A GNSS-based water vapor networking observation method and system

Abstract

The application provides a GNSS-based water vapor networking observation method and system, and relates to the technical field of meteorological observation, comprising obtaining first information and second information; obtaining a first feature vector corresponding to each GNSS receiver; obtaining a second feature vector; determining the correlation degree value of the first feature vector of each GNSS receiver and the second feature vector of the atmospheric water vapor information of a target region; and obtaining atmospheric water vapor information in a future preset time period. Through effective feature extraction and fusion of satellite data and atmospheric water vapor information collected by GNSS receivers in the water vapor networking, the observation of atmospheric water vapor is more accurate and real-time. Through principal component analysis to extract atmospheric influencing factors and combining a feature fusion model, the observation ability of water vapor information under complex atmospheric conditions is enhanced, and the regional atmospheric water vapor change can be observed in real time. Compared with the traditional meteorological observation station, the observation method has significant advantages in spatial distribution and time accuracy.

Description

Technical Field

[0001] This invention relates to the field of meteorological observation technology, and more specifically, to a GNSS-based water vapor network observation method and system. Background Technology

[0002] With the rapid development of Global Navigation Satellite System (GNSS) technology, GNSS-based methods for monitoring atmospheric water vapor have gradually emerged. GNSS signals experience delays when passing through the atmosphere due to atmospheric water vapor. Accurate measurement and analysis of these delays can reveal the atmospheric water vapor content. However, most existing GNSS-based water vapor monitoring methods rely on observations from single stations, only acquiring localized water vapor information at the observation point and failing to grasp the overall distribution and variation patterns of atmospheric water vapor within the region.

[0003] To achieve comprehensive, accurate, and real-time monitoring of atmospheric water vapor, constructing a GNSS-based networked water vapor observation system and method is of significant practical importance and urgent need. By networking multiple GNSS receivers, coordinating observations, and fusing data, it is expected to overcome the limitations of traditional observation methods and single-station GNSS observations, providing richer and more accurate atmospheric water vapor information for atmospheric science research and related applications. Summary of the Invention

[0004] The purpose of this invention is to provide a GNSS-based water vapor network observation method and system to improve the aforementioned problems. To achieve this objective, the technical solution adopted by this invention is as follows:

[0005] Firstly, this application provides a GNSS-based water vapor network observation method, including:

[0006] Acquire first information and second information. The first information is satellite data collected by the water vapor network in the target area and its corresponding time series. The second information is atmospheric water vapor data in the target area. The water vapor network includes multiple GNSS receivers.

[0007] Based on the first information, principal component analysis is performed to extract atmospheric influence factors, and the atmospheric influence factors are input into a preset feature transformation model for feature transformation to obtain the first feature vector corresponding to each GNSS receiver.

[0008] The second information is input into a preset feature fusion model for feature fusion to obtain a second feature vector;

[0009] Based on the first feature vector and the second feature vector, a correlation analysis is performed to determine the correlation value between the first feature vector of each GNSS receiver and the second feature vector of the atmospheric water vapor information of the target area.

[0010] The first feature vector, the second feature vector, and the correlation value of each GNSS receiver in the water vapor group are sent to a preset water vapor observation model for observation, so as to obtain atmospheric water vapor information of the target area in the future preset time period.

[0011] Secondly, this application also provides a GNSS-based water vapor network observation system, including:

[0012] The acquisition unit is used to acquire first information and second information. The first information is satellite data collected by the water vapor network in the target area and its corresponding time series. The second information is atmospheric water vapor data in the target area. The water vapor network includes multiple GNSS receivers.

[0013] The extraction unit is used to perform principal component analysis to extract atmospheric influence factors based on the first information, and input the atmospheric influence factors into a preset feature transformation model for feature transformation to obtain the first feature vector corresponding to each GNSS receiver.

[0014] The fusion unit is used to input the second information into a preset feature fusion model to perform feature fusion and obtain a second feature vector.

