is divided into four channels: ionospheric error, tropospheric error, multipath error and equipment drift error. An improved autoregressive motion-diffusion model is established for each channel. The model includes an autoregressive term, a moving average term, a drift term and a diffusion term.

[0075] S4. Parameter estimation is performed using batch training and a robust loss function to obtain the parameter set for each error channel;

[0076] S5. Construct a geographic diffusion map containing spatial grid nodes, determine edge weights based on the similarity of building occlusion and terrain between nodes, and use the spatial association relationship of the geographic diffusion map as a regularization constraint;

[0077] S6. Input real-time satellite observations and environmental characteristics, call the improved autoregressive mobile diffusion model to calculate the error prediction values ​​of each channel and perform weighted fusion to obtain the overall error correction amount and uncertainty;

[0078] S7. Based on the error correction amount and uncertainty, output the optimized positioning coordinates with confidence interval.

[0079] This invention achieves spatiotemporal collaborative optimization of satellite positioning errors through the fusion analysis of multi-source satellite observation data and environmental feature data. The method first establishes a unified spatiotemporal reference framework, performing time synchronization and spatial mapping on multi-source data to form a unified dataset suitable for modeling. Then, anomaly removal, interpolation completion, and feature standardization are performed on the data to generate multi-dimensional feature vectors. Improved autoregressive motion diffusion models are established for ionospheric, tropospheric, multipath, and equipment drift errors, respectively. Parameter estimation is performed using batch training and robust loss functions, and spatial regularization constraints from geographic diffusion maps are introduced into the models to maintain consistency between nodes. Finally, real-time observation data is used to calculate the predicted error values ​​for each channel and weighted fusion is performed to output a high-precision optimized positioning result with confidence intervals.

[0080] In this embodiment, step S1 includes:

[0081] S11. Collect multi-source historical data, including satellite observation residual data, pseudorange observation data, carrier phase observation data, satellite azimuth and elevation data, receiver latitude and longitude and elevation data, meteorological parameter data, ionospheric electron content data, and terrain and building obstruction data.

[0082] S12. Determine the spatial position coordinates of each satellite at the observation time based on the satellite ephemeris information, calculate the geometric distance data between the satellite and the receiver, and establish the spatial association between the satellite and the receiver;

[0083] S13. Perform time synchronization processing on data from satellite receivers, meteorological monitoring systems, ionospheric monitoring systems and geographic information databases, and align data from different sampling frequencies to the same time series using satellite observation time as a unified benchmark.

[0084] S14. Perform spatial mapping processing on the meteorological parameter data, total ionospheric electron content data, terrain and building occlusion data, project them onto a unified geographic reference coordinate system, and calculate the environmental characteristic parameters of the corresponding location based on the receiver's latitude, longitude and elevation.

[0085] S15. The time-synchronized satellite observation data and the spatially mapped environmental feature data are jointly registered according to the time index and spatial coordinates to form a unified dataset containing time, location, observation features and environmental features.

[0086] This invention establishes a high-precision, multi-source spatiotemporal data foundation for satellite positioning error optimization. By collecting historical data of various types, including satellite observations, meteorological data, ionospheric data, and geographical environmental data, and combining ephemeris calculations, it achieves spatial geometric association between satellites and receivers. After time synchronization and spatial mapping, data with different sampling frequencies and spatial resolutions are unified to the same spatiotemporal reference. Finally, through joint registration, a multidimensional dataset containing observational and environmental features is formed, providing highly consistent and complete data support for the training and error optimization of the improved autoregressive motion-diffusion model.

[0087] In this embodiment, step S2 includes:

[0088] S21. Perform integrity screening on the unified dataset and remove data records with a data missing rate exceeding a preset threshold.

[0089] S22. Establish independent data channels for different types of feature data in the unified dataset, construct time index sequences in each data channel, calculate the time interval and rate of change of adjacent observations, determine the location of abnormal observation points based on the time series continuity detection results, and use the sliding window interpolation method to complete the data of the abnormal points.

[0090] S23. Perform statistical smoothing on the interpolated data sequence to eliminate short-term fluctuations introduced by observation noise;

[0091] S24. Standardize the characteristic data of meteorological parameters, ionospheric electron content, satellite azimuth, elevation, receiver latitude and longitude and elevation of different dimensions, and map each feature to a unified scale of zero mean and unit variance.

[0092] S25. In the standardized feature space, verify the consistency of data from each data channel at the same observation time according to the unified time index, and delete data records that do not meet the time alignment requirements.

