A ZTD prediction method for large height difference areas based on GFS and GNSS

Abstract

The present invention proposes a ZTD prediction method for large height difference areas based on GFS and GNSS. First, based on historical GFS prediction data, the GFS turbulent mixing delay and GNSS turbulent mixing delay at the GNSS station location are obtained. Secondly, the XGBoost method is used to construct a non-linear mapping relationship between the GFS turbulent mixing delay and the GNSS turbulent mixing delay. Finally, the trained XGBoost method is used to correct the GFS data at the prediction time, achieving high-precision prediction of ZTD in large height difference areas. The method of the present invention does not require the participation of real-time GNSS ZTD. Only time, location, and GFS prediction data are needed to obtain high-precision ZTD prediction values through the constructed XGBoost model, which fully reflects the beneficial effects of the present invention.

Description

Technical Field

[0001] The present invention relates to the field of GNSS precise positioning, and particularly to a method for predicting ZTD in high-relief areas based on GFS and GNSS. Background Art

[0002] Tropospheric delay is one of the important error sources of the Global Navigation Satellite System (GNSS). The tropospheric delay in the zenith direction (Zenith tropospheric delay, ZTD) consists of two parts: the zenith hydrostatic delay (ZHD) and the zenith wet delay (ZWD). Among them, ZHD can be accurately calculated based on the Saastamoinen model using air pressure. For ZWD, due to the spatio-temporal variability of water vapor, it is difficult to achieve high-precision and high-spatial-resolution ZWD estimation. In the GNSS Precise Point Positioning (PPP) technology, empirical models are usually used to calculate ZHD, and ZWD or its unmodeled quantity is estimated as an unknown parameter. However, in areas with large height differences, the strong correlation between ZTD and height inevitably increases the convergence time of GNSS data processing, restricting the high-precision application of GNSS under near-real-time or real-time conditions.

[0003] At present, the most convenient way to obtain ZTD is to directly apply empirical models based on measured meteorological parameters, such as the Hopfield model, the Saastamoinen model, and the Black model. These models are usually constructed based on the ideal gas state equation and related assumptions, modeling ZHD and ZWD separately. By inputting measured meteorological parameters, the accuracy of the calculated ZTD is between decimeters and centimeters. However, due to the spatio-temporal variation of water vapor, the accuracy of the ZWD model in the empirical model based on measured meteorological parameters is usually limited. For historical data sets lacking direct meteorological observations or GNSS stations without co-located meteorological sensors, it is difficult to obtain measured meteorological parameters for ZTD estimation. Therefore, the input values of the empirical model based on measured meteorological parameters are generally the outputs of empirical meteorological parameter models such as GPT2w and GPT3, which will lead to a reduction in the accuracy of the estimated ZTD. To reduce the dependence on measured meteorological parameters, scholars have developed ZTD empirical models in the form of grids and spherical harmonic functions based on reanalysis data and radiosonde data, such as the IGGtrop series models, the GZTD series models, and the RGZTD model. The advantage of these models is that the ZTD at a specified location can be obtained by only inputting longitude, latitude, altitude, and time parameters. However, the ZTD empirical model is an approximation of the original data, which will inevitably lead to a reduction in accuracy. In addition, relying only on historical data without incorporating new observables limits the ability of the ZTD empirical model to capture the detailed characteristics of tropospheric delay changes.

[0004] With the development of GNSS, the estimation of regional ZTD can also be achieved based on the real-time PPP ZTD estimation values of surrounding GNSS stations through methods such as inverse distance weighting and spline interpolation. This method has relatively high requirements for the distribution of surrounding GNSS stations. In areas with large height differences, the calculated ZTD may be affected by the height differences between stations, resulting in large deviations. The forecast data based on Numerical Weather Models (NWM) has good spatial coverage and service continuity, and can provide good data supplementation for areas with sparse GNSS stations. The Global Forecast System (GFS) is a global numerical weather forecasting model developed and operated by the National Center for Environmental Prediction (NCEP) of the United States, providing global atmospheric and wave forecasts at a latitude-longitude grid resolution of 0.25°×0.25°. GFS runs forecasts at 00:00, 06:00, 12:00, and 18:00 UTC every day, generating hourly forecast outputs for 120 hours. Since the time delay for uploading forecast data is about 3 - 5 hours, using GFS forecast data for 6 - 11 hours can fully simulate the actual situation. This meets the requirements for real-time ZTD estimation in areas with large height differences using GFS. Previous studies have shown that combining GNSS and GFS data for regional ZTD estimation is an effective strategy. However, these methods usually require obtaining real-time GNSS ZTD estimation values at corresponding moments. For areas with large height differences, the distribution of GNSS stations is sparse and uneven, and there are also risks of unauthorized or data stream failures in regional real-time GNSS ZTD services. Therefore, it is necessary to construct a GFS correction model without the participation of real-time GNSS ZTD to achieve accurate forecasting of ZTD in areas with large height differences.