[0015] The first analysis unit is used to perform correlation analysis based on the first feature vector and the second feature vector to determine the correlation value between the first feature vector of each GNSS receiver and the second feature vector of the atmospheric water vapor information of the target area.

[0016] The observation unit is used to send the first feature vector, the second feature vector, and the correlation value of each GNSS receiver in the water vapor network to a preset water vapor observation model for observation, so as to obtain atmospheric water vapor information of the target area in the future preset time period.

[0017] The beneficial effects of this invention are as follows:

[0018] This invention achieves more accurate and real-time observation of atmospheric water vapor by effectively extracting and fusing features from satellite data and atmospheric water vapor information collected by GNSS receivers within a water vapor network. By extracting atmospheric influencing factors through principal component analysis and combining them with a feature fusion model, it can better process multi-source data, enhancing the ability to observe water vapor information under complex atmospheric conditions. It enables real-time observation of regional atmospheric water vapor changes. Compared to traditional meteorological observation stations, the observation method of this application has significant advantages in both spatial distribution and temporal accuracy.

[0019] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the GNSS-based water vapor network observation method described in this embodiment of the invention;

[0022] Figure 2 This is a schematic diagram of the GNSS-based water vapor network observation system described in an embodiment of the present invention.

[0023] The diagram is labeled as follows: 10, acquisition unit; 20, extraction unit; 30, fusion unit; 40, first analysis unit; 50, observation unit. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] Example 1:

[0027] This embodiment provides a GNSS-based water vapor network observation method.

[0028] See Figure 1 The figure shows that the method includes steps S10, S20, S30, S40 and S50.

[0029] Step S10. Obtain first information and second information. The first information is satellite data collected by the water vapor network in the target area and its corresponding time series. The second information is atmospheric water vapor data in the target area. The water vapor network includes multiple GNSS receivers.

[0030] Specifically, the water vapor network in this application consists of multiple GNSS receivers deployed at multiple ground stations. Over a large area, atmospheric water vapor content typically varies. Therefore, the spatial distribution of GNSS receivers directly affects the estimation accuracy of atmospheric water vapor information, and the relative positions of the receivers also influence their observation results. A denser GNSS receiver layout can more accurately capture spatial variations in atmospheric water vapor, while a sparse layout may lead to missing or inaccurate atmospheric water vapor information in some areas. To better observe atmospheric water vapor information in the target area, it is necessary to comprehensively consider the spatial distribution characteristics of each GNSS receiver in the water vapor network within the target area and their different contributions to atmospheric water vapor information. Therefore, this application considers the data collected by all GNSS receivers in the water vapor network within the target area for atmospheric water vapor information observation, ensuring full utilization of the data from each receiver, thereby improving the accuracy and comprehensiveness of the observation.

[0031] The first piece of information comprises all GNSS receivers included in the water vapor network within the target area, along with the satellite signal data acquired and the time-series information derived from this data. This data is helpful in analyzing the propagation delay and atmospheric effects of GNSS signals. The second piece of information refers to the specific atmospheric water vapor information of the target area, typically used to reflect the meteorological conditions of the target area, especially water vapor concentration. The combination of these two pieces of information provides the foundational data for subsequent atmospheric water vapor observation and analysis.

[0032] Step S20. Based on the first information, perform principal component analysis to extract atmospheric influence factors, and input the atmospheric influence factors into a preset feature transformation model for feature transformation to obtain the first feature vector corresponding to each GNSS receiver;

[0033] Specifically, the satellite data collected by the GNSS receivers in each water vapor network contains a lot of data, some of which can effectively reflect the water vapor information in the current atmosphere. Therefore, the most representative influencing factors, especially those related to water vapor, can be extracted from these multidimensional data. The extracted atmospheric influencing factors are then input into a preset feature transformation model to perform data standardization and dimensionality reduction, thereby improving the usability and accuracy of the data.

[0034] Specifically, step S20 includes steps S21 to S25:

[0035] Step S21. Perform principal component analysis on the satellite data in the first information. Calculate the covariance matrix of each feature in the satellite data and perform eigenvalue decomposition to obtain the simplified principal component feature set.