[0093] S26. After anomaly removal, interpolation completion, smoothing and standardization, and passing the consistency check, the data are renumbered and multi-dimensional feature vectors with observation features and environmental features as components are generated in time series order.

[0094] This invention ensures the integrity and consistency of model input data through multi-level data cleaning and feature construction. The method first performs integrity screening on a unified dataset, removing severely missing data records. Then, independent channels are established for different types of features, outliers are detected using time indexing and filled in using sliding window interpolation, while smoothing is combined to eliminate short-term noise interference. Next, standardization is performed on features of different dimensions, mapping meteorological, ionospheric, azimuth, and elevation features to a unified scale, and time indexing is used to verify multi-channel consistency, removing time-misaligned samples. Finally, a multi-dimensional feature vector arranged in chronological order is generated, providing a unified, robust, and highly comparable spatiotemporal feature input for subsequent multi-channel error modeling.

[0095] In this embodiment, step S3 includes:

[0096] S31. The multidimensional feature vector is divided into ionospheric feature set, tropospheric feature set, multipath feature set and equipment drift feature set according to the error source, which correspond to ionospheric error channel, tropospheric error channel, multipath error channel and equipment drift error channel respectively;

[0097] S32. Select the corresponding time series features in each error channel as input variables to establish an improved autoregressive moving-diffusion model, which includes an autoregressive term, a moving average term, a drift term, and a diffusion term.

[0098] S33. In the autoregressive term, historical observation sequences are extracted based on the input time series features, the difference between observation error values ​​between adjacent time steps is calculated, and a weighted linear combination of historical observation values ​​is generated according to a preset lag order to form a time prediction term for the current observation error.

[0099] The autoregressive structure is used to capture the temporal inertia and dynamic evolution of the positioning error, providing a time baseline for the subsequent calculation of the drift and diffusion terms;

[0100] S34. In the moving average term, calculate the difference between the observed residual and the predicted residual at each time step, and perform a weighted average of the difference with the mean of the residuals at the most recent several time steps to generate a noise correction term; the observed residual is the actual observed value minus the model predicted value; the predicted residual is the current predicted value minus the predicted value at the previous time step.

[0101] The moving average term suppresses the impact of random noise on prediction accuracy through a weighted smoothing operation of the residuals, thus maintaining the stability of the error sequence.

[0102] S35. In the drift term, a time variation function is constructed based on the environmental characteristics input, and meteorological parameters, ionospheric electron content and building shading angle are mapped as time driving factors, and their dynamic offset to the prediction error is calculated.

[0103] The mapping process includes: normalizing the temperature, humidity, and air pressure parameters in the meteorological data to form a continuous time series; using the rate of change of the total electron content in the ionosphere as an index of ionospheric disturbance intensity, calculating the average rate of change within the continuous time window to characterize the temporal stability of the ionosphere; periodically sampling the building obstruction angle, and calculating the obstruction index based on the overlap duration between the satellite azimuth angle and the obstruction angle. The time series of meteorological data, electron content, and obstruction index are input into a time-weighted function, which combines a sliding weighted average with periodic modulation, assigning higher weights to near-time data and lower weights to far-time data. The resulting time-driving factor reflects the comprehensive impact of environmental characteristics changing over time on the positioning error drift trend and is used to calculate the dynamic offset of the drift term output.

[0104] S36. In the diffusion term, the geographical distance and spatial correlation coefficient are calculated based on the spatial coordinates of adjacent receiver nodes, and the error values ​​in the spatial neighborhood are weighted and averaged according to the diffusion kernel weight to generate the spatial propagation term.

[0105] The spatial correlation coefficient is used to reflect the consistency of the changes in positioning errors between different nodes. Its calculation process includes: pairing the error time series of any two nodes within a selected time window, comparing the error change trends at each time point, and calculating the correlation index based on the consistency of the error difference. The correlation index value is between 0 and 1, and the larger the value, the more synchronous the error changes.

[0106] The diffusion kernel weight is used to describe the diffusion intensity of the error within the spatial neighborhood. Its determination method is as follows: a distance attenuation function is applied to the geographical distance between each node. Higher weights are assigned when the distance between nodes is less than a set scale parameter, and the weights decay to near zero when the distance exceeds three times the scale parameter. Finally, after normalization, the sum of the diffusion weights of all neighboring nodes is 1. Simultaneously, the spatial correlation coefficient is used as a weighting factor to introduce the adjustment effect of the similarity of error changes. The diffusion kernel weight obtained after distance attenuation and correlation adjustment reflects the propagation intensity of the positioning error in the spatial domain.

