A millimeter wave cloud detector data quality control method based on an XGBoost model

Abstract

This invention discloses a data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model, belonging to the field of millimeter-wave cloud measuring instrument data quality control technology. The method includes the following steps: S1, cloud measuring instrument data input; S2, multi-dimensional feature extraction: multi-dimensional feature extraction is performed on the data input in step S1 by calculating the time-domain coefficient of variation, extracting multivariate texture features, and considering echo energy distribution skewness; S3, preliminary quality control processing; S4, secondary verification based on power spectrum data; S5, outputting the filtered result. This invention extracts specific features for two types of interference echoes using multi-dimensional features—time-domain coefficient of variation, texture features, and energy distribution skewness—and utilizes the model to adaptively learn the interference patterns under different scenarios, achieving simultaneous identification and filtering of the two types of interference. Simultaneously, power spectrum recognition is introduced to further improve the filtering accuracy. This method eliminates the need for separate processing logic, simplifying the process while improving filtering accuracy and scenario adaptability.

Description

Technical Field

[0001] This invention relates to the field of data quality control technology for millimeter-wave cloud measuring instruments, and more specifically to the field of a data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model. Background Technology

[0002] Millimeter-wave cloud measuring instruments, as core equipment for observing the vertical structure of atmospheric cloud systems, possess high spatiotemporal resolution, strong penetration, and all-weather operation capabilities. They have been deployed on a large scale in meteorological observation, aviation support, and climate research. By transmitting millimeter-wave pulse signals and receiving the scattered echoes from cloud particles, they can accurately invert key parameters such as cloud top height, cloud thickness, and particle phase state, providing core data support for numerical weather prediction optimization, severe weather warning, and atmospheric radiation balance research. They are an indispensable key piece of equipment in the modern meteorological observation system.

[0003] However, in actual observations, the echo data from millimeter-wave cloud measuring instruments are severely affected by two types of typical interference signals, becoming the core challenge restricting data quality: First, clear-sky echoes, also known as clear-sky noise, are caused by atmospheric aerosol particles, water vapor condensates, or turbulent motion. The signal strength is weak but widely distributed, with poor spatiotemporal continuity and discrete energy. Moreover, it fluctuates irregularly with temperature, humidity, and aerosol concentration, making it easy to confuse with thin cloud echoes. Second, sidelobe echoes originate from the scattered signals from non-target areas such as the ground and buildings received by the antenna sidelobes. They are periodically distributed along the range library and have obvious abrupt changes in intensity. The interference is particularly prominent in complex terrain or near-ground observation scenarios, seriously affecting the accuracy of cloud parameter inversion.

[0004] Currently, the industry mostly adopts a separate processing approach for filtering out the two types of interference echoes, which has significant limitations: existing methods can only filter out one type of interference, either clear-sky echoes or sidelobe echoes, and cannot achieve simultaneous processing of both types of interference. This results in data quality control having to be performed step by step, which is cumbersome and inefficient. At the same time, methods such as thresholding and spatiotemporal continuity statistics for clear-sky echoes have low recognition rates for weak interference and are prone to misfiltering thin clouds. Fixed rule methods for sidelobe echoes rely on antenna parameters, resulting in poor versatility and difficulty in dealing with complex terrain interference. Furthermore, fixed thresholds and simple rules cannot adapt to interference changes of different sites, seasons, and terrains, and lack robustness, failing to meet the real-time quality control requirements of large-scale network observations. Summary of the Invention

[0005] The purpose of this invention is to provide a data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model in order to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention specifically adopts the following technical solution:

[0007] This invention provides a data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model, comprising the following steps:

[0008] S1, Input of cloud measuring instrument data;

[0009] S2. Multi-dimensional feature extraction: Multi-dimensional feature extraction is performed on the data input in step S1 by calculating the temporal coefficient of variation, extracting multivariable texture features, and measuring the echo energy distribution skewness.

