Hail prediction method, device, medium and program product

Abstract

The present application discloses a hail prediction method, device, medium, and program product, relating to the field of meteorological technology. The method comprises obtaining a neural network model; obtaining historical lightning characteristic variable data and historical hail characteristic variable data; dividing the historical lightning characteristic variable data and historical hail characteristic variable data into a training set and a validation set corresponding to the characteristic variables; training the neural network model using the training set and the validation set corresponding to the characteristic variables, and using the trained neural network model as a target neural network model; obtaining current lightning characteristic variable data; determining a dynamic weighted loss function based on the target neural network model; and predicting hail weather based on the current lightning characteristic variable data and the dynamic weighted loss function to obtain hail prediction parameters. The present application technology reduces misjudgments caused by predicting hail using radar reflectivity factors alone, and improves hail prediction accuracy by using lightning characteristic data as an indicator factor.

Description

Technical Field

[0001] The present application relates to the field of meteorological technology, and in particular to a hail prediction method, device, medium, and program product. Background Art

[0002] Current hail forecasts rely primarily on weather radar and numerical weather prediction models (ECMWF, European Centre for Medium-Range Weather Forecasts). These methods primarily rely on precipitation intensity, convective parameters (such as wind shear), and radar echoes (such as maximum reflectivity). However, these methods suffer from poor timeliness, high false alarm rates, and inaccurate estimates of convective intensity.

[0003] Because hail formation is usually short, while radar takes a long time to complete a volume scan, the information obtained by relying solely on traditional radar methods is not fast enough; hail identification relies solely on radar echoes, and the most commonly used non-polarized radars in business operations cannot provide information on the shape of particles within the cloud, making it easy to misjudge heavy precipitation as hail, resulting in a high false alarm rate; hail formation depends on the intensity of convection, but existing radar echo and convection parameter estimation methods cannot accurately reflect the energy distribution and airflow structure within the cloud. Summary of the Invention

[0004] The present application provides a hail prediction method, device, medium, and program product. The method includes obtaining a neural network model; obtaining historical lightning characteristic variable data and historical hail characteristic variable data; dividing the historical lightning characteristic variable data and historical hail characteristic variable data into a training set and a validation set corresponding to the characteristic variables; training the neural network model using the training set and the validation set corresponding to the characteristic variables, and using the trained neural network model as a target neural network model; obtaining current lightning characteristic variable data; determining a dynamic weighted loss function based on the target neural network model; and predicting hail weather based on the current lightning characteristic variable data and the dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: a hail occurrence probability parameter, a hail predicted occurrence time parameter, a hail predicted diameter parameter, and a hail predicted location parameter. The present application technology reduces misjudgments caused by predicting hail using radar reflectivity factors alone, and improves hail prediction accuracy by using lightning characteristic data indicator factors.

[0005] This application provides a hail prediction method, which includes:

[0006] Get the neural network model;

[0007] Obtain historical lightning characteristic variable data and historical hail characteristic variable data;

[0008] Divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0009] The neural network model is trained using the training set and validation set of the corresponding feature variables, and the trained neural network model is used as the target neural network model;

[0010] Get current lightning characteristic variable data;

[0011] Determine a dynamic weighted loss function based on the target neural network model;

[0012] Hail weather is predicted based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0013] The present application also provides a hail prediction device, comprising:

[0014] An acquisition module is used to acquire a neural network model; acquire historical lightning characteristic variable data and historical hail characteristic variable data; and acquire current lightning characteristic variable data;

[0015] A partitioning module is used to divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0016] The training module is used to train the neural network model using the training set and validation set of the corresponding feature variables, and use the trained neural network model as the target neural network model;

[0017] A determination module, used to determine a dynamic weighted loss function according to a target neural network model;

[0018] The prediction module is used to predict hail weather based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0019] The present application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for implementing the steps of a hail prediction method when executing the computer program. The method comprises:

[0020] Get the neural network model;

[0021] Obtain historical lightning characteristic variable data and historical hail characteristic variable data;

[0022] Divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0023] The neural network model is trained using the training set and validation set of the corresponding feature variables, and the trained neural network model is used as the target neural network model;

[0024] Get current lightning characteristic variable data;

[0025] Determine a dynamic weighted loss function based on the target neural network model;

[0026] Hail weather is predicted based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0027] The present application also provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, the steps of the hail prediction method are implemented. The method includes:

[0028] Get the neural network model;

[0029] Obtain historical lightning characteristic variable data and historical hail characteristic variable data;

[0030] Divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0031] The neural network model is trained using the training set and validation set of the corresponding feature variables, and the trained neural network model is used as the target neural network model;

[0032] Get current lightning characteristic variable data;

[0033] Determine a dynamic weighted loss function based on the target neural network model;

[0034] Hail weather is predicted based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0035] The present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the hail prediction method are implemented. The method includes:

[0036] Get the neural network model;

[0037] Obtain historical lightning characteristic variable data and historical hail characteristic variable data;

[0038] Divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0039] The neural network model is trained using the training set and validation set of the corresponding feature variables, and the trained neural network model is used as the target neural network model;

[0040] Get current lightning characteristic variable data;

[0041] Determine a dynamic weighted loss function based on the target neural network model;

[0042] Hail weather is predicted based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0043] The present invention provides a method that includes obtaining a neural network model; obtaining historical lightning characteristic variable data and historical hail characteristic variable data; dividing the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set corresponding to the characteristic variables; training the neural network model using the training set and the validation set corresponding to the characteristic variables, and using the trained neural network model as a target neural network model; obtaining current lightning characteristic variable data; determining a dynamic weighted loss function based on the target neural network model; and predicting hail weather based on the current lightning characteristic variable data and the dynamic weighted loss function to obtain hail prediction parameters. Therefore, the present invention reduces misjudgments caused by predicting hail using the radar reflectivity factor alone, and improves hail prediction accuracy by using lightning characteristic data as an indicator factor.