[0036] Specifically, by performing principal component analysis on the satellite data in the first information, calculating the covariance matrix of the features and performing eigenvalue decomposition, the main atmospheric-related influencing factors can be effectively extracted, the remaining irrelevant factors can be removed, and redundancy and noise information can be reduced.

[0037] S22. Based on the distance and time differences between sample points in the principal component feature set, construct a k-nearest neighbor graph. Nodes in the k-nearest neighbor graph represent sample points, and the weights of the edges represent the similarity values ​​between the corresponding sample points.

[0038] Specifically, constructing a k-nearest neighbor graph based on the distance and temporal differences between sample points in the principal component feature set can effectively reveal the similarity relationships between sample points. The k-nearest neighbor graph can not only reflect the spatial similarity of sample points, but also incorporate the temporal dimension to consider the mutual influence of sample points over time.

[0039] Specifically, step S22 includes steps S221 to S223:

[0040] Step S221. Calculate the distance difference and time difference between each sample point in the principal component feature set and construct vectors to obtain the distance vector and time vector respectively;

[0041] Specifically, step S221 includes steps S2211 to S2218:

[0042] Step S2211. Calculate the Euclidean distance between the sample point and the other sample points to obtain multiple first distances;

[0043] Step S2212. Determine the value within a preset distance range from multiple first distances as the reference distance;

[0044] Step S2213. Calculate the difference between the first distance and the reference distance corresponding to the sample point to obtain multiple second distances;

[0045] Step S2214. Construct a distance vector based on the second distance;

[0046] Step S2215. Calculate the time difference between the sample point and the other sample points to obtain multiple first differences;

[0047] Step S2216. Determine the value within a preset time range from multiple first differences as the benchmark difference;

[0048] Step S2217. Calculate the difference between the first difference corresponding to the sample point and the benchmark difference to obtain multiple second differences;

[0049] Step S2218. Construct a time vector based on the second difference;

[0050] Specifically, the Euclidean distance and time difference between sample points are calculated, a baseline value is determined based on a preset range, and the difference between the baseline value and the distance difference is used as a vector. This effectively improves the accuracy of similarity measurement, enhances the ability to capture spatiotemporal correlations, improves computational efficiency, and enhances model robustness, thus providing more accurate and effective support in complex spatiotemporal data analysis tasks. In this embodiment, the preset distance range is [9,13], and the preset time range is [2,8]. These values ​​can be adjusted according to actual conditions, and no special restrictions are imposed here.

[0051] Step S222. Based on the distance vector and time vector corresponding to the two sample points respectively, calculate the similarity value of the two sample points;

[0052] Specifically, the formula for calculating the similarity value between two sample points includes:

[0053]

[0054] Among them, S ab This represents the similarity value between the a-th sample point and the b-th sample point; This is the distance vector corresponding to the a-th sample point; Let be the time vector corresponding to the a-th sample point; This is the distance vector corresponding to the b-th sample point; Let a be the time vector corresponding to the b-th sample point; a ≠ b, a, b ∈ N, and N is the number of sample points.

[0055] Step S223. Construct a k-nearest neighbor graph based on similarity values, where the length of the edges in the k-nearest neighbor graph is a set threshold.

[0056] Specifically, this step analyzes the similarity of each sample point based on distance and time factors. By constructing a k-nearest neighbor graph, the system can more intuitively display the relationships between feature data, which is helpful for subsequent structural feature extraction.

[0057] Step S23. Perform graph convolution operation based on the k-nearest neighbor graph to pass the information of each sample point to the neighboring sample points, and combine the information of the neighboring sample points to update the features and obtain the updated sample point information.

[0058] Specifically, this step uses a structural feature extraction neural network algorithm to perform graph convolution operations. For the constructed k-nearest neighbor graph, the algorithm allows the information of each sample point to be passed to its neighboring sample points, and the information is updated by combining the information of the neighboring sample points.