[0107] S37. The time prediction term, noise correction term, environmental offset and spatial propagation term calculated by the autoregressive term, moving average term, drift term and diffusion term are jointly optimized to form an improved autoregressive moving diffusion model structure for each error channel.

[0108] The improved autoregressive motion-diffusion model in this invention introduces an extended structure of drift and diffusion terms on the basis of the traditional model. The drift term reflects the dynamic influence of environmental factors over time, while the diffusion term reflects the propagation correlation of errors between spatial neighborhoods. Through the combined effect of the time memory mechanism of the autoregressive term and the noise correction mechanism of the moving average term, the model achieves spatiotemporal joint modeling of positioning errors, effectively improving the stability and accuracy of positioning error optimization in complex environments.

[0109] In this embodiment, step S4 includes:

[0110] S41. Divide the multidimensional feature vectors into training and validation sets according to time order, and assign them to the corresponding improved autoregressive moving diffusion models according to channels.

[0111] S42. Load the training set data using batch training, input each batch of samples into the model, and calculate the residual sequence between the predicted output and the observed values.

[0112] S43. Define a robust loss function based on the residual sequence. The robust loss function adopts the Huber form. When the absolute value of the residual is less than the threshold δ, it is calculated as the squared error. When the absolute value of the residual is greater than the threshold δ, it is calculated as the linear error.

[0113] The threshold δ is adaptively set based on the quantiles of the residual distribution of the training data. A preferred method is to calculate it based on the quantiles of the absolute values ​​of the residuals in the previous training rounds, taking the 95th percentile of the absolute value distribution of the residuals as the threshold δ. This ensures that over 90% of normal samples are within the squared loss interval, and linear penalties are only applied to extreme outliers. During training, when the characteristics of the residual distribution change, the system can periodically recalculate the quantiles and dynamically adjust the threshold δ to maintain the stability of the loss function's suppression effect on outliers.

[0114] S44. With the goal of minimizing the robust loss function, gradient updates are performed on the autoregressive coefficients, moving average coefficients, drift function parameters, and diffusion function parameters, respectively. The gradient update adopts an adaptive learning rate mechanism.

[0115] The gradient update process employs an adaptive learning rate mechanism, which automatically adjusts the update step size based on the gradient change magnitude of each parameter in continuous iterations. For parameters with large gradient fluctuations, the learning rate is reduced to suppress oscillations, while for parameters with gentle gradient changes, the learning rate is increased to accelerate convergence, thereby ensuring the stable convergence of the autoregressive coefficients, moving average coefficients, drift function parameters, and diffusion kernel parameters in the multi-channel space.

[0116] S45. After each round of training, the model performance is evaluated based on the mean square error of the validation set residuals. When the performance improvement is lower than a preset threshold in several consecutive iterations, the training process is terminated.

[0117] S46. After training is completed, extract the final parameter set of each error channel. The parameter set includes the autoregressive coefficient vector, the moving average coefficient vector, the drift function time factor parameter, and the diffusion kernel space parameter.

[0118] In this invention, after each iteration of model training, the actual observations of the validation set are compared with the model's predicted output values ​​at each time step, and the prediction residuals at each time step are calculated. The prediction residuals are the difference between the actual observations and the predicted values. The prediction residuals of all samples in the validation set are squared and averaged to obtain the mean squared error of the validation set residuals. The mean squared error is used to reflect the overall prediction accuracy under the current model parameters.

[0119] In multiple consecutive training iterations, the mean squared error of the validation set residuals is recorded for each iteration, and the performance improvement between adjacent iterations, i.e., the percentage decrease in mean squared error, is calculated. When the percentage decrease in the mean squared error of the validation set residuals falls below a preset threshold in several consecutive iterations, the model performance is considered to have stabilized, and subsequent training iterations are stopped.

[0120] The threshold is dynamically determined based on the complexity of the training data and the residual variance, and is preferably set to a performance improvement of 1% to 2%. When the performance improvement is less than this threshold, it indicates that the model has reached convergence. At this point, terminating training can avoid overfitting and shorten the model convergence time.

[0121] In this embodiment, step S5 includes:

[0122] S51. Based on the latitude, longitude and elevation data of the receiver, determine the spatial distribution range of the receiver in the target area, divide the target geographical area into a set of regular spatial grid nodes, record the geographical coordinates, average elevation and local terrain feature parameters of each node, and form a set of nodes covering the target geographical area.