[0044] The technical solution of this application can effectively improve the forecast accuracy of hail size and falling area, and establish a hail formation mechanism analysis model based on lightning activity to better understand the growth process of hail in thunderstorms; and can develop an automated hail warning system, integrate the prediction model into the meteorological forecast system, and improve its practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0046] Figure 1A first flow chart of the hail prediction method provided in an embodiment of the present application;

[0047] Figure 2 A second flow chart of the hail prediction method provided in an embodiment of the present application;

[0048] Figure 3 A specific flow chart of the hail prediction method provided in an embodiment of the present application;

[0049] Figure 4 A structural diagram of a hail prediction device provided in an embodiment of the present application;

[0050] Figure 5 The exemplary systems provided for the embodiments of the present application can be used to implement the various embodiments described in the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0052] It should be noted that, in the description of this application, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. The terms "first," "second," etc., in this application are used to distinguish similar objects, and are not used to describe a particular order or sequence.

[0053] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0054] In conjunction with the specific application environment architecture or specific hardware architecture on which the execution of the hail prediction method depends, the specific application environment architecture or specific hardware architecture is described herein.

[0055] Hail is a severe weather phenomenon that can easily cause agricultural losses, traffic disruptions and property damage.

[0056] Recent research has shown a close relationship between lightning characteristics (polarity, altitude, frequency, etc.) and hail formation. In severe convective systems, a surge in the number of positive lightning strikes, an increase in the average altitude of lightning strikes, and a dramatic change in lightning frequency can all be precursors to hail formation. Therefore, combining lightning data with neural network technology is expected to overcome the limitations of traditional radar forecasts and improve the accuracy and lead time of hail warnings.

[0057] Furthermore, lightning data can provide information about cloud charge distribution, which is crucial for understanding the microphysical mechanisms of hail growth. For example, in supercell storms, an increase in positive lightning could indicate enhanced ice crystal growth in the upper layers of thunderstorms, leading to the formation of larger hailstones. Therefore, using artificial intelligence combined with lightning data for forecasting can effectively improve hail prediction capabilities.

[0058] The existing technology "Hail prediction method and system based on the jump characteristics of satellite-borne lightning observation data" is based on the change rate of lightning frequency generation, determines whether the lightning jump conditions are met, and then uses it to predict hail.

[0059] Disadvantages: single algorithm, low data information density, unreliable method, and failure to use the latest neural network related technologies.

[0060] The embodiment of the present application provides a hail prediction method, such as Figure 1 As shown, the method includes:

[0061] Get the neural network model;

[0062] Obtain historical lightning characteristic variable data and historical hail characteristic variable data;

[0063] Divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0064] The neural network model is trained using the training set and validation set of the corresponding feature variables, and the trained neural network model is used as the target neural network model;

[0065] Get current lightning characteristic variable data;

[0066] Determine a dynamic weighted loss function based on the target neural network model;

[0067] Hail weather is predicted based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0068] It can be understood that this application proposes a hail prediction method based on lightning characteristic data (positive and negative polarity, three-dimensional coordinates, frequency), using neural networks, and combining physical constraint modeling. This method can achieve: 1. Based on the multi-dimensional information development and change characteristics of lightning activity, a neural network model is used for training to obtain a hail prediction weight model to improve the accuracy of hail prediction; 2. On this basis, the model weight is further corrected by combining the polarized radar reflectivity; 3. The hail trajectory prediction is optimized in combination with wind field data parameters.

[0069] Based on the above method, the forecast accuracy of hail size and falling area can be effectively improved, and a hail formation mechanism analysis model based on lightning activity can be established to better understand the growth process of hail in thunderstorms; and an automated hail warning system can be developed, integrating the prediction model into the meteorological forecast system to improve its practical application value.

[0070] The embodiment of the present application provides a hail prediction method, such as Figure 2 As shown, the method includes:

[0071] This application proposes a hail warning method based on lightning characteristics and neural networks, which combines lightning observations, radar data and wind field parameters, and uses neural networks to predict the probability, size and landing area of ​​hail.

[0072] Step S01, obtaining a neural network model;

[0073] Obtain historical lightning characteristic variable data and historical hail characteristic variable data;

[0074] Divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0075] Here, data collection includes: lightning detection systems (such as LMA / BLNET), meteorological radar data, wind profiler radar data, reanalysis data (such as ERA5 / WRF forecast output), and hail record data (such as meteorological station information and user-provided hail information data collected through mobile devices. The data requires a sample set of 50 hail days and at least 150 non-hail days for thunderstorm weather).

[0076] Determine the spatial grid: For example, you can adopt a horizontal uniform latitude and longitude grid and a vertical uniform grid to unify multi-source data into grid data.

[0077] Step S02: converting historical lightning characteristic variable data and historical hail characteristic variable data into tensor data for recognition by a neural network model.

[0078] Feature extraction: Converts raw data collected from multiple sources into tensor data that can be input into the neural network.

[0079] Step S03: training the neural network model using the training set and validation set corresponding to the feature variables, and using the trained neural network model as the target neural network model.

[0080] Specifically, the division ratio: Under normal circumstances, the data set corresponding to the feature variable can be divided into 70% training set, 15% validation set, and 15% test set.

[0081] The lightning characteristic variables are extracted from the divided data set as input X and the hail characteristic variables are extracted as output Y.

[0082] Build and train a neural network model based on, for example, Transformer, and select an appropriate loss function based on actual conditions to train the model.

[0083] Step S031, cleaning the training set and validation set corresponding to the feature variables;

[0084] The neural network model is trained using the cleaned training set of the corresponding feature variables, and the loss function and optimizer of the neural network model are initialized;

[0085] Verify the neural network model through the validation set of the cleaned corresponding feature variables, and calculate the loss value and determination coefficient of the neural network model;

[0086] The loss value and determination coefficient of the neural network model are adjusted and optimized according to the training results.

