Ka-band radar echo attenuation correction method and system based on artificial intelligence

Abstract

The invention provides a Ka-band radar echo attenuation correction method and system based on artificial intelligence, and relates to the technical field of radars. Comprising the following steps: acquiring horizontal reflectivity factor data detected by a radar and rainfall intensity data at a corresponding observation moment; constructing a reflectivity factor attenuation correction model set; according to the rainfall intensity data, inputting the horizontal reflectivity factor data into a corresponding reflectivity factor attenuation correction model in a reflectivity factor attenuation correction model set to obtain a horizontal reflectivity factor attenuation correction result, the reflectivity factor attenuation correction model is a deep learning model obtained by training historical radar horizontal reflectivity factor samples based on different rain intensity conditions. According to the method, the problems of relatively large error and uncertainty of a correction result caused by limitation of a traditional empirical formula method in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the field of radar technology, and in particular to an artificial intelligence-based Ka-band radar echo attenuation correction method and system. Background Art

[0002] Clouds play a crucial role in the atmospheric radiation balance, the water cycle, and climate change. They are also crucial targets for weather forecasting and weather modification. Therefore, the detection and study of cloud macro and micro parameters are of great significance. Currently, millimeter-wave cloud radar is one of the primary means of observing clouds. Compared to centimeter-wave radar, millimeter-wave radar has a shorter wavelength and higher spatial resolution and sensitivity, enabling more accurate detection of cloud micro parameters and characterization of cloud internal physical structure. However, radar echo signals are weak due to factors such as absorption by atmospheric water vapor and oxygen, attenuation by cloud particles (especially liquid water within clouds), water film on the radar antenna, Mie scattering, and supersaturation. Therefore, attenuation correction of the echo signals is necessary to improve data precision and accuracy and restore more realistic echo information. Attenuation correction is crucial in weather radar and cannot be ignored.

[0003] The traditional method uses fixed empirical coefficients for attenuation correction. Due to limited parameters and insufficient adaptability, the correction results have large errors and uncertainties. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide a Ka-band radar echo attenuation correction method and system based on artificial intelligence. The present invention solves the limitations of the traditional empirical formula method in the existing technology, which leads to large errors and uncertainties in the correction results.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A Ka-band radar echo attenuation correction method based on artificial intelligence, comprising:

[0007] Obtain the horizontal reflectivity factor data detected by radar and the rainfall intensity data at the corresponding observation time;

[0008] Constructing a reflectivity factor attenuation correction model set;

[0009] According to the rainfall intensity data, the horizontal reflectivity factor data is input into the corresponding reflectivity factor attenuation correction model in the reflectivity factor attenuation correction model set to obtain a horizontal reflectivity factor attenuation correction result, wherein the reflectivity factor attenuation correction model is a deep learning model trained based on historical radar horizontal reflectivity factor samples under different rainfall intensity conditions.

[0010] Preferably, it also includes:

[0011] Calculate the corresponding deviation, root mean square error and mean absolute error according to the horizontal reflectivity factor attenuation correction result;

[0012] The horizontal reflectivity factor attenuation correction result is evaluated according to the deviation, root mean square error and mean absolute error to obtain an evaluation result.

[0013] Preferably, the obtaining of horizontal reflectivity factor data detected by radar includes:

[0014] Determine the horizontal polarization of electromagnetic waves emitted by the target radar;

[0015] Calculating the backscattered signal intensity of the transmitted horizontally polarized electromagnetic wave;

[0016] Horizontal reflectivity factor data is determined according to the backscattered signal intensity.

[0017] Preferably, the process of constructing the reflectivity factor attenuation correction model set includes:

[0018] Collect and synchronize the raw horizontal reflectivity factor data of Ka-band radar and C-band radar;

[0019] Performing quality control on the original horizontal reflectivity factor data and dividing the samples into several categories according to rainfall intensity;

[0020] A sliding window method is used, using the horizontal reflectivity factor data of C wave as labels and the horizontal reflectivity factor data of the first 5, 10, and 15 distance libraries of Ka band as input features to construct corresponding training samples;

[0021] Various samples are input into the preset deep network for training to obtain a reflectivity factor attenuation correction model set.

