A novel error coupling compensation model for multi-factor learning networks

Abstract

The present invention provides a novel multi-factor learning network error coupling compensation model, which relates to the field of data processing technology, including: using the signal-to-noise ratio inversion method in the shore-based GNSS-R algorithm to calculate the sea surface height change sequence without atmospheric delay; extracting the tide station sequence data in the area where the GNSS station is located, and unifying the sea surface height time series obtained by GNSS inversion and the tide station sequence to the same time scale; subtracting the sea surface height sequence results of the tide station from the sea surface height sequence results obtained by shore-based GNSS-R inversion, thereby obtaining the sea surface height error caused by atmospheric refraction; using atmospheric factors and spatiotemporal factors as input data, using a deep learning algorithm to learn them, thereby constructing a shore-based GNSS-R atmospheric delay deep learning correction network. The present invention not only avoids the complexity of traditional physical modeling, but also can better adapt to various atmospheric conditions.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a novel multi-factor learning network error coupling compensation model. Background Art

[0002] When GNSS-IR measures sea level, the signal will be affected by atmospheric refraction when passing through the atmosphere, resulting in a deviation between the measurement result and the actual sea level. This deviation will change with weather, season, geographic location and other factors, so it is difficult to accurately correct it using traditional physical models.

[0003] In order to solve this problem, researchers have conducted a lot of research and proposed various atmospheric delay correction methods. However, these methods are usually based on physical models and require detailed measurement and modeling of atmospheric conditions, which is often difficult to achieve in practical applications. In addition, these methods can usually only correct specific atmospheric conditions and are difficult to adapt to the complex and changeable atmospheric environment. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a novel multi-factor learning network error coupling compensation model, which not only avoids the complexity of traditional physical modeling, but also can better adapt to various atmospheric conditions.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0006] A novel multi-factor learning network error coupling compensation model, including:

[0007] The sea surface height change sequence without atmospheric delay is calculated using the signal-to-noise ratio inversion method in the shore-based GNSS-R algorithm;

[0008] Extract the tide gauge station sequence data within the area covered by the GNSS station, and unify the sea surface height time series obtained by GNSS inversion and the tide gauge station sequence to the same time scale;

[0009] The sea surface height error caused by atmospheric refraction is obtained by subtracting the true sea surface height sequence results of the tide gauge station from the sea surface height sequence results obtained by shore-based GNSS-R inversion.

[0010] Taking atmospheric factors and spatiotemporal factors as input data, deep learning algorithms are used to learn them, so as to construct a shore-based GNSS-R atmospheric delay deep learning correction network.

[0011] Furthermore, the sea surface height change sequence without adding atmospheric delay is calculated by using the signal-to-noise ratio inversion method in the shore-based GNSS-R algorithm, including:

[0012] Collect data from specific shore-based GNSS receiving stations, including GNSS direct signals and signals reflected from the sea surface; pre-process the collected data;

[0013] Calculate the signal-to-noise ratio using the direct signal and reflected signal received by the GNSS receiver;

[0014] Based on the relationship between SNR and sea level, the incoherent terms in SNR are fitted using a second-order polynomial and deducted to obtain the oscillation term of SNR. The oscillation frequency of SNR is analyzed to derive the antenna height, which is then converted into sea level height changes.

[0015] The sea surface height data calculated in continuous time are integrated into a time series, and the calculated sea surface height change series is not corrected for atmospheric delay.

[0016] Furthermore, the tide gauge station sequence data in the area covered by the GNSS station are extracted, and the sea surface height time series obtained by GNSS inversion and the tide gauge station sequence are unified to the same time scale, including:

[0017] Determine the GNSS station area, and find and identify all tide stations within the calibrated GNSS station area;

[0018] Obtain the sea surface height data of tide gauge stations from relevant institutions or public data sources, and perform quality checks on the tide gauge station data obtained; analyze the time scale of the sea surface height time series and tide gauge station series obtained by GNSS inversion, including the sampling frequency, start time and end time of the data. If the sampling frequencies of the two sequences are different, perform data interpolation or resampling; assign a unified timestamp to each data point in the two sequences.