[0123] S52. Calculate the geographical distance between any two spatial grid nodes, and determine the spatial association weight based on the similarity of building occlusion and terrain between the nodes. The spatial association weight is calculated using a Gaussian radial function. Defined as:

[0124] ;

[0125] in, Represents a node Spatial correlation weights; It is an exponential function; Represents a node The square of the geographical distance; Scale parameters representing distance similarity; Represents a node Average elevation; Represents a node Average elevation; Scale parameters indicating elevation similarity; Represents a node The building's obstruction angle; Represents a node The building's obstruction angle; Scale parameters indicating similarity in occlusion;

[0126] S53. Construct a geographic diffusion graph based on the node set and the corresponding spatial association weights. In the geographic diffusion graph, nodes are used as vertices and spatial association weights are used as edge weights, forming a weighted undirected graph structure that reflects geographic proximity and environmental similarity.

[0127] S54. Convert the node connection relationships in the geographic diffusion map into matrix form to obtain the node connection relationship matrix. Element corresponding node With nodes Spatial correlation weights;

[0128] S55. Introduce the node connection relationship matrix into the constraints of model training, and perform constraint correction in each round of parameter update according to the weights corresponding to the matrix elements. Calculate the squared difference term of the prediction error between adjacent nodes and accumulate it into a space regularization term.

[0129] S56. After training, output the node spatial relationship and weight matrix data, and generate a structured result file containing node index, spatial distance and associated weight.

[0130] This invention incorporates the node connection matrix into the model training constraints. By comparing the prediction error differences between adjacent nodes in space, the square of the error difference between each node is calculated and weighted according to the association weights between nodes to form a spatial regularization term. This spatial regularization term reflects the degree of consistency of prediction results of different nodes in space. During the model parameter update process, it participates in iterative optimization simultaneously with the main loss function, ensuring that nodes with similar geographical locations and environmental characteristics maintain continuity and smoothness in error estimation, thereby reducing abrupt error changes caused by local noise or individual abnormal nodes.

[0131] In this embodiment, step S6 includes:

[0132] S61. Acquire real-time satellite observation data and environmental feature data. The real-time satellite observation data includes pseudorange observations, carrier phase observations, satellite azimuth and elevation angles. The environmental feature data includes meteorological parameters, ionospheric electron content, terrain and building obstruction angles.

[0133] S62. Perform feature mapping and index matching on the real-time satellite observation data and environmental feature data according to the multi-dimensional feature dimension structure of the model input;

[0134] S63. Input the real-time feature vectors into the improved autoregressive motion diffusion models corresponding to the ionospheric error channel, tropospheric error channel, multipath error channel and equipment drift error channel respectively, and calculate the error prediction values ​​of each channel.

[0135] S64. Determine the weights based on the error variance of each channel model during the verification phase, perform weighted fusion on each error prediction value, and obtain the overall error correction amount;

[0136] S65. Using the node connection matrix and the corresponding spatial association weights, perform spatial smoothing on the fusion result to adjust the consistency of error correction between adjacent nodes;

[0137] S66. Calculate the uncertainty of the overall error correction amount based on the results of weighted fusion and spatial smoothing, and output the overall error correction amount and uncertainty.

[0138] In this invention, real-time satellite observation data is acquired through a dynamic acquisition module of a satellite receiver, with a sampling period set to 1 to 10 seconds. Environmental feature data is updated in real time by meteorological sensors, an ionospheric monitoring system, and a geographic information database. Before being input into the model, the real-time data undergoes feature mapping and index matching based on the multi-dimensional feature structure established during the training phase, aligning the observation data with environmental features through satellite numbers and timestamps. Each error channel model loads the parameter set obtained during the training phase during system initialization, including autoregressive coefficients, moving average coefficients, drift function time factor parameters, and diffusion kernel spatial parameters. During model runtime, real-time feature vectors are input into the model channel by channel, outputting the corresponding error prediction values. The system determines weights based on the error variance of each channel during the validation phase. The weights are inversely proportional to the variance and are normalized before being used for weighted fusion to obtain the overall error correction. When calculating the uncertainty of the overall error correction, the comprehensive uncertainty value is determined based on the prediction residual variance of each error channel during the validation phase and the fusion weights. The residual variance of each channel is propagated and accumulated according to the square of its weight during weighted fusion; the resulting weighted variance is used to characterize the stability of the overall error correction. The square root of this weighted variance is used as an estimate of the overall uncertainty to quantify the confidence level of the fusion result. This uncertainty value changes dynamically with the prediction performance and weights of each channel, reflecting the reliability range of the real-time error correction.