[0010] S3. Preliminary quality control process:

[0011] S31. Construct a labeled dataset: Select cloud instrument data from different observation stations, different seasons, and different time periods, and carry out manual labeling processing; the labeled dataset covers typical weather scenarios, and the labeled data includes cloud echo, clear sky echo, sidelobe echo, core echo, and undetermined echo type;

[0012] S32. Training based on the XGBoost model: The features extracted in step S2 are processed in step S31 and then fed into the XGBoost model for training.

[0013] S33. Preliminary echo classification based on XGBoost model: The XGBoost model will preliminarily identify the base data and classify the base data into cloud echo, clear sky echo, sidelobe echo and undetermined echo.

[0014] S4. Secondary verification based on power spectrum data: Based on the prediction results of the XGBoost model, identification is performed based on power spectrum data. The purpose is to make a final classification of the undetermined echoes identified by the XGBoost model. The identification methods include the main lobe energy ratio method and the main lobe peak sharpness method.

[0015] S5. Output filtered result: According to the judgment criteria, the undetermined echo identified by the XGBoost model is determined to be one of cloud echo, clear sky echo or sidelobe echo.

[0016] In one implementation, the specific method for calculating the time-domain variation coefficient in step S2 is as follows:

[0017] Interference echoes from cloud measuring instruments, such as sidelobe echoes and clear-sky echoes, have a distinct characteristic: they are less continuous and uniform than cloud echoes. To extract this characteristic of interference echoes, we propose the time-domain variation coefficient as a feature quantity.

[0018] In step S2, the specific method for calculating the time-domain variation coefficient is as follows:

[0019] Data for which time-domain coefficient of variation characteristic quantities need to be calculated include reflectance. and linear depolarization ratio ;

[0020] For reflectivity In other words:

[0021] Take a certain distance library continuous The reflectance data at each time point, and the echo data sequence are as follows: ;

[0022] Calculate the median in the sequence :

[0023] ;

[0024] Calculate the absolute deviation of the median after calibration :

[0025] ;

[0026] In the formula, The calibration coefficient is adjusted in conjunction with the cloud measuring instrument's transmission power;

[0027] Calculate the time-domain coefficient of variation :

[0028] ;

[0029] In the formula, To avoid the denominator being 0;

[0030] For linear depolarization ratio In other words:

[0031] Take a certain distance library continuous The linear depolarization ratio data at each time point, and the echo data sequence are as follows: ;

[0032] Calculate the median in the sequence :

[0033] ;

[0034] Calculate the absolute deviation of the median after calibration :

[0035] ;

[0036] In the formula, The calibration coefficient is adjusted in conjunction with the cloud measuring instrument's transmission power;

[0037] Calculate the time-domain coefficient of variation :

[0038] ;

[0039] In the formula, To avoid the denominator being 0.

[0040] In one implementation, in step S2, a multivariate texture feature extraction method is designed to extract the discontinuous and non-uniform features of the cloud meter interference echo. The specific method is as follows:

[0041] Define a Local window, for cloud reflectance ,speed Linear depolarization ratio Perform the following calculations;

[0042] For cloud reflectance In other words:

[0043] The formulas for calculating the mean and variance are as follows:

[0044] , ;

[0045] entropy The calculation formula is as follows:

[0046] , ;

[0047] In the formula, The first one in the display window Class value appearance Second-rate;

[0048] gradient magnitude The calculation formula is as follows:

[0049] ;

[0050] In the formula, This is the center value of the local window;

[0051] For cloud measuring instrument speed In other words:

[0052] The formulas for calculating the mean and variance are as follows:

[0053] , ;

[0054] entropy The calculation formula is as follows:

[0055] , ;

[0056] In the formula, The first one in the display window Class value appearance Second-rate;

[0057] gradient magnitude The calculation formula is as follows:

[0058] ;

[0059] In the formula, This is the center value of the local window;

[0060] For the linear depolarization ratio of cloud measuring instruments In other words:

[0061] The formulas for calculating the mean and variance are as follows:

[0062] , ;

[0063] entropy The calculation formula is as follows:

[0064] , ;

[0065] In the formula, The first one in the display window Class value appearance Second-rate;

[0066] gradient magnitude The calculation formula is as follows:

[0067] ;

[0068] In the formula, This is the center value of the local window.

