Rainfall prediction method and device fusing water vapor field and rainfall field by Kalman filtering

Abstract

This application provides a Kalman filter-based rainfall prediction method that fuses water vapor and rainfall fields. The method includes: establishing multiple satellite-to-ground microwave links for transmitting GNSS and communication satellite data; acquiring GNSS and communication satellite data from these links; performing atmospheric precipitability inversion on each of the multiple links based on the GNSS data, and constructing a two-dimensional water vapor field based on the inversion results; performing rainfall intensity inversion on each link based on the communication satellite data, and constructing a two-dimensional rainfall field based on the inversion results; fusing the two-dimensional water vapor and rainfall fields using a Kalman filter to obtain a fused field, and then transforming the fused field to obtain a rainfall prediction field. This method, by fusing the water vapor and rainfall fields to form a rainfall prediction field, can predict the entire rainfall process with high accuracy.

Description

Technical Field

[0001] This application belongs to the field of rainfall prediction, and in particular relates to a Kalman filter method and apparatus for rainfall prediction that fuses water vapor field and rainfall field. Background Technology

[0002] In recent years, extreme weather events have become more frequent, placing higher demands on the accuracy and real-time performance of short- and medium-term rainfall monitoring and forecasting. Numerical weather prediction (NWP), based on atmospheric dynamic equations, excels at medium- and long-term systematic forecasting but is less effective for short-term forecasting and exhibits a lag in responding to the triggering and evolution of small- and medium-scale convective rainfall. Radar extrapolation predicts rainfall paths by analyzing changes in radar echo intensity, making it suitable for short- and nowcasting, but it lacks the ability to identify weak rainfall echoes and is significantly affected by terrain obstruction and Doppler velocity estimation errors. Satellite composite cloud image inversion combines cloud top temperature, infrared, and microwave channel information from geostationary and polar-orbiting satellites, offering broad coverage, but its spatiotemporal resolution and inversion accuracy are limited.

[0003] However, existing rainfall monitoring and forecasting methods still have shortcomings. They cannot monitor and forecast the entire rainfall process, or their accuracy is low when monitoring the entire rainfall process using existing methods. Therefore, it is necessary to provide a more comprehensive rainfall monitoring and forecasting method. Summary of the Invention

[0004] In view of this, this application provides a method and apparatus for rainfall prediction by fusing water vapor field and rainfall field using Kalman filter, which aims to fuse water vapor field and rainfall field to form rainfall prediction field, thereby providing a more comprehensive description of the entire rainfall process.

[0005] In a first aspect, this application provides a Kalman filter-based rainfall prediction method that fuses water vapor and rainfall fields, including: Multiple satellite-to-ground microwave links are established, which are used to transmit GNSS data and communication satellite data; Acquire GNSS data and communication satellite data from multiple satellite-to-ground microwave links; Atmospheric precipitation was retrieved from multiple satellite-to-ground microwave links based on GNSS data, and a two-dimensional water vapor field was constructed based on the retrieval results. Rainfall intensity was retrieved from multiple satellite-to-ground microwave links based on communication satellite data, and a two-dimensional rainfall field was constructed based on the retrieval results. Kalman filtering is used to fuse the two-dimensional water vapor field and the two-dimensional rainfall field to obtain a fused field, and the fused field is then transformed to obtain the rainfall prediction field.

[0006] Optionally, the steps of retrieving atmospheric precipitation data from multiple satellite-to-ground microwave links based on GNSS data and constructing a two-dimensional water vapor field based on the retrieval results include: Acquire satellite clock bias data and Based on GNSS data and satellite clock bias data, the receiver position, clock bias, and total tropospheric delay are calculated. The GNSS data includes carrier phase. and pseudo-distance The receiver position is used to establish a two-dimensional field, and the clock error is used to correct the observation time deviation of GNSS data. Meteorological data from the ground receiving station of the microwave communication link corresponding to the observation time of GNSS data is acquired, and the atmospheric dry delay is calculated using the dry delay model. The meteorological parameters include air pressure, temperature, station latitude, and altitude. Based on the total tropospheric delay and atmospheric dry delay The atmospheric humidity delay was calculated. And according to atmospheric humidity delay The precipitable water content on multiple satellite-to-ground microwave links was calculated separately. A two-dimensional water vapor field is constructed based on the calculated precipitation on multiple satellite-to-ground microwave links.