[0019] Furthermore, the true sea surface height sequence results of the tide gauge station are subtracted from the sea surface height sequence results obtained by shore-based GNSS-R inversion to obtain the sea surface height error caused by atmospheric refraction, including:

[0020] Perform difference calculation for each pair of corresponding data points in the GNSS-R inversion sequence and the tide gauge station sequence, that is, subtract the sea surface height data obtained by GNSS-R inversion from the real sea surface height data of the tide gauge station, and calculate for each time point to obtain a series of error values ​​to form an error sequence;

[0021] Conduct statistical analysis on the error sequence and calculate its mean, standard deviation, maximum and minimum statistical characteristics;

[0022] The distribution of errors is displayed in graphs. The error sequence is analyzed to identify the error components related to atmospheric refraction. The calculated error sequence and related statistical characteristics and visual graphs are recorded and saved.

[0023] Furthermore, atmospheric factors and spatiotemporal factors are used as input data and deep learning algorithms are used to learn them, so as to construct a shore-based GNSS-R atmospheric delay deep learning correction network, including:

[0024] Acquire atmospheric factor data related to shore-based GNSS-R measurements, including atmospheric temperature, humidity, air pressure, and state parameters of the ionosphere and troposphere; collect geographic location information and timestamp data of the measurement points;

[0025] The collected atmospheric factor and spatiotemporal factor data are cleaned, and the preprocessed atmospheric factor and spatiotemporal factor data are used as the input data of the convolutional neural network. The output of the convolutional neural network is the target variable, that is, the corrected value of the atmospheric delay or the corrected sea surface height value;

[0026] The preprocessed data is divided into training set, validation set and test set. The training set is used to train the convolutional neural network, the validation set is used to adjust the parameters of the convolutional neural network and monitor the training process, and the test set is used to evaluate the performance of the convolutional neural network. The convolutional neural network is trained using the training set, and the parameters of the convolutional neural network are continuously optimized through the back propagation algorithm. The trained convolutional neural network is applied to actual shore-based GNSS-R measurements to correct the atmospheric delay.

[0027] Furthermore, atmospheric factors include temperature, pressure and electron content; and spatiotemporal factors include longitude, latitude and time.

[0028] The above solution of the present invention includes at least the following beneficial effects:

[0029] By correcting the atmospheric delay through a deep learning algorithm, the sea level measurement error caused by atmospheric refraction can be significantly reduced, thereby improving the measurement accuracy of GNSS-IR. Traditional physical model correction methods are often limited to specific atmospheric conditions, while the deep learning algorithm used in the present invention can automatically extract features by learning a large amount of data, and has stronger adaptability to different atmospheric conditions. Compared with traditional methods that require detailed measurement and modeling of atmospheric conditions, the present invention automatically learns and corrects errors in a data-driven manner, simplifies the operation process, and reduces the difficulty of implementation. The method of the present invention can be combined with a real-time GNSS-IR measurement system to achieve real-time and accurate measurement of sea level, meeting the real-time requirements of applications such as marine environmental monitoring and disaster warning. The deep learning model has good scalability, and with the accumulation of data and continuous training of the model, the correction accuracy and range can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of a novel multi-factor learning network error coupling compensation model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0032] like Figure 1 As shown, the embodiment of the present invention proposes a novel multi-factor learning network error coupling compensation model, including:

[0033] The sea surface height change sequence without atmospheric delay is calculated using the signal-to-noise ratio inversion method in the shore-based GNSS-R algorithm;

[0034] Extract the tide gauge station sequence data within the area covered by the GNSS station, and unify the sea surface height time series obtained by GNSS inversion and the tide gauge station sequence to the same time scale;

[0035] The sea surface height error caused by atmospheric refraction is obtained by subtracting the true sea surface height sequence results of the tide gauge station from the sea surface height sequence results obtained by shore-based GNSS-R inversion.

[0036] Taking atmospheric factors and spatiotemporal factors as input data, deep learning algorithms are used to learn them, so as to construct a shore-based GNSS-R atmospheric delay deep learning correction network.