[0022] Preferably, the preset deep network includes:

[0023] Three parallel long short-term memory network layers, each receiving three input sequences;

[0024] Three independent fully connected layers, corresponding to LSTM output dimensionality reduction;

[0025] Vector concatenation layer, concatenates the three-way fully connected results into a one-dimensional vector;

[0026] The linear layer regresses the spliced vector and outputs the horizontal reflectivity factor attenuation correction result.

[0027] A Ka-band radar echo attenuation correction system based on artificial intelligence, comprising:

[0028] An acquisition module is used to obtain the horizontal reflectivity factor data detected by the radar and the rainfall intensity data at the corresponding observation time;

[0029] A construction module for constructing a reflectivity factor attenuation correction model set;

[0030] A correction module is used to input the horizontal reflectivity factor data into the corresponding reflectivity factor attenuation correction model in the reflectivity factor attenuation correction model set according to the rainfall intensity data to obtain a horizontal reflectivity factor attenuation correction result, wherein the reflectivity factor attenuation correction model is a deep learning model trained based on historical radar horizontal reflectivity factor samples under different rainfall intensity conditions.

[0031] The present invention discloses the following technical effects:

[0032] The present invention provides an artificial intelligence-based Ka-band radar echo attenuation correction method, comprising: obtaining horizontal reflectivity factor data detected by radar and rainfall intensity data corresponding to the observation time; constructing a reflectivity factor attenuation correction model set; and inputting the horizontal reflectivity factor data into the corresponding reflectivity factor attenuation correction model in the reflectivity factor attenuation correction model set based on the rainfall intensity data to obtain a horizontal reflectivity factor attenuation correction result, wherein the reflectivity factor attenuation correction model is a deep learning model trained based on historical radar horizontal reflectivity factor samples under different rainfall intensities. The present invention replaces traditional empirical formulas with a data-driven intelligent model and utilizes the powerful feature extraction and nonlinear fitting capabilities of a deep learning algorithm to significantly improve the accuracy and stability of the attenuation correction results. It also possesses excellent generalization performance and can adapt to the attenuation correction needs under different precipitation conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0034] Figure 1 A flow chart of a method for correcting Ka-band radar echo attenuation based on artificial intelligence provided by an embodiment of the present invention;

[0035] Figure 2 A schematic diagram of a network architecture for radar echo attenuation correction provided by an embodiment of the present invention;

[0036] Figure 3Schematic diagram of the horizontal reflectivity factor density distribution of Ka-band radar and C-band radar before (ac) and after (df) R0-R2 correction provided in an embodiment of the present invention;

[0037] Figure 4 Schematic diagram of the horizontal reflectivity factor density distribution of Ka-band radar and C-band radar before (ac) and after (df) correction of R3-R5 provided in an embodiment of the present invention;

[0038] Figure 5 This is an example diagram of the Ka-band radar, C-band radar, and Ka-band radar provided in an embodiment of the present invention after correction by the reflectivity factor attenuation correction model. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] like Figure 1 As shown, the present invention provides a Ka-band radar echo attenuation correction method based on artificial intelligence, comprising:

[0042] Step 100: Obtaining horizontal reflectivity factor data detected by radar and rainfall intensity data corresponding to the observation time;

[0043] Step 200: constructing a reflectivity factor attenuation correction model set;

[0044] Step 300: Input the horizontal reflectivity factor data into the corresponding reflectivity factor attenuation correction model in the reflectivity factor attenuation correction model set according to the rainfall intensity data to obtain a horizontal reflectivity factor attenuation correction result, wherein the reflectivity factor attenuation correction model is a deep learning model trained based on historical radar horizontal reflectivity factor samples under different rainfall intensity conditions.

[0045] Specifically, the execution subject of this method can be an electronic device, or a radar echo attenuation correction device set in the electronic device. The radar echo attenuation correction device can be implemented through software, hardware, or a combination of both.

[0046] Specifically, horizontal reflectivity factor data detected by radar and rainfall intensity data corresponding to the observation time are obtained. Radar is an electronic device that detects targets using electromagnetic waves. The radar can be, for example, a Ka-band millimeter-wave radar, which is not limited in this embodiment.