[0059] The formulas for updating sample point information include:

[0060]

[0061] Among them, s' i The updated sample point information for the i-th sample point; s i For the sample point information of the i-th sample point; s j For the sample point information of the j-th sample point; X ij Let be the correlation value between the i-th sample point and the j-th sample point; W is the weight matrix; δ is the activation function; and N is the number of sample points.

[0062] Specifically, step S23 includes steps S231 to S233:

[0063] Step S231. Use the similarity value matrix as the weight matrix, and input the weight matrix and the preset activation function into the input layer of the graph convolutional neural network;

[0064] Specifically, using the similarity matrix as the weight matrix means that the similarity between sample points is quantified as the strength of connections in a graph convolutional network. In a graph convolutional neural network, these similarity values ​​determine the degree of information transfer between sample points, which is represented by the edge weights in the graph. By inputting this weight matrix along with a predefined activation function into the input layer of the graph convolutional neural network, the network can dynamically update the feature representation of each sample point based on the similarity between them. The activation function helps the model introduce non-linear relationships in this process, making feature updates more flexible and thus improving the network's ability to express complex data patterns.

[0065] Step S232. The information of each sample point is weighted and summed with the information transmitted from the adjacent sample points, and then linearly transformed through the weight matrix to obtain the feature data information after linear transformation;

[0066] Specifically, this step can effectively aggregate local neighborhood information, so that each sample point not only retains its own feature information, but also updates its own features by combining the information of its neighboring sample points.

[0067] Step S233. Based on the activation function, perform nonlinear mapping on the linearly transformed feature data information to obtain the updated sample point information.

[0068] Specifically, this step, through the nonlinear mapping of the activation function, enables the system to learn and capture more complex patterns and features in the data, effectively improving the expressive power of the neural network.

[0069] Step S24. Perform pooling operation on all updated sample point information to obtain multiple overall sample information;

[0070] Step S25. Send the overall sample information after pooling to the fully connected layer to obtain the first feature vector;

[0071] Specifically, pooling is performed on all updated sample point information. By extracting the maximum or average value of features globally, the dimensionality of the data is reduced while retaining important feature information, making the features more concise and representative. The fully connected layer further processes the pooled features, using linear transformations and nonlinear activation functions to generate the first feature vector for each GNSS receiver. This first feature vector represents a refined expression of all sample information within the GNSS receiver and can be used as input for subsequent tasks.

[0072] Step S30. Input the second information into the preset feature fusion model to perform feature fusion and obtain the second feature vector;

[0073] Specifically, by inputting the second piece of information into a pre-defined feature fusion model for feature fusion, atmospheric water vapor information from multiple sources or dimensions can be effectively integrated to extract the most representative features. Through feature fusion, the model can retain key information while reducing redundancy and noise, thereby improving the data's expressive power.

[0074] Specifically, step S30 includes steps S31 to S33:

[0075] Step S31. Analyze and process the second information, wherein the second information is divided into multiple levels, each level has a sequential inclusion relationship from top to bottom, and a judgment matrix is ​​constructed based on the importance of the data corresponding to each pair of levels.

[0076] Step S32. Calculate the weight of each level in the judgment matrix based on the AHP algorithm to obtain the relative importance value of each data in the second information;

[0077] Step S33. Sort the relative importance values ​​in order of magnitude, and extract and fuse the features of the data that are ranked before the set position to obtain the second feature vector;

[0078] Specifically, the second information is divided into multiple levels, and the weight of data at each level is calculated based on the AHP algorithm. This allows for a scientific assessment of the relative importance of each data point within the entire information system, enabling reasonable ranking based on data importance. Subsequently, data features with higher importance are extracted and fused in a predetermined order to obtain a concise and discriminative second feature vector. This approach not only improves the efficiency of feature selection but also ensures that the most critical data is retained, contributing to improved accuracy and robustness of subsequent observations and analyses, while reducing interference from redundant information.

