Satellite-borne GNSS-R sea surface wind speed inversion method based on multilayer iterative weighted model
Through multi-layer iterative weighted model and segmented modeling technology, the problem of large wind speed inversion error in deep learning models in the high wind speed interval is solved, the accuracy and reliability of sea surface wind speed inversion are improved, and different environmental conditions are adapted to.
Patent Information
- Application Number
- CN202411435363.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-15
- Publication Date
- 2024-11-29
AI Technical Summary
The existing deep learning models have large error differences in different wind speed intervals in sea surface wind speed inversion, especially in high wind speed intervals with low inversion accuracy and difficult to distinguish between effective signals and noise, resulting in large errors in wind speed intervals with wind speed intervals.
Using the satellite-borne GNSS-R sea surface wind speed inversion method based on multi-layer iterative weighting model, the segmented wind speed interval inversion model is constructed through segmented modeling and iterative calculation, and the weighting mechanism of multiple segmented models is used to reduce the single model's Error, improve wind speed inversion accuracy.
It improves the accuracy and reliability of wind speed inversion, especially in the high wind speed range, reduces errors, enhances the applicability and flexibility of the model, and adapts to different environments and conditions.
Smart Images

Figure CN119045004A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of satellite remote sensing, and specifically is a satellite-borne GNSS-R sea surface wind speed inversion method based on a multi-layer iterative weighted model. Background Art
[0002] Global Navigation Satellite System-Reflectometry (GNSS-R) is a passive remote sensing technology that uses signals transmitted from satellites and reflected on the sea surface to obtain ocean surface information. The characteristics of the reflected signal (such as delay, frequency offset, signal strength, etc.) are affected by the roughness of the sea surface, which is closely related to the wind speed. Therefore, the wind speed on the sea surface can be indirectly inferred by analyzing the reflected signal. Traditional sea surface wind speed measurements usually rely on active remote sensing instruments (such as synthetic aperture radar, microwave scatterometer) carried by meteorological buoys or satellites. However, these methods are expensive and have limited coverage. GNSS-R greatly reduces equipment costs by using existing global navigation satellite signals and can provide continuous monitoring over a large area. With the rapid development of deep learning, researchers have begun to use models such as deep neural networks and convolutional neural networks to process GNSS-R data. By automatically extracting and analyzing complex features in the reflected signal, the deep learning model can learn the nonlinear relationship between the reflected signal and the sea surface wind speed under large data volumes, significantly improving the inversion accuracy.
[0003] Although the deep learning-based satellite GNSS-R model has played an important role in the inversion of sea surface wind speed, there are still some shortcomings and challenges: 1. The existing deep learning model uses the full wind speed range data for model training, and uses the trained model to estimate wind speed. Due to the large difference in the proportion of samples in different wind speed ranges in the training samples, the wind speed inversion errors in different wind speed ranges are quite different, showing a trend that the wind speed inversion error gradually increases with the increase of wind speed, especially the inversion error in the high wind speed range is low. 2. In a high wind speed environment, the noise in the reflected signal may increase, and the full wind speed range model is difficult to distinguish between the effective signal and noise in the high wind speed range, further reducing the accuracy of wind speed inversion in the high wind speed range. Summary of the invention
[0004] The present invention aims to solve the deficiencies of the above-mentioned prior art and proposes a satellite-borne GNSS-R sea surface wind speed inversion method based on a multi-layer iterative weighted model, in order to improve the accuracy of wind speed inversion and reduce the wind speed inversion error in high wind speed ranges, thereby providing a new idea for sea surface wind speed inversion with high temporal and spatial resolution.