[0047] Furthermore, the reflectivity factor attenuation correction model is a deep learning model trained based on historical radar horizontal reflectivity factor samples under different rain intensity conditions. The reflectivity factor attenuation correction model is a model for correcting radar echo attenuation.

[0048] Specifically, after obtaining the trained six reflectivity factor attenuation correction models R0-R5, the horizontal reflectivity factor is used as the input of the reflectivity factor attenuation correction model. According to the rainfall intensity at the corresponding moment, the reflectivity factor attenuation correction model of different rainfall intensity intervals is input. Then, the reflectivity factor attenuation correction model processes the input horizontal reflectivity factor to obtain the echo attenuation correction result output by the reflectivity factor attenuation correction model.

[0049] For example, when it is determined that the sample is a horizontal reflectivity factor sample, model training can be performed based on the horizontal reflectivity factor as the input of the reflectivity factor attenuation correction model, so that the obtained reflectivity factor attenuation correction model processes the horizontal reflectivity factor to obtain the echo attenuation correction result output by the reflectivity factor attenuation correction model.

[0050] Furthermore, the obtaining of horizontal reflectivity factor data detected by radar includes:

[0051] Determine the horizontal polarization of electromagnetic waves emitted by the target radar;

[0052] Calculating the backscattered signal intensity of the transmitted horizontally polarized electromagnetic wave;

[0053] Horizontal reflectivity factor data is determined according to the backscattered signal intensity.

[0054] Specifically, the horizontal reflectivity factor is a parameter used in weather radar to describe the effect of precipitation particles on electromagnetic wave propagation. It represents the power intensity of horizontally polarized electromagnetic waves emitted by the radar after being scattered by precipitation particles (such as raindrops and hail). The horizontal reflectivity factor is obtained by transmitting horizontally polarized electromagnetic waves from the radar and measuring the strength of the backscattered signal.

[0055] Furthermore, the process of constructing the reflectivity factor attenuation correction model set includes:

[0056] Collect and synchronize the raw horizontal reflectivity factor data of Ka-band radar and C-band radar;

[0057] Performing quality control on the original horizontal reflectivity factor data and dividing the samples into several categories according to rainfall intensity;

[0058] A sliding window method is used, using the horizontal reflectivity factor data of C wave as labels and the horizontal reflectivity factor data of the first 5, 10, and 15 distance libraries of Ka band as input features to construct corresponding training samples;

[0059] Various samples are input into the preset deep network for training to obtain a reflectivity factor attenuation correction model set.

[0060] Specifically, the reflectivity factor attenuation correction model is trained based on the following steps: obtaining horizontal reflectivity factor data and rainfall intensity data at the corresponding observation time, and making horizontal reflectivity factor samples under six different rainfall intensities R0-R5; determining the network architecture based on the horizontal reflectivity factor samples; optimizing and training the network architecture based on the horizontal reflectivity factor samples to obtain six reflectivity factor attenuation correction models R0-R5; the horizontal reflectivity factor samples are samples for network construction.

[0061] Obtaining a horizontal reflectivity factor sample includes: synchronously obtaining original horizontal reflectivity factor data of a Ka-band radar and a C-band radar; wherein the Ka-band and the C-band are different bands; performing quality control preprocessing on the original horizontal reflectivity factor data of the Ka-band radar and the C-band radar to form a valid horizontal reflectivity factor data set; interpolating the horizontal reflectivity factor data of the C-band radar into the horizontal reflectivity factor set of the Ka-band radar at the same time, using the horizontal reflectivity factor of the C-band radar as label data, and the horizontal reflectivity factors of the Ka-band radar 5, 10, and 15 bins before the bin where the label is located as input factors of a model, constructing a sample, and sliding the data bin by bin along the radial direction of the Ka-band radar according to a sliding window method to form multiple samples to obtain the horizontal reflectivity factor sample.

[0062] The original horizontal reflectivity factor set of the Ka-band radar is the set of horizontal reflectivity factors at different distance libraries in the Ka-band.

[0063] The distance bin refers to a small unit divided by distance along a ray direction in radar echo signal processing. The number of distance bins may be 5, 10, or 15, for example, and this embodiment does not limit this.

