Inversion method and system of troposphere mixed Kalman filter-neural network based on satellite-borne GNSS (Global Navigation Satellite System) data

Abstract

The invention provides a troposphere mixed Kalman filter-neural network inversion method and system based on satellite-borne GNSS data, and belongs to the field of meteorological data processing. The problems that in an existing troposphere inversion technology, a single atmosphere model is difficult to accurately describe all environment conditions, and the atmosphere model is low in precision, low in calculation efficiency and poor in real-time performance are solved. According to the method, the meteorological parameters are estimated by using the neural network, the optimal neural network estimation parameters are found through comparison, so that the accurate meteorological parameters are estimated, the parameters serve as an observation equation of a Kalman filter, time sequence meteorological parameter tracking is achieved, and inversion of the troposphere is achieved through iterative tracking. According to the method, the troposphere inversion precision, efficiency and robustness are remarkably improved, atmospheric interference can be effectively reduced, the satellite navigation positioning precision is improved, a weather forecast and communication system is optimized, meanwhile, high real-time performance and adaptability are achieved, and wide application prospects and remarkable economic benefits are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological data processing, and in particular, to an inversion method and system based on a tropospheric hybrid Kalman filter-neural network using spaceborne GNSS data. Background Art

[0002] With the continuous development and wide application of global navigation satellite system (GNSS) technology, GNSS-based tropospheric inversion technology plays a crucial role in multiple fields, especially in improving the accuracy of satellite navigation systems, enhancing weather forecasting capabilities, and supporting environmental monitoring. Specifically, during the propagation of GNSS signals, they are significantly affected by tropospheric atmospheric conditions, especially the refraction effect of water vapor, which has a non-negligible impact on the signal propagation delay. The change in water vapor content, especially the distribution of precipitable water vapor (PWV) in the atmosphere, has a profound impact on the propagation path and speed of GNSS signals. By accurately inverting the meteorological parameters of the troposphere, especially the water vapor content, the propagation delay of GNSS signals can be effectively corrected, thereby significantly improving the positioning accuracy. Especially in areas with large meteorological changes and high humidity environments, the positioning error caused by atmospheric delay can be significantly reduced, and the reliability and stability of satellite navigation systems can be enhanced.

[0003] However, although GNSS-based tropospheric inversion technology has great potential and application prospects, it still faces a series of challenges and technical problems in practical applications, which restrict the improvement of its accuracy and efficiency. First of all, the accuracy of the atmospheric model is still a core problem in current technology. Most existing atmospheric models are established based on simplified assumptions or regional empirical data. Although they can provide certain estimations, there are often large errors when facing complex tropospheric conditions, especially the influence of extreme weather phenomena or non-uniform atmospheres. Since the troposphere is a highly dynamic system, its physical properties such as temperature, humidity, and pressure have complex changes in space and time, and a single atmospheric model is difficult to accurately describe all environmental conditions.

[0004] Secondly, the computational complexity and real-time issues also plague the wide application of tropospheric inversion technology. Since tropospheric inversion models usually need to process a large amount of observational data and simultaneously perform dynamic model updates and parameter estimations, the computational amount is extremely large. Especially in real-time application scenarios, how to reduce the computational complexity to meet the real-time processing requirements while ensuring the inversion accuracy is still an urgent problem to be solved. Efficient algorithm optimization, especially machine learning and data-driven intelligent algorithms, may be the key to breaking through this bottleneck.

[0005] In summary, although the GNSS-based tropospheric inversion technology has shown great potential in applications in multiple fields, its development still faces many challenges, especially in aspects such as the accuracy of atmospheric models, computational efficiency, non-linear processing, and accuracy improvement. Summary of the Invention

[0006] The technical problem to be solved by the present invention is:

[0007] To solve the problems existing in the existing tropospheric inversion technology, where a single atmospheric model is difficult to accurately describe all environmental conditions, and the atmospheric model has low accuracy, low computational efficiency, and poor real-time performance.