[0007] Optionally, the steps of retrieving rainfall intensity for each satellite-to-ground microwave link based on communication satellite data and constructing a two-dimensional rainfall field based on the retrieval results include: Based on communication satellite data, a sunny / rainy season classification result is determined using a sunny / rainy season differentiation model; the communication satellite data includes microwave signals from communication satellites; the sunny / rainy season classification result includes sunny or rainy season. Based on the classification results of sunny and rainy seasons and the currently acquired communication satellite data, rain-induced attenuation is obtained through an attenuation prediction model. During the rainy season, the effect of removing the melt layer on rain-induced attenuation is considered when solving for rain-induced attenuation. Based on the rain-induced attenuation and rainfall intensity inversion model, the rainfall intensity on multiple satellite-to-ground microwave links was calculated. A two-dimensional rainfall field is constructed based on the rainfall intensity on multiple satellite-to-ground microwave links obtained from the calculation.

[0008] Optionally, the methods for constructing a two-dimensional water vapor field or a two-dimensional rainfall field include Kriging interpolation, inverse distance weighting, and spatial variational interpolation.

[0009] Optionally, the Kalman filter is an extended Kalman filter or an unscented Kalman filter.

[0010] Optionally, the steps for transforming the fused field to obtain the rainfall prediction field include: At the spatial level, several convolutional neural networks are used to map the fused field to obtain high-dimensional spatial features, which serve as the first branch; At the temporal level, the fusion field from multiple moments is input into the Long Short-Term Memory network in chronological order to capture the dynamic trend of the fusion field's evolution. , as the second branch; Extract several physical characteristic parameters from the fused field. As the third branch; The features of the three branches are fused in series at a high level and mapped through a fully connected network or a hybrid density network to obtain the rainfall prediction field at different times.

[0011] Optionally, the fully connected network is a fully connected network with residual connections or a fully connected network with an attention mechanism.

[0012] Secondly, this application provides a rainfall prediction device that fuses water vapor field and rainfall field using Kalman filtering, comprising: The data transmission module is used to establish multiple satellite-to-ground microwave links, which are used to transmit GNSS data and communication satellite data. The data acquisition module is used to acquire GNSS data and communication satellite data from multiple satellite-to-ground microwave links; The water vapor field construction module is used to invert atmospheric precipitation based on GNSS data for multiple satellite-to-ground microwave links, and construct a two-dimensional water vapor field based on the inversion results. The rainfall field construction module is used to perform rainfall intensity inversion on multiple satellite-to-ground microwave links based on communication satellite data, and construct a two-dimensional rainfall field based on the inversion results; The rainfall prediction field construction module is used to fuse a two-dimensional water vapor field and a two-dimensional rainfall field using Kalman filtering to obtain a fused field, and then transform the fused field to obtain the rainfall prediction field.

[0013] Thirdly, this application provides an electronic device, including a rainfall prediction device that fuses water vapor field and rainfall field using Kalman filtering as described above.

[0014] Fourthly, this application provides a computer-readable storage medium storing at least one piece of program code, which is executed by a processor to implement the rainfall prediction method for fusing water vapor field and rainfall field as described in any of the preceding claims.

[0015] The beneficial effects of the technical solution provided in this application include: This application provides a Kalman filter-based rainfall prediction method that fuses water vapor and rainfall fields. By simultaneously acquiring GNSS and communication satellite data, water vapor fields on multiple satellite-to-ground microwave links are retrieved from the GNSS data, and rainfall fields on the same links are retrieved from the communication satellite data. The water vapor and rainfall fields are then fused to obtain a rainfall prediction field. This prediction field allows for the prediction and monitoring of the entire rainfall process, and compared to existing methods that rely solely on GNSS data for rainfall prediction, this method offers higher prediction accuracy. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a rainfall prediction method that fuses water vapor and rainfall fields using Kalman filtering, as provided in an embodiment of this application. Figure 2 A flowchart illustrating a rainfall prediction method that fuses water vapor and rainfall fields using Kalman filtering, as provided in an embodiment of this application. Figure 3 A flowchart illustrating a method for constructing a two-dimensional water vapor field according to an embodiment of this application; Figure 4 A flowchart illustrating a method for constructing a two-dimensional rainfall field according to an embodiment of this application; Figure 5 A schematic diagram illustrating the construction process of a rainfall prediction field provided in an embodiment of this application; Figure 6 A structural block diagram of a rainfall prediction device that fuses water vapor field and rainfall field using Kalman filtering, provided in an embodiment of this application; Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of this application.

[0018] The attached figures are labeled as follows: 11: Data transmission module; 12: Data acquisition module; 13: Water vapor field construction module; 14: Rainfall field construction module; 15: Rainfall prediction field construction module.

[0019] 21: Processor; 22: Memory. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] GNSS meteorological technology has developed rapidly in recent years. It mainly operates in the L-band (approximately 1 GHz to 2 GHz), which is highly sensitive to atmospheric water vapor refraction. It can retrieve atmospheric water vapor content by measuring the total tropospheric delay, thus providing information for rainfall forecasting.