[0087] Specifically, the collected data is organized into the required input and output for neural network model training, including lightning data extraction of lightning characteristic variables, including:

[0088] Lightning-related characteristic variables: ① Lightning time ( T l ): Absolute timestamp (or normalized time); ② Polarity ( E l ): positive / negative lightning (0 / 1 encoding); ③ three-dimensional coordinates ( L3 l ): latitude and longitude + altitude, or rasterized into a 3D spatial distribution map; ④ minute-level frequency ( F l ): can be counted as the number of lightning strikes per unit time within the grid;

[0089] Output hail feature variables: including: ①Whether hail occurs ( TF h ): 0 / 1 binary classification; ② Hail diameter ( D h ): can be used as a regression target; ③ Hail falling point ( L2h ): used for subsequent landing area prediction; ④ Time ( T h ): can be used as the target event time window.

[0090] Hail prediction: Historical data is divided into training and validation sets according to certain rules, and then input into a neural network model (e.g., based on Transformer) for training. The input is all lightning characteristic variable data, and the output is all hail characteristic variable data. Combined with a targeted loss function, the prediction of all hail characteristic variables is completed.

[0091] Step S04, obtaining current lightning characteristic variable data;

[0092] Determine the dynamic weighted loss function according to the target neural network model.

[0093] Step S041, the dynamic weighted loss function is L total :

[0094] ;

[0095] in, L TF is the hail binary classification loss function, L T is the hail time prediction loss function, L LOC is the hail location prediction loss function, L D Loss function for hail diameter prediction;

[0096] is the time weighting coefficient, is the position weighting coefficient, is the diameter weighting coefficient; 、 、 、 is the loss weight coefficient.

[0097] Specifically, a. Input data: all lightning characteristic variables:

[0098] b. Neural Network Model: Transformer

[0099] c. Output data: Contains four prediction targets, representing different task types: ① hail presence or absence: classification; ② hail time, probability, diameter, and location regression;

[0100] d. Loss function design:

[0101] Simultaneously predicting all hail variables (occurrence, time, location, and diameter) requires the neural network model to simultaneously perform classification (hail occurrence) and regression (time, location coordinates, and diameter). However, conventional weighted loss functions can cause these four tasks to interfere with each other, preventing convergence. Furthermore, because the variables predicted in this task are causally related, an optimized cascade prediction function module is designed.

[0102] Design a dynamically weighted loss function to ensure that the loss of each task is weighted according to the completion of the previous tasks, and the weight is dynamically adjusted according to the prediction quality of each task (for example, through prediction error or confidence measurement).

[0103] The specific formula is as follows: Multi-task joint training, the total loss function is as follows:

[0104] ;

[0105] ;

[0106] Step S0411, hail binary classification loss function ;

[0107] in, is the predicted probability of hail occurrence, α is the hail category weight, and γ is an adjustable parameter.

[0108] Step S0412: hail time prediction loss function ;

[0109] in, is the predicted time of hail occurrence, This is the actual time when the hail occurred.

[0110] Step S0413: hail diameter prediction loss function ;

[0111] in, is the predicted hail diameter, is the actual hailstone diameter.

[0112] Step S0414: hail position prediction loss function ;

[0113] in, are the hail forecast location coordinates, are the actual location coordinates of the hailstone.

[0114] Step S0415, time weighting coefficient ;

[0115] Position weighting coefficient ;

[0116] Diameter weighting factor ;

[0117] in, 、 、 is the regulating factor, TF is the time-frequency prediction value, Tpredict is the time prediction accuracy value, LOCpredict is the location prediction accuracy value.

[0118] Specifically, the four loss functions are:

[0119] is the hail binary classification loss function. When there are few hail event samples, Focal Loss is used to improve the recognition ability of rare events (hail occurrence):

[0120] ;

[0121] is the predicted probability, α is the class weight, and γ is an adjustable parameter.

[0122] The hail time prediction loss function (time regression) is used to predict the most likely time point for hail to occur (unit: minutes). The robust regression loss function Smooth L1 can be used to balance stability and anti-outlier capabilities:

[0123] ;

[0124] is the prediction time, It is real time;

[0125] is the hail diameter prediction loss function (regression), which uses logarithmic scale regression and Smooth L1 loss:

[0126] ;

[0127] is the predicted hail diameter, is the actual hailstone diameter;

[0128] is the hail location prediction loss function, taking coordinate point direct regression (MSE):

[0129] ;

[0130] is the predicted position coordinate, are the true hailstone coordinates.

[0131] Weighting coefficient 、 、 It can be defined as follows, where 、 、 It is a regulating factor used to control the rate of change of the weight of each task; for each task, the better the prediction effect of the current task, the higher the weight coefficient of the subsequent task. 、 、 will be larger, thereby increasing its contribution to the total loss.

[0132] ;

[0133] ;

[0134] ;

[0135] Step S05 , predicting hail weather based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter, and hail prediction position parameter.

[0136] Step S06, optimizing hail prediction position parameters based on wind field data and current lightning characteristic variable data;

[0137] Obtain wind field data, current lightning characteristic variable data, and hail prediction location parameters;

[0138] Calculate the target horizontal wind speed based on wind field data;

[0139] Obtaining a machine learning algorithm, wherein the machine learning algorithm includes a linear regression algorithm;

[0140] Calculate hail motion trajectories based on the machine learning algorithm, current lightning characteristic variable data, and hail prediction location parameters, and adjust the machine learning algorithm parameters based on the calculation results.

[0141] Retrain the adjusted machine learning algorithm based on the target horizontal wind speed;

[0142] The hail movement trajectory is optimized based on the trained machine learning algorithm to generate a list of areas affected by hail.