[0064] The original horizontal reflectivity factor set of the C-band radar is the set of horizontal reflectivity factors at different distance libraries in the C-band, and is used as the true value of the echo.

[0065] Preprocessing, for example, may include performing quality control, normalization, and other processing on the original horizontal reflectivity factor set of the Ka-band radar and the original horizontal reflectivity factor set of the C-band radar, respectively. Quality control may include, for example, filtering out non-meteorological echoes, ground objects, and isolated clutter, etc., which is not limited in this embodiment.

[0066] Furthermore, the preset deep network includes:

[0067] Three parallel long short-term memory network layers, each receiving three input sequences;

[0068] Three independent fully connected layers, corresponding to LSTM output dimensionality reduction;

[0069] Vector concatenation layer, concatenates the three-way fully connected results into a one-dimensional vector;

[0070] The linear layer regresses the spliced vector and outputs the horizontal reflectivity factor attenuation correction result.

[0071] Specifically, Figure 2 This is a schematic diagram of the network architecture for echo attenuation correction of the radar provided by the present invention. In a specific embodiment, the network architecture is determined according to the horizontal reflectivity factor samples, including: inputting the horizontal reflectivity factor samples into the long short-term memory network layers of three basic architectures respectively; inputting the output results of the long short-term memory network layers of the three basic architectures into the fully connected layers of the three basic architectures respectively to obtain three output results; splicing the results in the full connections of the three basic architectures to form a one-dimensional vector; inputting the one-dimensional vector into the linear layer of the basic architecture to obtain the final attenuation correction result of the horizontal reflectivity factor; and further optimizing and training the architecture according to the attenuation correction result and the labels in the samples to determine the network architecture.

[0072] The basic architecture is a pre-set, untrained basic network architecture, which includes a long short-term memory network, a fully connected layer, and a linear layer, which is not limited in this embodiment.

[0073] The basic architecture includes three long short-term memory network layers, three fully connected layers, and finally outputs the echo attenuation correction result through a linear layer. The basic architecture can be, for example, a Ka-band radar echo attenuation correction framework designed based on LSTM, which is not limited in this embodiment.

[0074] In this step, the output nodes of the long short-term memory network can be 128, and the fully connected layer can be, for example, an output layer that passes through a 64-dimensional vector, a 32-dimensional vector, and a 1-dimensional vector respectively. This embodiment does not limit this.

[0075] For example, the effective values of horizontal reflectivity factor samples from 5, 10, and 15 distance bins can be used as the three input factors of the model. These are then passed through three long short-term memory networks and three fully connected layers, and finally concatenated and input into a linear layer to obtain the final echo attenuation correction result. ReLU is used as the activation function, and Dropout is used to prevent overfitting; this is not limited in this embodiment.

[0076] Furthermore, it also includes:

[0077] Calculate the corresponding deviation, root mean square error and mean absolute error according to the horizontal reflectivity factor attenuation correction result;

[0078] The horizontal reflectivity factor attenuation correction result is evaluated according to the deviation, root mean square error and mean absolute error to obtain an evaluation result.

[0079] Specifically, after obtaining the horizontal reflectivity factor attenuation correction result, the horizontal reflectivity factor attenuation correction result may be further evaluated to obtain an evaluation result.

[0080] Evaluation indicators such as deviation, root mean square error and mean absolute error are calculated based on the correction results of horizontal reflectivity factor attenuation, and the correction results of horizontal reflectivity factor attenuation are evaluated by ratio deviation, root mean square error and mean absolute error.

[0081] The calculation of BIAS is shown in formula (1), the calculation of root mean square error RMSE is shown in formula (2), and the calculation of mean absolute error MAE is shown in formula (3).

[0082]

[0083] Where is the C-band echo intensity (true value), is the corrected Ka-band echo intensity (predicted value), and n represents the total number of range bins. BIAS measures the difference between the corrected Ka-band echo intensity and the C-band echo intensity. A smaller BIAS value indicates a better overall correction effect. A smaller RMSE value indicates a smaller difference between the corrected Ka-band echo intensity and the C-band echo intensity. Unlike BIAS, MAE is less susceptible to outliers because it only considers the average of absolute values.