[0022] Meanwhile, observing the rain attenuation effect of microwave signals using satellite-to-ground links has become an emerging method for remote sensing of rainfall. Satellite-to-ground links typically operate in the Ka (12GHz~18GHz) / Ku band (27GHz~40GHz). Microwaves experience power attenuation due to absorption and scattering in the rainfall medium, and the rainfall rate can be retrieved by analyzing the rainfall attenuation through the link.

[0023] GNSS satellite-based precipitable water content retrieval and communication link-based rain attenuation monitoring offer complementary advantages throughout the entire rainfall process: the former utilizes L-band signals to retrieve atmospheric water vapor content, suitable for monitoring water vapor accumulation before rainfall, but insensitive to signal attenuation during rainfall itself; the latter, based on Ka / Ku band microwave links, is highly sensitive to signal attenuation caused by rainfall, enabling real-time retrieval of rainfall intensity. This application fuses the information from both methods as multi-source data input, thereby constructing a full-cycle, high spatiotemporal resolution monitoring and forecasting system from water vapor accumulation to rainfall evolution, providing technical support for large-scale, multi-scale rainfall early warning and forecasting.

[0024] Figure 1 A flowchart illustrating a rainfall prediction method that fuses water vapor and rainfall fields using Kalman filtering, provided in an embodiment of this application. See also... Figure 1 ,include: S101. Establish multiple satellite-to-ground microwave links, which are used to transmit GNSS data and communication satellite data.

[0025] In one example, multiple satellite-to-ground microwave links are established in step S101 to acquire GNSS data and communication data from multiple different locations, which will facilitate the subsequent construction of a two-dimensional field.

[0026] In one example, a satellite-to-ground link is established between GNSS and communication satellites, and a joint receiving station is deployed to achieve synchronous reception of GNSS L-band and communication satellite Ka / Ku-band signals.

[0027] Of course, in the satellite-to-ground link construction and receiving station deployment phase of this application, in addition to using Ka / Ku band communication satellites, they can also be equivalently replaced by X-band communication satellites; in addition, in addition to using low Earth orbit (LEO) communication satellites, they can also be equivalently replaced by geostationary orbit (GEO) communication satellites.

[0028] Establishing a space-to-ground link between GNSS and communication satellites requires the unified deployment of joint receiving stations on the ground to achieve high spatial consistency in observation coverage of the target area. The GNSS receiving module receives L-band signals from multiple constellations such as GPS and BeiDou, while a synchronously deployed microwave receiving module receives Ka / Ku band downlink signals from low-Earth orbit or geostationary orbit communication satellites. By coordinating site selection, ephemeris planning, antenna polarization, and beam direction control, the GNSS and communication links form an overlapping observation area on the ground. This constructs a space-to-ground observation network that fuses precipitation precursors and real-time information, providing a spatially matched data foundation for building comprehensive rainfall monitoring and forecasting models.

[0029] S102. Acquire GNSS data and communication satellite data from multiple satellite-to-ground microwave links.

[0030] In one example, step S102 includes: The GNSS receiving module and microwave receiving module arranged in step S101 are used to synchronously receive GNSS data and communication satellite data from multiple satellite-to-ground microwave link locations.

[0031] S103. Based on GNSS data, atmospheric precipitation can be inverted for multiple satellite-to-ground microwave links, and a two-dimensional water vapor field is constructed based on the inversion results.

[0032] See Figure 3 In one example, step S103 includes: S1031. Obtain satellite clock bias data and Based on GNSS data and satellite clock bias data, the receiver position, clock bias, and total tropospheric delay are calculated. The GNSS data includes carrier phase. and pseudo-distance The receiver position is used to establish a two-dimensional field, and the clock error is used to correct the observation time deviation of GNSS data.

[0033] In one example, step S1031 includes: The raw GNSS data (including carrier phase) recorded by the receiver and pseudo-distance (and accompanying precise ephemeris and satellite clock bias data) As input, using a PPP (Precise Point Positioning) or multi-station network-RTK framework, the receiver position is simultaneously estimated using solution software (such as GIPSY-OASIS, BERNESE). Clock difference and total tropospheric delay .

[0034] in, .

[0035] Total tropospheric delay It includes two parts: dry delay and wet delay, i.e. .

[0036] In other embodiments provided in this application, during the GNSS water vapor field inversion process, the PPP / Network RTK solution can be equivalently replaced by Dynamic PPP based on multi-frequency Kalman filtering, or by using a virtual reference station (VRS) differential network solution.

[0037] S1032. Obtain meteorological data from the ground receiving station of the microwave communication link corresponding to the observation time of the GNSS data, and calculate the atmospheric dry delay using the dry delay model. The meteorological parameters include air pressure, temperature, station latitude, and altitude.