[0143] Specifically, obtain 500 hPa wind field data: obtain wind speed and direction data at the 500 hPa altitude layer, ensuring that these data cover the corresponding geographical area and have a high enough temporal resolution;

[0144] Match 500 hPa wind data with other existing meteorological data (such as radar reflectivity, lightning activity, ground temperature, humidity, etc.) in time and space for unified analysis;

[0145] Based on 500 hPa wind field data, calculate key indicators that may affect the movement path of hail, such as vertical wind shear (wind speed differences between different altitude layers), horizontal wind speed, wind direction changes, etc.

[0146] Observe the temporal and spatial trends of wind fields and identify factors that may cause changes in the direction of hail movement;

[0147] According to the currently observed hailstone location and size, initial conditions are set, including but not limited to the starting position and speed of the hailstone, and the surrounding wind speed and direction;

[0148] Use appropriate physical models or machine learning algorithms, combined with 500 hPa wind data and other relevant meteorological data, to simulate the movement of hail in the atmosphere. Consider using particle tracking technology to simulate the movement paths of single or multiple hail particles.

[0149] Continuously adjust and optimize physical model parameters based on actual observations to improve prediction accuracy;

[0150] The key indicators extracted from the 500 hPa wind field data are added as new feature variables to the existing feature set;

[0151] Retrain the prediction model using the updated feature set to ensure that the newly added features actually help improve model performance. You can evaluate the effect of model improvements through methods such as cross-validation.

[0152] Use the trained model to predict the hail path in the future and generate a list of areas that may be affected;

[0153] Assess the potential risk level of each region based on the predicted hailfall area and intensity information;

[0154] Release accurate hail warning information to the public in a timely manner and guide residents to take necessary protective measures.

[0155] Through the above steps, 500 hPa wind field data can be effectively integrated into the hail prediction system, thereby more accurately predicting the hailfall area and reducing disaster losses.

[0156] Step S061, calculating the predicted wind direction and speed using the Kalman filter wind speed budget formula and wind field data;

[0157] The actual measured wind direction and speed and the predicted wind direction and speed are filtered according to the Kalman filter method to obtain the target horizontal wind speed;

[0158] The predicted wind direction and speed are calculated using the Kalman filter wind speed budget formula and wind field data, including:

[0159] Get the dynamic model of wind speed F k , control input matrix B k , control vector ,time k ,exist k -1 The best wind speed estimate at time ;

[0160] By formula: , calculated at time k ,based on k -1 Predicted wind direction and speed at time ;

[0161] Get in k The best estimate error covariance at time -1 , process noise covariance matrix Q k ;

[0162] By formula: , calculated at time k ,exist k -1 prediction error covariance matrix .

[0163] At the same time, the predicted wind direction and speed are slightly adjusted through the prediction error covariance matrix.

[0164] Specifically, the Kalman filter method is used to filter the actual measured wind direction and speed with the predicted wind direction and speed to obtain an optimal estimate to improve the forecast effect;

[0165] The Kalman filter wind speed budget formula includes the predicted current wind speed and the estimation error, including the wind speed prediction and prediction error covariance:

[0166] ;

[0167] ;

[0168] in It's time k Forecast wind speed (based on k -1 moment data); B k is the control input matrix;F k is the dynamic model of wind speed, Q k is the process noise covariance;

[0169] The update step is to use observations to correct the prediction, which includes three parts: Kalman gain, wind speed correction, and error covariance update:

[0170] ;

[0171] ;

[0172] ;

[0173] in, It's time k The actual observed wind speed; H k is the observation model (usually H k = 1, directly observed wind speed); R k is the observation noise covariance; K k is the Kalman gain, which determines the importance of new observations relative to the predicted values.

[0174] Step S07, obtaining polarization radar data, wherein the polarization radar data includes a reflectivity factor, a differential reflectivity, a specific differential phase, and a correlation coefficient;

[0175] Obtaining microphysical characteristic data of hail particles based on polarization radar data, and a mapping relationship between the polarization radar data and the microphysical characteristic data of hail particles;

[0176] The hail prediction diameter parameters are optimized based on the mapping relationship between the microphysical characteristic data of hail particles, polarization radar data and the microphysical characteristic data of hail particles;

[0177] The microphysical characteristics of hail particles are obtained based on polarization radar data, including:

[0178] Determine the hail presence value based on the correlation coefficient and reflectivity factor;

[0179] The diameter of the hail is determined based on the relationship between the differential reflectivity and the hail size, and the responsivity of a specific differential phase;

[0180] The density and shape of hail are determined based on the changing curves of differential reflectivity and correlation coefficient;

[0181] The mapping relationship between polarization radar data and microphysical characteristic data of hail particles is obtained based on the polarization radar data, including:

[0182] Create a convolutional neural network model;

[0183] The convolutional neural network model is trained using historical polarization radar data and historical microphysical characteristic data of hail particles to obtain the mapping relationship between the polarization radar data and the microphysical characteristic data of hail particles.

[0184] The convolutional neural network model is improved based on the optimization results of hail prediction diameter parameters.

[0185] Specifically, polarimetric radar data can provide characteristic quantities of the microphysical characteristics of hail particles (such as size, density, and shape). Combined with existing state prediction quantities, the actual situation and the prediction can be combined to provide an optimal budget and improve the forecast effect.

[0186] Polarimetric radar acquires detailed information about targets by emitting horizontally and vertically polarized electromagnetic waves and receiving echo signals. It can provide information about precipitation particle types (such as raindrops, snowflakes, hail, etc.) and their microphysical properties.

[0187] Observation data collected using S-band or C-band dual-polarization weather radar include reflectivity factor (Z), differential reflectivity (Zdr), specific differential phase (Kdp), correlation coefficient (ρhv), etc.

[0188] Remove noise and non-meteorological echoes, and correct errors caused by ground clutter, sidelobe effects, etc.

[0189] Develop specialized algorithms based on polarization parameters to distinguish hail from other precipitation types, for example, using low ρhv values ​​combined with high Z values ​​to indicate the presence of hail;

[0190] The diameter of hail is estimated by using the relationship between Zdr and hail size, as well as the response characteristics of Kdp in the case of large particles.

