Artificial Intelligence Dual-Polarization Radar Quantitative Precipitation Estimation Method with Enhanced Data Features

Abstract

The present invention discloses an artificial intelligence dual-polarization radar quantitative precipitation estimation method with enhanced data features, including: S1, obtaining dual-polarization radar observation data and constructing a three-dimensional spatial grid structure data set; S2, constructing a double Gaussian probability density function to enhance the data features of a single variable in the three-dimensional spatial grid structure data set; S3, constructing a multi-dimensional joint probability density function based on the double Gaussian probability density function to perform joint data feature enhancement on all variables in the three-dimensional spatial grid structure data set; S4, training an artificial intelligence model that reflects the relationship between radar data and precipitation data. The present invention innovatively uses the probability density function feature to describe the rolling change of dual-polarization radar variables within one hour. By constructing the three-dimensional joint probability density function feature, not only can the distribution characteristics of dual-polarization radar variables be extracted, but also the spatio-temporal evolution law of precipitation particles can be effectively captured.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, in particular to the field of weather radar data processing, and specifically relates to a data feature enhanced artificial intelligence dual-polarization radar quantitative precipitation estimation method. Background Art

[0002] Dual-Polarization Radar (Dual-Pol Radar) is a more advanced meteorological radar technology than traditional single-polarization radar. Traditional radar uses only one polarization signal (usually horizontal polarization) to detect the reflection characteristics of precipitation particles, while dual-polarization radar sends and receives horizontal and vertical polarization signals at the same time. By comparing the reflection differences between the two polarization signals, dual-polarization radar can provide information about the shape, size, phase (such as rain, snow, hail, etc.) and distribution of precipitation particles. This additional information gives dual-polarization radar a clear advantage in weather forecasting, precipitation intensity estimation, storm monitoring and disaster warning, and can significantly improve the precision and accuracy of precipitation estimation in complex weather systems.

[0003] Quantitative Precipitation Estimation (QPE) is a key technology for meteorological and hydrological monitoring. Traditional meteorological radars only use reflectivity factors to establish precipitation estimation formulas, resulting in large inaccuracies in precipitation estimates. Dual-polarization radars can provide more abundant atmospheric cloud physics information, thereby further improving the accuracy of precipitation estimates. Current dual-polarization radars mainly use reflectivity factors, differential reflectivity, and differential phase shift rates to establish precipitation estimation formulas. This technical method usually underestimates heavy precipitation because the precipitation process itself is highly dynamic and uncertain. The precipitation estimation formula established only by directly observing radar parameters cannot accurately describe the physical characteristics of precipitation cloud generation and disappearance.

[0004] Therefore, how to extract effective precipitation evolution features from dual-polarization radar data and establish a stable and reliable precipitation estimation model is a technical problem that needs to be solved urgently. In addition, there are differences in precipitation mechanisms under different geographical and climatic conditions. How to build a universal quantitative precipitation estimation method to adapt to complex and changeable weather systems is also an important issue. These challenges involve multiple technical fields such as data processing, feature extraction, statistical modeling, and machine learning. It is necessary to innovatively integrate multiple methods to break through the existing technical bottlenecks and realize the high-precision and high-efficiency conversion of radar signals into quantitative precipitation estimation products. Summary of the invention

[0005] The present invention aims to solve one of the technical problems existing in the related art at least to a certain extent.

[0006] The object of the present invention is to provide an artificial intelligence dual-polarization radar quantitative precipitation estimation method with enhanced data features, construct a three-dimensional joint probability density function to perform additional processing on radar data, enhance the cloud precipitation physical characteristics in the data, provide more characteristic variables for artificial intelligence training, and thus significantly improve the accuracy of dual-polarization radar quantitative precipitation estimation.