[0005] Given the complexity of the spatio-temporal variation of ZTD in areas with large height differences and the characteristics of sparse distribution of GNSS stations, how to achieve accurate forecasting of ZTD with high precision and high spatial resolution through GNSS historical data and GFS forecast products is a key problem that urgently needs to be solved, which has important theoretical research and practical application significance for real-time high-precision applications of GNSS. Summary of the Invention

[0006] In order to improve the effectiveness of real-time high-precision applications of GNSS in areas with large height differences, the present invention proposes a method for forecasting ZTD in areas with large height differences based on GFS and GNSS, which solves the problem that real-time GNSS ZTD participation is required for regional ZTD estimation in previous methods, and can also be applied to areas with sparse GNSS stations. The technical solutions provided by the present invention are as follows:

[0007] A ZTD prediction method for large height difference areas based on GFS and GNSS, comprising the following steps:

[0008] S1. Calculate the historical data of GNSS ZTD and GFS stratified ZTD at the positions of all GNSS stations in the study area;

[0009] S2. Obtain the GNSS turbulent mixing delay and GFS turbulent mixing delay at the positions of GNSS stations based on the GFS stratified ZTD;

[0010] S3. Construct a mapping model from GFS turbulent mixing delay to GNSS turbulent mixing delay based on XGBoost, and predict the ZTD at a specified position and altitude through the GFS vertical stratification delay at the prediction time and the trained XGBoost model.

[0011] Preferably, step S1 is specifically as follows:

[0012] S11. Calculate the GNSS ZTD at the positions of all GNSS stations through precise point positioning PPP;

[0013] S12. Calculate the GFS ZTD values of all pressure layers below 10 km at the positions of all grid points in the historical GFS forecast data in the study area;

[0014] S13. Calculate the GFS ZTD values of all pressure layers below 10 km at the positions of GNSS stations by using the inverse distance weighting method based on the GFS ZTD values of the four GFS grid points closest to the GNSS station positions to obtain the GFS stratified ZTD.

[0015] Preferably, the calculation method of the GFS ZTD value in S12 is as follows:

[0016] Use the Saastamoinen model to calculate the ZTD at the highest pressure layer of GFS top , and the water vapor content at the highest pressure layer of GFS is assumed to be zero:

[0017]

[0018] In the formula, P top and h top are the air pressure and geodetic height of the highest pressure layer respectively, is the latitude of the grid point;

[0019] Adopt the stratified integration method to calculate the ZTD between the top layer and other layers:

[0020]

[0021] In the formula, k1, k2 and k3 are atmospheric refractive index constants, e i,ave , Ti,ave and rh i,ave are the average water vapor pressure, temperature, and relative humidity between the i-th layer and the (i + 1)-th layer; ds is the thickness between the i-th layer and the (i + 1)-th layer. Adding ZTD top and ZTD part yields the ZTD of the specified pressure layer.

[0022] Preferably, step S2 is specifically as follows:

[0023] S21. Use the exponential function model to describe the vertical variation of ZTD:

[0024] ZTD α = TD α + ZTD α0 e βh + ε α

[0025] In the formula, ZTD α , TD α and ZTD α0 are the tropospheric delay, turbulent mixing delay, and vertical stratification delay at a height of 0 m at position α, respectively. ε α is the unmodeled residual at position α, β is the model coefficient, and ZTD α0 e βh is the vertical stratification delay component; Based on the GFS stratified ZTD, using the exponential function model and the least squares method to estimate the turbulent mixing delay TD α and the model coefficient β, so as to obtain the GFS vertical stratification delay and GFS turbulent mixing delay at the GNSS station location;

[0026] S22. Subtract the GFS vertical stratification delay at the corresponding position from the GNSS ZTD to obtain the GNSS turbulent mixing delay.