[0079] Step S40. Based on the first feature vector and the second feature vector, perform correlation analysis to determine the correlation value between the first feature vector of each GNSS receiver and the second feature vector of the atmospheric water vapor information of the target area;

[0080] Specifically, correlation analysis based on the first and second feature vectors can effectively assess the relationship between geographic information and atmospheric water vapor information in the area where the GNSS receiver is located.

[0081] Specifically, step S40 includes steps S41 to S43:

[0082] Step S41. Based on the mean transformation method, the first eigenvector and the second eigenvector are dimensionless to obtain the dimensionless first eigenvector and the second eigenvector respectively.

[0083] Specifically, the differences between all first eigenvectors and corresponding second eigenvectors in each water vapor network are eliminated through dimensionless processing. Dimensionless processing includes maximum-minimum normalization, standard deviation normalization, and maximum absolute value normalization, etc., without special restrictions here.

[0084] Step S42. Calculate the relationship coefficients based on the dimensionless first and second eigenvectors to obtain the relationship coefficients between the dimensionless first and second eigenvectors.

[0085] Step S43. Calculate the correlation value between the dimensionless first eigenvector and the second eigenvector based on the relation coefficient;

[0086] Specifically, the formula for calculating the relationship coefficient is as follows:

[0087]

[0088] Where x1 is the dimensionless first feature vector; x2 is the dimensionless second feature vector; t1 is the first information acquisition time information corresponding to the first feature vector; t2 is the second information acquisition time information corresponding to the second feature vector; σ is the resolution coefficient, with a value range of [1,0]; γ is the relationship coefficient between the dimensionless first feature vector and the second feature vector.

[0089] By quantifying the correlation between the first and second eigenvectors, we can provide a valid basis for subsequent observation, modeling, or decision-making, and help the model better understand the intrinsic relationship between the two eigenvectors.

[0090] Step S50. Send the first feature vector, the second feature vector, and the correlation value of each GNSS receiver in the water vapor network to the preset water vapor observation model for observation, and obtain the atmospheric water vapor information of the area where the GNSS receiver is located in the future preset time period;

[0091] Specifically, the first feature vectors and corresponding second feature vectors of multiple GNSS receivers within the same water vapor network are sent to a pre-defined water vapor observation model. Based on the input feature vectors and correlation values, the model uses specific algorithms and principles to analyze and calculate the water vapor conditions at the locations of the GNSS receivers within the water vapor network. This yields atmospheric water vapor information for a predetermined time period at the receiver's location, such as water vapor content and vertical distribution. The resulting prediction data can provide crucial decision-making support in various fields, particularly in addressing extreme weather, resource management, environmental protection, and public safety.

[0092] Example 2:

[0093] like Figure 2 As shown, this embodiment provides a GNSS-based water vapor network observation system, which includes:

[0094] The acquisition unit 10 is used to acquire first information and second information. The first information is satellite data collected by the water vapor network in the target area and its corresponding time series. The second information is atmospheric water vapor data in the target area. The water vapor network includes multiple GNSS receivers.

[0095] Extraction unit 20 is used to extract atmospheric influence factors based on the first information by performing principal component analysis, and input the atmospheric influence factors into a preset feature transformation model for feature transformation to obtain the first feature vector corresponding to each GNSS receiver;

[0096] The fusion unit 30 is used to input the second information into a preset feature fusion model to perform feature fusion and obtain a second feature vector;

[0097] The first analysis unit 40 is used to perform correlation analysis based on the first feature vector and the second feature vector to determine the correlation value between the first feature vector of each GNSS receiver and the second feature vector of the atmospheric water vapor information of the target area.

[0098] The observation unit 50 is used to send the first feature vector, the second feature vector, and the correlation value of each GNSS receiver in the water vapor network to the preset water vapor observation model for observation, so as to obtain atmospheric water vapor information of the target area in the future preset time period.