[0038] In one example, step S1032 includes: To obtain high-precision total tropospheric delay Then, the air pressure corresponding to the observation time is extracted from ground meteorological stations or numerical weather prediction products. ,temperature Station latitude and altitude Meteorological parameters, and atmospheric dry delay calculated based on the dry delay model. .

[0039] As some examples provided in this application, dry delay models may include the Saastamoinen model, the Black model, the Hopfield model, and the AN model.

[0040] In one example, using the Saastamoinen model, atmospheric dry delay... The calculation process is as follows:

[0041] If systematic biases are found in historical comparisons, the system can also incorporate correction coefficients into the model parameters or calculation results using linear regression or least squares methods, and the corrected atmospheric dry delay can be applied. As output:

[0042] in, This indicates the result calculated by the dry delay model. value, , These are the correction parameters obtained from the regression.

[0043] S1033, Based on the total tropospheric delay and atmospheric dry delay The atmospheric humidity delay was calculated. And according to atmospheric humidity delay The precipitable water content on multiple satellite-to-ground microwave links was calculated separately.

[0044] In one example, step S1033 includes: In obtaining atmospheric dry delay Then, the total tropospheric delay will be... Subtract atmospheric dry delay Atmospheric humidity delay can be obtained. .

[0045] Utilizing temperature and water vapor pressure profile Calculate the weighted average temperature using the following formula. :

[0046] This represents the temperature function of the satellite-to-ground microwave link at different altitudes. This represents the vapor pressure function of the satellite-to-ground microwave link at different altitudes. Indicates altitude.

[0047] Then the water vapor conversion factor can be obtained. :

[0048] in, and These are atmospheric physical constants. This represents the coefficient after temperature weighting correction. The constant of water vapor. Let this be the density of liquid water. Then, ultimately, it can be... Converted to precipitation :

[0049] S1034. Based on the calculated rainfall on multiple satellite-to-ground microwave links, construct a two-dimensional water vapor field.

[0050] In some examples provided in this application, the methods for constructing two-dimensional water vapor fields include Kriging interpolation, inverse distance weighting, and spatial variational interpolation.

[0051] In one example, a two-dimensional water vapor field is constructed using Kriging interpolation. Step S1034 includes: After completing the water vapor inversion for multiple satellite-to-ground microwave links, discrete point-like rainfall information has been obtained along the paths of each link. To achieve continuous and visualized rainfall distribution within a region, it is necessary to extend these non-uniformly distributed point data into a high spatial resolution two-dimensional water vapor field. Therefore, this application employs Kriging interpolation for spatial interpolation to construct a two-dimensional water vapor field. The specific steps are as follows: Step 1: Data preprocessing and spatiotemporal registration.

[0052] Time alignment: Since the sampling rates of GNSS and communication link data are different, the precipitable water and rainfall rate data are first resampled in time, and missing measurement points are interpolated to make up for them.

[0053] Error removal: Sliding window statistics and outlier detection are used to remove observations that significantly deviate from the physical range; Geographic mapping: All site data are uniformly mapped to a unified projected coordinate system, and an observation-grid cell spatial index is established.

[0054] The second step is the interpolation algorithm and spatial field generation.

[0055] The entire satellite-to-ground link coverage area is divided into We have n grids, assuming uniform rainfall intensity and water vapor distribution within each grid. We neglect variations in rainfall intensity along vertical height. Then, the nth grid... Rain attenuation in a satellite-to-ground link is represented as follows:

[0056] in, Indicates the first The number of grids that a link traverses. This represents the horizontal projection length of the link on the ground. Indicates the first The attenuation at each grid point is expressed in dB / km. Indicates the link is at the 1st Projected length within each grid, This indicates the satellite pitch angle (the angle between the satellite link and the horizontal plane).

[0057] For locations not covered by satellite-to-ground links The water vapor in this area It can be represented as:

[0058] in, Represents the weighting coefficient, indicating and The degree of correlation can be solved by the following system of linear equations:

[0059] in, It is a Lagrange multiplier. This represents the semivariance between a known point n and itself. Represents a known point And the point to be estimated The semivariance between them for and The semivariance function between these two positions, which is the expected value of the difference between the values ​​at these two positions, is defined as:

[0060] in, It is the mathematical expectation, representing the average value over multiple samples or pairs of observations.

[0061] S104. Based on communication satellite data, rainfall intensity was inverted for multiple satellite-to-ground microwave links, and a two-dimensional rainfall field was constructed based on the inversion results.

[0062] See Figure 4 In one example, step S104 includes: S1041. Based on communication satellite data, determine the weather classification result using a weather-season differentiation model; the communication satellite data includes microwave signals from communication satellites; the weather classification result includes sunny season or rainy season.