[0027] Preferably, step S3 is specifically as follows:

[0028] S31. Construct the XGBoost model framework. Use the day of the year, hour, longitude, latitude, and GFS turbulent mixing delay of historical data as inputs, and the GNSS turbulent mixing delay as the output for model training. The GNSS turbulent mixing delay is used to correct the accuracy of the GFS turbulent mixing delay to obtain the trained XGBoost_TD model;

[0029] S32. Obtain GFS forecast data at a specified location and altitude. Calculate the GFS turbulence mixing delay and GFS vertical stratification delay at the specified location and altitude according to the GFS forecast data at the forecast time using the formula in S21. Use the day of the year, hour, longitude, latitude, and GFS turbulence mixing delay at the forecast time as the input of the trained XGBoost_TD model to obtain the corrected GFS turbulence mixing delay. Add the GFS vertical stratification delay at the forecast time and the corrected GFS turbulence mixing delay to obtain the ZTD forecast value.

[0030] Preferably, the input data form of the XGBoost model is a two-dimensional matrix of sample×5, and the output data is a vector of sample×1, where sample is the number of samples of historical data.

[0031] Preferably, the five-fold cross-validation method is used to determine the optimal parameters of the XGBoost model. The dataset is divided into five parts, and every four parts are used as training data, and the remaining one is used as validation data. The finally trained model is the model corresponding to the parameter set that minimizes the root mean square error of all validation data.

[0032] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: By separating the vertical stratification delay and the turbulence mixing delay, the adaptability of the XGBoost model to areas with large height differences and areas with sparse GNSS stations is improved; the trained XGBoost model does not require the participation of real-time GNSS ZTD and has better adaptability, which is of great significance for the high-precision forecast of ZTD in areas with large height differences. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0034] Figure 1 is the overall flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0036] To make the above objects, features, and effects of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0037] Example 1: A method for predicting ZTD in areas with large height differences based on GFS and GNSS, as Figure 1 shown, includes the following steps:

[0038] S1. Calculate the historical data of GNSS ZTD and GFS stratified ZTD at the positions of all GNSS stations in the study area. Specifically:

[0039] S11. Calculate the GNSS ZTD value y at the positions of all GNSS stations through precise point positioning PPP ZTDs .

[0040] S12. Calculate the GFS ZTD values of all pressure layers below 10 km at the positions of all grid points in the historical GFS forecast data in the study area; the calculation method of the GFS ZTD value is as follows:

[0041] Use the Saastamoinen model to calculate the ZTD at the highest pressure layer of GFS top , and the water vapor content at the highest pressure layer of GFS is assumed to be zero:

[0042]

[0043] In the formula, P top and h top are the air pressure and geodetic height of the highest pressure layer respectively, is the latitude of the grid point;

[0044] Adopt the stratified integration method to calculate the ZTD between the top layer and other layers:

[0045]

[0046] In the formula, k1, k2 and k3 are the atmospheric refractive index constants, e i,ave , T i,ave and rh i,ave are the average water vapor pressure, temperature and relative humidity between the i-th layer and the i+1-th layer; ds is the thickness between the i-th layer and the i+1-th layer, add ZTD top and ZTD part to obtain the ZTD of the specified pressure layer.

[0047] S13. Calculate the GFS ZTD values of all pressure layers below 10 km at the positions of GNSS stations by using the inverse distance weighting method based on the GFS ZTD values of the four GFS grid points closest to the positions of GNSS stations to obtain the GFS stratified ZTD.