[0139] In this embodiment, step S7 includes:

[0140] S71. Obtain the overall error correction amount and uncertainty, as well as the real-time satellite positioning results;

[0141] S72. Decompose the overall error correction into coordinates and map them to the latitude, longitude and elevation components of the satellite positioning results respectively.

[0142] S73. Calculate the correction value for each coordinate component, and superimpose the correction value onto the real-time positioning result according to the time index to generate the corrected positioning coordinates;

[0143] S74. Calculate the confidence interval range of the corrected positioning coordinates based on the total uncertainty, and determine the upper and lower limits of the coordinates according to the set confidence level;

[0144] S75. Output the optimized positioning coordinates with confidence intervals, and record the positioning timestamp, correction amount, and uncertainty parameters.

[0145] In this invention, when calculating the confidence interval range of the corrected positioning coordinates, the width of the confidence interval is determined based on the uncertainty of the overall error correction. When calculating the confidence interval of the corrected positioning coordinates, the uncertainty is multiplied by a confidence coefficient to determine the upper and lower limit offsets, which are then added to the three-dimensional components of the corrected coordinates to obtain the upper and lower confidence limits of the corrected positioning coordinates. The confidence coefficient is adaptively set according to the target confidence level, which can be any value between 90% and 99%. The confidence interval is used to characterize the reliability range of the corrected positioning result, enabling the positioning system to have error confidence assessment capabilities while outputting optimized results.

[0146] Example 1:

[0147] To verify the feasibility and superiority of this invention in practical satellite positioning applications, it was applied to a high-precision geographic mapping project in a complex terrain area of ​​a certain province. This area includes mountains, hills, plains, and urban building clusters, with significant terrain undulations and dense building obstruction. Traditional GNSS positioning methods are prone to positioning drift and signal anomalies in this environment. A typical scenario area of ​​approximately 250 square kilometers was selected for the test, and 32 GNSS receiver nodes were deployed, 20 for training data acquisition and 12 for independent verification. Each receiver acquired satellite observation data including pseudorange observations, carrier phase, satellite azimuth and elevation angles, and simultaneously collected meteorological temperature and humidity, air pressure, ionospheric electron content, and building obstruction angle information. The sampling frequency was 1 Hz, and the continuous acquisition time was 72 hours.

[0148] In the data preprocessing stage, the system first performs time synchronization and spatial mapping on multi-source data to ensure that data from different sources are aligned under a unified spatiotemporal reference. For approximately 3.7% of the anomalous observation records detected during outlier detection, a sliding window interpolation method is used to complete the data, and a statistical smoothing algorithm is employed to reduce short-term observation fluctuations. After standardization, the average variance of the data is reduced to 18.6% of the original, significantly improving feature stability. Next, an error channel input is constructed using multi-dimensional feature vectors, divided into four feature sets: ionosphere, troposphere, multipath, and equipment drift. These are then input into an improved autoregressive mobile diffusion model for modeling.

[0149] During model training, the first 60 hours of data were used as the training set, and the last 12 hours of data were used as the validation set. Each channel model employed batch training and Huber robust loss function optimization, with the residual threshold δ adaptively set to 0.85 based on the residual distribution quantiles. An adaptive learning rate mechanism was used during training, with an initial learning rate of 0.001, which was automatically halved after three consecutive validation epochs if there was no significant decrease in the validation error. The average number of convergence epochs was 48, and the final average training error for the four channels decreased to 6.3% of the original value.

[0150] In constructing the spatial association constraints, the receiver distribution range was divided into 625 spatial grid nodes, with an average spacing of approximately 500 meters between each node. The spatial association weights between nodes were calculated using a Gaussian radial function, where the geographic distance influence factor σ was set to 1000 meters, building occlusion similarity weighted at 0.4, and terrain similarity weighted at 0.6. By constructing a geographic diffusion map and introducing a spatial regularization term, the model's error propagation consistency in the boundary region was improved by approximately 22%.

[0151] In the real-time prediction phase, the model is input with new satellite observations and environmental characteristics data within the past 72 hours, and independently predicts four error channels. The predicted values ​​for each channel are fused using variance weighting to obtain the overall error correction, and the uncertainty is estimated using the square root variance weighted method of the fused results. The final output is the positioning coordinates with confidence intervals, and the confidence level range is adjusted according to the dynamic uncertainty, averaging 94.5%.