[0063] In one example, step S1042 includes: In the process of precipitation inversion based on satellite-to-ground link microwave signals, accurate identification of sunny and rainy conditions is the primary step in the entire technical approach. It is necessary to accurately identify the rainy season in the time-series signal to provide a reliable basis for subsequent attenuation baseline modeling and precipitation inversion. The principle of the sunny / rainy season differentiation model is mainly based on the difference in microwave signal attenuation between sunny and rainy seasons, using machine learning to automatically distinguish between sunny and rainy seasons in real time. Based on the preprocessed microwave signal, multi-dimensional signal feature vectors such as standard deviation, trend, and information entropy are extracted. These multi-dimensional signal feature vectors are used as input to the machine learning model, outputting real-time sunny / rainy classification results, which serve as input conditions for subsequent attenuation baseline calculation and precipitation inversion modules.

[0064] In one example, in the rainfall inversion of communication satellite microwave links, the rainy season classification model can not only use machine learning models such as random forest and SVM, but also set an empirical threshold based on the historical rainy season attenuation distribution of the link or combine external meteorological monitoring (such as ground rain gauges and radar echoes) for dual discrimination.

[0065] S1042. Based on the classification results of sunny and rainy seasons and the currently acquired communication satellite data, the rain-induced attenuation is obtained through the attenuation prediction model. During the rainy season, the influence of the melt layer on rain-induced attenuation is removed during the process of solving for rain-induced attenuation.

[0066] In one example, step S1043 includes: In one example, the decay prediction model includes any one of LSTM networks, GRU, causal convolutional networks (TCN), or autoregressive moving average models (ARIMA).

[0067] For example, the attenuation prediction model is an LSTM network.

[0068] In one example, for historical link signals during sunny periods, a Long Short-Term Memory (LSTM) network is used to model the normal attenuation characteristics of the link under non-rainy conditions, generating a dynamic attenuation baseline.

[0069] The formula for calculating microwave signal power at the ground end is as follows:

[0070] In the formula, It refers to the satellite antenna's transmit power. It refers to the gain of the satellite transmitting antenna. It is the gain of the ground-side receiving antenna. It is free-space propagation attenuation. It is tropospheric attenuation.

[0071] It should be noted that, It refers to the time changing The function, This indicates that at a certain moment... The above , , , Both represent functions that change over time.

[0072] The calculation formula is:

[0073] in, Frequency, in MHz; The distance traveled is measured in km.

[0074] Known satellite transmitter antenna power With gain and the transmitting antenna power at the ground receiving end. With gain Free space propagation attenuation The tropospheric attenuation can be calculated from the above formula. . This includes attenuation caused by oxygen decay, water vapor decay, liquid water in clouds decay, scintillation decay, precipitation decay, and other factors. The sum of attenuations excluding rainfall attenuation is called the attenuation baseline. Once the attenuation baseline is determined, rainfall-induced attenuation can be calculated. .

[0075] The attenuation baseline can be modeled and predicted using an LSTM (Long Short-Term Memory) network model. First, based on the output of the sunny / rainy season classification model, the sunny season... As input to the model, the LSTM network dynamically adjusts its internal parameters based on non-rainfall attenuation and historical state information, thereby achieving accurate prediction of the attenuation baseline at future times. During the rainfall period, due to... Since the data already includes rain decay, the LSTM network relies solely on previously learned time-series patterns and internal state information to continue predicting the decay baseline for the next time step. .

[0076] It is important to note that for rainfall occurring in stratiform clouds, the attenuation caused by the melting layer must be considered; that is, the effect of the melting layer needs to be removed when calculating rain-induced attenuation. The attenuation coefficient in the melting layer... The calculation formula (unit: dB / km) is as follows:

[0077] in, This indicates the distance the object fell within the melt layer. Let be the extinction area of ​​the particle. The equivalent diameter of the melted particles, The equivalent diameter of the raindrop. The maximum equivalent diameter of a raindrop. The smallest equivalent diameter of a raindrop. This represents the volume fraction of snow particles in the melted particles. , The particle densities are those of snow particles and raindrops, respectively. express The particle spectrum distribution.

[0078] In other embodiments provided in this application, melting layer correction can be performed in addition to the physical extinction model, using the ITU-RP.840 cloud attenuation empirical formula combined with radar echo top height correction, or linear regression correction based on satellite microwave cloud probe liquid water path (LWP).

[0079] S1043. Based on the rain-induced attenuation and rainfall intensity inversion model, the rainfall intensity on multiple satellite-to-ground microwave links is calculated respectively.

[0080] In one example, step S1043 includes: After identifying the rainy / sunny season and determining the attenuation baseline, the rain attenuation component in the microwave link can be accurately extracted. To further convert rainfall attenuation into rainfall intensity, a neural network-based rainfall intensity inversion model was constructed to achieve a nonlinear mapping from signal features to precipitation (RR).