[0099] In one specific embodiment disclosed in this application, the extraction unit 20 includes:

[0100] The second analysis unit is used to perform principal component analysis based on the satellite data in the first information. By calculating the covariance matrix of each feature in the satellite data and performing eigenvalue decomposition, a simplified principal component feature set is obtained.

[0101] The first construction unit is used to construct a k-nearest neighbor graph based on the distance and time differences between sample points in the principal component feature set. Nodes in the k-nearest neighbor graph represent sample points, and the weights of the edges represent the similarity values ​​between the corresponding sample points.

[0102] The transfer unit is used to perform graph convolution operations based on the k-nearest neighbor graph, passing the information of each sample point to the neighboring sample points, and combining the information of the neighboring sample points to update the features and obtain the updated sample point information.

[0103] The pooling unit is used to perform pooling operations on all updated sample point information to obtain multiple overall sample information.

[0104] The sending unit is used to send the overall sample information after pooling to the fully connected layer to obtain the first feature vector.

[0105] In one specific embodiment disclosed in this application, the first building unit includes:

[0106] The first calculation unit is used to calculate the distance difference and time difference between each sample point in the principal component feature set and construct vectors to obtain the distance vector and time vector respectively.

[0107] The second calculation unit is used to calculate the similarity value between two sample points based on the distance vector and time vector corresponding to the two sample points respectively.

[0108] The second building unit is used to construct a k-nearest neighbor graph based on similarity values, where the length of the edges in the k-nearest neighbor graph is a set threshold.

[0109] In one specific embodiment disclosed in this application, the fusion unit 30 includes:

[0110] The third analysis unit is used to analyze and process the second information. The second information is divided into multiple levels, and each level has a sequential inclusion relationship from top to bottom. The importance of the data corresponding to each pair of levels is compared to construct a judgment matrix.

[0111] The judgment unit is used to calculate the weight of each level in the judgment matrix based on the AHP algorithm, so as to obtain the relative importance value of each data in the second information;

[0112] The sorting unit is used to sort the relative importance values ​​in order of magnitude, and to extract and fuse the features of the data that are ranked before the set position to obtain the second feature vector.

[0113] In one specific embodiment disclosed in this application, the first computing unit includes:

[0114] The third calculation unit is used to calculate the Euclidean distance between the sample point and the other sample points to obtain multiple first distances;

[0115] The first determining unit is used to determine a value within a preset distance range from a plurality of first distances, as a reference distance;

[0116] The fourth calculation unit is used to calculate the difference between the first distance and the reference distance corresponding to the sample point, and obtain multiple second distances;

[0117] The first constituent unit is used to construct a distance vector based on the second distance;

[0118] The fifth calculation unit is used to calculate the time difference between the sample point and the other sample points, and obtain multiple first differences;

[0119] The second determining unit is used to determine the value within a preset time range from multiple first differences as the benchmark difference;

[0120] The sixth calculation unit is used to calculate the difference between the first difference corresponding to the sample point and the benchmark difference, and obtain multiple second differences;

[0121] The second constituent unit is used to construct a time vector based on the second difference.

[0122] In one specific embodiment disclosed in this application, the transmission unit includes:

[0123] As a unit, it is used to take the similarity value matrix as the weight matrix and input the weight matrix and the preset activation function into the input layer of the graph convolutional neural network;

[0124] The weighted unit is used to sum the information of each sample point with the information passed from the neighboring sample points, and then perform a linear transformation through the weight matrix to obtain the linearly transformed feature data information.

[0125] The mapping unit is used to perform nonlinear mapping on the linearly transformed feature data based on the activation function to obtain updated sample point information.

[0126] In one specific embodiment disclosed in this application, the first analysis unit 40 includes:

[0127] The processing unit is used to perform dimensionless processing on the first eigenvector and the second eigenvector based on the mean transformation method, so as to obtain the dimensionless first eigenvector and the second eigenvector respectively.

[0128] The seventh calculation unit is used to calculate the relationship coefficient based on the dimensionless first eigenvector and the second eigenvector, and to obtain the relationship coefficient between the dimensionless first eigenvector and the second eigenvector.