[0007] To achieve the above object, the present invention provides an artificial intelligence dual-polarization radar quantitative precipitation estimation method with enhanced data features, including:

[0008] S1. Obtain dual-polarization radar observation data, preprocess the dual-polarization radar observation data, and construct a three-dimensional spatial grid structure data set;

[0009] S2. Construct a bi-Gaussian probability density function to enhance the data features of a single variable in the three-dimensional spatial grid structure data set;

[0010] S3. Based on the bi-Gaussian probability density function, construct a multi-dimensional joint probability density function to perform joint data feature enhancement on all variables in the three-dimensional spatial grid structure data set, and obtain a three-dimensional spatial grid structure enhanced feature data set;

[0011] S4. Use the three-dimensional spatial grid structure enhanced feature data set as a characteristic variable and the precipitation data of the corresponding grid as a target variable to train an artificial intelligence model reflecting the relationship between radar data and precipitation data for artificial intelligence quantitative precipitation estimation.

[0012] A further preferred technical solution of the present invention is that step S1 of obtaining the dual-polarization radar observation data, preprocessing the dual-polarization radar observation data, and constructing a three-dimensional spatial grid structure data set specifically includes:

[0013] S11. Extract azimuth, elevation, and radial distance information from the dual-polarization radar observation data;

[0014] S12. According to the azimuth, elevation, and radial distance information, map the dual-polarization radar observation data into a three-dimensional spatial grid structure;

[0015] S13. Establish a data set of three variables, namely reflectivity factor, differential reflectivity, and differential phase shift rate, in each spatial grid to form a three-variable dual-polarization radar data set;

[0016] S14. Perform standardization processing on each variable of the three-variable dual-polarization radar data set to obtain the three-dimensional spatial grid structure data set.

[0017] Preferably, in step S13, a data set of three variables, namely, reflectivity factor, differential reflectivity, and differential phase shift rate, is established in each spatial grid to form a three-variable dual-polarization radar data set, which is expressed as:

[0018] ;

[0019] Each variable of the three-variable dual-polarization radar data set is normalized to obtain the three-dimensional spatial grid structure data set, which is expressed as:

[0020] ;

[0021] In the formula, represents the dual-polarization radar spatial grid data set, and the subscript represents the variable in the data set; respectively represent the horizontal and vertical coordinates of the space, is the range of the horizontal and vertical coordinates; is the reflectivity factor, is the differential reflectivity, is the differential phase shift rate; represents the three-dimensional spatial grid structure data set after normalization, and are the mean and variance of each variable, respectively.

[0022] Preferably, in step S2, a double Gaussian probability density function is constructed to enhance the data characteristics of a single variable in the three-dimensional spatial grid structure data set; specifically, it includes:

[0023] S21. For the three-dimensional spatial grid data set, select the rolling update observation data within a continuous time window;

[0024] S22. Fit a single variable of the rolling update observation data using a Gaussian mixture model to obtain a double Gaussian distribution function;

[0025] S23. According to the double Gaussian distribution function, calculate the weight, the first Gaussian mean and the second Gaussian mean, and the first Gaussian variance and the second Gaussian variance, which are used to describe the probability distribution characteristics of the radar variable during this time period, and generate the enhanced feature data of a single variable.

[0026] Preferably, during the process of fitting a single variable of the rolling update observation data using a Gaussian mixture model, it is assumed that the radar data obeys a Gaussian mixture model composed of the linear weighting of two normal distributions. The expression of the double Gaussian distribution function is as follows:

[0027] ;

[0028] In the formula, is a double Gaussian distribution function, and are the first Gaussian function and the second Gaussian function, is the double Gaussian weight, and are the mean parameters of the first Gaussian function and the mean parameters of the second Gaussian function, and are the variance parameters of the first Gaussian function and the variance parameters of the second Gaussian function.