[0081] The neural network model comprises an input layer, hidden layers, and an output layer. This model takes rainfall attenuation and multi-dimensional signal feature vectors as input. Through a nonlinear activation function, the neural network automatically extracts deep patterns from the input features, learning the mapping relationship between rainfall attenuation and rainfall intensity. The number of hidden layers and neurons is flexibly configured according to the amount of data and prediction accuracy requirements to achieve the model's optimal expressive power. During training, historical rain gauge measurements from the same period are used as supervision labels. The neural network continuously iterates and optimizes the loss function, adjusting internal weights to improve prediction accuracy. The trained model can be directly applied to newly received satellite signal data, outputting rainfall intensity values ​​at the corresponding time and path in real time. Compared to traditional empirical models, this method has stronger nonlinear fitting capabilities and cross-regional adaptability, improving the spatial continuity and accuracy of rainfall retrieval.

[0082] The rainfall intensity on multiple satellite-to-ground microwave links was obtained by using a pre-trained rainfall intensity inversion model.

[0083] In other embodiments provided in this application, in addition to multi-layer neural networks, rainfall intensity inversion can also employ ensemble learning (XGBoost, LightGBM) or graph neural network (GNN) models.

[0084] S1044. Based on the rainfall intensity on multiple satellite-to-ground microwave links obtained from the solution, construct a two-dimensional rainfall field.

[0085] In some examples provided in this application, the methods for constructing two-dimensional rainfall fields include Kriging interpolation, inverse distance weighting, and spatial variational interpolation.

[0086] In one example, a two-dimensional rainfall field is constructed using Kriging interpolation. Please refer to step S1034 for details. The construction method of the two-dimensional rainfall field is the same as that of the two-dimensional water vapor field, and will not be elaborated here.

[0087] S105. A fused field is obtained by fusing the two-dimensional water vapor field and the two-dimensional rainfall field using Kalman filtering, and the fused field is then transformed to obtain the rainfall prediction field.

[0088] In one example, step S105 includes: S1051. A fused field is obtained by fusing the two-dimensional water vapor field and the two-dimensional rainfall field using Kalman filtering.

[0089] In one example, the Kalman filter is an extended Kalman filter or an unscented Kalman filter.

[0090] The multi-source fusion data uses extended / unscented Kalman filtering as the core algorithm framework. Either extended Kalman filtering (EKF) or unscented Kalman filtering (UKF) can be selected to fuse two types of data: precipitable water and rainfall intensity.

[0091] In the first example provided in this application, the steps for fusing the two-dimensional water vapor field and the two-dimensional rainfall field using extended Kalman filtering are as follows: Standard Kalman filtering requires linear system equations, but real atmospheric system models exhibit nonlinear characteristics. The core idea of ​​extended Kalman filtering is to linearize the nonlinear system model using Taylor expansion. exist Expanded to:

[0092] in, For function exist nth derivative, Expand the remainder of Taylor's terms.

[0093] In the case of high-dimensional variables, the Taylor expansion of the above equation is derived as follows:

[0094] In the formula: This indicates the nonlinear function at the expansion point. The function value at that point, Indicates the reference point for the Taylor expansion. This represents the higher-order remainder term in the Taylor expansion.

[0095] Let the Jacobian matrix be:

[0096] in, Given an m-dimensional state vector, It is an n-dimensional state vector.

[0097] According to the water vapor field and rainfall intensity Constructing state vectors The evolution of the system state can be described by a nonlinear function:

[0098] in, express The state vector at time t, For system noise, Indicates a time step. This indicates a Taylor expansion.

[0099] The observation equation is expressed as:

[0100] in, Indicates in The observed value at that location (also the measured satellite link signal). Indicates and Unrelated noise The observation function (also derived from the state vector) represents the observation function. ) to observed value (mapping function).

[0101] By performing a Taylor series expansion of the nonlinear vector function and representing it with a Jacobian matrix, it is possible to... To achieve system linearization, the discrete form of the extended Kalman filter equation is derived:

[0102] in, This represents the optimal posterior state estimate at time k+1. Represents the Jacobian matrix. Represents the Jacobian matrix. Indicates in Kalman gain at that location.

[0103] Then the fusion field That is, the result obtained from the above formula. .

[0104] In the second example provided in this application, the steps for fusing the two-dimensional water vapor field and the two-dimensional rainfall field using unscented Kalman filtering are as follows: Unscented Kalman filtering constructs a set of representative sigma points in the state space through unscented transformation, avoiding the problem of first-order linearization approximation of nonlinear systems, thereby improving the accuracy of characterizing nonlinear propagation characteristics while ensuring computational efficiency.

[0105] According to the water vapor field and rainfall intensity Constructing state vectors Then the initial state of the state vector. Covariance for:

[0106]

[0107] in, Represents the mathematical expectation. Represents the prior estimate of the initial state. This represents the transpose of a matrix.