[0037] In an embodiment of the present invention, by using a deep learning algorithm to correct the atmospheric delay, the present invention can significantly reduce the sea surface height measurement error caused by atmospheric refraction, thereby effectively improving the measurement accuracy of the GNSS-IR technology. Traditional correction methods are usually sensitive to specific atmospheric conditions, while the deep learning model of the present invention can adaptively handle various complex and changeable atmospheric environments by learning a variety of data sets, thereby greatly enhancing the versatility and practicality of the correction method. The present invention simplifies the data processing process by integrating GNSS-R inversion data and tide station data and unifying them to the same time scale, thereby improving the efficiency of data utilization and the efficiency of correction work. By comparing and subtracting the real sea surface height data of the tide station with the GNSS-R inversion data, the present invention can accurately quantify the sea surface height error caused by atmospheric refraction, providing an accurate and reliable basis for subsequent error correction. Since the deep learning model has a strong learning and generalization ability, with the continuous accumulation of data and the continuous optimization of the model, the correction method of the present invention is expected to be applied to a wider range of scenarios, promoting the in-depth application of GNSS-IR technology in the fields of ocean monitoring and climate research.

[0038] Furthermore, the sea surface height change sequence without adding atmospheric delay is calculated by using the signal-to-noise ratio inversion method in the shore-based GNSS-R algorithm, including:

[0039] Collect data from specific shore-based GNSS receiving stations, including GNSS direct signals and signals reflected from the sea surface; pre-process the collected data;

[0040] Calculate the signal-to-noise ratio using the direct signal and reflected signal received by the GNSS receiver;

[0041] Based on the relationship between SNR and sea level, the incoherent terms in SNR are fitted using a second-order polynomial and deducted to obtain the oscillation term of SNR. The oscillation frequency of SNR is analyzed to derive the antenna height, which is then converted into sea level height changes.

[0042] The sea surface height data calculated in continuous time are integrated into a time series, and the calculated sea surface height change series is not corrected for atmospheric delay.

[0043] In an embodiment of the present invention, by collecting direct signals and reflected signals of a specific shore-based GNSS receiving station and preprocessing these data, the present invention can effectively extract high-quality signal data and provide accurate basic data for subsequent sea level inversion. The present invention uses the direct signals and reflected signals received by the GNSS receiver to calculate the signal-to-noise ratio (SNR), and based on the relationship between the SNR and the sea level, accurately derives the antenna height through second-order polynomial fitting and oscillation term analysis, and then converts it into sea level change. This method can more accurately reflect the actual change of the sea level and improve the accuracy of the inversion. By integrating the sea level data calculated in continuous time into a time series, the present invention can provide continuous and real-time sea level change monitoring data. This is of great significance to the fields of marine environment monitoring, marine disaster warning, etc., and helps to timely discover and respond to changes in the marine environment. The method of the present invention is based on the shore-based GNSS-R algorithm, and through a series of automated data processing steps, the conversion from raw signal data to sea level change sequence is realized. This processing method simplifies the complex process in traditional sea level measurement and improves the efficiency and automation of data processing.

[0044] The following is a detailed explanation of each step and a description of the specific implementation process:

[0045] To collect data from a specific shore-based GNSS receiving station, first select one or more GNSS receiving stations located on the shore. These receiving stations should be able to receive GNSS direct signals and signals reflected from the sea surface. Use the equipment of these receiving stations to continuously receive direct signals from GNSS satellites and signals reflected from the sea surface. The received data should contain information such as the amplitude, phase, and timestamp of the signal. The received raw data should be stored in a local or remote server in an appropriate format for subsequent processing and analysis.

[0046] Preprocess the collected data to remove outliers, noise and interference signals in the collected data to ensure the accuracy and reliability of the data, and perform time synchronization on the direct signal and the reflected signal to ensure that they are consistent in time scale; convert the cleaned and synchronized data into a format suitable for subsequent algorithm processing, such as extracting signal data for a specific time period and performing necessary normalization or standardization.

[0047] The amplitude or power information of the direct signal and the reflected signal are respectively extracted from the preprocessed data. The signal-to-noise ratio (SNR) is calculated based on the extracted signal information. The SNR is defined as the logarithmic form of the ratio of the direct signal power to the reflected signal power, and is used to characterize the strength and quality of the signal. Optionally, the calculated SNR is smoothed to reduce the impact of random errors and noise and improve the stability of the SNR.