[0008] The technical solution adopted by the present invention to solve the above technical problems:

[0009] The present invention provides an inversion method of a tropospheric hybrid Kalman filter-neural network based on spaceborne GNSS data, including the following steps:

[0010] S100. Obtain measurement data through spaceborne GNSS;

[0011] S200. Initialize the neural network parameters and train the neural network;

[0012] S300. Use the neural network to generate estimated atmospheric temperature, humidity, and pressure profiles, including calculating the absorption coefficient of oxygen molecules, the absorption coefficient of water vapor, the brightness temperature of atmospheric microwave radiation, and constructing an atmospheric microwave radiation transfer model;

[0013] S400. Take the estimated atmospheric temperature, humidity, and pressure profiles generated by the neural network as the observation equation, input the measurement data obtained in step S100 into the Kalman filter, generate a time series parameter sequence for realizing the time series tracking of tropospheric meteorological parameters, and substitute the solution obtained after iterative calculation into the atmospheric microwave transmission model to calculate the fitness value;

[0014] S500. When the termination condition is satisfied, return the estimated meteorological parameters in step S300, that is, the atmospheric temperature, humidity, and pressure profiles, to realize atmospheric inversion.

[0015] Further, in step S100, the collected measurement data includes pressure, temperature, humidity, atmospheric refraction delay, refractive index, signal refraction angle, time delay of the signal path, satellite elevation angle, and relative geometric position of the satellite.

[0016] Further, in step S200, it includes initializing the neural network parameters, and through iterative optimization of the neural network parameters, finding the optimal neural network parameters to realize the training of the neural network.

[0017] Further, in step S300, it includes,

[0018] S310. Calculate the absorption coefficient of oxygen molecules by using the characteristics of atmospheric microwave absorption.

[0019] Oxygen molecules exhibit two absorption regions in the microwave band, including an absorption line with a single spectral line structure and a resonance absorption band centered on radiation, which is composed of several absorption lines. The contribution of the oxygen line is expressed as:

[0020]

[0021] where f is the receiving frequency, f 0 is 60 GHz, T is the atmospheric temperature, and γ is the original frequency of oxygen molecules; is the absorption coefficient of oxygen molecules; P represents the atmospheric pressure;

[0022] The spectral line width parameter is:

[0023]

[0024] Use the above formula to calculate the absorption coefficient of oxygen molecules.

[0025] Furthermore, in step S300, it also includes S320. Calculate the absorption coefficient of water vapor.

[0026] Within the microwave band, the absorption coefficient of water vapor below 100 GHz is expressed by the following formula:

[0027]

[0028] In the formula, γ 1 is the original frequency of water molecules, ρ v is the water vapor density, P represents the atmospheric pressure, dB is the atmospheric radiation brightness temperature, km is the atmospheric radiation coefficient, and T is the atmospheric temperature.

[0029] Furthermore, in step S300, it also includes S330. Calculate the atmospheric microwave radiation brightness temperature.

[0030]

[0031] In the formula, is the observed brightness temperature of the microwave radiometer, θ is the zenith angle, q is the frequency of the observation channel, T δ is the cosmic background brightness temperature, T z is the atmospheric temperature at altitude z, and dB is the atmospheric radiation brightness temperature.

[0032] Furthermore, in step S300, it also includes S350. Construct an atmospheric microwave radiation transfer model.

[0033] Using the MonoRTM model to calculate the temperature, humidity, and pressure profiles of radiosonde data, obtaining the simulated observed brightness temperature, and thus retrieving the atmospheric temperature and humidity profiles; among them, the simulated brightness temperature of each observation channel is;

[0034]

[0035] In the formula, h is the mixed function of atmospheric temperature T', humidity H, and pressure V, representing the set; g is the implicit function of the observation zenith angle θ and the wave number v; is the coupling operator.

[0036] Furthermore, in step S400, when calculating the model using the Kalman filter, it includes:

[0037]

[0038] P k∣k =(I - K k *H k )*P k∣k-1

[0039] In the formula, is the measurement residual, K k is the optimal Kalman gain, S k is the measurement residual covariance, z k is a measurement value of the true state at time k, represents the optimal estimate value at time k, P k∣k is the covariance matrix of the estimate value at time k, H k is the observation model, mapping the true space state into the observation space, R k is the covariance matrix of the observation noise.

[0040] A tropospheric hybrid Kalman filter-neural network inversion system based on spaceborne GNSS data, the system has program modules corresponding to the above steps, and executes the steps in the above tropospheric hybrid Kalman filter-neural network inversion method based on spaceborne GNSS data when running.