[0108] Calculate deterministic sampling points:

[0109] In the formula, Indicates the central sampling point. Indicates the first One sampling point, Indicates the index number of the sampling point. Indicates in The optimal estimate of the state vector at time step [time]. Indicates in Covariance at time, The dimension representing the state variable. The parameter representing the control of the sampling point distribution satisfies the following formula:

[0110] In the formula, This represents the scaling parameter that controls the distribution range. This represents the second-order parameter of the moderating distribution.

[0111] The weight corresponding to each sampling point is:

[0112] In the formula, the parameter , and All of them can be adjusted. and These represent the mean weight and the covariance weight, respectively.

[0113] Then state variables Error covariance Observed variables Predicted value , , for:

[0114]

[0115]

[0116] Covariance of observed variables for:

[0117] Cross-covariance between measured values ​​and state variables for:

[0118] Then the gain for:

[0119] Then the estimated value of the state variable and error covariance Represented as:

[0120]

[0121] in, This is the desired fusion field.

[0122] Step S1052: Transform the fused field to obtain the rainfall prediction field.

[0123] See Figure 5 In one example, step S1052 includes: The first step is to use several convolutional neural networks at the spatial level to map the fused field to obtain high-dimensional spatial features, which serve as the first branch.

[0124] In other embodiments, the CNN for spatial branch feature extraction can be equivalent to ResNet, U-Net, or Graph Convolutional Network (GCN).

[0125] The second step, at the temporal level, involves inputting the fusion field from multiple moments into the Long Short-Term Memory network in chronological order to capture the dynamic trends of the fusion field's evolution. , as the second branch.

[0126] The third step is to extract several physical characteristic parameters from the fused field. , as the third branch.

[0127] The fourth step is to connect and fuse the features of the three branches in a high-level array, and then map them through a fully connected network or a hybrid density network to obtain the rainfall prediction field at different times.

[0128] In one example, the fully connected network is a fully connected network with residual connections or a fully connected network with an attention mechanism.

[0129] Figure 6 This is a structural block diagram of a rainfall prediction device that fuses water vapor and rainfall fields using Kalman filtering, provided as an embodiment of this application. See also... Figure 6 ,include: Data transmission module 11 is used to establish multiple satellite-to-ground microwave links, which are used to transmit GNSS data and communication satellite data; Data acquisition module 12 is used to acquire GNSS data and communication satellite data from multiple satellite-to-ground microwave links; The water vapor field construction module 13 is used to perform atmospheric precipitation inversion on multiple satellite-to-ground microwave links based on GNSS data, and to construct a two-dimensional water vapor field based on the inversion results. Rainfall field construction module 14 is used to perform rainfall intensity inversion on multiple satellite-to-ground microwave links based on communication satellite data, and construct a two-dimensional rainfall field based on the inversion results; The rainfall prediction field construction module 15 is used to fuse the two-dimensional water vapor field and the two-dimensional rainfall field using Kalman filtering to obtain a fused field, and then transform the fused field to obtain the rainfall prediction field.

[0130] It should be noted that, since the specific execution method has already been described in the preceding steps, therefore, Figure 6 The method steps performed by the provided device can be directly referred to the foregoing. Figures 1 to 5 The contents of the record will not be elaborated here.

[0131] Figure 7 This is a structural block diagram of an electronic device provided according to an embodiment of this application. See also... Figure 7 Electronic devices may include Figure 6The aforementioned Kalman filter-based rainfall prediction device fuses water vapor and rainfall fields. Typically, the electronic equipment includes a processor 21 and a memory 22. The processor 21 may include one or more processing cores, such as a 4-core processor or an 8-core processor. The processor 21 can be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor is used to process data in the wake-up state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state. The memory 22 may include one or more computer-readable storage media, which may be non-transitory. The memory 22 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage medium in memory 22 is used to store at least one instruction, which is executed by processor 21 to implement the Kalman filter fusion of water vapor field and rainfall field precipitation prediction method provided by an electronic device in the method embodiments of this application.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for rainfall prediction by Kalman filter fusion of water vapor field and rainfall field, characterized in that, include: Multiple satellite-to-ground microwave links are established, which are used to transmit GNSS data and communication satellite data; Acquire GNSS data and communication satellite data from multiple satellite-to-ground microwave links; Atmospheric precipitation was retrieved from multiple satellite-to-ground microwave links based on GNSS data, and a two-dimensional water vapor field was constructed based on the retrieval results. Rainfall intensity was retrieved from multiple satellite-to-ground microwave links based on communication satellite data, and a two-dimensional rainfall field was constructed based on the retrieval results. Kalman filtering is used to fuse the two-dimensional water vapor field and the two-dimensional rainfall field to obtain a fused field, and the fused field is then transformed to obtain the rainfall prediction field.