[0048] S2. Obtain the GNSS turbulent mixing delay and GFS turbulent mixing delay at the positions of GNSS stations based on the GFS stratified ZTD. The specific steps of S2 are as follows:

[0049] S21. Describe the vertical variation of ZTD using an exponential function model:

[0050] ZTD α = TD α + ZTD α0 e βh + ε α

[0051] In the formula, ZTD α , TD α and ZTD α0 are the tropospheric delay, turbulent mixing delay at position α, and vertical stratification delay at a height of 0 m respectively. ε is the unmodeled residual, β is the model coefficient, and ZTD α0 e βh is the vertical stratification delay component; Based on the GFS stratified ZTD using an exponential function model, the least squares method is used to estimate the turbulent mixing delay TD α and the model coefficient β, so as to obtain the GFS vertical stratification delay y SDg and the GFS turbulent mixing delay y TDg at the GNSS station location;

[0052] S22. Subtract the GFS vertical stratification delay y ZTDs at the corresponding position from the GNSS ZTD value y SDg to obtain the GNSS turbulent mixing delay y TDs .

[0053] S3. Build a mapping model from GFS turbulent mixing delay to GNSS turbulent mixing delay based on XGBoost, and predict the ZTD at a specified position and height through the GFS vertical stratification delay at the prediction time and the trained XGBoost model. Step S3 is specifically as follows:

[0054] S31. Build the XGBoost model framework, and use the day of the year, hour, longitude, latitude of historical data, and the GFS turbulent mixing delay y TDg as the input, and the GNSS turbulent mixing delay y TDs as the output for model training. The GNSS turbulent mixing delay is used to correct the accuracy of the GFS turbulent mixing delay to obtain the trained XGBoost model. Because the accuracy of the GNSS turbulent mixing delay at the same position is better and the accuracy of the GFS turbulent mixing delay is lower. The trained model is a mapping from GFS turbulent mixing delay to GNSS turbulent mixing delay. During use, when the GFS turbulent mixing delay at other positions is input, the output of the trained XGBoost model becomes a corrected value, which is equivalent to an improvement in the accuracy of the GFS turbulent mixing delay.

[0055] The input data format of the XGBoost model is a two-dimensional matrix of sample×5, and the output data is a vector of sample×1, where sample is the number of samples of historical data. The five-fold cross-validation method is used to determine the optimal parameters of the XGBoost model. The dataset is divided into five parts, four of which are used as training data, and the remaining one is used as validation data. The finally trained model is the model corresponding to the parameter set that minimizes the root mean square error of all validation data.

[0056] S32. Obtain the GFS forecast data at the specified location and altitude. Calculate the GFS turbulent mixing delay and GFS vertical stratification delay at the specified location and altitude according to the GFS forecast data at the forecast time using the S21 formula. Use the day of the year, hour, longitude, latitude, and GFS turbulent mixing delay at the forecast time as the input of the trained XGBoost model to obtain the corrected GFS turbulent mixing delay. Add the GFS vertical stratification delay and the corrected GFS turbulent mixing delay at the forecast time to obtain the ZTD forecast value.

[0057] Example 2: To evaluate the performance of the proposed ZTD estimation method in high-relief areas, Yunnan Province, China is regarded as a typical high-relief area, and GNSS and GFS data are used to train and validate the model. The GNSS historical data is the ZTD products generated by 26 GNSS stations of the China Crustal Movement Observation Network from 2016 to 2022; the GFS data is the GFS 6-hour and 9-hour forecast data with a resolution of 0.25°×0.25° downloaded from the NCEP GFS 0.25° global forecast grid historical archive (https: / / rda.ucar.edu / datasets / d084001 / ). By combining the GFS 6-hour and 9-hour forecast data, GFS forecast data with a time interval of 3 hours from 2016 to 2022 is formed. According to the GNSS historical data and GFS historical data, the GNSS turbulent mixing delay time series and GFS turbulent mixing delay time series at the positions of 26 GNSS stations are calculated. After quality control, the input data of the XGBoost model is a two-dimensional matrix of 427079×5, and the output data is a vector matrix of 427079×1. The five-fold cross-validation method is used to determine the best parameters of the XGBoost model, and the trained XGBoost_TD model is obtained.