[0029] Preferably, in step S21, for the three-dimensional spatial grid data set, rolling updated observation data within a continuous time window is selected; specifically including:

[0030] Obtain the dual-polarization radar observation data at a preset time interval;

[0031] Divide the dual-polarization radar observation data into multiple continuous time windows, where each time window contains multiple continuous dual-polarization radar 6-minute interval volume scan data;

[0032] For each time window, extract the reflectivity factor, differential reflectivity, and differential phase shift rate in the three-dimensional spatial grid data set;

[0033] Generate rolling updated observation data for the corresponding time window according to the reflectivity factor, differential reflectivity, and differential phase shift rate.

[0034] Preferably, step S3 is based on the double Gaussian probability density function to construct a multi-dimensional joint probability density function, and perform joint data feature enhancement on all variables in the three-dimensional spatial grid structure data set to obtain a three-dimensional spatial grid structure enhanced feature data set; including:

[0035] S31. For the reflectivity factor, differential reflectivity, and differential phase shift rate in the three-dimensional spatial grid data set, perform joint modeling using a three-dimensional Gaussian mixture model;

[0036] S32. Iteratively update the weights, means, and covariance matrices of the three-dimensional Gaussian mixture model through the expectation-maximization algorithm to obtain a converged three-dimensional joint probability density function;

[0037] S33. Extract the weights, means, and covariance of the converged three-dimensional joint probability density function as the three-dimensional spatial grid structure enhanced feature data set.

[0038] Preferably, in step S31, the joint modeling using a three-dimensional Gaussian mixture model is expressed as:

[0039] ;

[0040] ;

[0041] In the formula, is a three-dimensional double Gaussian probability density function, is a single Gaussian three-dimensional probability density function containing three radar variables, represents the subscript of the first or second Gaussian function; is the mean of each dual-polarization variable, is the covariance between each dual-polarization variable.

[0042] Preferably, in step S32, the weights, means, and covariance matrix of the three-dimensional Gaussian mixture model are iteratively updated by the expectation-maximization algorithm to obtain a converged three-dimensional joint probability density function; it includes:

[0043] Initialize the weights, means, and covariance matrix of the three-dimensional Gaussian mixture model;

[0044] According to the current parameters, calculate the posterior probability that each data point belongs to each Gaussian component;

[0045] Update the weights, means, and covariance matrix according to the posterior probability;

[0046] If the updated parameters meet the convergence condition, output the three-dimensional joint probability density function;

[0047] If the convergence condition is not met, continue iterative update.

[0048] Preferably, in step S4, a random forest algorithm is used to establish the relationship between feature variables and target variables; it includes:

[0049] Using the grid reflectivity factor, differential reflectivity, differential phase shift rate, weights of the three-dimensional double Gaussian probability density function, three first Gaussian means, three second Gaussian means, three first Gaussian variances, three second Gaussian variances, covariance between reflectivity factor and differential reflectivity, covariance between differential reflectivity and differential phase shift rate, covariance between reflectivity factor and differential phase shift rate, a total of 19 variables, as the random forest training feature variables;

[0050] Using the precipitation observation data corresponding to the longitude and latitude of the target variable grid as the target variable;

[0051] Generate a new time window feature data pair every 6 minutes to perform real-time rolling update and training on the feature variables.

[0052] Beneficial effects: The present invention innovatively uses probability density function features to describe the rolling changes of dual-polarization radar variables within one hour. By constructing three-dimensional joint probability density function features, not only can the distribution characteristics of dual-polarization radar variables be extracted, but also the spatio-temporal evolution law of precipitation particles can be effectively captured. Specifically, the present invention obtains dual-polarization radar observation data, constructs a three-dimensional spatial grid data set and performs normalization processing, uses a rolling time window and a double Gaussian mixture model to fit parameters such as reflectivity factor, constructs a three-dimensional joint probability density function and extracts precipitation evolution features, and optimizes to obtain a set of characteristic variables using the expectation maximization algorithm. An artificial intelligence training set is constructed by combining precipitation observation data, and a relationship model between characteristic variables and precipitation is established using the random forest algorithm, finally realizing quantitative precipitation estimation and prediction. The present invention innovatively integrates technologies such as dual-polarization radar data processing, probability statistics modeling, and machine learning, effectively improving the accuracy and timeliness of precipitation estimation, and providing a new technical means for fields such as meteorological forecasting and hydrological monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a flowchart of the artificial intelligence dual-polarization radar quantitative precipitation estimation method with data feature enhancement according to the present invention;

