CN-122017906-A - 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, which comprises the steps of S1, obtaining multi-source historical data to form a unified data set, S2, carrying out abnormal rejection, interpolation complementation and standardization on the data to construct a multi-dimensional feature vector, S3, dividing positioning errors into four channels of an ionosphere, a troposphere, a multipath and equipment drift error, respectively establishing an improved autoregressive movement diffusion model, S4, carrying out parameter estimation by utilizing batch training and a steady loss function, S5, constructing a geographic diffusion map of space grid nodes, determining side weights by using similarity among the nodes and taking the side weights as regular constraint, S6, inputting real-time satellite observation and environmental characteristics, calculating error predicted values of each channel, weighting and fusing to obtain total error correction quantity and uncertainty, and S7, outputting optimized positioning coordinates with confidence intervals. The invention obviously improves satellite positioning precision, stability and anti-interference performance.

Inventors

• LI GUANGHUI

Assignees

• 山东鑫达建安工程有限公司

Dates

Publication Date

20260512

Application Date

20260226

Claims (8)

1. 1. The satellite positioning error optimization method based on big data analysis is characterized by comprising the following steps: s1, acquiring multi-source historical data, and performing time synchronization and space mapping to form a unified data set; s2, carrying out abnormal elimination, interpolation completion and feature standardization on the unified data set to construct a multidimensional feature vector; S3, dividing the positioning error into four channels of an ionosphere error, a troposphere error, a multipath error and a device drift error, and respectively establishing an improved autoregressive movement diffusion model, wherein the improved autoregressive movement diffusion model comprises an autoregressive term, a moving average term, a drift term and a diffusion term; S4, parameter estimation is carried out by utilizing batch training and a steady loss function, and a parameter set of each error channel is obtained; S5, constructing a geographic diffusion diagram containing space grid nodes, determining an edge weight according to the building shielding similarity and the terrain similarity among the nodes, and taking the space association relation of the geographic diffusion diagram as a regular constraint; S6, inputting real-time satellite observation and environmental characteristics, calling the improved autoregressive moving diffusion model to calculate error prediction values of all channels, and carrying out weighted fusion to obtain total error correction and uncertainty; And S7, outputting optimized positioning coordinates with confidence intervals based on the total error correction quantity and the uncertainty.
2. 2. The method for optimizing satellite positioning errors based on big data analysis according to claim 1, wherein the step S1 comprises: s11, acquiring multi-source historical data, wherein the multi-source historical data comprise satellite observation residual data, pseudo-range observation data, carrier phase observation data, satellite azimuth angle and pitch angle data, longitude and latitude and elevation data of a receiver, meteorological parameter data, ionosphere electronic content data and topography and building shielding data; S12, determining the space position coordinates of each satellite at the observation time according to satellite ephemeris information, calculating geometric distance data between the satellite and a receiver, and establishing space association between the satellite and the receiver; s13, performing time synchronization processing on data from a satellite receiver, a meteorological monitoring system, an ionosphere monitoring system and a geographic information database, and aligning the data with different sampling frequencies to the same time sequence by taking satellite observation time as a unified reference; s14, performing space mapping processing on the meteorological parameter data, the separation layer electronic content data and the terrain and building shielding data, projecting the data to a unified geographic reference coordinate system, and calculating environmental characteristic parameters of corresponding positions according to longitude and latitude and elevation of a receiver; And S15, carrying out joint registration on the satellite observation data after time synchronization and the environmental characteristic data after space mapping according to the time index and the space coordinates to form a unified data set containing time, position, observation characteristics and environmental characteristics.
3. 3. The method for optimizing satellite positioning errors based on big data analysis according to claim 1, wherein the step S2 comprises: S21, carrying out integrity screening on the unified data set, and eliminating data records with data loss rate exceeding a preset threshold; S22, establishing independent data channels for different types of characteristic data in the unified data set, establishing a time index sequence in each data channel, calculating the time interval and the change rate of adjacent observation values, determining the position of an abnormal observation point according to the continuity detection result of the time sequence, and carrying out data complementation on the abnormal observation point by adopting a sliding window interpolation method; s23, carrying out statistical smoothing treatment on the data sequence subjected to interpolation complementation, and eliminating short-time fluctuation introduced by observation noise; S24, respectively carrying out standardized processing on characteristic data of meteorological parameters, ionosphere electron content, satellite azimuth angle, pitch angle, longitude and latitude of a receiver and different dimensions of elevation, and mapping each characteristic to a unified dimension of zero mean and unit variance; S25, checking the data consistency of each data channel at the same observation time according to the unified time index in the standardized feature space, and deleting the data records which do not meet the time alignment requirement; s26, renumbering the data which is processed by anomaly rejection, interpolation completion, smoothing and standardization and passes consistency verification, and generating multidimensional feature vectors with observation features and environmental features as components according to a time sequence.
4. 4. The method for optimizing satellite positioning errors based on big data analysis according to claim 1, wherein the step S3 comprises: S31, dividing the multidimensional feature vector into an ionosphere feature set, a troposphere feature set, a multipath feature set and a device drift feature set according to error sources, wherein the ionosphere error channel, the troposphere error channel, the multipath error channel and the device drift error channel are respectively corresponding to the multidimensional feature vector; s32, selecting corresponding time sequence features in each error channel as input variables, and establishing an improved autoregressive moving diffusion model, wherein the improved autoregressive moving diffusion model comprises an autoregressive term, a moving average term, a drift term and a diffusion term; S33, in the autoregressive term, a historical observation sequence is extracted according to the input time sequence characteristics, differential calculation is carried out on the observation error values between adjacent time steps, and a weighted linear combination of the historical observation values is generated according to a preset hysteresis order, so that a time prediction term of the current observation error is formed; S34, in the moving average term, calculating the difference value between the observation residual error and the prediction residual error of each time step, and carrying out weighted average on the difference