[0058] In terms of accuracy verification, the observation data of 130 GNSS stations from the continuous operation reference station network in Yunnan Province, China, during the period from September 17 to December 21, 2023, were processed using the post - processing PPP strategy, and the obtained GNSS ZTD data was used as the reference true value for verification. At the same time, an XGBoost_ZTD model (the day - of - year, hour, longitude, latitude, height, and GFS ZTD of historical data were used as inputs, and GNSS ZTD was used as the output) and a GRNN model (the longitude, latitude, and GFS turbulent mixing delay at the prediction time were used as inputs, and GNSS turbulent mixing delay was used as the output) were constructed for comparison. The specific experimental scheme is as follows: (1) the original GFS ZTD prediction value; (2) the XGBoost_ZTD model; (3) the GRNN model; (4) the XGBoost_TD model. It should be noted that the experimental platform of this embodiment is the Python 3.8 software under the Windows operating system, and the analysis results of this embodiment will not be affected by the operating system and the Python software version. Table 1 gives the statistical data of the experimental results.

[0059] Table 1

[0060]

[0061] The experimental results show that the average biases of the original GFS ZTD prediction value, the XGBoost_ZTD model, the GRNN model, and the XGBoost_TD model are - 22.76 mm, 22.85 mm, - 4.10 mm, and 1.11 mm respectively, and the root - mean - square errors RMS are 27.17 mm, 36.55 mm, 18.03 mm, and 16.46 mm respectively. Among the four schemes, the XGBoost_ZTD model not only did not improve the accuracy of GFS ZTD prediction, but also introduced significant outliers. This is mainly because the XGBoost_ZTD model did not separate the vertical variation characteristics of ZTD, resulting in a lack of representativeness of its training results in areas with large height differences. Compared with the original GFS ZTD prediction value, the accuracy of the GRNN model increased by about 33.64%. However, from the statistical results of extreme values, it can be found that the error range of the GRNN model exceeded the error range of the original GFS ZTD prediction value. This is because the distribution of GNSS stations participating in training is sparse, and the GRNN model has only one hyperparameter, which is prone to overfitting of the model.

[0062] Compared with the original GFS ZTD prediction value, the XGBoost_ZTD model, and the GRNN model, the accuracy of the XGBoost_TD model has increased by approximately 39.42%, 54.96%, and 8.71% respectively. The XGBoost_TD model not only uses the exponential function to separate the vertical stratification delay component of ZTD but also adjusts the model parameters through a large amount of historical data, reducing the local optimal risk caused by the sparse distribution of GNSS stations involved in training, thereby obtaining the ZTD prediction value with the highest accuracy.

[0063] In summary, the ZTD prediction method for high-relief areas based on GFS and GNSS proposed in the present invention has good performance. By separating the vertical stratification delay and the turbulent mixing delay, the adaptability of the XGBoost model to high-relief areas and areas with sparse GNSS stations is improved. The trained XGBoost model does not require the participation of real-time GNSS ZTD and has better adaptability, which is of great significance for the high-precision prediction of ZTD in high-relief areas.

[0064] Embodiment 3:

[0065] The computer-readable storage medium of this embodiment stores a computer program, and when the program is executed by a processor, it implements the steps in a ZTD prediction method for high-relief areas based on GFS and GNSS in Embodiment 1.

[0066] The computer-readable storage medium of this embodiment can be an internal storage unit of the terminal, such as the hard disk or memory of the terminal; the computer-readable storage medium of this embodiment can also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card, a secure digital card, a flash card, etc. equipped on the terminal; further, the computer-readable storage medium can also include both the internal storage unit and the external storage device of the terminal.

[0067] The computer-readable storage medium of this embodiment is used to store the computer program and other programs and data required by the terminal, and the computer-readable storage medium can also be used to temporarily store the data that has been output or will be output.

[0068] Embodiment 4:

[0069] The computer device of this embodiment includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in a ZTD prediction method for high-relief areas based on GFS and GNSS in Embodiment 1.