[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme:
[0006] The invention discloses a method for inverting sea surface wind speed using a satellite-borne GNSS-R based on a multi-layer iterative weighted model, which comprises the following steps:
[0007] S1. After obtaining and preprocessing the satellite-borne GNSS-R data, n satellite-borne GNSS-R feature value sequences are extracted from it, where the jth satellite-borne GNSS-R feature value sequence X j ={ },in, represents the jth satellite-borne GNSS-R eigenvalue sequence X j The i-th eigenvalue in , M represents the total number of eigenvalues of each satellite-borne GNSS-R eigenvalue sequence, j=1,2,…,n;
[0008] According to the acquisition time Time of the satellite-borne GNSS-R data, the sea surface wind speed data set is synchronously obtained, and according to the position of the mirror reflection point (SPLat, SPLon) in the satellite-borne GNSS-R data, the j-th satellite-borne GNSS-R eigenvalue sequence X j After temporal and spatial matching with the sea surface wind speed dataset, the sea surface wind speed corresponding to the jth satellite-borne GNSS-R eigenvalue sequence is obtained. , so that X j and The jth data sample is formed, and the n data samples form the full wind speed interval data set D;
[0009] The full wind speed interval data set D is randomly divided into the full wind speed interval training set D according to a certain ratio. train , full wind speed interval verification set D val And the full wind speed range test set D test ;
[0010] S2, using the full wind speed interval training set D train The deep learning model is trained to obtain the inversion model F for the entire wind speed range;
[0011] S3, the full wind speed interval verification set D val All satellite-borne GNSS-R feature value sequences {X k val |k=1,2,…,m} is input into the full wind speed interval inversion model F for processing to obtain the predicted satellite-borne GNSS-R wind speed sequence { |k=1,2,…,m}, where X k val Indicates D val The kth satellite-borne GNSS-R eigenvalue sequence in, represents the kth predicted satellite-borne GNSS-R wind speed, and m represents D val The total number of data samples in ;
[0012] S4, based on the full wind speed range verification set D val The middle sea surface wind speed series |k=1,2,…,m}, statistically analyze the inversion model F for D in the full wind speed range val Total wind speed retrieval error ;
[0013] According to a certain wind speed interval p, the full wind speed interval consisting of the maximum sea surface wind speed and the minimum sea surface wind speed in the full wind speed interval data set D is divided into t wind speed intervals, and the mean error { |s=1,2,…t}; where represents the mean error of the sth wind speed interval;
[0014] S5, according to the mean error of the sth wind speed interval The total wind speed inversion error , the full wind speed interval is re-divided into q segmented wind speed intervals, and they are respectively denoted as { |v=1,2,…q}; where represents the vth segment wind speed interval, ;
[0015] S6, according to q segmented wind speed intervals { |v=1,2,…q}, for D train Classify and obtain the segmented training set corresponding to the q segmented wind speed intervals |v=1,2,…q}, where Indicates the vth segment wind speed interval The corresponding segmented training set;
[0016] use The deep learning model is trained to obtain the vth segment wind speed interval inversion model H v , thus obtaining the q segmented wind speed interval inversion model H v |v=1,2,…q};
[0017] S7, D test Middle Satellite-borne GNSS-R characteristic value sequence X l test The full wind speed interval inversion model F is input for processing to obtain the predicted Satellite-borne GNSS-R wind speed ;
[0018] S8. According to The mean error corresponding to the wind speed interval ,calculate Satellite-borne GNSS-R characteristic value sequence X l test The wind speed estimation interval [ , + ), denoted as R l 0 ;
[0019] S9, define the current number of iterations as c, and initialize c=1;
[0020] S10, initialize v=1;
[0021] S11. Calculate the wind speed estimation interval R in the c-1th iteration l c-1 and the vth segment wind speed interval The overlap degree in the cth iteration , and judge >0 is true, if true, then the vth segment wind speed interval inversion model H v As the candidate inversion model for the cth iteration, As H v Weight ; Otherwise, after assigning v+1 to v, return to S11 until v>q, thereby obtaining several candidate inversion models and their weights after screening in the cth iteration;
[0022] S12, D test X in l test The predicted satellite-borne GNSS-R wind speed sequences output by each candidate inversion model in the cth iteration are weighted and averaged with the weights corresponding to the candidate inversion model in the cth iteration to obtain X l test The inverted value of sea surface wind speed in the cth iteration ;
[0023] S13. According to The mean error corresponding to the wind speed interval , calculate the wind speed range in the cth iteration [ , + ), denoted as R l c ;
[0024] S14. Judgment The wind speed inversion with round c-1 Is the absolute value of the difference less than a certain limit? If it is less than a certain limit, it means that the sea surface wind speed inversion value under the cth iteration is the optimal value and outputs it, ending the process; otherwise, assign c+1 to c and return to S10 for sequential execution. When c=1, initialize .
[0025] An electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the satellite-borne GNSS-R sea surface wind speed inversion method, and the processor is configured to execute the program stored in the memory.
[0026] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the satellite-borne GNSS-R sea surface wind speed inversion method when the computer program is executed by a processor.