[0069] In one implementation, in step S2, the core difference between the cloud meter interference echo and the cloud echo lies not only in their spatiotemporal continuity / uniformity, but also in the morphological characteristics of their energy distribution. The energy distribution of the cloud echo is closer to a symmetrical normal distribution, while the interference echo, lacking stable scattering support, exhibits significant asymmetry in its energy distribution. A method for calculating the skewness of the echo energy distribution based on Poisson distribution fitting is proposed.

[0070] Inverse data transformation: Since the Poisson distribution describes discrete integer random variables, while the reflectance in the base data... Logarithmic reflectance is continuous non-linear data, so it needs to be converted into linear data first.

[0071] ;

[0072] Poisson distribution parameter estimation: The calculation involves inputting data from an entire profile, denoted as M. Then:

[0073] ;

[0074] Echo energy distribution skewness calculate:

[0075] ;

[0076] In the formula, , .

[0077] In one implementation, the specific method for constructing the label dataset in step S31 is as follows:

[0078] Clear-sky echo annotation can be performed using a 2-8 hour historical data backtracking method, combined with the characteristics of echo intensity fluctuating with aerosol concentration.

[0079] Side lobe echoes are labeled by starting with the pulse splicing position and combining the characteristics of the side lobe echoes, such as the periodic intensity abrupt change along the distance library;

[0080] Cloud echoes are labeled as 0, clear sky echoes as 1, side lobe echoes as 2, and undetermined echoes as 3. Undetermined echoes are echoes whose type is difficult to determine during manual annotation. To ensure model accuracy, undetermined echoes are not included in subsequent model training.

[0081] In one implementation, the specific method for training the XGBoost model in step S32 is as follows:

[0082] Time-domain variation coefficient Multivariate texture features include mean ,variance ,entropy Gradient magnitude echo energy distribution skewness , Put it into the XGBoost model for training;

[0083] During XGBoost model training, although undetermined echoes were removed, uneven distribution of cloud echoes, clear sky echoes, and sidelobe echoes still existed; cloud echoes accounted for 60%, clear sky echoes for 30%, and sidelobe echoes for 10%. Therefore, class balancing weights were introduced into the XGBoost model's native cross-entropy loss. :

[0084] ;

[0085] In the formula, ,in, The total number of samples, For the first Number of class samples; since the sidelobe echo sample size is the smallest, the echo weights are assigned to the sidelobes. echo weights of clear-sky echoes Cloud echoes have the largest sample size, so they are assigned echo weights. ;

[0086] This is the regularization term for the tree, responsible for balancing loss optimization and model generalization. The number of leaf nodes. and Parameters used to control complexity;

[0087] For the sample The true label, if the sample Belongs to the echo, then ,otherwise ;

[0088] Predict samples for the model Belongs to the The probability of an echo-like signal, taking values ​​in the range (0,1) and satisfying the following conditions: .

[0089] In one implementation, step S33 involves inferring a preliminary echo classification based on the XGBoost model, specifically as follows:

[0090] Based on the trained XGBoost model, preliminary quality control processing is performed on the cloud measuring instrument base data. The XGBoost model will perform preliminary identification of the base data and classify the base data into cloud echo, clear sky echo, sidelobe echo and undetermined echo.