[0054] Figure 2 is a comparison chart of the effects of radar quantitative precipitation estimation using the traditional method and the method of the present invention;

[0055] Figure 2 In (a) is the quantitative precipitation estimation result chart of the traditional method, Figure 2 In (b) is the quantitative precipitation estimation result chart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention, and they should not be construed as limiting the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. In the description of the present invention, it should be understood that the terms used are only for the purpose of description and cannot be construed as indicating or implying relative importance.

[0057] The following will be combined with Figure 1 - Figure 2 to describe the artificial intelligence dual-polarization radar quantitative precipitation estimation method with data feature enhancement provided by the present invention.

[0058] Embodiment 1: This embodiment provides an artificial intelligence dual-polarization radar quantitative precipitation estimation method with enhanced data features. It generally includes four parts. One is the acquisition and processing of dual-polarization radar observation data; the second is the method for constructing the probability density function of the dual-polarization radar; the third is the method for constructing the joint probability density function of three variables; and the fourth is the method for constructing a rolling-updated artificial intelligence quantitative precipitation estimation data set.

[0059] The specific steps are as Figure 1 shown, including:

[0060] Step 1: Method for acquiring and preprocessing dual-polarization radar data.

[0061] Specifically, it includes:

[0062] Extract the azimuth, elevation, and radial distance information from the dual-polarization radar observation data;

[0063] According to the azimuth, elevation, and radial distance information, map the dual-polarization radar observation data into a three-dimensional spatial grid structure;

[0064] Establish a data set of three variables, namely reflectivity factor, differential reflectivity, and differential phase shift rate, in each spatial grid to form a three-variable dual-polarization radar data set, denoted as:

[0065] ;

[0066] Perform standardization processing on each variable of the three-variable dual-polarization radar data set to eliminate the influence of different variable dimension differences and improve the Gaussian model fitting quality in the probability density function construction step, and obtain a three-dimensional spatial grid structure data set, denoted as:

[0067] ;

[0068] In the formula, represents the dual-polarization radar spatial grid data set, and the subscript indicates the variable in the data set; respectively represent the horizontal and vertical coordinates of the space, is the range of the horizontal and vertical coordinates; is the reflectivity factor, is the differential reflectivity, is the differential phase shift rate; represents the three-dimensional spatial grid structure data set after standardization processing, and are the mean and variance of each variable respectively.

[0069] Step 2: Method for window partitioning of dual-polarization radar observation data.

[0070] To extract the statistical parameters reflecting the precipitation evolution characteristics from radar observation data, the dual-polarization radar observation data updated at 6-minute intervals within each hour are divided into data windows with 10 consecutive time steps as a group (corresponding to 60 minutes of rolling data within 1 hour), and each window contains the radar observation values at 10 observation time steps.

[0071] Step 3: Establish a dual-Gaussian function model for a single variable.

[0072] Use the Gaussian Mixture Model (GMM) to perform fitting analysis on the dual-polarization radar variables within the rolling time window. Apply the GMM algorithm to each group of data to fit the dual-Gaussian distribution function, and calculate the weights, the first and second Gaussian means, and the first and second Gaussian variances of the dual-Gaussian function for each radar variable, a total of 5 variables, which are used to describe the probability distribution characteristics of the radar variable during this time period. During the modeling process, it is assumed that the radar data obeys a Gaussian mixture model composed of the linear weighting of two normal distributions, and its expression is as follows:

[0073] ;

[0074] In the formula, is the dual-Gaussian function, and are the first and second Gaussian functions, is the dual-Gaussian weight, and are the mean parameters of the first and second Gaussian functions, and are the variance parameters of the first and second Gaussian functions.