[0048] Based on the SNR inversion of sea surface height changes, the incoherent terms in the SNR are fitted using a second-order polynomial or other appropriate mathematical model, and these incoherent terms are deducted from the original SNR to obtain the oscillation terms of the SNR. The oscillation terms of the SNR after deducting the incoherent terms are subjected to spectrum analysis to identify their main oscillation frequencies. This oscillation frequency is closely related to the sea surface height changes; based on the identified oscillation frequency and other related parameters (such as satellite elevation angle, receiver position, etc.), the height of the antenna relative to the sea surface is derived using an appropriate mathematical model; the derived antenna height is converted into sea surface height changes. This usually involves some geometric transformation and correction processes to ensure that the obtained sea surface height change data is accurate.

[0049] Integrate the time series data, integrate the sea surface height data calculated in continuous time in chronological order to form a time series data set, use appropriate data interpolation methods to fill in the missing data or outliers that may exist in the time series, and smooth the entire time series to improve the continuity and stability of the data, output the integrated time series data in an appropriate format and store it on a local or remote server for subsequent analysis and application. It should be noted that the sea surface height change sequence obtained at this time is the original data that has not been corrected for atmospheric delay.

[0050] Through the detailed explanation of the above steps and the description of the specific implementation process, we can more clearly understand how to use the signal-to-noise ratio inversion method in the shore-based GNSS-R algorithm to calculate the sea surface height change sequence without adding atmospheric delay.

[0051] Furthermore, the tide gauge station sequence data in the area covered by the GNSS station are extracted, and the sea surface height time series obtained by GNSS inversion and the tide gauge station sequence are unified to the same time scale, including:

[0052] Determine the GNSS station area, and find and identify all tide stations within the calibrated GNSS station area;

[0053] Obtain the sea surface height data of tide gauge stations from relevant institutions or public data sources, and perform quality checks on the tide gauge station data obtained; analyze the time scale of the sea surface height time series and tide gauge station series obtained by GNSS inversion, including the sampling frequency, start time and end time of the data. If the sampling frequencies of the two sequences are different, perform data interpolation or resampling; assign a unified timestamp to each data point in the two sequences.

[0054] In an embodiment of the present invention, by accurately matching the time scales of GNSS inversion data and tide gauge data, the present invention ensures that the comparative analysis between the two is performed in the same time frame. This time alignment can significantly improve the accuracy of data comparison, thereby more accurately evaluating the reliability and accuracy of GNSS inversion results. The present invention realizes the unified processing of data with different sampling frequencies through data interpolation or resampling technology. This data fusion capability not only broadens the diversity of data sources, but also enables the information between different data sources to be effectively integrated; by assigning a unified timestamp to each data point in the two sequences, the present invention simplifies the data processing process and makes data query, retrieval and analysis more efficient. The uniformity of this timestamp also helps to discover outliers and missing values ​​in the data, so as to clean and correct the data in a timely manner and improve the overall quality of the data. The implementation of the present invention enables GNSS inversion data and tide gauge data to be collaboratively analyzed on the same time scale.

[0055] The following is a detailed explanation of each step and a description of the specific implementation process:

[0056] Based on the location information of the GNSS station, determine a specific geographical area, which will serve as the basis for searching for tide gauges; use geographic information system (GIS) tools or online map services to search within the defined area to find existing tide gauges. These tools usually provide the function of searching for specific types of facilities by geographic location; filter out relevant information about tide gauges from the search results, such as name, location coordinates, operating agency, etc., to ensure that all tide gauges located in the GNSS station area are accurately identified.

[0057] Determine which relevant institutions or public data sources to obtain the sea surface height data of tide gauge stations from. These data sources may be the State Oceanic Administration, scientific research institutions or international data sharing platforms; access the selected data sources and download the sea surface height data of tide gauge stations according to the required time range and format. These data are usually provided in the form of time series. Perform quality checks on the downloaded tide gauge station data, including checking the completeness, continuity and accuracy of the data. For example, check whether there are missing values, outliers or unexpected data fluctuations. If data quality problems are found, such as missing values ​​or outliers, take appropriate data cleaning measures, such as interpolation filling, outlier removal or data smoothing, to ensure the reliability of the data.