[0070] In this embodiment, the processor may be a central processing unit, or may also be other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor, etc.; the memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0071] Those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0072] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A ZTD prediction method for large elevation difference areas based on GFS and GNSS, characterized in that: The following steps are involved: S1, calculate the historical data of GNSS ZTD and GFS layered ZTD at all GNSS station locations in the study area; S2, based on the GFS layered ZTD, obtain the GNSS turbulence mixing delay and GFS turbulence mixing delay at the GNSS station location, specifically: S21, using the exponential function model to describe the vertical change of ZTD: ZTD α =TD α +ZTD α0 in βh +ε α Where ZTD α , TD α and ZTD α0 are the tropospheric delay at position α, the turbulent mixing delay and the vertical stratification delay at a height of 0 m, ε α is the unmodeled residual at position α, β is the model coefficient, and ZTD α0 e βh is the vertical stratified delay component; based on the GFS stratified ZTD, the exponential function model is used to estimate the turbulent mixing delay TD using the least squares method α and model coefficient β, thereby obtaining the GFS vertical stratification delay and GFS turbulent mixing delay at the GNSS station location; S22, subtract the GNSS ZTD from the GFS vertical layered delay at the corresponding position to obtain the GNSS turbulent mixing delay; S3, based on XGBoost, builds a mapping model from GFS turbulent mixing delay to GNSS turbulent mixing delay, and predicts the ZTD of the specified location and altitude through the GFS vertical stratified delay at the forecast time and the trained XGBoost model.

2. The ZTD prediction method for large elevation difference areas based on GFS and GNSS according to claim 1, characterized in that: Step S1 is specifically as follows: S11, calculate the GNSS ZTD at all GNSS station locations by precise point positioning PPP; S12, calculate the GFS ZTD values ​​of all pressure layers below 10 km at all grid point locations in the historical GFS forecast data in the study area; S13, according to the GFS ZTD values ​​of the four GFS grid points closest to the GNSS station location, the GFS ZTD values ​​of all pressure layers below 10 km at the GNSS station location are calculated using the inverse distance weighted method to obtain the GFS layered ZTD.

3. The ZTD prediction method for large elevation difference areas based on GFS and GNSS according to claim 2, characterized in that: The calculation method of GFS ZTD value in S12 is as follows: Calculation of ZTD at the highest pressure layer of the GFS using the Saastamoinen model top , the water vapor content at the highest pressure layer of the GFS is assumed to be zero: Where P top and h top They are the air pressure of the highest atmospheric layer and the earth's height. is the latitude of the grid point; The ZTD between the top layer and other layers is calculated using the hierarchical integration method: Where k1, k2 and k3 are atmospheric refractive index constants, e i,ave , T i,ave and rh i,ave is the average water vapor pressure, temperature and relative humidity between the i-th layer and the i+1-th layer; ds is the thickness between the i-th layer and the i+1-th layer. top and ZTD part Add them together to get the ZTD of the specified pressure layer.

4. The ZTD prediction method for large elevation difference areas based on GFS and GNSS according to claim 2, characterized in that: Step S3 is specifically as follows: S31, construct the XGBoost model framework, take the annual accumulated day, hour, longitude, latitude and GFS turbulence mixing delay of historical data as input, and GNSS turbulence mixing delay as output for model training. GNSS turbulence mixing delay is used to correct the accuracy of GFS turbulence mixing delay, and obtain the trained XGBoost_TD model; S32, obtain the GFS forecast data of the specified location and altitude, calculate the GFS turbulent mixing delay and GFS vertical stratification delay of the specified location and altitude according to the GFS forecast data at the forecast time using the formula of S21, use the annual accumulated day, hour, longitude, latitude and GFS turbulent mixing delay at the forecast time as the input of the trained XGBoost_TD model, obtain the corrected GFS turbulent mixing delay, add the GFS vertical stratification delay and the corrected GFS turbulent mixing delay at the forecast time, and obtain the ZTD forecast value.

5. The ZTD prediction method for large elevation difference areas based on GFS and GNSS according to claim 4, characterized in that: The XGBoost model input data is in the form of a two-dimensional matrix of sample×5, and the output data is a vector of sample×1, where sample is the number of samples of historical data.

6. The ZTD prediction method for large elevation difference areas based on GFS and GNSS according to claim 5, characterized in that: The five-fold cross-validation method is used to determine the optimal parameters of the XGBoost model. The data set is divided into five parts, four of which are used as training data and the remaining one is used as validation data. The final trained model is the model corresponding to the parameter set that minimizes the root mean square error of all validation data.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of a ZTD prediction method for large elevation difference areas based on GFS and GNSS as described in any one of claims 1 to 6 are implemented.

8. A computer device comprising a processor, a memory and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of a ZTD prediction method for large elevation difference areas based on GFS and GNSS are implemented as described in any one of claims 1-6.

Images