[0075] Step 4: Establish a three-dimensional dual-Gaussian function model for the joint of three variables.

[0076] Based on the data feature enhancement method of the three-dimensional joint probability density function.

[0077] Based on the dual-Gaussian probability density function, construct a three-dimensional joint probability density function to perform joint data feature enhancement on all variables in the three-dimensional space grid structure dataset, and obtain a three-dimensional space grid structure enhanced feature dataset.

[0078] In order to accurately describe various detailed characteristics of the cloud precipitation physical process in radar data, in this embodiment, based on the dual-Gaussian form probability density function, a three-dimensional joint probability density function is further constructed to enhance data features. Using the three-dimensional dual-Gaussian function model (3D Gaussian Mixture Model, 3D-GMM), perform joint modeling on three key dual-polarization parameters of the radar (reflectivity factor, differential reflectivity, differential phase shift rate), which is expressed as:

[0079] ;

[0080] ;

[0081] Wherein, is a three-dimensional double Gaussian probability density function, is a single Gaussian three-dimensional probability density function containing three radar variables, represents the subscript of the first or second Gaussian function; is the mean of each dual-polarization variable, is the covariance between each dual-polarization variable.

[0082] To solve the model parameters, the Expectation Maximization algorithm is adopted. The posterior probability of each data point belonging to each Gaussian component is calculated according to the current parameters, and the weights, means, and covariance matrices are updated according to these posterior probabilities. After multiple iterations until the parameters converge, a double Gaussian mixture probability density function that can fully describe the three-dimensional characteristics of radar observation data is finally obtained. Still taking every continuous 10 radar observation times as a sliding time window, a sample matrix containing three variables is constructed, where each sample matrix contains 10 observation time data, and each row of samples contains the values of 3 variables. The total number of extractable feature quantities is: 3 radar parameters, 1 dual Gaussian function weight variable, 6 mean variables, 6 variance variables, and 3 covariance variables, totaling 19 feature variables.

[0083] Compared with independently modeling each variable separately, this method can simultaneously consider the correlation and co-variation characteristics between variables, which helps to extract more accurate and rich statistical feature information. This method can provide richer physical features for distinguishing echo features of different cloud precipitation types in subsequent artificial intelligence-based quantitative precipitation estimation.

[0084] Step Five: A dual-polarization radar artificial intelligence quantitative precipitation estimation method with rolling update.

[0085] The 19 three-dimensional joint probability density function parameter grid data obtained are used as feature variables for artificial intelligence training (including: grid reflectivity factor, differential reflectivity, differential phase shift rate, three-dimensional double Gaussian probability density function weights, three first Gaussian means, three second Gaussian means, three first Gaussian variances, three second Gaussian variances, covariance between reflectivity factor and differential reflectivity, covariance between differential reflectivity and differential phase shift rate, covariance between reflectivity factor and differential phase shift rate), and the precipitation data corresponding to the grid are used as target variables. In order to include the evolution characteristics of precipitation cloud systems, new feature data for each hour window are generated every 6 minutes to achieve real-time rolling update. The random forest algorithm is used to establish the relationship between feature variables and target variables, and the relationship model between radar data and precipitation data is trained to finally realize quantitative precipitation estimation.