[0152] To verify the beneficial effects of the present invention, during the experimental phase, the positioning accuracy of the method of the present invention was compared with that of the traditional method in complex terrain areas. The experimental results are shown in Table 1:

[0153] Table 1. Comparison of positioning accuracy between the method of this invention and traditional algorithms in complex terrain areas.

[0154] Test area type Ionospheric TEC Changes (TECU) Multipath interference intensity (dB) Average error (m) of traditional method The average error (m) of the method of this invention Error reduction rate (%) mountain forest area 18.2 -14.5 3.27 0.82 74.9 Urban building area 11.6 -20.7 2.98 0.63 78.8 River valley plain area 9.3 -12.1 2.46 0.55 77.6 High-altitude sparsely populated areas 15.8 -16.3 3.04 0.71 76.6 Overall Average 13.7 -15.9 2.84 0.67 76.4

[0155] As shown in Table 1, the present invention significantly outperforms traditional differential GPS positioning methods in various complex geographical environments. Firstly, in mountainous and forested areas, where satellite signals are severely affected by terrain obstruction and multipath effects, the average error of traditional differential GPS reaches as high as 3.27 meters. The present invention, through multi-source data fusion and diffusion modeling, reduces the average error to 0.82 meters, a reduction of 74.9%, and improves positioning stability by over 60%. In urban built-up areas, where signal obstruction is high and traditional algorithms are significantly affected by multipath interference, the present invention, by introducing a geographic diffusion map and spatial regularization constraints, achieves smooth error propagation between nodes, reducing the positioning error from 2.98 meters to 0.63 meters, a reduction of 78.8%, demonstrating strong robustness. In river valleys and plains, where the terrain is relatively open but humidity gradients and ground reflection exist, the average error of traditional algorithms is 2.46 meters, while the present invention's error is only 0.55 meters, a reduction of 77.6%, indicating that the model maintains high-precision prediction performance even in weakly obstructed scenarios. In sparsely populated high-altitude areas, traditional algorithms are unstable due to changes in air pressure and ionospheric disturbances. The improved autoregressive motion-diffusion model of this invention, by introducing drift and diffusion terms, effectively characterizes the relationship between ionospheric changes and spatial propagation, reducing the positioning error from 3.04 meters to 0.71 meters. Overall results show that the average error of this invention is 0.67 meters, a 76.4% reduction compared to traditional methods, and an average improvement of 63.9% in positioning stability across different terrain environments.

[0156] Overall, the method of this invention shows the most significant performance improvement under conditions of high ionospheric variation and high building obstruction, indicating its stronger adaptability to nonlinear and non-stationary environments. Thanks to batch training and the introduction of a robust loss function, the model maintains stable parameter convergence even under noise interference. Simultaneously, spatial regularization constraints and geographic diffusion mechanisms ensure consistent error correction in the spatial domain, avoiding the problem of discontinuous error propagation in boundary regions encountered by traditional methods. In summary, this invention achieves significant improvements in positioning accuracy, stability, and anti-interference capabilities, verifying its practicality and technological advancement in high-precision positioning optimization under complex environments.

[0157] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A satellite positioning error optimization method based on big data analysis, characterized in that, Includes the following steps: S1. Acquire historical data from multiple sources and perform time synchronization and spatial mapping to form a unified dataset; S2. Perform anomaly removal, interpolation completion, and feature standardization on the unified dataset to construct a multidimensional feature vector; S3. The positioning error is divided into four channels: ionospheric error, tropospheric error, multipath error and equipment drift error. An improved autoregressive motion-diffusion model is established for each channel. The improved autoregressive motion-diffusion model includes an autoregressive term, a moving average term, a drift term and a diffusion term. S4. Parameter estimation is performed using batch training and a robust loss function to obtain the parameter set for each error channel; S5. Construct a geographic diffusion map containing spatial grid nodes, determine edge weights based on the similarity of building occlusion and terrain between nodes, and use the spatial association relationship of the geographic diffusion map as a regularization constraint; S6. Input real-time satellite observations and environmental characteristics, call the improved autoregressive mobile diffusion model to calculate the error prediction values ​​of each channel and perform weighted fusion to obtain the overall error correction amount and uncertainty; S7. Based on the total error correction amount and uncertainty, output the optimized positioning coordinates with confidence interval.