[0191] By comprehensively considering the variation patterns of Zdr and ρhv, the density and shape of hail particles can be inferred; for example, hail with higher sphericity usually exhibits lower Zdr values;

[0192] The microphysical features extracted from polarimetric radar data are added as input variables to the existing prediction model;

[0193] Select an appropriate model architecture (e.g., random forest, support vector machine, convolutional neural network, etc.) and train it on historical datasets to establish a mapping relationship between polarimetric radar signatures and hail diameter and intensity levels;

[0194] Cross-validation methods were used to evaluate model performance and hyperparameters were adjusted to optimize prediction accuracy;

[0195] Apply the trained model to real-time polarimetric radar data streams to dynamically update hailfall area forecasts and intensity level predictions;

[0196] In addition to polarimetric radar data, information from multiple sources, such as satellite remote sensing and ground observation stations, can be integrated to further improve prediction accuracy;

[0197] Based on feedback from actual hail events, the network model is continuously adjusted and improved to ensure its adaptability and reliability;

[0198] Through the above method, the unique advantages of polarization radar products can be effectively utilized to deeply explore the microphysical properties of hail particles, thereby significantly improving the ability to predict hail diameter and intensity level; this will not only help enhance the effectiveness of hail warning systems, but also provide more accurate services for industries such as agriculture and aviation.

[0199] Step S08: automatically generate hail weather warning information based on the hail prediction result and send it to the meteorological center.

[0200] Here, as Figure 3 As shown, this application collects data: hail-related data from various sources, including but not limited to radar reflectivity, lightning activity, and meteorological fields (such as wind speed, temperature, humidity, etc.);

[0201] Ensure that all data sources have consistent temporal and spatial resolutions for subsequent processing;

[0202] Data cleaning: remove invalid or abnormal data points and fill missing values;

[0203] Standardization / normalization: converting data of different magnitudes to the same magnitude to improve the efficiency and accuracy of model training;

[0204] Time series alignment: ensure that all feature variables are aligned in time to facilitate subsequent analysis;

[0205] Radar reflectivity feature variable extraction: Calculate the statistical characteristics of the radar reflectivity factor (Z), such as maximum value, average value, standard deviation, etc.; extract the reflectivity distribution characteristics of a specific altitude layer; Lightning feature variable extraction: Count the number of lightning strikes per unit time; Analyze the relationship between lightning location and radar echo; Meteorological field feature variable extraction: Extract the spatiotemporal variation characteristics of meteorological elements such as wind speed, wind direction, temperature, and humidity; Calculate parameters closely related to hail formation, such as vertical wind shear and CAPE (convective effective potential energy); Model selection and design: Select a neural network model suitable for time series data processing, such as Transformer and LSTM; Combine the above-extracted feature variables into the model input vector; Use Use historical data to train the model and optimize model parameters so that it can accurately predict hail characteristic variables; design a cascade prediction model based on the different stages of hail formation and gradually refine the prediction results; use the prediction results of the previous stage as input to the next stage to improve prediction accuracy; based on the output of the neural network model, predict characteristic variables such as the probability, intensity, and location of hail occurrence; set reasonable thresholds to convert continuous prediction results into binary or multi-classification hail warning information; conduct rationality checks on the prediction results to avoid false alarms or omissions; promptly release hail warning information through various channels (such as text messages, websites, apps, etc.); collect user feedback on warning information for continuous optimization and improvement of the model.

[0206] The above steps, from data collection to the final release of warning information, constitute a complete hail prediction system.

[0207] The technical solution of the present application reduces the misjudgment caused by forecasting hail based solely on the radar reflectivity factor, uses lightning data indicator factors to improve the hail recognition rate, and can also construct a richer lightning-hail physical model; since both are the result of the action of different ice phase particles, the maximum hail diameter can be inferred based on the lightning-hail relationship; the hail trajectory is calculated in combination with the wind field, thereby improving the accuracy of the predicted landing area.

[0208] The core innovation of this technical solution lies in the combination of lightning observation data and neural network model prediction, providing an efficient and accurate hail warning method.

[0209] In addition, the method further comprises:

[0210] Optimize the hail time prediction loss function (time regression), including:

[0211] Use an asymmetric loss function to increase sensitivity to the direction of prediction errors;

[0212] Asymmetric loss function Loss = α·error, if error>0 (prediction is too late); β·error, if error≤0 (prediction is too early);

[0213] in:

[0214] α: controls the penalty for late predictions;

[0215] β: controls the penalty for early predictions;

[0216] You can set α>β to emphasize "it is better to report early than late";

[0217] Use a learning rate scheduler to dynamically adjust α and β;

[0218] Set α and β as learnable parameters;

[0219] Introducing temporal uncertainty modeling through probability output + Gaussian NLL;

[0220] Modify the neural network model structure so that it outputs two values: mean μ and standard deviation σ;

[0221] Define Gaussian NLL Loss loss function;

[0222] ;

[0223] Where: y is the true value, μ is the predicted mean, is the standard deviation of the prediction;

[0224] Perform forward propagation of the Gaussian NLL Loss loss function, calculate the loss and backpropagate the gradient;

[0225] Evaluate the uncertainty of the neural network model based on the standard deviation;

[0226] Introducing time series modeling to enhance time modeling capabilities;

[0227] The LSTM (Long Short-Term Memory) model and the GRU (Gated Recurrent Unit) model are used to process long sequence data, alleviating the vanishing gradient problem in traditional RNNs.

[0228] Use causal convolutions and dilated convolutions to capture long-term model dependencies without losing temporal order;

[0229] Indicators such as MAE / RMSE / R² are added to better evaluate the performance of the hail time prediction loss function.

[0230] It is understandable that this can be achieved by increasing the sensitivity to the direction of time errors (such as penalizing evening reports more severely than morning reports); introducing uncertainty modeling (probabilistic output); combining time series modeling to enhance temporal consistency; and introducing an attention mechanism to capture key time points, thereby improving the accuracy and stability of the neural network model's hail time prediction.