[0027] Compared with the prior art, the advantages of the present invention are:
[0028] 1. The present invention realizes segmented modeling for different wind speed intervals and proposes to construct a segmented wind speed interval inversion model, thereby improving the applicability of the model in each wind speed interval;
[0029] 2. The present invention uses the wind speed inversion result and the inversion error to obtain the estimated range of the wind speed, and improves the wind speed inversion accuracy through repeated iterative calculations;
[0030] 3. The weighting mechanism of the inversion model using multiple segmented wind speed intervals in the present invention can effectively reduce the error of a single model and enhance the reliability of the wind speed inversion results;
[0031] 4. The present invention is flexible and adaptable, and can flexibly adjust the combination of wind speed range and model according to different environments and conditions. It has strong adaptability and helps to maintain good performance in different application scenarios, especially under high wind speeds. It provides a new idea for the problem of large error in inversion of sea surface wind speed in high wind speed range faced by existing research. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 Flow chart for implementing the present invention;
[0033] Figure 2 Flowchart for building and training wind speed inversion model;
[0034] Figure 3 This is the inversion result diagram of the wind speed inversion model. DETAILED DESCRIPTION
[0035] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Figure 1 and Figure 2 , a method for inverting sea surface wind speed using a spaceborne GNSS-R based on a multi-layer iterative weighted model, comprising the following steps:
[0036] S1. After obtaining and preprocessing the satellite-borne GNSS-R data, n satellite-borne GNSS-R feature value sequences are extracted from it, where the jth satellite-borne GNSS-R feature value sequence X j ={ },in, represents the jth satellite-borne GNSS-R eigenvalue sequence X j The i-th eigenvalue in , M represents the total number of eigenvalues of each satellite-borne GNSS-R eigenvalue sequence, j=1,2,…,n;
[0037] According to the acquisition time Time of the satellite-borne GNSS-R data, the sea surface wind speed data set is synchronously obtained, and according to the position of the mirror reflection point (SPLat, SPLon) in the satellite-borne GNSS-R data, the j-th satellite-borne GNSS-R eigenvalue sequence X j After temporal and spatial matching with the sea surface wind speed dataset, the sea surface wind speed corresponding to the jth satellite-borne GNSS-R eigenvalue sequence is obtained. , so that X j and The j-th data sample is constituted, and the n data samples constitute the full wind speed interval data set D; the sea surface wind speed data set used in this embodiment is the ERA5 sea surface wind speed data, and the data preprocessing content is to match the CYGNSS data with the ERA5 sea surface wind speed data in time and space, eliminate Nan values or negative values, and filter the data according to the quality control flag to obtain the required delay Doppler map and characteristic parameters, where the characteristic parameters include normalized dual-static radar cross section, front slope, geographic location information, etc.
[0038] The full wind speed interval data set D is randomly divided into the full wind speed interval training set D according to a certain ratio. train , full wind speed interval verification set D val And the full wind speed range test set D test In this embodiment, the ratio of the training set, validation set, and test set is 6:2:2;
[0039] S2, using the full wind speed interval training set D train The deep learning model is trained to obtain the inversion model F for the entire wind speed range;
[0040] S3, the full wind speed interval verification set D val All satellite-borne GNSS-R feature value sequences {Xk val |k=1,2,…,m} is input into the full wind speed interval inversion model F for processing to obtain the predicted satellite-borne GNSS-R wind speed sequence { |k=1,2,…,m}, where X k val Indicates D val The kth satellite-borne GNSS-R eigenvalue sequence in, represents the kth predicted satellite-borne GNSS-R wind speed sequence, and m represents D val The total number of data samples in ;
[0041] S4, based on the full wind speed range verification set D val The middle sea surface wind speed series |k=1,2,…,m}, statistically analyze the inversion model F for D in the full wind speed range val The total wind speed inversion error ;
[0042] According to a certain wind speed interval p, the full wind speed interval consisting of the maximum sea surface wind speed and the minimum sea surface wind speed in the full wind speed interval data set D is divided into t wind speed intervals, and the wind speed prediction error { |s=1,2,…t}; where Represents the wind speed prediction error in the sth wind speed interval.