2. The method according to claim 1, wherein the Kalman filter fusion of water vapor field and rainfall field is used for rainfall prediction. The steps for retrieving atmospheric precipitation data from multiple satellite-to-ground microwave links based on GNSS data and constructing a two-dimensional water vapor field based on the retrieval results include: Acquiring satellite clock error data And And according to the GNSS data and the satellite clock error data, the receiver position, clock error and total delay of the troposphere are calculated The GNSS data includes carrier phase And pseudo-range The receiver position is used to establish a two-dimensional field, and the clock error is used to correct the observation time deviation of the GNSS data Obtaining meteorological data at a ground receiving station of a microwave communication link corresponding to an observation time of GNSS data, and obtaining atmospheric dry delay through a dry delay model ; the meteorological parameters include air pressure, temperature, station dimension and altitude According to the tropospheric total delay and the atmospheric dry delay , the atmospheric wet delay is calculated, and according to the atmospheric wet delay , the precipitable water on a plurality of satellite-ground microwave links is calculated respectively; A two-dimensional water vapor field is constructed based on the calculated precipitation on multiple satellite-to-ground microwave links.

3. The method according to claim 2, wherein the Kalman filter fusion of water vapor field and rainfall field is used for rainfall prediction. The steps for retrieving rainfall intensity from each satellite-to-ground microwave link based on communication satellite data and constructing a two-dimensional rainfall field based on the retrieval results include: Based on communication satellite data, a sunny / rainy season classification result is determined using a sunny / rainy season differentiation model; the communication satellite data includes microwave signals from communication satellites; the sunny / rainy season classification result includes sunny or rainy season. According to the sunny and rainy period classification result and the current communication satellite data, rain attenuation is obtained through a rain attenuation prediction model ; wherein, during the rainy period, the influence of the melting layer on the rain attenuation is removed in the process of solving the rain attenuation. Based on the rain-induced attenuation and rainfall intensity inversion model, the rainfall intensity on multiple satellite-to-ground microwave links was calculated. A two-dimensional rainfall field is constructed based on the rainfall intensity on multiple satellite-to-ground microwave links obtained from the calculation.

4. The rainfall prediction method based on Kalman filtering fusion of water vapor field and rainfall field according to any one of claims 1 to 3, characterized in that, Methods for constructing two-dimensional water vapor fields or two-dimensional rainfall fields include Kriging interpolation, inverse distance weighting, and spatial variational interpolation.

5. The rainfall prediction method based on Kalman filtering fusion of water vapor field and rainfall field according to any one of claims 1 to 3, characterized in that, The Kalman filter is either an extended Kalman filter or an unscented Kalman filter.

6. The rainfall prediction method based on Kalman filtering fusion of water vapor field and rainfall field according to any one of claims 1 to 3, characterized in that, The steps to transform the fused field to obtain the rainfall prediction field include: At the spatial level, several convolutional neural networks are used to map the fused field to obtain high-dimensional spatial features, which serve as the first branch; At the temporal level, the fusion field from multiple moments is input into the Long Short-Term Memory network in chronological order to capture the dynamic trend of the fusion field's evolution. , as the second branch; Extract several physical characteristic parameters from the fused field. As the third branch; The features of the three branches are fused in series at a high level and mapped through a fully connected network or a hybrid density network to obtain the rainfall prediction field at different times.

7. The rainfall prediction method based on Kalman filtering fusion of water vapor field and rainfall field according to claim 6, characterized in that, The fully connected network is either a fully connected network with residual connections or a fully connected network with an attention mechanism.

8. A rainfall prediction device that fuses water vapor field and rainfall field using Kalman filtering, characterized in that, include: The data transmission module is used to establish multiple satellite-to-ground microwave links, which are used to transmit GNSS data and communication satellite data. The data acquisition module is used to acquire GNSS data and communication satellite data from multiple satellite-to-ground microwave links; The water vapor field construction module is used to invert atmospheric precipitation based on GNSS data for multiple satellite-to-ground microwave links, and construct a two-dimensional water vapor field based on the inversion results. The rainfall field construction module is used to perform rainfall intensity inversion on multiple satellite-to-ground microwave links based on communication satellite data, and construct a two-dimensional rainfall field based on the inversion results; The rainfall prediction field construction module is used to fuse a two-dimensional water vapor field and a two-dimensional rainfall field using Kalman filtering to obtain a fused field, and then transform the fused field to obtain the rainfall prediction field.

9. An electronic device, characterized in that, The rainfall prediction device includes the Kalman filter fusion of water vapor field and rainfall field as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one piece of program code, which is executed by a processor to implement the Kalman filter fusion method for rainfall prediction as described in any one of claims 1 to 7.

Images