value between the observation residual error and the prediction residual error of each time step and the residual error average value of a plurality of latest time steps to generate a noise correction term, wherein the observation residual error is a real observation value minus a model prediction value; S35, in the drift item, a time change function is built according to environmental characteristic input, meteorological parameters, ionosphere electron content and building shielding angles are mapped into time driving factors, and dynamic offset of the time driving factors to prediction errors is calculated; S36, in the diffusion term, calculating a geographic distance and a spatial correlation coefficient according to the spatial coordinates of adjacent receiver nodes, and carrying out weighted average on error values in the spatial neighborhood according to diffusion kernel weights so as to generate a spatial propagation term; and S37, performing joint optimization on the time prediction term, the noise correction term, the environment offset and the space propagation term obtained by calculating the autoregressive term, the moving average term, the drift term and the diffusion term to form an improved autoregressive moving diffusion model structure of each error channel.
5. 5. The method for optimizing satellite positioning errors based on big data analysis according to claim 1, wherein the step S4 comprises: s41, dividing the multidimensional feature vector into a training set and a verification set according to a time sequence, and distributing the training set and the verification set to a corresponding improved autoregressive moving diffusion model according to a channel; S42, loading training set data in a batch training mode, inputting a model to each batch of samples, and calculating a residual sequence between the prediction output and the observed value; S43, defining a robust loss function according to the residual sequence, wherein the robust loss function adopts a Huber form, and is calculated according to a square error when the absolute value of the residual is smaller than a threshold delta, and is calculated according to a linear error when the absolute value of the residual is larger than the threshold delta, and the threshold delta is adaptively set according to the quantile of the residual distribution of training data; s44, respectively performing gradient update on the autoregressive coefficient, the moving average coefficient, the drift function parameter and the diffusion function parameter by taking the minimization of the robust loss function as a target, wherein the gradient update adopts an adaptive learning rate mechanism; S45, after each round of training is finished, evaluating the performance of the model according to the mean square error of the residual error of the verification set, and terminating the training process when the improvement amplitude of the continuous several rounds of iteration is lower than a preset threshold value; s46, extracting a final parameter set of each error channel after training is completed, wherein the final parameter set comprises an autoregressive coefficient vector, a moving average coefficient vector, a drift function time factor parameter and a diffusion kernel space parameter.
6. 6. The method for optimizing satellite positioning errors based on big data analysis according to claim 2, wherein the step S5 comprises: S51, determining the spatial distribution range of the receiver in the target area according to the longitude and latitude and elevation data of the receiver, dividing the target geographic area into a regular space grid node set, and recording the geographic coordinates, the average elevation and the local terrain characteristic parameters of each node to form a node set covering the target geographic area; S52, calculating the geographic distance between any two space grid nodes, and determining a space association weight according to the building shielding similarity and the terrain similarity among the nodes, wherein the space association weight is calculated according to a Gaussian radial function; S53, constructing a geographic diffusion diagram according to the node set and the corresponding spatial association weight, wherein in the geographic diffusion diagram, nodes are used as vertexes, the spatial association weight is used as an edge weight, and a weighted undirected diagram structure reflecting geographic adjacency and environmental similarity is formed; S54, converting the node connection relation in the geographic diffusion diagram into a matrix form to obtain a node connection relation matrix, wherein the node connection relation matrix is the first node connection relation matrix Element corresponding node And node Is a spatial correlation weight of (1); s55, introducing the node connection relation matrix into constraint conditions of model training, executing constraint correction in each round of parameter updating according to weights corresponding to matrix elements, calculating a difference square term of adjacent node prediction errors, and accumulating the difference square term into a space regular term; and S56, outputting node spatial relationship and weight matrix data after training is finished, and generating a structured result file containing node indexes, spatial distances and associated weights.
7. 7. The method for optimizing satellite positioning errors based on big data analysis according to claim 6, wherein the step S6 includes: S61, acquiring real-time satellite observation data and environment characteristic data, wherein the real-time satellite observation data comprises a pseudo-range observation value, a carrier phase observation value, a satellite azimuth angle and a pitch angle, and the environment characteristic data comprises meteorological parameters, ionosphere electronic content, topography and a building shielding angle; S62, performing feature mapping and index matching on the real-time satellite observation data and the environment feature data according to a multidimensional feature dimension structure input by a model; s63, respectively inputting the real-time multidimensional feature vectors into an ionosphere error channel, a troposphere error channel, a multipath error channel and an improved autoregressive movement diffusion model corresponding to the equipment drift error channel, and calculating error prediction values of all channels; S64, determining weights according to the error variances of the channel models in the verification stage, and performing weighted fusion on the error prediction values to obtain an overall error correction quantity; S65, performing spatial smoothing on the fusion result by using the node connection relation matrix and the corresponding spatial association weight, and adjusting the consistency of error correction between adjacent nodes; s66, calculating uncertainty of the total error correction according to the weighted fusion and the spatial smoothing result, and outputting the total error correction and the uncertainty.
8. 8. The method for optimizing satellite positioning errors based on big data analysis according to claim 1, wherein the step S7 comprises: S71, acquiring the total error correction and uncertainty and a real-time satellite positioning result; S72, carrying out coordinate decomposition on the total error correction quantity, wherein the total error correction quantity corresponds to latitude, longitude and elevation components of a satellite positioning result respectively; S73, calculating a correction value for each coordinate component, and overlapping the correction value to a real-time positioning result according to a time index to generate a corrected positioning coordinate; s74, calculating a confidence interval range of the corrected positioning coordinates according to the overall uncertainty, and determining an upper limit and a lower limit of the coordinates according to the set confidence level; And S75, outputting the optimized positioning coordinates with the confidence interval, and recording the positioning time stamp, the correction quantity and the uncertainty parameter.