[0043] S5. Determine the mean error of the sth wind speed interval The total wind speed inversion error , the full wind speed interval is re-divided into q segmented wind speed intervals, and they are respectively denoted as { |v=1,2,…q}; where represents the vth segment wind speed interval, ; In this embodiment, it is divided into 4 segment intervals, namely [0, 4], (4, 8], (8,12], (12, Inf )
[0044] S6, according to q segmented wind speed intervals { |v=1,2,…q}, for D train Classify and obtain the segmented training set corresponding to q segmented wind speed intervals { |v=1,2,…q}, where Indicates the vth segment wind speed interval The corresponding segmented training set;
[0045] use The deep learning model is trained to obtain the vth segment wind speed interval inversion model H v , thus obtaining q segmented wind speed interval inversion models {H v |v=1,2,…q};
[0046] S7, D test Middle Satellite-borne GNSS-R characteristic value sequence X l test The full wind speed interval inversion model F is input for processing to obtain the predicted Satellite-borne GNSS-R wind speed series ;
[0047] S8. According to The mean error corresponding to the wind speed interval ,calculate Satellite-borne GNSS-R characteristic value sequence X l test The wind speed estimation interval [ , + ), denoted as R l 0 .
[0048] S9, define the current number of iterations as c, and initialize c=1;
[0049] S10, initialize v=1;
[0050] S11. Calculate the wind speed estimation interval R in the c-1th iteration l c-1 and the vth segment wind speed interval The overlap degree in the cth iteration , and judge >0 is true, if true, then the vth segment wind speed interval inversion model H v As the candidate inversion model for the cth iteration, As H v Weight ; Otherwise, after assigning v+1 to v, return to S11 until v>q, thereby obtaining several candidate inversion models and their weights after screening in the cth iteration;
[0051] S12, D test X in l testThe predicted satellite-borne GNSS-R wind speed sequences output by each candidate inversion model in the cth iteration are weighted and averaged with the weights corresponding to the candidate inversion model in the cth iteration to obtain X l test The inverted value of sea surface wind speed in the cth iteration For example, the total wind speed inversion error of model F in the full wind speed range is =1.6m / s, the predicted value of the test set obtained by the basic model is 9m / s, so [7.4m / s, 10.6m / s] is considered to be the wind speed estimation interval of the eigenvalue sequence corresponding to the test set. The wind speed intervals corresponding to the segmented wind speed interval inversion models H1, H2, H3 and H4 are [0, 4], (4, 8], (8, 12], (12, Inf ), then the wind speed estimation interval of the eigenvalue sequence is [7.4m / s, 10.6m / s] and and The intersection weights are (8-7.4) / 3.2=0.187 and (10.6-8) / 3.2=0.813 respectively. Finally, H2(X l test )×0.187+ H3(X l test )×0.813 as the final predicted value.
[0052] S13. According to The mean error corresponding to the wind speed interval , calculate the wind speed range in the cth iteration [ , + ), denoted as R l c ;
[0053] S14. Judgment The wind speed inversion with round c-1 Is the absolute value of the difference less than a certain limit? In this embodiment, the limit is set to 0.01 m / s. If it is less than a certain limit, it means that the sea surface wind speed inversion value under the cth iteration is the optimal value and outputs it, ending the process; otherwise, assign c+1 to c and return to S10 for sequential execution. When c=1, initialize .
[0054] Figure 3The comparison results of the CYGNSS sea surface wind speed obtained by the method of the present invention and the ERA5 sea surface wind speed dataset based on the spaceborne GNSS-R data provided by the CYGNSS satellite constellation are shown in the figure. It can be seen from the figure that the root mean square error of the wind speed inversion is 1.387m / s, and the deviation is -0.0193, which has a high wind speed inversion accuracy.
[0055] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.
[0056] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.