[0041] A computer-readable storage medium, the computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of the tropospheric hybrid Kalman filter-neural network inversion method based on spaceborne GNSS data when called by a processor.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] The present invention combines a hybrid model of a Kalman filter and a neural network, significantly improving the accuracy, efficiency, and robustness of tropospheric inversion. The tracking ability based on the Kalman filter can effectively reduce atmospheric interference, solve the problem of strong dependence on the atmosphere in existing inversion methods, improve satellite navigation and positioning accuracy, optimize meteorological forecasting and communication systems, and at the same time has strong real-time performance and adaptability, with broad application prospects and significant economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flowchart of an inversion method of a tropospheric hybrid Kalman filter-neural network based on spaceborne GNSS data in an embodiment of the present invention;

[0045] Figure 2 It is a flowchart of the Kalman filter in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention will be given in conjunction with the accompanying drawings.

[0047] Specific Embodiment 1: Combining Figure 1 and Figure 2 As shown, the present invention provides an inversion method of a tropospheric hybrid Kalman filter-neural network based on spaceborne GNSS data, including the following steps:

[0048] S100. Obtain measurement data through spaceborne GNSS, including pressure, temperature, humidity, atmospheric refraction delay, refractive index, signal refraction angle, time delay of the signal path, satellite elevation angle, and relative geometric position of the satellite;

[0049] S200. Initialize the neural network parameters and train the neural network;

[0050] Including initializing the neural network parameters, and finding the optimal neural network parameters through iterative optimization of the neural network parameters to achieve the training of the neural network;

[0051] S300. Use the neural network to generate an estimated atmospheric temperature, humidity, and pressure profile, specifically including,

[0052] S310. Utilize the atmospheric microwave absorption characteristics, and the absorption coefficient of oxygen molecules is calculated by the following method:

[0053] Oxygen molecules exhibit two main absorption regions in the microwave band; one of them is an absorption line with a single spectral line structure, and its wavelength is 2.53 mm; the other absorption region is a resonance absorption band centered at 5 mm, which is composed of several absorption lines. The contribution of the oxygen line can be expressed as:

[0054]

[0055] Among them, f is the received frequency, and f 0 is 60 GHz, T is the atmospheric temperature, and γ is the original frequency of oxygen molecules; is the absorption coefficient of oxygen molecules; P represents the atmospheric pressure;

[0056] Spectral line width parameter:

[0057]

[0058] The above formula is used to calculate the absorption coefficient of oxygen molecules;

[0059] S320, Calculation of water vapor absorption coefficient:

[0060] In the microwave band (1 - 300 GHz), water vapor has rotational absorption spectral lines at specific frequencies, namely 22.235 GHz and 183.31 GHz; when the frequency is greater than 200 GHz, other spectral lines also affect microwave absorption; the contributions of spectral lines above 300 GHz are combined into a "residual term" to simplify the calculation; for frequencies below 100 GHz, the low-frequency approximation method is used to calculate the absorption coefficient, and the water vapor absorption coefficient below 100 GHz can be expressed by the following formula:

[0061]

[0062] In the formula, γ 1 is the original frequency of water molecules, ρ v is the water vapor density, P represents the atmospheric pressure, dB is the atmospheric radiation brightness temperature, km is the atmospheric radiation coefficient, and T is the atmospheric temperature;

[0063] S330, Calculation of atmospheric microwave radiation brightness temperature:

[0064]

[0065] In the formula, is the observed brightness temperature of the microwave radiometer, θ is the zenith angle, q is the frequency of the observation channel, and T δ is the cosmic background brightness temperature, and T z is the atmospheric temperature at altitude z, and dB is the atmospheric radiation brightness temperature;

[0066] It can be seen from the above formula that the contribution of the atmosphere to the observed brightness temperature of the microwave radiometer comes from two parts: one part is the brightness temperature of cosmic background radiation, and the other part is the brightness temperature of the atmosphere itself that enters the microwave radiometer after being attenuated by the lower atmosphere;

[0067] S340, The atmospheric microwave radiation transfer model MonoRTM is:

[0068] The MonoRTM model is an atmospheric radiative transfer model developed by the Atmospheric and Environmental Research (AER) in the United States and is applicable to microwave band simulations. The water vapor absorption in this model uses the Voigt line shape, and the relevant parameters are sourced from the HITRAN database (such as the pressure broadening coefficient, temperature broadening coefficient, and self-broadening coefficient of water vapor). The water vapor absorption spectrum uses the MTCKD model. This model takes into account the pressure broadening and self-broadening effects of oxygen, nitrogen, carbon dioxide, and ozone.