[0086] To verify the effect of the method of the present invention, the radar data and precipitation observation data from May to September 2022 are used as artificial intelligence training data in the present invention, and the differences in quantitative precipitation estimation effects between the method of the present invention and traditional methods are compared. The results are as Figure 2 shown. In the effect comparison, the radar data is used as the feature variable, and the precipitation observation data is used as the target variable. The random forest algorithm is used to train with 100 random forest trees. The traditional algorithm only uses 3 variables, namely the reflectivity factor, differential reflectivity and differential phase shift rate of the dual-polarization radar, as feature variables to estimate precipitation. The results show that there is a large underestimation in the traditional method for heavy precipitation, as Figure 2 shown in (a) below. It is relatively discrete in the scatter plot of the estimated value and the actual value. Using the data processing method of the present invention can significantly improve the accuracy of quantitative precipitation estimation, as Figure 2 shown in (b) below. The root mean square error of quantitative precipitation estimation in the traditional scheme is reduced from 5.9 mm to 3.2 mm, and the accuracy is improved by 46%. Especially for short-term heavy precipitation (hourly precipitation > 20 mm), the method of the present invention has a high estimation accuracy through the dual-polarization radar.

[0087] The verification results prove that the method of the present invention significantly improves the accuracy of quantitative precipitation estimation. The present invention aims at the problem of the lack of physical characteristics of precipitation cloud generation and dissipation in current quantitative precipitation estimation. A feature enhancement method for artificial intelligence training data of dual-polarization radar quantitative precipitation estimation based on three-dimensional joint double Gaussian probability density function is established. The main advantages are as follows:

[0088] Feature enhancement: The probability density function is introduced to process the radar data, enhancing the cloud precipitation physical characteristics of the data, effectively expanding the input parameters of the artificial intelligence model, and improving the model's perception ability of precipitation characteristics. For the first time, the probability density function characteristics of dual-polarization radar variables are used for quantitative precipitation estimation. Especially by constructing features through a rolling time window, it can better reflect the dynamic evolution process of radar variables.

[0089] Three-dimensional joint feature modeling: Construct a joint probability density function of three variables, which not only extracts the distribution characteristics of each variable, but also captures the spatio-temporal evolution law of precipitation particles, improving the physical interpretability and prediction ability of the model. The experimental verification results show that this method has a significant improvement in the accuracy of quantitative precipitation estimation, especially in complex severe convective weather conditions.

[0090] Embodiment 2: This embodiment provides a non-transitory computer-readable storage medium, on which computer instructions are stored, and the computer instructions cause the computer to execute the artificial intelligence dual-polarization radar quantitative precipitation estimation method with enhanced data features. The method includes the following steps:

[0091] S1. Obtain the dual-polarization radar observation data, preprocess the dual-polarization radar observation data, and construct a three-dimensional spatial grid structure data set;

[0092] S2. Construct a bi-Gaussian probability density function to enhance the data features of a single variable in the three-dimensional spatial grid structure data set;

[0093] S3. Based on the bi-Gaussian probability density function, construct a multi-dimensional joint probability density function to enhance the joint data features of all variables in the three-dimensional spatial grid structure data set, and obtain a three-dimensional spatial grid structure enhanced feature data set;

[0094] S4. Use the three-dimensional spatial grid structure enhanced feature data set as the feature variable and the precipitation data of the corresponding grid as the target variable to train an artificial intelligence model reflecting the relationship between radar data and precipitation data for artificial intelligence quantitative precipitation estimation.

[0095] Embodiment 3: This embodiment provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus. The processor can call the logical instructions in the memory to execute the artificial intelligence dual-polarization radar quantitative precipitation estimation method with enhanced data features. The method includes the following steps:

[0096] S1. Obtain the dual-polarization radar observation data, preprocess the dual-polarization radar observation data, and construct a three-dimensional spatial grid structure data set;

[0097] S2. Construct a bi-Gaussian probability density function to enhance the data features of a single variable in the three-dimensional spatial grid structure data set;

[0098] S3. Construct a multi-dimensional joint probability density function based on the double Gaussian probability density function, perform joint data feature enhancement on all variables in the three-dimensional spatial grid structure dataset, and obtain an enhanced feature dataset of the three-dimensional spatial grid structure;

[0099] S4. Use the enhanced feature dataset of the three-dimensional spatial grid structure as the feature variable and the precipitation data of the corresponding grid as the target variable to train an artificial intelligence model reflecting the relationship between radar data and precipitation data for artificial intelligence quantitative precipitation estimation.