[0129] The eighth calculation unit is used to calculate the correlation value of the dimensionless first eigenvector and the second eigenvector based on the relation coefficient.

[0130] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0131] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0132] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A GNSS-based water vapor network observation method, characterized in that, include: Acquire first information and second information. The first information is satellite data collected by the water vapor network in the target area and its corresponding time series. The second information is atmospheric water vapor data in the target area. The water vapor network includes multiple GNSS receivers. Based on the first information, principal component analysis is performed to extract atmospheric influence factors, and the atmospheric influence factors are input into a preset feature transformation model for feature transformation to obtain the first feature vector corresponding to each GNSS receiver. The second information is input into a preset feature fusion model for feature fusion to obtain a second feature vector; Based on the first feature vector and the second feature vector, a correlation analysis is performed to determine the correlation value between the first feature vector of each GNSS receiver and the second feature vector of the atmospheric water vapor information of the target area. The first feature vector, the second feature vector, and the correlation value of each GNSS receiver in the water vapor network are sent to a preset water vapor observation model for observation, so as to obtain atmospheric water vapor information of the target area in the future preset time period; Specifically, the correlation analysis based on the first feature vector and the second feature vector determines the correlation value between the first feature vector of each GNSS receiver and the second feature vector of the target area, including: The first eigenvector and the second eigenvector are dimensionless based on the mean transformation method to obtain the dimensionless first eigenvector and the second eigenvector, respectively. The relationship coefficient is calculated based on the dimensionless first and second eigenvectors, and the formula for calculating the relationship coefficient is as follows: ； in, This is the first eigenvector after dimensionless transformation; This is the dimensionless second eigenvector; The first information acquisition time information corresponding to the first feature vector; This refers to the acquisition time information of the second information corresponding to the second feature vector; The resolution coefficient has a range of values. ; The relationship coefficients between the first and second eigenvectors after dimensionless transformation; The correlation value between the dimensionless first eigenvector and the second eigenvector is calculated based on the relation coefficient.

2. The GNSS-based water vapor network observation method according to claim 1, characterized in that... Based on the first information, principal component analysis is performed to extract atmospheric influence factors, and these atmospheric influence factors are input into a preset feature transformation model for feature transformation to obtain a first feature vector corresponding to each GNSS receiver, including: Principal component analysis is performed on the satellite data in the first information. By calculating the covariance matrix of each feature in the satellite data and performing eigenvalue decomposition, a simplified principal component feature set is obtained. Based on the distance and time differences between sample points in the principal component feature set, a k-nearest neighbor graph is constructed. In the k-nearest neighbor graph, nodes represent sample points, and the weights of the edges represent the similarity values ​​between the corresponding sample points. Graph convolution is performed based on the k-nearest neighbor graph to pass the information of each sample point to the neighboring sample points, and feature updates are performed by combining the information of the neighboring sample points to obtain the updated sample point information. All updated sample point information is pooled to obtain multiple overall sample information; The overall sample information after pooling is sent to the fully connected layer to obtain the first feature vector.

3. The GNSS-based water vapor network observation method according to claim 2, characterized in that... Based on the distance and time differences between sample points in the principal component feature set, a k-nearest neighbor graph is constructed, including: Calculate the distance difference and time difference between each sample point in the principal component feature set and construct vectors to obtain the distance vector and time vector respectively; Based on the distance vector and time vector corresponding to the two sample points respectively, the similarity value of the two sample points is calculated; The k-nearest neighbor graph is constructed based on the similarity value, and the length of the edge in the k-nearest neighbor graph is a set threshold.

4. The GNSS-based water vapor network observation method according to claim 1, characterized in that... The second information is input into a preset feature fusion model for feature fusion to obtain a second feature vector, including: The second information is analyzed and processed, wherein the second information is divided into multiple levels, each level has a sequential inclusion relationship from top to bottom, and a judgment matrix is ​​constructed based on the importance comparison of the data corresponding to each pair of levels. The weight of each level in the judgment matrix is ​​calculated based on the AHP algorithm to obtain the relative importance value of each data in the second information; The relative importance values ​​are sorted in order of magnitude, and the data that are ranked before the set position are subjected to feature extraction and fusion to obtain the second feature vector.