[0069] The MonoRTM model is written in FORTRAN language and provided to users in source code form. Users need to complete the compilation themselves in a Linux environment. The model uses files as the interface, and in the input file, the temperature, air pressure, altitude, humidity, liquid water content of each layer of the atmosphere, as well as the number of channels, channel wave numbers, and observation zenith angle parameter information of the microwave radiometer are set. Through calculation, the simulated brightness temperature of each observation channel can be obtained.

[0070]

[0071] In the formula, h is the mixed function of the atmospheric temperature T', humidity H, and pressure V, representing the set; g is the implicit function of the observation zenith angle θ and the wave number v. is the coupling operator;

[0072] The present invention uses the MonoRTM model to calculate the temperature, humidity, and pressure profiles of radiosonde data, obtain the simulated observation brightness temperature, and thus invert the atmospheric temperature and humidity profiles.

[0073] S400. Take the atmospheric temperature, humidity, and pressure profiles estimated by the neural network as the observation equation, and input the measurement data obtained in step S100 into the Kalman filter. In this process, it includes initializing the Kalman parameters, generating a time series parameter sequence for realizing the time series tracking of tropospheric meteorological parameters, and substituting the solution obtained after iterative calculation into the atmospheric microwave transmission equation to calculate the fitness value.

[0074] When calculating the model using the Kalman filter, it specifically includes:

[0075]

[0076] P k∣k =(I - K k *H k )*P k∣k-1

[0077] In the formula, is the measurement residual, K k is the optimal Kalman gain, S k is the measurement residual covariance, zk is a measurement of the true state at time k, represents the optimal estimate at time k, P k∣k is the covariance matrix of the estimate at time k, H k is the observation model that maps the true space state to the observation space, R k is the covariance matrix of the observation noise;

[0078] Based on the above formula, the observation equation parameters input through the neural network are implemented to track meteorological parameters;

[0079] S500. When the algorithm termination condition is met, the estimated meteorological parameters, i.e., the atmospheric temperature, humidity, and pressure profiles, are returned to achieve atmospheric inversion.

[0080] Specific Embodiment 2: The present invention proposes an inversion system of a tropospheric hybrid Kalman filter-neural network based on spaceborne GNSS data. This system has program modules corresponding to the above steps and executes the steps in the above inversion method of the tropospheric hybrid Kalman filter-neural network based on spaceborne GNSS data when running.

[0081] Other combinations and connection relationships in this embodiment are the same as those in Specific Embodiment 1.

[0082] Specific Embodiment 3: The present invention proposes a computer-readable storage medium that stores a computer program. The computer program is configured to implement the steps of the inversion method of the tropospheric hybrid Kalman filter-neural network based on spaceborne GNSS data when called by a processor.

[0083] Other combinations and connection relationships in this embodiment are the same as those in Specific Embodiment 1.

[0084] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art of the present invention can make various changes and modifications without departing from the spirit and scope of the present disclosure, and these changes and modifications will all fall within the protection scope of the present invention.

Claims

1. A tropospheric hybrid Kalman filter-neural network inversion method based on space-borne GNSS data, characterized in that: The following steps are involved: S100, obtain measurement data through satellite-borne GNSS; S200, initializing neural network parameters and training the neural network; S300, using a neural network to generate an estimated atmospheric temperature, humidity and pressure profile, including calculating the absorption coefficient of oxygen molecules, the absorption coefficient of water vapor, the brightness temperature of atmospheric microwave radiation, and constructing an atmospheric microwave radiation transmission model; S400, using the atmospheric temperature, humidity and pressure profile estimated by the neural network as the observation equation, inputting the measurement data obtained in step S100 into the Kalman filter, generating a time series parameter sequence for realizing the time series tracking of tropospheric meteorological parameters, and bringing the solution obtained after iterative calculation into the atmospheric microwave transmission model to calculate the fitness value; S500. When the termination condition is met, return to the estimated meteorological parameters of step S300, that is, the atmospheric temperature, humidity and pressure profile, to achieve atmospheric inversion.