[0091] In one implementation, in step S4, the main lobe energy ratio refers to the ratio of the energy in the main lobe region to the total energy in the power spectrum data, directly quantifying the concentration of echo signal energy. The main lobe energy ratio is identified as follows:

[0092] First, determine the peak value of the main lobe. ,by Taking the corresponding spectral velocity as the center, extend 10% to both sides, i.e. This interval is the effective main lobe interval, denoted as . ;

[0093] Integrating and summing the power spectral density values ​​over the effective main lobe region: ,in For spectral data The corresponding power spectral value at that location;

[0094] Integral summation of the power spectral values ​​over the full spectrum: ;

[0095] The final comparison yields the main lobe energy percentage: ,in This is used to avoid the denominator being 0.

[0096] In one implementation, in step S4, the main lobe peak sharpness refers to the ratio of the main lobe peak value to the average total energy in the power spectrum, quantifying the prominence of the main lobe peak. The specific details of the main lobe peak sharpness identification method are as follows:

[0097] The total average energy is calculated as follows: ,in Number of all spectral points;

[0098] The main lobe peak sharpness is calculated as follows: ,in This is used to avoid the denominator being 0.

[0099] In one implementation, step S5 is as follows:

[0100] For the undetermined echoes identified by the model:

[0101] like and If so, it is determined to be a cloud echo;

[0102] like and This is then determined to be a clear-sky echo;

[0103] like and It is then determined to be a sidelobe echo.

[0104] The beneficial effects of this invention are as follows:

[0105] This invention proposes a data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model. It extracts specific features for two types of interference echoes by using multi-dimensional features such as time-domain variation coefficient, texture features, and energy distribution skewness. The method then utilizes the model to adaptively learn interference patterns under different scenarios, achieving simultaneous identification and filtering of the two types of interference. Furthermore, power spectrum recognition is introduced to further improve filtering accuracy. This method eliminates the need for separate processing logic, simplifying the process while enhancing filtering accuracy and scenario adaptability, effectively overcoming the shortcomings of existing technologies. Attached Figure Description

[0106] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0107] Figure 1 This is a flowchart of a data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model, according to the present invention.

[0108] Figure 2 This is a diagram of the label.

[0109] Figure 3 This is the initial identification result of the model.

[0110] Figure 4 It is a complete radial power spectrum.

[0111] Figure 5 This is the cloud echo power spectrum.

[0112] Figure 6 This is the power spectrum of the sidelobe echo.

[0113] Figure 7 This is a power spectrum diagram of clear-sky echoes.

[0114] Figure 8 This is one of the final results shown.

[0115] Figure 9 This is the second part of the final effect display. Detailed Implementation

[0116] To make the technical problems, technical solutions, and technical effects of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0117] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0118] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0119] In the description of the embodiments of the present invention, it should be noted that the terms "inner", "outer", "upper", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the present invention.

[0120] Example 1

[0121] This invention provides a data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model, comprising the following steps:

[0122] S1, Input of cloud measuring instrument data;

[0123] S2. Multi-dimensional feature extraction: Multi-dimensional feature extraction is performed on the data input in step S1 by calculating the temporal coefficient of variation, extracting multivariable texture features, and measuring the echo energy distribution skewness.

[0124] The specific method for calculating the time-domain coefficient of variation is as follows:

[0125] Interference echoes from cloud measuring instruments, such as sidelobe echoes and clear-sky echoes, have a distinct characteristic: they are less continuous and uniform than cloud echoes. To extract this characteristic of interference echoes, we propose the time-domain variation coefficient as a feature quantity.