[0231] The hail prediction method provided in the embodiment of the present application can be further improved and optimized without departing from the technical solution of the present application, and these improvements and optimizations should also be considered as the scope of protection of the present application.

[0232] This application can also provide and optimize the construction of hail-lightning correlation models, and use model interpretation for the study of physical mechanisms; combine with atmospheric models to expand to other disaster predictions: severe convective rainstorms, thunderstorms and strong winds and other disaster predictions, expand the scope of application of the technical solution of this application; expand to other fields beyond meteorology, such as agricultural insurance, aviation safety, etc.

[0233] The beneficial effects of the technical solution provided by the embodiments of the present application are:

[0234] The technology of the present application reduces misjudgments caused by forecasting hail using radar reflectivity factors alone, and improves the prediction accuracy of hail through lightning characteristic data indicator factors.

[0235] The technical solution of this application can effectively improve the forecast accuracy of hail size and falling area, and establish a hail formation mechanism analysis model based on lightning activity to better understand the growth process of hail in thunderstorms; and can develop an automated hail warning system, integrate the prediction model into the meteorological forecast system, and improve its practical application value.

[0236] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method.

[0237] The embodiment of the present application also provides a hail prediction device, such as Figure 4 As shown, the device includes: an acquisition module, a division module, a conversion module, a training module, a determination module, a prediction module, and an optimization module.

[0238] In this embodiment, the acquisition module is used to acquire a neural network model; acquire historical lightning characteristic variable data and historical hail characteristic variable data; acquire current lightning characteristic variable data;

[0239] A partitioning module is used to divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0240] The training module is used to train the neural network model using the training set and validation set of the corresponding feature variables, and use the trained neural network model as the target neural network model;

[0241] A determination module, used to determine a dynamic weighted loss function according to a target neural network model;

[0242] The prediction module is used to predict hail weather based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0243] In this embodiment, the conversion module is used to convert historical lightning characteristic variable data and historical hail characteristic variable data into tensor data for recognition by a neural network model.

[0244] In one embodiment, the training module is used to clean the training set and the validation set corresponding to the feature variables;

[0245] The neural network model is trained using the cleaned training set of the corresponding feature variables, and the loss function and optimizer of the neural network model are initialized;

[0246] Verify the neural network model through the validation set of the cleaned corresponding feature variables, and calculate the loss value and determination coefficient of the neural network model;

[0247] The loss value and determination coefficient of the neural network model are adjusted and optimized according to the training results.

[0248] In one embodiment, a module is used to determine the dynamic weighted loss function. L total :

[0249] ;

[0250] in, L TF is the hail binary classification loss function, L T is the hail time prediction loss function, L LOC is the hail location prediction loss function, L D Loss function for hail diameter prediction;

[0251] is the time weighting coefficient, is the position weighting coefficient, is the diameter weighting coefficient; 、 、 、 is the loss weight coefficient.

[0252] In one embodiment, the determination module is used for the hail binary classification loss function ;

[0253] in, is the predicted probability of hail occurrence, α is the hail category weight, and γ is an adjustable parameter;

[0254] Hail time prediction loss function ;

[0255] in, is the predicted time of hail occurrence, It is the actual time when the hailstorm occurred;

[0256] Hail diameter prediction loss function ;

[0257] in, is the predicted hail diameter, is the actual hailstone diameter;

[0258] Hail location prediction loss function ;

[0259] in, are the hail forecast location coordinates, are the actual location coordinates of the hailstone.

[0260] In one embodiment, a determination module for a time weighting coefficient ;

[0261] Position weighting coefficient ;

[0262] Diameter weighting factor ;

[0263] in, 、 、 is the regulating factor, TF is the time-frequency prediction value, Tpredict is the time prediction accuracy value, LOCpredict is the location prediction accuracy value.

[0264] In one embodiment, the optimization module is used to obtain wind field data, current lightning characteristic variable data, and hail prediction location parameters;

[0265] Calculate the target horizontal wind speed based on wind field data;

[0266] Obtaining a machine learning algorithm, wherein the machine learning algorithm includes a linear regression algorithm;

[0267] Calculate hail motion trajectories based on the machine learning algorithm, current lightning characteristic variable data, and hail prediction location parameters, and adjust the machine learning algorithm parameters based on the calculation results.

[0268] Retrain the adjusted machine learning algorithm based on the target horizontal wind speed;

[0269] The hail movement trajectory is optimized based on the trained machine learning algorithm to generate a list of areas affected by hail.

[0270] In one embodiment, the optimization module is used to calculate the predicted wind direction and wind speed using a Kalman filter wind speed budget formula and wind field data;

[0271] The actual measured wind direction and speed and the predicted wind direction and speed are filtered according to the Kalman filter method to obtain the target horizontal wind speed;

[0272] The predicted wind direction and speed are calculated using the Kalman filter wind speed budget formula and wind field data, including:

[0273] Get the dynamic model of wind speed F k , control input matrix B k , control vector ,time k ,exist k -1 The best wind speed estimate at time ;

[0274] By formula: , calculated at time k ,based on k -1 Predicted wind direction and speed at time ;

[0275] Get in k The best estimate error covariance at time -1 , process noise covariance matrix Q k ;

[0276] By formula: , calculated at time k ,exist k -1 prediction error covariance matrix .