5. The GNSS-based water vapor network observation method according to claim 3, characterized in that... Calculate the distance difference and time difference between each sample point in the principal component feature set and construct vectors to obtain the distance vector and time vector, respectively, including: Calculate the Euclidean distance between the sample point and the other sample points to obtain multiple first distances; The value within a preset distance range is determined from multiple first distances and used as the reference distance; Calculate the difference between the first distance corresponding to the sample point and the reference distance to obtain multiple second distances; The distance vector is constructed based on the second distance; Calculate the time difference between the sample point and the other sample points to obtain multiple first differences; The value within a preset time range is determined from multiple first differences and used as the benchmark difference; Calculate the difference between the first difference corresponding to the sample point and the benchmark difference to obtain multiple second differences; The time vector is constructed based on the second difference.

6. The GNSS-based water vapor network observation method according to claim 3, characterized in that... Based on the k-nearest neighbor graph, a graph convolution operation is performed to pass the information of each sample point to its neighboring sample points, and the information of the neighboring sample points is combined to update the features, resulting in updated sample point information, including: The similarity value matrix is ​​used as the weight matrix, and the weight matrix and the preset activation function are input into the input layer of the graph convolutional neural network. The information of each sample point is weighted and summed with the information passed from the adjacent sample points, and then linearly transformed through the weight matrix to obtain the linearly transformed feature data information. The feature data information after linear transformation is nonlinearly mapped based on the activation function to obtain updated sample point information.

7. The GNSS-based water vapor network observation method according to claim 6, characterized in that... The formula for updating sample point information includes: ； in, For the first Updated sample point information for each sample point; For the first Sample point information for each sample point; For the first Sample point information for each sample point; For the first The updated correlation value of sample point information for each sample point; This is the weight matrix; For activation functions; This represents the number of sample points.

8. A GNSS-based water vapor network observation system, characterized in that, include: The acquisition unit is used to acquire first information and second information. The first information is satellite data collected by the water vapor network in the target area and its corresponding time series. The second information is atmospheric water vapor data in the target area. The water vapor network includes multiple GNSS receivers. The extraction unit is used to perform principal component analysis to extract atmospheric influence factors based on the first information, and input the atmospheric influence factors into a preset feature transformation model for feature transformation to obtain the first feature vector corresponding to each GNSS receiver. The fusion unit is used to input the second information into a preset feature fusion model to perform feature fusion and obtain a second feature vector. The first analysis unit is used to perform correlation analysis based on the first feature vector and the second feature vector to determine the correlation value between the first feature vector of each GNSS receiver and the second feature vector of the atmospheric water vapor information of the target area. The observation unit is used to send the first feature vector, the second feature vector, and the correlation value of each GNSS receiver in the water vapor network to a preset water vapor observation model for observation, so as to obtain atmospheric water vapor information of the target area in the future preset time period; Specifically, the correlation analysis based on the first feature vector and the second feature vector determines the correlation value between the first feature vector of each GNSS receiver and the second feature vector of the target area, including: The first eigenvector and the second eigenvector are dimensionless based on the mean transformation method to obtain the dimensionless first eigenvector and the second eigenvector, respectively. The relationship coefficient is calculated based on the dimensionless first and second eigenvectors, and the formula for calculating the relationship coefficient is as follows: ； in, This is the first eigenvector after dimensionless transformation; This is the dimensionless second eigenvector; The first information acquisition time information corresponding to the first feature vector; This refers to the acquisition time information of the second information corresponding to the second feature vector; The resolution coefficient has a range of values. ; The relationship coefficients between the first and second eigenvectors after dimensionless transformation; The correlation value between the dimensionless first eigenvector and the second eigenvector is calculated based on the relation coefficient.

Images