[0100] In addition, when the logical instructions in the above-mentioned memory are implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical disks, etc., which can store program codes.

[0101] Embodiment 4: This embodiment provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the artificial intelligence dual-polarization radar quantitative precipitation estimation method for data feature enhancement, and this method includes the following steps:

[0102] S1. Obtain dual-polarization radar observation data, preprocess the dual-polarization radar observation data, and construct a three-dimensional spatial grid structure dataset;

[0103] S2. Construct a double Gaussian probability density function and perform data feature enhancement on a single variable in the three-dimensional spatial grid structure dataset;

[0104] S3. Construct a multi-dimensional joint probability density function based on the double Gaussian probability density function, perform joint data feature enhancement on all variables in the three-dimensional spatial grid structure dataset, and obtain an enhanced feature dataset of the three-dimensional spatial grid structure;

[0105] S4. Using the enhanced feature dataset of the three-dimensional spatial grid structure as the feature variable and the precipitation data of the corresponding grid as the target variable, train an artificial intelligence model that reflects the relationship between radar data and precipitation data for artificial intelligence quantitative precipitation estimation.

[0106] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0107] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An artificial intelligence dual-polarization radar quantitative precipitation estimation method with enhanced data features, characterized in that, Including: S1. Obtain dual-polarization radar observation data, preprocess the dual-polarization radar observation data, and construct a three-dimensional spatial grid structure data set, specifically including: S11. Extract azimuth, elevation, and radial distance information from the dual-polarization radar observation data; S12. Map the dual-polarization radar observation data into a three-dimensional spatial grid structure according to the azimuth, elevation, and radial distance information; S13. Establish data sets of three variables, namely reflectivity factor, differential reflectivity, and differential phase shift rate, in each spatial grid to form a three-variable dual-polarization radar data set; S14. Standardize each variable of the three-variable dual-polarization radar data set to obtain the three-dimensional spatial grid structure data set; S2. Construct a double Gaussian probability density function to enhance the data characteristics of a single variable in the three-dimensional spatial grid structure data set; specifically including: S21. For the three-dimensional spatial grid structure data set, select rolling updated observation data within a continuous time window; S22. Fit a single variable of the rolling updated observation data using a Gaussian mixture model to obtain a double Gaussian distribution function; S23. According to the double Gaussian distribution function, calculate the weight, the first Gaussian mean and the second Gaussian mean, the first Gaussian variance and the second Gaussian variance, which are used to describe the probability distribution characteristics of the radar variable within this continuous time window, and generate enhanced feature data of a single variable; S3. Based on the double Gaussian probability density function, construct a multi-dimensional joint probability density function to jointly enhance the data characteristics of all variables in the three-dimensional spatial grid structure data set, and obtain a three-dimensional spatial grid structure enhanced feature data set; including: S31. For the reflectivity factor, differential reflectivity, and differential phase shift rate in the three-dimensional spatial grid structure data set, perform joint modeling using a three-dimensional Gaussian mixture model, expressed as: In the formula, is a three-dimensional double Gaussian probability density function, is a single Gaussian three-dimensional probability density function containing three radar variables, and k represents the subscript of the first or second Gaussian function; μ k is the mean of each dual-polarization variable, and Σ k is the covariance between each dual-polarization variable; S32. Iteratively update the weight, mean, and covariance matrix of the three-dimensional Gaussian mixture model through the expectation-maximization algorithm to obtain a converged three-dimensional joint probability density function; S33. Extract the weight, mean, and covariance of the converged three-dimensional joint probability density function as the three-dimensional spatial grid structure enhanced feature data set; S4. Use the three-dimensional spatial grid structure enhanced feature data set as a feature variable and the precipitation data of the corresponding grid as a target variable to train an artificial intelligence model that reflects the relationship between radar data and precipitation data for artificial intelligence quantitative precipitation estimation.