2. The satellite positioning error optimization method based on big data analysis according to claim 1, characterized in that, Step S1 includes: S11. Collect multi-source historical data, including satellite observation residual data, pseudorange observation data, carrier phase observation data, satellite azimuth and elevation data, receiver latitude and longitude and elevation data, meteorological parameter data, ionospheric electron content data, and terrain and building obstruction data. S12. Determine the spatial position coordinates of each satellite at the observation time based on the satellite ephemeris information, calculate the geometric distance data between the satellite and the receiver, and establish the spatial association between the satellite and the receiver; S13. Perform time synchronization processing on data from satellite receivers, meteorological monitoring systems, ionospheric monitoring systems and geographic information databases, and align data from different sampling frequencies to the same time series using satellite observation time as a unified benchmark. S14. Perform spatial mapping processing on the meteorological parameter data, delamination electron content data, and terrain and building occlusion data, project them onto a unified geographic reference coordinate system, and calculate the environmental characteristic parameters of the corresponding location based on the receiver's latitude, longitude, and elevation. S15. The time-synchronized satellite observation data and the spatially mapped environmental feature data are jointly registered according to the time index and spatial coordinates to form a unified dataset containing time, location, observation features and environmental features.

3. The satellite positioning error optimization method based on big data analysis according to claim 1, characterized in that, Step S2 includes: S21. Perform integrity screening on the unified dataset and remove data records with a data missing rate exceeding a preset threshold. S22. Establish independent data channels for different types of feature data in the unified dataset, construct time index sequences in each data channel, calculate the time interval and rate of change of adjacent observations, determine the location of abnormal observation points based on the time series continuity detection results, and use the sliding window interpolation method to complete the data of the abnormal observation points. S23. Perform statistical smoothing on the interpolated data sequence to eliminate short-term fluctuations introduced by observation noise; S24. Standardize the characteristic data of meteorological parameters, ionospheric electron content, satellite azimuth, elevation, receiver latitude and longitude and elevation of different dimensions, and map each feature to a unified scale of zero mean and unit variance. S25. In the standardized feature space, verify the consistency of data from each data channel at the same observation time according to the unified time index, and delete data records that do not meet the time alignment requirements. S26. After anomaly removal, interpolation completion, smoothing and standardization, and passing the consistency check, the data are renumbered and multi-dimensional feature vectors with observation features and environmental features as components are generated in time series order.

4. The satellite positioning error optimization method based on big data analysis according to claim 1, characterized in that, Step S3 includes: S31. The multidimensional feature vector is divided into ionospheric feature set, tropospheric feature set, multipath feature set and equipment drift feature set according to the error source, which correspond to ionospheric error channel, tropospheric error channel, multipath error channel and equipment drift error channel respectively; S32. Select the corresponding time series features in each error channel as input variables to establish an improved autoregressive moving diffusion model. The improved autoregressive moving diffusion model includes an autoregressive term, a moving average term, a drift term, and a diffusion term. S33. In the autoregressive term, historical observation sequences are extracted based on the input time series features, the difference between observation error values ​​between adjacent time steps is calculated, and a weighted linear combination of historical observation values ​​is generated according to a preset lag order to form a time prediction term for the current observation error. S34. In the moving average term, calculate the difference between the observed residual and the predicted residual at each time step, and perform a weighted average of the difference between the observed residual and the predicted residual at each time step and the mean of the residuals at the most recent several time steps to generate a noise correction term; the observed residual is the actual observed value minus the model predicted value; the predicted residual is the current predicted value minus the predicted value at the previous time step. S35. In the drift term, a time variation function is constructed based on the environmental characteristics input, and meteorological parameters, ionospheric electron content and building shading angle are mapped to time driving factors, and the dynamic offset of the time driving factors to the prediction error is calculated. S36. In the diffusion term, the geographical distance and spatial correlation coefficient are calculated based on the spatial coordinates of adjacent receiver nodes, and the error values ​​in the spatial neighborhood are weighted and averaged according to the diffusion kernel weight to generate the spatial propagation term. S37. The time prediction term, noise correction term, environmental offset and spatial propagation term calculated by the autoregressive term, moving average term, drift term and diffusion term are jointly optimized to form an improved autoregressive moving diffusion model structure for each error channel.