Description

Satellite positioning error optimization method based on big data analysis Technical Field The invention relates to the technical field of big data analysis and satellite navigation positioning, in particular to a satellite positioning error optimization method based on big data analysis. Background With the rapid development of high-precision positioning, automatic driving, geographic information acquisition, disaster monitoring and other applications, the positioning precision and reliability requirements of a satellite navigation system are continuously improved. In order to eliminate errors generated in the satellite signal transmission and reception process, the prior researches generally adopt methods such as differential positioning, model correction or empirical compensation, and the like to model and correct main error sources such as ionospheric delay, tropospheric delay, multipath effect, receiver drift, and the like. However, in a practical complex environment, the satellite signal propagation path is commonly affected by factors such as weather, ionosphere activity, topography, building shielding and the like, so that various errors show strong nonlinearity, time variability and spatial correlation, and the complex space-time coupling characteristic is difficult to accurately characterize by a traditional single-channel model or an independent error correction method. Most of the existing satellite positioning error optimization methods are based on empirical parameters or static models, such as estimating delay by ionosphere dual-frequency observation, performing delay correction based on troposphere empirical formulas, reducing multipath interference through geometric path analysis, and the like. The method generally assumes that error items are mutually independent and stable in time change, and omits dynamic association among multi-source environment characteristics, so that the error correction effect is obviously reduced in areas with complex terrain or frequent electromagnetic disturbance. In addition, the existing model generally depends on a small amount of ground reference stations or single type observation data, and is difficult to effectively fuse meteorological monitoring, ionosphere detection and geographic environment information, so that the input dimension of the model is limited, and the spatial resolution is insufficient. In terms of data processing, the traditional error modeling flow mainly adopts a single-source time sequence analysis or simple statistical smoothing method, and lacks a unified management and space-time alignment mechanism for multi-source heterogeneous data. Different sampling frequencies, different observation periods, and time reference differences across the sensing system often result in data misalignment, feature inconsistencies, and sample loss. The existing compensation strategies aiming at the problems mostly adopt linear interpolation or mean filling, and are difficult to maintain the dynamic continuity and abrupt change characteristics of error signals, so that the stability and the accuracy of subsequent modeling are affected. In addition, in the model training process, the conventional least squares regression or mean square error optimization function is highly sensitive to abnormal noise samples, and parameter deviation or over fitting phenomenon is easy to occur. For the processing of spatial correlation, most of the existing error correction models are only based on point-to-point distance or area average to carry out spatial smoothing, and the non-uniform propagation effect caused by geographic environment, building shielding and terrain difference cannot be considered, and a spatial constraint mechanism is not introduced in a training stage, so that under the condition of boundary area or non-uniform node distribution, error propagation is unstable and spatial consistency is poor. Overall, the existing satellite positioning error correction method generally has the problems of poor data synchronism, weak model adaptability, insufficient space constraint, uncertainty evaluation deficiency and the like, and is difficult to realize high-precision and stable real-time positioning optimization in a multi-source complex environment. Therefore, how to provide a satellite positioning error optimization method based on big data analysis is a problem that needs to be solved by those skilled in the art. Disclosure of Invention The invention aims to provide a satellite positioning error optimization method based on big data analysis, which realizes unified modeling of multi-source satellite observation and environmental characteristics by constructing an improved autoregressive movement diffusion model and introducing spatial regular constraint, improves the stability of parameter estimation by utilizing a steady loss function and a self-adaptive training mechanism, realizes spatial smoothing and error consistency correction by combining a geographi