2. The method for inversion of tropospheric hybrid Kalman filter-neural network based on satellite-borne GNSS data according to claim 1, characterized in that: In step S100, the collected measurement data include pressure, temperature, humidity, atmospheric refraction delay, refractive index, signal refraction angle, time delay of signal path, satellite altitude angle and relative geometric position of satellite.

3. The method for inversion of tropospheric hybrid Kalman filter-neural network based on space-borne GNSS data according to claim 1, characterized in that: In step S200, the neural network parameters are initialized, and the neural network parameters are iteratively optimized to find the optimal neural network parameters and implement neural network training.

4. The method of claim 3 for inversion of tropospheric hybrid Kalman filter-neural network based on satellite-borne GNSS data, characterized in that: In step S300, it includes: S310, using the atmospheric microwave absorption characteristics, calculate the oxygen molecule absorption coefficient, Oxygen molecules show two absorption regions in the microwave band, including an absorption line with a single spectral line structure and a centrally radiated resonant absorption band, which is composed of several absorption lines. The contribution of the oxygen line is expressed as: Among them, f is the receiving frequency, f0 is 60GHz, T is the atmospheric temperature, and γ is the original frequency of oxygen molecules; is the absorption coefficient of oxygen molecules; P represents the atmospheric pressure; The line width parameters are: The above formula is used to calculate the absorption coefficient of oxygen molecules.

5. The method of claim 4 for inversion of tropospheric hybrid Kalman filter-neural network based on satellite-borne GNSS data, characterized in that: In step S300, the method further includes: S320, calculating a water vapor absorption coefficient; In the microwave band, the water vapor absorption coefficient below 100 GHz is expressed by the following formula: Where γ1 is the original frequency of water molecules, ρ v is the water vapor density, P is the atmospheric pressure, dB is the atmospheric radiation brightness temperature, km is the atmospheric radiation coefficient, and T is the atmospheric temperature.

6. The method of claim 5 for inversion of tropospheric hybrid Kalman filter-neural network based on satellite-borne GNSS data, characterized in that: In step S300, the process further includes: S330, calculating the atmospheric microwave radiation brightness temperature; In the formula, is the brightness temperature observed by the microwave radiometer, θ is the zenith angle, q is the observation channel frequency, T δ is the cosmic background brightness temperature, T z is the atmospheric temperature at height z, and dB is the atmospheric radiation brightness temperature.

7. The method of claim 6 for inversion of tropospheric hybrid Kalman filter-neural network based on satellite-borne GNSS data, characterized in that: In step S300, it also includes, S350, constructing an atmospheric microwave radiation transmission model, The MonoRTM model is used to calculate the temperature, humidity and pressure profiles of the sounding data, and the simulated observation brightness temperature is obtained, thereby inverting the atmospheric temperature and humidity profiles; among which, the simulated brightness temperature of each observation channel is; Where h is a mixed function of atmospheric temperature T', humidity H, and pressure V, representing a set; g is an implicit function of the observed zenith angle θ and the wave number v; is the coupling operator.

8. The method of claim 7 for inversion of tropospheric hybrid Kalman filter-neural network based on satellite-borne GNSS data, characterized in that: In step S400, when the Kalman filter is used to calculate the model, it includes: In the formula, To measure the residual, K k is the optimal Kalman gain, S k To measure the residual covariance, z k is a measurement value of the real state at time k, represents the optimal estimate at time k, P k∣k is the covariance matrix of the estimated value at time k, H k is the observation model, which maps the real space state into the observation space, R k is the covariance matrix of the observation noise.

9. A tropospheric hybrid Kalman filter-neural network inversion system based on spaceborne GNSS data, characterized by: The system has a program module corresponding to the steps of any one of claims 1 to 8, and executes the steps in the above-mentioned tropospheric hybrid Kalman filter-neural network inversion method based on satellite-borne GNSS data when running.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of the tropospheric hybrid Kalman filter-neural network inversion method based on space-borne GNSS data as described in any one of claims 1 to 8 when called by a processor.

Images