[0277] In one embodiment, the optimization module is configured to obtain polarization radar data, wherein the polarization radar data includes a reflectivity factor, a differential reflectivity, a specific differential phase, and a correlation coefficient;

[0278] Obtaining microphysical characteristic data of hail particles based on polarization radar data, and a mapping relationship between the polarization radar data and the microphysical characteristic data of hail particles;

[0279] The hail prediction diameter parameters are optimized based on the mapping relationship between the microphysical characteristic data of hail particles, polarization radar data and the microphysical characteristic data of hail particles;

[0280] The microphysical characteristics of hail particles are obtained based on polarization radar data, including:

[0281] Determine the hail presence value based on the correlation coefficient and reflectivity factor;

[0282] The diameter of the hail is determined based on the relationship between the differential reflectivity and the hail size, and the responsivity of a specific differential phase;

[0283] The density and shape of hail are determined based on the changing curves of differential reflectivity and correlation coefficient;

[0284] The mapping relationship between polarization radar data and microphysical characteristic data of hail particles is obtained based on the polarization radar data, including:

[0285] Create a convolutional neural network model;

[0286] The convolutional neural network model is trained using historical polarization radar data and historical microphysical characteristic data of hail particles to obtain the mapping relationship between the polarization radar data and the microphysical characteristic data of hail particles.

[0287] The convolutional neural network model is improved based on the optimization results of hail prediction diameter parameters.

[0288] Specifically, this application can improve the accuracy of hail diameter prediction based on a hail prediction model based on lightning characteristics (polarity + height + frequency), combined with radar data and convection parameters; and accurately predict the hail fall area by combining a method of calculating hail trajectories with wind field data.

[0289] The beneficial effects of the technical solution provided by the embodiments of the present application are:

[0290] The technology of the present application reduces misjudgments caused by forecasting hail using radar reflectivity factors alone, and improves the prediction accuracy of hail through lightning characteristic data indicator factors.

[0291] The technical solution of this application can effectively improve the forecast accuracy of hail size and falling area, and establish a hail formation mechanism analysis model based on lightning activity to better understand the growth process of hail in thunderstorms; and can develop an automated hail warning system, integrate the prediction model into the meteorological forecast system, and improve its practical application value.

[0292] For the description of the features in the embodiment corresponding to the hail prediction device, reference can be made to the relevant description of the embodiment corresponding to the hail prediction method, and no further details will be given here.

[0293] An embodiment of the present application further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps of an embodiment of a hail prediction method, the method comprising:

[0294] Get the neural network model;

[0295] Obtain historical lightning characteristic variable data and historical hail characteristic variable data;

[0296] Divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0297] The neural network model is trained using the training set and validation set of the corresponding feature variables, and the trained neural network model is used as the target neural network model;

[0298] Get current lightning characteristic variable data;

[0299] Determine a dynamic weighted loss function based on the target neural network model;

[0300] Hail weather is predicted based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0301] like Figure 5 As shown, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the steps of the hail prediction method embodiment when running, the method comprising:

[0302] Get the neural network model;

[0303] Obtain historical lightning characteristic variable data and historical hail characteristic variable data;

[0304] Divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0305] The neural network model is trained using the training set and validation set of the corresponding feature variables, and the trained neural network model is used as the target neural network model;

[0306] Get current lightning characteristic variable data;

[0307] Determine a dynamic weighted loss function based on the target neural network model;

[0308] Hail weather is predicted based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0309] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0310] An embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the hail prediction method embodiment are implemented. The method includes:

[0311] Get the neural network model;

[0312] Obtain historical lightning characteristic variable data and historical hail characteristic variable data;

[0313] Divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0314] The neural network model is trained using the training set and validation set of the corresponding feature variables, and the trained neural network model is used as the target neural network model;

[0315] Get current lightning characteristic variable data;

[0316] Determine a dynamic weighted loss function based on the target neural network model;

[0317] Hail weather is predicted based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0318] The embodiments of the present application further provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the computer program implements the steps in the embodiment of the hail prediction method, including:

[0319] Get the neural network model;

[0320] Obtain historical lightning characteristic variable data and historical hail characteristic variable data;

[0321] Divide the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of the corresponding characteristic variables;

[0322] The neural network model is trained using the training set and validation set of the corresponding feature variables, and the trained neural network model is used as the target neural network model;

[0323] Get current lightning characteristic variable data;

[0324] Determine a dynamic weighted loss function based on the target neural network model;

[0325] Hail weather is predicted based on current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: hail occurrence probability parameter, hail prediction occurrence time parameter, hail prediction diameter parameter and hail prediction location parameter.

[0326] The technology of the present application reduces misjudgments caused by forecasting hail using radar reflectivity factors alone, and improves the prediction accuracy of hail through lightning characteristic data indicator factors.

[0327] The technical solution of this application can effectively improve the forecast accuracy of hail size and falling area, and establish a hail formation mechanism analysis model based on lightning activity to better understand the growth process of hail in thunderstorms; and can develop an automated hail warning system, integrate the prediction model into the meteorological forecast system, and improve its practical application value.

[0328] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0329] The above describes in detail the hail prediction method, device, medium, and program product provided by this application. Specific examples are used herein to illustrate the principles and implementation methods of this application. The description of the above examples is intended only to facilitate understanding of the method and core concepts of this application. It should be noted that those skilled in the art may make various improvements and modifications to this application without departing from the principles of this application, and such improvements and modifications also fall within the scope of protection of this application.