[0126] The specific method for calculating the time-domain coefficient of variation is as follows:

[0127] Data for which time-domain coefficient of variation characteristic quantities need to be calculated include reflectance. and linear depolarization ratio ;

[0128] For reflectivity In other words:

[0129] Take a certain distance library continuous The reflectance data at each time point, and the echo data sequence are as follows: ;

[0130] Calculate the median in the sequence :

[0131] ;

[0132] Calculate the absolute deviation of the median after calibration :

[0133] ;

[0134] In the formula, The calibration coefficient is adjusted in conjunction with the cloud measuring instrument's transmission power;

[0135] Calculate the time-domain coefficient of variation :

[0136] ;

[0137] In the formula, To avoid the denominator being 0;

[0138] For linear depolarization ratio In other words:

[0139] Take a certain distance library continuous The linear depolarization ratio data at each time point, and the echo data sequence are as follows: ;

[0140] Calculate the median in the sequence :

[0141] ;

[0142] Calculate the absolute deviation of the median after calibration :

[0143] ;

[0144] In the formula, The calibration coefficient is adjusted in conjunction with the cloud measuring instrument's transmission power;

[0145] Calculate the time-domain coefficient of variation :

[0146] ;

[0147] In the formula, To avoid the denominator being 0.

[0148] To extract the discontinuous and non-uniform features of interference echoes from cloud measuring instruments, a multivariate texture feature extraction method is designed, as follows:

[0149] Define a Local window, for cloud reflectance ,speed Linear depolarization ratio Perform the following calculations;

[0150] For cloud reflectance In other words:

[0151] The formulas for calculating the mean and variance are as follows:

[0152] , ;

[0153] entropy The calculation formula is as follows:

[0154] , ;

[0155] In the formula, The first one in the display window Class value appearance Second-rate;

[0156] gradient magnitude The calculation formula is as follows:

[0157] ;

[0158] In the formula, This is the center value of the local window;

[0159] For cloud measuring instrument speed In other words:

[0160] The formulas for calculating the mean and variance are as follows:

[0161] , ;

[0162] entropy The calculation formula is as follows:

[0163] , ;

[0164] In the formula, The first one in the display window Class value appearance Second-rate;

[0165] gradient magnitude The calculation formula is as follows:

[0166] ;

[0167] In the formula, This is the center value of the local window;

[0168] For the linear depolarization ratio of cloud measuring instruments In other words:

[0169] The formulas for calculating the mean and variance are as follows:

[0170] , ;

[0171] entropy The calculation formula is as follows:

[0172] , ;

[0173] In the formula, The first one in the display window Class value appearance Second-rate;

[0174] gradient magnitude The calculation formula is as follows:

[0175] ;

[0176] In the formula, This is the center value of the local window.

[0177] The core difference between cloud echo interference and cloud echo lies not only in their spatiotemporal continuity / uniformity but also in the morphological characteristics of their energy distribution. Cloud echo energy distribution is closer to a symmetrical normal distribution, while interference echo energy distribution exhibits significant asymmetry due to the lack of a stable scattering body. A method for calculating echo energy distribution skewness based on Poisson distribution fitting is proposed.

[0178] Inverse data transformation: Since the Poisson distribution describes discrete integer random variables, while the reflectance in the base data... Logarithmic reflectance is continuous non-linear data, so it needs to be converted into linear data first.

[0179] ;

[0180] Poisson distribution parameter estimation: The calculation involves inputting data from an entire profile, denoted as M. Then:

[0181] ;

[0182] Echo energy distribution skewness calculate:

[0183] ;

[0184] In the formula, , .

[0185] S3. Preliminary quality control process:

[0186] S31. Constructing a Labeled Dataset: Select cloud meter data from different observation stations, seasons, and time periods, and perform manual annotation processing; the labeled dataset covers typical weather scenarios, and the labeled data includes cloud echoes, clear sky echoes, sidelobe echoes, core echoes, and undetermined echo types; the specific method for constructing the labeled dataset is as follows:

[0187] Clear-sky echo annotation can be performed using a 2-8 hour historical data backtracking method, combined with the characteristics of echo intensity fluctuating with aerosol concentration.

[0188] Side lobe echoes are labeled by starting with the pulse splicing position and combining the characteristics of the side lobe echoes, such as the periodic intensity abrupt change along the distance library;

[0189] Cloud echoes are labeled as 0, clear sky echoes as 1, side lobe echoes as 2, and undetermined echoes as 3. Undetermined echoes are echoes whose type is difficult to determine during manual annotation. To ensure model accuracy, undetermined echoes are not included in subsequent model training.