[0058] The sea surface height time series obtained by GNSS inversion is compared with the tide gauge series in terms of time scale, including comparison of the sampling frequency (such as hourly, daily, etc.), start time and end time of the two. If the sampling frequencies of the two series are different, a common sampling frequency is selected as needed, and the data is interpolated or resampled. The interpolation method can be linear interpolation, polynomial interpolation or other suitable interpolation algorithms to ensure the consistency of the two series in time scale.

[0059] Generate a unified timestamp for each data point in both series. This timestamp should be accurate enough to reflect the actual observation time of the data point, usually accurate to the second or millisecond level. Make sure that the timestamps of the GNSS inversion data and the tide gauge data are completely aligned, that is, each data point corresponds to the same time point. In this way, in subsequent analysis and comparison, it can be ensured that the two series are conducted in the same time frame.

[0060] Furthermore, the true sea surface height sequence results of the tide gauge station are subtracted from the sea surface height sequence results obtained by shore-based GNSS-R inversion to obtain the sea surface height error caused by atmospheric refraction, including:

[0061] Perform difference calculation for each pair of corresponding data points in the GNSS-R inversion sequence and the tide gauge station sequence, that is, subtract the sea surface height data obtained by GNSS-R inversion from the real sea surface height data of the tide gauge station, and calculate for each time point to obtain a series of error values ​​to form an error sequence;

[0062] Conduct statistical analysis on the error sequence and calculate its mean, standard deviation, maximum and minimum statistical characteristics;

[0063] The distribution of errors is displayed in graphs. The error sequence is analyzed to identify the error components related to atmospheric refraction. The calculated error sequence and related statistical characteristics and visual graphs are recorded and saved.

[0064] In the embodiment of the present invention, by comparing the true sea level height sequence of the tide gauge station with the sea level height sequence obtained by shore-based GNSS-R inversion and calculating the difference between the two, the sea level height error caused by atmospheric refraction can be obtained. The beneficial effects of this method include:

[0065] It can directly quantify the size of the error affected by atmospheric refraction when the GNSS-R technology is used to invert the sea surface height; by statistically analyzing the error sequence, we can gain an in-depth understanding of the distribution characteristics, fluctuation range and change patterns of the error, which helps to identify and separate the error components related to atmospheric refraction. Based on the error analysis results, the GNSS-R inversion algorithm can be improved in a targeted manner to reduce or compensate for the error caused by atmospheric refraction and improve the accuracy of sea surface height inversion. By identifying and correcting the atmospheric refraction error, the overall quality of the GNSS-R inversion data can be improved. The recorded and saved error sequences, statistical characteristics and visualization charts can provide valuable data and reference information for subsequent research, and promote scientific research and technological development in related fields.

[0066] The specific implementation process is as follows:

[0067] Ensure that the GNSS-R inverted sea surface height sequence and the tide gauge station real sea surface height sequence that have been quality checked and time-scaled are unified have been obtained; perform difference calculations on each pair of corresponding data points in the two sequences, that is, subtract the sea surface height data obtained by GNSS-R inversion from the real sea surface height data of the tide gauge station; perform the above calculations at each time point to obtain a series of error values, which constitute an error sequence; perform statistical analysis on the error sequence, and calculate its statistical characteristics such as mean, standard deviation, maximum and minimum values ​​to describe the overall level and fluctuation range of the error; use charts (such as histograms, line charts or scatter plots, etc.) to show the distribution of errors to help intuitively understand the changing laws and characteristics of the errors; by analyzing the error sequence in the chart, try to identify specific error patterns or periodic changes related to atmospheric refraction; organize the calculated error sequence and related statistical characteristic data into documents or database records; through the above implementation process, the sea surface height errors caused by atmospheric refraction can be systematically analyzed and recorded.