Claims

1. A hail prediction method, characterized in that: The method comprises: Get the neural network model; Obtain historical lightning characteristic variable data and historical hail characteristic variable data; Dividing the historical lightning characteristic variable data and the historical hail characteristic variable data into a training set and a validation set of corresponding characteristic variables; Training the neural network model using the training set and validation set of the corresponding feature variables, and using the trained neural network model as the target neural network model; Get current lightning characteristic variable data; Determining a dynamic weighted loss function according to the target neural network model; hail weather is predicted based on the current lightning characteristic variable data and a dynamic weighted loss function to obtain hail prediction parameters, wherein the hail prediction parameters include: a hail occurrence probability parameter, a hail prediction occurrence time parameter, a hail prediction diameter parameter, and a hail prediction location parameter; Determining a dynamic weighted loss function according to the target neural network model includes: The dynamic weighted loss function is L total : L total =λ1L TF +λ2·w T (L T )+λ3·w LOC (L LOC )+λ4·w D (L D ); Among them, L TF is the hail classification loss function, L T is the hail time prediction loss function, L LOC is the hail location prediction loss function, L D Loss function for hail diameter prediction; w T is the time weighting coefficient, w LOC is the position weighting coefficient, w D is the diameter weighting coefficient; λ1, λ2, λ3, λ4 are the loss weighting coefficients; The dynamic weighted loss function is used to ensure that the loss of each task is weighted according to the completion of the previous tasks, and the weight is dynamically adjusted according to the prediction quality of each task; The time weighting coefficient w T =1 / (1+exp(-α1(TF-0.5))); The position weighting coefficient w LOC =1 / (1+exp(-α2(Tpredict-0.5))); The diameter weighting coefficient w D =1 / (1+exp(-α3(Tpredict+LOCpredict-1))); Among them, α1, α2, and α3 are adjustment factors used to control the weight change rate of each task; TF is the time frequency prediction value, Tpredict is the time prediction accuracy value, and LOCpredict is the location prediction accuracy value.

2. The hail prediction method according to claim 1, characterized in that: Before training the neural network model using the training set and the validation set of the corresponding feature variables, the method includes: The historical lightning characteristic variable data and the historical hail characteristic variable data are converted into tensor data recognized by the neural network model.

3. The hail prediction method according to claim 1, characterized in that: The training of the neural network model using the training set and the validation set of the corresponding feature variables includes: Cleaning the training set and validation set of the corresponding feature variables; Training the neural network model using the cleaned training set of the corresponding feature variables, and initializing the loss function and optimizer of the neural network model; Verify the neural network model using the validation set of the cleaned corresponding feature variables, and calculate the loss value and determination coefficient of the neural network model; The loss value and determination coefficient of the neural network model are adjusted and optimized according to the training results.

4. The hail prediction method according to claim 1, wherein: The hailstone binary classification loss function L TF =-α(1-p t ) γ log(p t ): Among them, p t is the predicted probability of hail occurrence, α is the hail category weight, and γ is an adjustable parameter.

5. The hail prediction method according to claim 1, characterized in that: The hail time prediction loss function in, is the predicted time of hail occurrence, This is the actual time when the hail occurred.

6. The hail prediction method according to claim 1, characterized in that: The hail diameter prediction loss function in, is the predicted hail diameter, is the actual hailstone diameter.

7. The hail prediction method according to claim 1, characterized in that: The hail position prediction loss function in, are the hail forecast location coordinates, are the actual location coordinates of the hailstone.

8. The hail prediction method according to claim 1, wherein: After predicting hail weather based on the current lightning characteristic variable data and the dynamic weighted loss function to obtain hail prediction parameters, the method includes: Obtain wind field data, current lightning characteristic variable data, and hail prediction location parameters; Calculating a target horizontal wind speed according to the wind field data; Obtaining a machine learning algorithm, wherein the machine learning algorithm includes a linear regression algorithm; Calculating hail motion trajectories based on the machine learning algorithm, current lightning characteristic variable data, and hail prediction position parameters, and adjusting parameters of the machine learning algorithm based on the calculation results; retraining the adjusted machine learning algorithm according to the target horizontal wind speed; The hail movement trajectory is optimized according to the trained machine learning algorithm to generate a list of areas affected by the hail.

9. The hail prediction method according to claim 8, characterized in that: The calculating the target horizontal wind speed according to the wind field data includes: Calculate the predicted wind direction and speed using the Kalman filter wind speed budget formula and wind field data; The actual measured wind direction and speed and the predicted wind direction and speed are filtered according to the Kalman filter method to obtain the target horizontal wind speed; The wind direction and wind speed are calculated and predicted by the Kalman filter wind speed budget formula and wind field data, including: obtaining the dynamic model F of wind speed k , control input matrix B k , control vector u k , time k, the best wind speed estimate at time k-1 By formula: Calculate the wind direction and speed at time k based on the prediction at time k-1 Get the best estimation error covariance P at time k-1 k-1|k-1 , process noise covariance matrix Q k ; By formula: Calculate the prediction error covariance matrix P at time k and time k-1 k|k-1 .

10. The hail prediction method according to claim 1, characterized in that: After predicting hail weather based on the current lightning characteristic variable data and the dynamic weighted loss function to obtain hail prediction parameters, the method further includes: Acquiring polarization radar data, wherein the polarization radar data includes a reflectivity factor, a differential reflectivity, a specific differential phase, and a correlation coefficient; obtaining microphysical characteristic data of hail particles and a mapping relationship between the polarization radar data and the microphysical characteristic data of hail particles according to the polarization radar data; Optimizing the predicted hail diameter parameter according to a mapping relationship between the microphysical characteristic data of the hail particles, the polarization radar data, and the microphysical characteristic data of the hail particles; The obtaining of microphysical characteristic data of hail particles according to the polarization radar data includes: Determining a hail presence value based on the correlation coefficient and the reflectivity factor; determining the diameter of the hailstone according to the relationship between the differential reflectivity and the hailstone size and the responsivity of the specific differential phase; Determining the density and shape of the hailstone according to the variation curve of the differential reflectivity and the correlation coefficient; The obtaining, based on the polarization radar data, a mapping relationship between the polarization radar data and the microphysical characteristic data of hail particles includes: Create a convolutional neural network model; The convolutional neural network model is trained using historical polarization radar data and historical microphysical characteristic data of hail particles to obtain a mapping relationship between the polarization radar data and the microphysical characteristic data of hail particles; the convolutional neural network model is improved based on the optimization results of the hail prediction diameter parameters.

11. An electronic device, characterized in that: include: memory for storing computer programs; A processor, configured to implement the steps of the hail prediction method according to any one of claims 1 to 10 when executing the computer program.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the hail prediction method according to any one of claims 1 to 10 are implemented.

13. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the hail prediction method according to any one of claims 1 to 10 are implemented.

Images