[0190] S32. Training based on the XGBoost model: The features extracted in step S2 are processed in step S31 and then fed into the XGBoost model for training. The specific method of training based on the XGBoost model is as follows:

[0191] Time-domain variation coefficient Multivariate texture features include mean ,variance ,entropy Gradient magnitude echo energy distribution skewness , Put it into the XGBoost model for training;

[0192] During XGBoost model training, although undetermined echoes were removed, uneven distribution of cloud echoes, clear sky echoes, and sidelobe echoes still existed; cloud echoes accounted for 60%, clear sky echoes for 30%, and sidelobe echoes for 10%. Therefore, class balancing weights were introduced into the XGBoost model's native cross-entropy loss. :

[0193] ;

[0194] In the formula, ,in, The total number of samples, For the first Number of class samples; since the sidelobe echo sample size is the smallest, the echo weights are assigned to the sidelobes. echo weights of clear-sky echoes Cloud echoes have the largest sample size, so they are assigned echo weights. ;

[0195] This is the regularization term for the tree, responsible for balancing loss optimization and model generalization. The number of leaf nodes. and Parameters used to control complexity;

[0196] For the sample The true label, if the sample Belongs to the echo, then ,otherwise ;

[0197] Predict samples for the model Belongs to the The probability of an echo-like signal, taking values ​​in the range (0,1) and satisfying the following conditions: .

[0198] S33. Preliminary Echo Classification Based on XGBoost Model: The XGBoost model performs preliminary identification of the base data, classifying it into cloud echoes, clear-sky echoes, sidelobe echoes, and undetermined echoes. The preliminary echo classification is inferred based on the XGBoost model, as follows:

[0199] Based on the trained XGBoost model, preliminary quality control processing is performed on the cloud measuring instrument base data. The XGBoost model will perform preliminary identification of the base data and classify the base data into cloud echo, clear sky echo, sidelobe echo and undetermined echo.

[0200] S4. Secondary verification based on power spectrum data: Based on the prediction results of the XGBoost model, identification is performed based on power spectrum data. The purpose is to make a final classification of the undetermined echoes identified by the XGBoost model. The identification methods include the main lobe energy ratio method and the main lobe peak sharpness method.

[0201] The main lobe energy ratio refers to the ratio of the energy in the main lobe region to the total energy in the power spectrum data. It directly quantifies the concentration of energy in the echo signal. The main lobe energy ratio is identified as follows:

[0202] First, determine the peak value of the main lobe. ,by Taking the corresponding spectral velocity as the center, extend 10% to both sides, i.e. This interval is the effective main lobe interval, denoted as . ;

[0203] Integrating and summing the power spectral density values ​​over the effective main lobe region: ,in For spectral data The corresponding power spectral value at that location;

[0204] Integral summation of the power spectral values ​​over the full spectrum: ;

[0205] The final comparison yields the main lobe energy percentage: ,in This is used to avoid the denominator being 0.

[0206] The main lobe peak sharpness refers to the ratio of the main lobe peak value to the average total energy in the power spectrum. It quantifies the prominence of the main lobe peak. The specific details of the main lobe peak sharpness identification method are as follows:

[0207] The total average energy is calculated as follows: ,in Number of all spectral points;

[0208] The main lobe peak sharpness is calculated as follows: ,in This is used to avoid the denominator being 0.

[0209] S5. Output filtered result: For the undetermined echo identified by the XGBoost model:

[0210] like and If so, it is determined to be a cloud echo;

[0211] like and This is then determined to be a clear-sky echo;

[0212] like and It is then determined to be a sidelobe echo.