2. The method for quantitatively estimating precipitation of an artificial intelligence dual-polarization radar with enhanced data features according to claim 1, wherein In step S13, data sets of three variables, namely reflectivity factor, differential reflectivity, and differential phase shift rate, are established in each spatial grid to form a three-variable dual-polarization radar data set, expressed as: Standardize each variable of the three-variable dual-polarization radar data set to obtain the three-dimensional spatial grid structure data set, expressed as: where X j represents the dual-polarization radar spatial grid dataset, and the subscript j represents the variable in the dataset; x and y represent the spatial horizontal and vertical coordinates respectively, and N is the range of the horizontal and vertical coordinates; Z H is the reflectivity factor, Z DR is the differential reflectivity, K DP is the differential phase shift rate; represents the three-dimensional spatial grid structure dataset after standardization processing, μ j and σ j are the mean and variance of each variable respectively.

3. The method for quantitatively estimating precipitation of an artificial intelligence dual-polarization radar with enhanced data features according to claim 1, wherein In the process of fitting a single variable of the rolling update observation data with a Gaussian mixture model, it is assumed that the radar data obeys a Gaussian mixture model composed of the linear weighting of two normal distributions. The expression of the double Gaussian distribution function is as follows: G(x) = a·G1(μ1,σ1)+(1 - a)·G2(μ2,σ2); Wherein, G(x) is a double Gaussian distribution function, G1 and G2 are the first Gaussian function and the second Gaussian function, a is the double Gaussian weight, μ1 and μ2 are the mean parameters of the first Gaussian function and the second Gaussian function, and σ1 and σ2 are the variance parameters of the first Gaussian function and the second Gaussian function.

4. The method for quantitatively estimating precipitation of an artificial intelligence dual-polarization radar with enhanced data characteristics according to claim 1, wherein In step S21, for the three-dimensional spatial grid structure data set, select the rolling update observation data within a continuous time window; specifically including: Obtain the dual-polarization radar observation data at a preset time interval; Divide the dual-polarization radar observation data into multiple continuous time windows, where each time window contains multiple continuous dual-polarization radar 6-minute interval volume scan data; For each time window, extract the reflectivity factor, differential reflectivity, and differential phase shift rate in the three-dimensional spatial grid structure data set; Generate the rolling update observation data for the corresponding time window according to the reflectivity factor, differential reflectivity, and differential phase shift rate.

5. The method for quantitatively estimating precipitation of an artificial intelligence dual-polarization radar with enhanced data characteristics according to claim 1, characterized in that, In step S32, the weights, means, and covariance matrices of the three-dimensional Gaussian mixture model are iteratively updated through the expectation-maximization algorithm to obtain a converged three-dimensional joint probability density function; including: Initialize the weights, means, and covariance matrices of the three-dimensional Gaussian mixture model; Calculate the posterior probability that each data point belongs to each Gaussian component according to the current parameters; Update the weights, means, and covariance matrices according to the posterior probability; If the updated parameters meet the convergence condition, output the three-dimensional joint probability density function; If the convergence condition is not met, continue iterative update.

6. The method for quantitatively estimating precipitation of an artificial intelligence dual-polarization radar with enhanced data characteristics according to claim 1, wherein In step S4, use the random forest algorithm to establish the relationship between feature variables and target variables; including: Use a total of 19 variables, namely grid reflectivity factor, differential reflectivity, differential phase shift rate, weights of the three-dimensional double Gaussian probability density function, three first Gaussian means, three second Gaussian means, three first Gaussian variances, three second Gaussian variances, covariance between reflectivity factor and differential reflectivity, covariance between differential reflectivity and differential phase shift rate, and covariance between reflectivity factor and differential phase shift rate, as the random forest training feature variables; Use the precipitation observation data corresponding to the longitude and latitude of the target variable grid as the target variable; Generate a new time window feature data pair every 6 minutes to perform real-time rolling update and training on the feature variables.

Images