[0068] Furthermore, atmospheric factors and spatiotemporal factors are used as input data and deep learning algorithms are used to learn them, so as to construct a shore-based GNSS-R atmospheric delay deep learning correction network, including:

[0069] Acquire atmospheric factor data related to shore-based GNSS-R measurements, including atmospheric temperature, humidity, air pressure, and state parameters of the ionosphere and troposphere; collect geographic location information and timestamp data of the measurement points;

[0070] The collected atmospheric factor and spatiotemporal factor data are cleaned, and the preprocessed atmospheric factor and spatiotemporal factor data are used as the input data of the convolutional neural network. The output of the convolutional neural network is the target variable, that is, the corrected value of the atmospheric delay or the corrected sea surface height value;

[0071] The preprocessed data is divided into training set, validation set and test set. The training set is used to train the convolutional neural network, the validation set is used to adjust the parameters of the convolutional neural network and monitor the training process, and the test set is used to evaluate the performance of the convolutional neural network. The convolutional neural network is trained using the training set, and the parameters of the convolutional neural network are continuously optimized through the back propagation algorithm. The trained convolutional neural network is applied to actual shore-based GNSS-R measurements to correct the atmospheric delay. Atmospheric factors include temperature, pressure and electron content. Spatiotemporal factors include longitude, latitude and time.

[0072] In the embodiment of the present invention, by constructing a shore-based GNSS-R atmospheric delay deep learning correction network and introducing atmospheric factors and spatiotemporal factors as input data, the following beneficial effects can be achieved:

[0073] The deep learning algorithm can automatically learn the complex relationship between atmospheric delay and input factors, so as to more accurately predict and correct atmospheric delay and improve the accuracy of GNSS-R sea surface height measurement. Since a variety of atmospheric factors and spatiotemporal factors are taken into account, the constructed deep learning model has stronger generalization ability and can adapt to GNSS-R measurements under different environmental conditions. The deep learning model can realize automatic atmospheric delay correction, reduce manual intervention and data processing time, and improve work efficiency. The improved quality of the corrected GNSS-R data will help expand the application scope of GNSS-R technology in fields such as ocean monitoring and climate change research.

[0074] The specific implementation process is as follows:

[0075] Collect atmospheric factor data related to shore-based GNSS-R measurements, including temperature, humidity, air pressure, and state parameters of the ionosphere and troposphere. These data can be obtained from meteorological observation stations, satellite remote sensing or other relevant data sources. At the same time, collect the geographic location information (longitude, latitude) and timestamp data of the measurement point to ensure comprehensive coverage of spatiotemporal factors.

[0076] Clean the collected atmospheric and spatiotemporal factor data, remove outliers, missing values, and duplicate values, and ensure data quality. Standardize or normalize the data to eliminate the impact of different dimensions and numerical ranges on model training. Select an appropriate deep learning framework (such as TensorFlow, PyTorch, etc.) to build a convolutional neural network (CNN) model. Design the network structure of CNN, including convolutional layers, pooling layers, fully connected layers, etc., to adapt to the characteristics of input data and the needs of output targets. Use the preprocessed atmospheric and spatiotemporal factor data as the input of CNN, and set the output target as the corrected value of atmospheric delay or the corrected sea surface height value. Divide the preprocessed data into training set, validation set, and test set. Usually, random partitioning or chronological partitioning can be used. Use the training set to train the CNN model, calculate the output value through forward propagation, and compare it with the true value to obtain the error. Use the backpropagation algorithm and gradient descent optimizer to continuously adjust the parameters of the CNN model to minimize the training error. Evaluate the performance of the model on the validation set, and adjust the model parameters (such as learning rate, batch size, etc.) or network structure based on the evaluation results to optimize the model performance. Use the test set to test the trained CNN model to evaluate its generalization ability and correction effect on independent data. Make necessary adjustments and optimizations to the model based on the test results to ensure its reliability and stability in practical applications. Apply the trained CNN model to actual shore-based GNSS-R measurements, automatically correct the atmospheric delay, and improve the accuracy of sea surface height measurements.