Claims

1. A data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model, characterized in that, Includes the following steps: S1, Input of cloud measuring instrument data; S2. Multi-dimensional feature extraction: Multi-dimensional feature extraction is performed on the data input in step S1 by calculating the temporal coefficient of variation, extracting multivariable texture features, and measuring the echo energy distribution skewness. S3. Preliminary quality control process: S31. Construct a labeled dataset: Select cloud instrument data from different observation stations, different seasons, and different time periods, and carry out manual labeling processing; the labeled dataset covers typical weather scenarios, and the labeled data includes cloud echo, clear sky echo, sidelobe echo, core echo, and undetermined echo type; S32. Training based on the XGBoost model: The features extracted in step S2 are processed in step S31 and then fed into the XGBoost model for training. S33. Preliminary echo classification based on XGBoost model: The XGBoost model will preliminarily identify the base data and classify the base data into cloud echo, clear sky echo, sidelobe echo and undetermined echo. S4. Secondary verification based on power spectrum data: Based on the prediction results of the XGBoost model, identification is performed based on power spectrum data. The purpose is to make a final classification of the undetermined echoes identified by the XGBoost model. The identification methods include the main lobe energy ratio method and the main lobe peak sharpness method. S5. Output filtered result: According to the judgment criteria, the undetermined echo identified by the XGBoost model is determined to be one of cloud echo, clear sky echo or sidelobe echo.

2. The data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model according to claim 1, characterized in that, In step S2, the specific method for calculating the time-domain variation coefficient is as follows: Data for which time-domain coefficient of variation characteristic quantities need to be calculated include reflectance. and linear depolarization ratio ; For reflectivity In other words: Take a certain distance library continuous The reflectance data at each time point, and the echo data sequence are as follows: ; Calculate the median in the sequence : ； Calculate the absolute deviation of the median after calibration : ； In the formula, The calibration coefficient is adjusted in conjunction with the cloud measuring instrument's transmission power; Calculate the time-domain coefficient of variation : ； In the formula, To avoid the denominator being 0; For linear depolarization ratio In other words: Take a certain distance library continuous The linear depolarization ratio data at each time point, and the echo data sequence are as follows: ; Calculate the median in the sequence : ； Calculate the absolute deviation of the median after calibration : ； In the formula, The calibration coefficient is adjusted in conjunction with the cloud measuring instrument's transmission power; Calculate the time-domain coefficient of variation : ； In the formula, To avoid the denominator being 0.

3. The data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model according to claim 2, characterized in that, In step S2, to extract the discontinuous and non-uniform features of the cloud meter interference echo, a multivariate texture feature extraction method is designed, as follows: Define a Local window, for cloud reflectance ,speed Linear depolarization ratio Perform the following calculations; For cloud reflectance In other words: The formulas for calculating the mean and variance are as follows: ， ； entropy The calculation formula is as follows: ， ； In the formula, The first one in the display window Class value appearance Second-rate; gradient magnitude The calculation formula is as follows: ； In the formula, This is the center value of the local window; For cloud measuring instrument speed In other words: The formulas for calculating the mean and variance are as follows: ， ； entropy The calculation formula is as follows: ， ； In the formula, The first one in the display window Class value appearance Second-rate; gradient magnitude The calculation formula is as follows: ； In the formula, This is the center value of the local window; For the linear depolarization ratio of cloud measuring instruments In other words: The formulas for calculating the mean and variance are as follows: ， ； entropy The calculation formula is as follows: ， ； In the formula, The first one in the display window Class value appearance Second-rate; gradient magnitude The calculation formula is as follows: ； In the formula, This is the center value of the local window.