Claims
1. A method for inverting sea surface wind speed using satellite-based GNSS-R based on a multi-layer iterative weighted model, characterized in that: The following steps are involved: S1. After obtaining and preprocessing the satellite-borne GNSS-R data, n satellite-borne GNSS-R feature value sequences are extracted from it, where the jth satellite-borne GNSS-R feature value sequence X j ={ },in, represents the jth satellite-borne GNSS-R eigenvalue sequence X j The i-th eigenvalue in , M represents the total number of eigenvalues of each satellite-borne GNSS-R eigenvalue sequence, j=1,2,…,n; According to the acquisition time Time of the satellite-borne GNSS-R data, the sea surface wind speed data set is synchronously obtained, and according to the position of the mirror reflection point (SPLat, SPLon) in the satellite-borne GNSS-R data, the j-th satellite-borne GNSS-R eigenvalue sequence X j After temporal and spatial matching with the sea surface wind speed dataset, the sea surface wind speed corresponding to the jth satellite-borne GNSS-R eigenvalue sequence is obtained. , so that X j and The jth data sample is formed, and the n data samples form the full wind speed interval data set D; The full wind speed interval data set D is randomly divided into the full wind speed interval training set D according to a certain ratio. train , full wind speed interval verification set D val And the full wind speed range test set D test ; S2, using the full wind speed interval training set D train The deep learning model is trained to obtain the inversion model F for the entire wind speed range; S3, the full wind speed interval verification set D val All satellite-borne GNSS-R feature value sequences {X k val |k=1,2,…,m} is input into the full wind speed interval inversion model F for processing to obtain the predicted satellite-borne GNSS-R wind speed sequence { |k=1,2,…,m}, where X k val Indicates D val The kth satellite-borne GNSS-R eigenvalue sequence in, represents the kth predicted satellite-borne GNSS-R wind speed, and m represents D val The total number of data samples in ; S4, based on the full wind speed range verification set D val The middle sea surface wind speed series |k=1,2,…,m}, statistically analyze the inversion model F for D in the full wind speed range val Total wind speed retrieval error ; According to a certain wind speed interval p, the full wind speed interval consisting of the maximum sea surface wind speed and the minimum sea surface wind speed in the full wind speed interval data set D is divided into t wind speed intervals, and the mean error { |s=1,2,…t}; where represents the mean error of the sth wind speed interval; S5, according to the mean error of the sth wind speed interval The total wind speed inversion error , the full wind speed interval is re-divided into q segmented wind speed intervals, and they are respectively denoted as { |v=1,2,…q}; where represents the vth segment wind speed interval, ; S6, according to q segmented wind speed intervals { |v=1,2,…q}, for D train Classify and obtain the segmented training set corresponding to the q segmented wind speed intervals |v=1,2,…q}, where Indicates the vth segment wind speed interval The corresponding segmented training set; use The deep learning model is trained to obtain the vth segment wind speed interval inversion model H v , thus obtaining the q segmented wind speed interval inversion model H v |v=1,2,…q}; S7, D test Middle Satellite-borne GNSS-R characteristic value sequence X l test The full wind speed interval inversion model F is input for processing to obtain the predicted Satellite-borne GNSS-R wind speed ; S8. According to The mean error corresponding to the wind speed interval ,calculate Satellite-borne GNSS-R characteristic value sequence X l test The wind speed estimation interval [ , + ), denoted as R l 0 ; S9, define the current number of iterations as c, and initialize c=1; S10, initialize v=1; S11. Calculate the wind speed estimation interval R in the c-1th iteration l c-1 and the vth segment wind speed interval The overlap degree in the cth iteration , and judge >0 is true, if true, then the vth segment wind speed interval inversion model H v As the candidate inversion model for the cth iteration, As H v Weight ; Otherwise, after assigning v+1 to v, return to S11 until v>q, thereby obtaining several candidate inversion models and their weights after screening in the cth iteration; S12, D test X in l test The predicted satellite-borne GNSS-R wind speed sequences output by each candidate inversion model in the cth iteration are weighted and averaged with the weights corresponding to the candidate inversion model in the cth iteration to obtain X l test The inverted value of sea surface wind speed in the cth iteration ; S13. According to The mean error corresponding to the wind speed interval , calculate the wind speed range in the cth iteration [ , + ), denoted as R l c ; S14. Judgment The wind speed inversion of round c-1 Is the absolute value of the difference less than a certain limit? If it is less than a certain limit, it means that the sea surface wind speed inversion value under the cth iteration is the optimal value and outputs it, ending the process; otherwise, assign c+1 to c and return to S10 for sequential execution. When c=1, initialize .
2. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the satellite-borne GNSS-R sea surface wind speed inversion method according to claim 1, and the processor is configured to execute the program stored in the memory.
3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the satellite-borne GNSS-R sea surface wind speed inversion method of claim 1 are executed.
Citation Information
Cited By
Satellite-borne GNSS-R multi-system multi-polarization sea surface wind speed estimation method and system
CN121476637A
Spaceborne GNSS-R Multi-System Multi-Polarization Sea Surface Wind Speed Estimation Method and System
CN121476637B
Self-adaptive wet troposphere path delay inversion method and system fusing wind speed overlapping partition modeling, multi-scale collaborative sample equalization and model soft fusion
CN122508958A