[0084] For example, as shown in Table 1, various evaluation index results of the horizontal reflectivity factor before correction and the horizontal reflectivity factor after correction by the reflectivity factor attenuation correction model are shown.

[0085] Table 1

[0086]

[0087]

[0088] As shown in Table 1, by comparing the BIAS, MAE, and RMSE before and after the horizontal reflectivity factor correction, the effect of the reflectivity factor attenuation correction model is significantly improved.

[0089] Figure 3 、 Figure 4 These are the reflectivity factor density distributions of Ka-band radar and C-band radar before R0-R2 and R3-R5 corrections and after the reflectivity factor attenuation correction model is corrected.

[0090] Under the condition of no precipitation (R0 interval), the reflectivity factor distribution of Ka band and C band is basically the same ( Figure 3 a), indicating that the Ka band attenuation is weak at this time. After the technical solution of the present invention is revised ( Figure 3 d), the consistency between the two is further improved.

[0091] Under drizzle conditions (R1 interval), the Ka-band horizontal reflectivity factor shows obvious attenuation deviation ( Figure 3 b). After the technical solution of the present invention is revised ( Figure 3 e), the bias is significantly improved, with the mean absolute error (MAE) reduced from 4.66dBZ to 2.87dBZ, and the root mean square error (RMSE) is reduced simultaneously.

[0092] As the precipitation intensity increases to light rain level (R2-R3 range), the attenuation of the uncorrected data increases ( Figure 3 c, Figure 4 a), RMSE can reach 10dBZ. After processing by the algorithm in this paper ( Figure 3 f, Figure 4 d), the consistency between the reflectivity factor distribution and the C-band reference value is significantly improved, and the RMSE is reduced to about 5dBZ.

[0093] In moderate to heavy rain conditions (R4-R5 range), Ka-band signal attenuation is more serious ( Figure 4 b-4c), the RMSE of the uncorrected data in the R5 interval is as high as 19.78dBZ. After correction by the algorithm in this paper ( Figure 4 e-4f), the reflectivity factor deviation is effectively compensated, and all error indicators are significantly reduced (Table 1), verifying the stable correction performance of the scheme of the present invention under different precipitation intensities.

[0094] Figure 5The image shows the Ka-band radar data, C-band radar data, and the Ka-band radar data corrected by the reflectivity factor attenuation correction model on May 8, 2016. The horizontal axis is time (from 9:00 to 23:00), the vertical axis is altitude (unit: kilometers, range: approximately 0-12 kilometers), and the color represents the reflectivity factor (dBZ).

[0095] contrast Figure 5 a(Ka band) and Figure 5 As can be seen from b (C band), there are differences in overall morphology and color distribution between the two. At around 11:00, 12:00, 16:00, 19:00, and 21-22:00, the C band shows strong echoes at certain altitudes, while the Ka band echo intensity is weaker than the C band. This indicates that at these times and altitudes, the Ka band is affected by factors such as water vapor and cloud particles in the atmosphere and the echo is attenuated. After processing with the reflectivity factor attenuation correction model ( Figure 5 c) The echo intensity of Ka-band radar data is more consistent with that of C-band, which effectively corrects the intensity loss caused by attenuation and improves data accuracy.

[0096] The present invention provides an artificial intelligence-based Ka-band radar echo attenuation correction method. The method obtains horizontal reflectivity factor data detected by radar. Based on real-time rainfall intensity information, the horizontal reflectivity factor is input into a reflectivity factor attenuation correction model for the corresponding rainfall intensity range for processing, resulting in an echo attenuation correction result output by the reflectivity factor attenuation correction model. The reflectivity factor attenuation correction model is a deep learning model trained based on historical radar horizontal reflectivity factor samples under different rainfall intensities. The reflectivity factor attenuation correction model is a model for performing radar echo attenuation correction. The technical solution of the present invention addresses the limitations of conventional empirical formula methods in the prior art. Conventional methods use fixed empirical coefficients for attenuation correction. Due to limited parameters and insufficient adaptability, the correction results contain large errors and uncertainties. To address this issue, the present invention processes the radar's horizontal reflectivity factor using a trained reflectivity factor attenuation correction model to obtain a radar echo attenuation correction result. The reflectivity factor attenuation correction model exhibits good stability and generalization capabilities, thereby improving the accuracy of radar echo attenuation correction results.