4. The data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model according to claim 3, characterized in that, In step S2, the core difference between the cloud meter interference echo and the cloud echo lies not only in their spatiotemporal continuity / uniformity but also in the morphological characteristics of their energy distribution. The energy distribution of the cloud echo is closer to a symmetrical normal distribution, while the interference echo, lacking stable scattering support, exhibits a significant asymmetry in its energy distribution. A method based on Poisson distribution fitting to calculate the skewness of the echo energy distribution is proposed. Inverse data transformation: Since the Poisson distribution describes discrete integer random variables, while the reflectance in the base data... Logarithmic reflectance is continuous non-linear data, so it needs to be converted into linear data first. ； Poisson distribution parameter estimation: The calculation involves inputting data from an entire profile, denoted as M. Then: ； echo energy distribution skewness calculate: ； In the formula, , .

5. The data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model according to claim 4, characterized in that, In step S31, the specific method for constructing the label dataset is as follows: Clear-sky echo annotation uses a 2-8 hour historical data backtracking method, combined with the characteristic of echo intensity fluctuating with aerosol concentration; Side lobe echoes are labeled by starting with the pulse splicing position and combining the periodic intensity abrupt change characteristics of the side lobe echoes along the distance library; Cloud echoes are marked as 0, clear sky echoes as 1, sidelobe echoes as 2, and undetermined echoes as 3. Undetermined echoes are echoes whose type is difficult to determine during manual annotation. To ensure model accuracy, undetermined echoes are not included in subsequent model training.

6. The data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model according to claim 5, characterized in that, In step S32, the specific method for training the XGBoost model is as follows: Time-domain variation coefficient Multivariate texture features include mean ,variance ,entropy Gradient magnitude echo energy distribution skewness , Put it into the XGBoost model for training; During XGBoost model training, although undetermined echoes were removed, uneven distribution of cloud echoes, clear sky echoes, and sidelobe echoes still existed; therefore, class balancing weights were introduced into the native cross-entropy loss of the XGBoost model. : ； In the formula, ,in, The total number of samples, For the first Number of samples in each class; This is the regularization term for the tree, responsible for balancing loss optimization and model generalization. The number of leaf nodes. and Parameters used to control complexity; For the sample The true label, if the sample Belongs to the echo, then ,otherwise ; Predict samples for the model Belongs to the The probability of an echo-like signal, taking values ​​in the range (0,1) and satisfying the following conditions: .

7. The data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model according to claim 6, characterized in that, In step S33, the initial classification of the echo is inferred based on the XGBoost model, as follows: Based on the trained XGBoost model, preliminary quality control processing is performed on the cloud measuring instrument base data. The XGBoost model will perform preliminary identification of the base data and classify the base data into cloud echo, clear sky echo, sidelobe echo and undetermined echo.

8. The data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model according to claim 7, characterized in that, In step S4, the main lobe energy ratio refers to the ratio of the energy in the main lobe region to the total energy in the power spectrum data. It directly quantifies the concentration of the echo signal energy. The main lobe energy ratio is identified as follows: First, determine the peak value of the main lobe. ,by Taking the corresponding spectral velocity as the center, extend 10% to both sides, i.e. This interval is the effective main lobe interval, denoted as . ; Integrating and summing the power spectral density values ​​over the effective main lobe region: ,in For spectral data The corresponding power spectral value at that location; Integral summation of the power spectral values ​​over the full spectrum: ; The final comparison yields the main lobe energy percentage: ,in This is used to avoid the denominator being 0.

9. A data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model according to claim 8, characterized in that, In step S4, the main lobe peak sharpness refers to the ratio of the main lobe peak value to the average total energy in the power spectrum, quantifying the prominence of the main lobe peak. The specific details of the main lobe peak sharpness identification method are as follows: The total average energy is calculated as follows: ,in Number of all spectral points; The main lobe peak sharpness is calculated as follows: ,in This is used to avoid the denominator being 0.

10. A data quality control method for millimeter-wave cloud measuring instruments based on the XGBoost model according to claim 9, characterized in that, The specific details of step S5 are as follows: For the undetermined echoes identified by the model: like and If so, it is determined to be a cloud echo; like and This is then determined to be a clear-sky echo; like and It is then determined to be a sidelobe echo.

Images