[0077] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A novel multi-factor learning network error coupling compensation model, characterized in that: include: The sea surface height change sequence without atmospheric delay is calculated using the signal-to-noise ratio inversion method in the shore-based GNSS-R algorithm; Extract the tide gauge station sequence data within the area covered by the GNSS station, and unify the sea surface height time series obtained by GNSS inversion and the tide gauge station sequence to the same time scale; The sea surface height error caused by atmospheric refraction is obtained by subtracting the true sea surface height sequence results of the tide gauge station from the sea surface height sequence results obtained by shore-based GNSS-R inversion. Taking atmospheric factors and spatiotemporal factors as input data and the corrected value of atmospheric delay or the corrected sea surface height value as output, the deep learning algorithm is used to learn them, so as to construct a shore-based GNSS-R atmospheric delay deep learning correction network; The sea surface height change sequence without atmospheric delay is calculated by using the signal-to-noise ratio inversion method in the shore-based GNSS-R algorithm, including: Collect data from a specific shore-based GNSS receiving station, including GNSS direct signals and signals reflected from the sea surface; pre-process the collected data, wherein the specific shore-based GNSS receiving station has the ability to receive GNSS direct signals and signals reflected from the sea surface; Calculate the signal-to-noise ratio using the direct signal and reflected signal received by the GNSS receiver; Based on the relationship between SNR and sea level, the incoherent terms in SNR are fitted using a second-order polynomial, and the incoherent terms are deducted from the original SNR to obtain the oscillation terms of SNR. The oscillation terms of SNR after deducting the incoherent terms are subjected to spectrum analysis to identify the oscillation frequency. The oscillation frequency of SNR is analyzed to derive the antenna height, which is then converted into changes in sea level height. The sea surface height data calculated in continuous time are integrated into a time series, and the calculated sea surface height change series is not corrected for atmospheric delay.

2. The novel multi-factor learning network error coupling compensation model according to claim 1 is characterized in that: Extract the tide gauge station sequence data within the area covered by the GNSS station, and unify the sea surface height time series obtained by GNSS inversion and the tide gauge station sequence to the same time scale, including: Determine the GNSS station area, and find and identify all tide stations within the calibrated GNSS station area; The sea surface height data of tide gauge stations are obtained from public data sources, and the quality of the tide gauge station data obtained is checked. The time scale of the sea surface height time series and tide gauge station series obtained by GNSS inversion is analyzed, including the sampling frequency, start time and end time of the data. If the sampling frequencies of the two sequences are different, data interpolation or resampling is performed. A unified timestamp is assigned to each data point in the two sequences.

3. The novel multi-factor learning network error coupling compensation model according to claim 2 is characterized in that: The sea surface height sequence results of the tide gauge station are subtracted from the sea surface height sequence results obtained by shore-based GNSS-R inversion to obtain the sea surface height error caused by atmospheric refraction, including: Perform difference calculation for each pair of corresponding data points in the GNSS-R inversion sequence and the tide gauge station sequence, that is, subtract the sea surface height data obtained by GNSS-R inversion from the real sea surface height data of the tide gauge station, and calculate for each time point to obtain a series of error values ​​to form an error sequence; Conduct statistical analysis on the error sequence and calculate its mean, standard deviation, maximum and minimum statistical characteristics; The distribution of errors is displayed in graphs. The error sequence is analyzed to identify the error components related to atmospheric refraction. The calculated error sequence and related statistical characteristics and visual graphs are recorded and saved.

4. The novel multi-factor learning network error coupling compensation model according to claim 3 is characterized in that: Taking atmospheric factors and spatiotemporal factors as input data, deep learning algorithms are used to learn them, so as to build a shore-based GNSS-R atmospheric delay deep learning correction network, including: Acquire atmospheric factor data related to shore-based GNSS-R measurements, including atmospheric temperature, humidity, air pressure, and state parameters of the ionosphere and troposphere; collect geographic location information and timestamp data of the measurement points; The collected atmospheric factor and spatiotemporal factor data are cleaned, and the preprocessed atmospheric factor and spatiotemporal factor data are used as the input data of the convolutional neural network. The output of the convolutional neural network is the target variable, that is, the corrected value of the atmospheric delay or the corrected sea surface height value; The preprocessed data is divided into training set, validation set and test set. The training set is used to train the convolutional neural network, the validation set is used to adjust the parameters of the convolutional neural network and monitor the training process, and the test set is used to evaluate the performance of the convolutional neural network. The convolutional neural network is trained using the training set, and the parameters of the convolutional neural network are continuously optimized through the back propagation algorithm. The trained convolutional neural network is applied to actual shore-based GNSS-R measurements to correct the atmospheric delay.

5. The novel multi-factor learning network error coupling compensation model according to claim 4 is characterized in that: Atmospheric factors include temperature, pressure and electron content; spatiotemporal factors include longitude, latitude and time.

Images