5. The satellite positioning error optimization method based on big data analysis according to claim 1, characterized in that, Step S4 includes: S41. Divide the multidimensional feature vectors into training and validation sets according to time order, and assign them to the corresponding improved autoregressive moving diffusion models according to channels. S42. Load the training set data using batch training, input each batch of samples into the model, and calculate the residual sequence between the predicted output and the observed values. S43. Define a robust loss function based on the residual sequence. The robust loss function adopts the Huber form. When the absolute value of the residual is less than the threshold δ, it is calculated as squared error. When the absolute value of the residual is greater than the threshold δ, it is calculated as linear error. The threshold δ is adaptively set according to the quantiles of the residual distribution of the training data. S44. With the goal of minimizing the robust loss function, gradient updates are performed on the autoregressive coefficients, moving average coefficients, drift function parameters, and diffusion function parameters, respectively. The gradient update adopts an adaptive learning rate mechanism. S45. After each round of training, the model performance is evaluated based on the mean square error of the validation set residuals. When the performance improvement is lower than a preset threshold in several consecutive iterations, the training process is terminated. S46. After training is completed, extract the final parameter set of each error channel. The final parameter set includes the autoregressive coefficient vector, the moving average coefficient vector, the drift function time factor parameter, and the diffusion kernel space parameter.

6. The satellite positioning error optimization method based on big data analysis according to claim 2, characterized in that, Step S5 includes: S51. Based on the latitude, longitude and elevation data of the receiver, determine the spatial distribution range of the receiver in the target area, divide the target geographical area into a set of regular spatial grid nodes, record the geographical coordinates, average elevation and local terrain feature parameters of each node, and form a set of nodes covering the target geographical area. S52. Calculate the geographical distance between any two spatial grid nodes, and determine the spatial association weight based on the similarity of building occlusion and terrain between nodes. The spatial association weight is calculated based on the Gaussian radial function. S53. Construct a geographic diffusion graph based on the node set and the corresponding spatial association weights. In the geographic diffusion graph, nodes are used as vertices and spatial association weights are used as edge weights, forming a weighted undirected graph structure that reflects geographic proximity and environmental similarity. S54. Convert the node connection relationships in the geographic diffusion map into matrix form to obtain the node connection relationship matrix. The node connection relationship matrix is... Element corresponding node With nodes Spatial correlation weights; S55. Introduce the node connection relationship matrix into the constraints of model training, and perform constraint correction in each round of parameter update according to the weights corresponding to the matrix elements. Calculate the squared difference term of the prediction error between adjacent nodes and accumulate it into a space regularization term. S56. After training, output the node spatial relationship and weight matrix data, and generate a structured result file containing node index, spatial distance and associated weight.

7. The satellite positioning error optimization method based on big data analysis according to claim 6, characterized in that, Step S6 includes: S61. Acquire real-time satellite observation data and environmental feature data. The real-time satellite observation data includes pseudorange observations, carrier phase observations, satellite azimuth and elevation angles. The environmental feature data includes meteorological parameters, ionospheric electron content, terrain and building obstruction angles. S62. Perform feature mapping and index matching on the real-time satellite observation data and environmental feature data according to the multi-dimensional feature dimension structure of the model input; S63. Input the real-time multidimensional feature vectors into the improved autoregressive motion diffusion models corresponding to the ionospheric error channel, tropospheric error channel, multipath error channel and equipment drift error channel respectively, and calculate the error prediction values ​​of each channel. S64. Determine the weights based on the error variance of each channel model during the verification phase, perform weighted fusion on each error prediction value, and obtain the overall error correction amount; S65. Using the node connection matrix and the corresponding spatial association weights, perform spatial smoothing on the fusion result to adjust the consistency of error correction between adjacent nodes; S66. Calculate the uncertainty of the overall error correction amount based on the results of weighted fusion and spatial smoothing, and output the overall error correction amount and uncertainty.

8. The satellite positioning error optimization method based on big data analysis according to claim 1, characterized in that, Step S7 includes: S71. Obtain the overall error correction amount and uncertainty, as well as the real-time satellite positioning results; S72. Perform coordinate decomposition on the overall error correction amount, corresponding to the latitude, longitude and elevation components of the satellite positioning result respectively; S73. Calculate the correction value for each coordinate component, and superimpose the correction value onto the real-time positioning result according to the time index to generate the corrected positioning coordinates; S74. Calculate the confidence interval range of the corrected positioning coordinates based on the overall uncertainty, and determine the upper and lower limits of the coordinates according to the set confidence level. S75. Output the optimized positioning coordinates with confidence intervals, and record the positioning timestamp, correction amount, and uncertainty parameters.

Images