[0097] The following describes an artificial intelligence-based Ka-band radar echo attenuation correction device provided by the present invention. The radar echo attenuation correction device described below and the radar echo attenuation correction method described above can be referenced to each other.

[0098] This implementation also provides an artificial intelligence-based Ka-band radar echo attenuation correction system, including:

[0099] An acquisition module is used to obtain the horizontal reflectivity factor data detected by the radar and the rainfall intensity data at the corresponding observation time;

[0100] A construction module for constructing a reflectivity factor attenuation correction model set;

[0101] A correction module is used to input the horizontal reflectivity factor data into the corresponding reflectivity factor attenuation correction model in the reflectivity factor attenuation correction model set according to the rainfall intensity data to obtain a horizontal reflectivity factor attenuation correction result, wherein the reflectivity factor attenuation correction model is a deep learning model trained based on historical radar horizontal reflectivity factor samples under different rainfall intensity conditions.

[0102] More specifically, in an exemplary embodiment, the apparatus further includes a model training module. The model training module is configured to: obtain horizontal reflectivity factor data and rainfall intensity data corresponding to the observation time, and generate horizontal reflectivity factor samples at six different rainfall intensities (R0-R5); determine a network architecture based on the horizontal reflectivity factor samples; and optimize and train the network architecture based on the horizontal reflectivity factor samples to obtain six reflectivity factor attenuation correction models (R0-R5).

[0103] In an example embodiment, a model training module obtains horizontal reflectivity factor samples, specifically for: synchronously obtaining original horizontal reflectivity factor data of a Ka-band radar and a C-band radar; wherein the Ka-band and the C-band are different bands; performing quality control preprocessing on the original horizontal reflectivity factor data of the Ka-band radar and the C-band radar to form a valid horizontal reflectivity factor data set; interpolating the horizontal reflectivity factor data of the C-band radar into the horizontal reflectivity factor set of the Ka-band radar at the same time, using the horizontal reflectivity factor of the C-band radar as label data, and the horizontal reflectivity factors of the Ka-band radar 5, 10, and 15 bins before the label is located in the bin as input factors of the model, constructing a sample, and sliding along the radial direction of the Ka-band radar bin by bin according to the sliding window method to form multiple samples to obtain the horizontal reflectivity factor sample.

[0104] In one exemplary embodiment, a model training module determines effective horizontal reflectivity factor data for the Ka-band radar and effective horizontal reflectivity factor data for the C-band radar based on the raw horizontal reflectivity factor data for the Ka-band radar and the raw horizontal reflectivity factor data for the C-band radar. Specifically, quality control preprocessing includes classifying the horizontal reflectivity factors into six rainfall intensity levels, R0-R5, based on synchronized rainfall intensity observation data. When matching Ka-band and C-band radar profile data, only horizontal reflectivity factor data with continuous profile data segments no shorter than 33 bins (990 meters) is retained to avoid the influence of clutter.

[0105] In an example embodiment, the model training module determines the network architecture based on the horizontal reflectivity factor samples, and is specifically used to: input the horizontal reflectivity factor samples into the long short-term memory network layers of the three basic architectures respectively; input the output results of the long short-term memory network layers of the three basic architectures into the fully connected layers of the three basic architectures respectively to obtain three output results; splice the results in the full connections of the three basic architectures to form a one-dimensional vector; input the one-dimensional vector into the linear layer of the basic architecture to obtain the final attenuation correction result of the horizontal reflectivity factor; further optimize and train the architecture according to the attenuation correction result and the labels in the samples to determine the network architecture.

[0106] The device of this embodiment can be used to execute the method of any embodiment in the embodiment of the echo attenuation correction method side of the Ka-band radar. Its specific implementation process and technical effects are similar to those in the embodiment of the radar echo attenuation correction method side. For details, please refer to the detailed description in the embodiment of the radar echo attenuation correction method side, which will not be repeated here.

[0107] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0108] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A Ka-band radar echo attenuation correction method based on artificial intelligence, characterized in that: include: Obtain the horizontal reflectivity factor data detected by radar and the rainfall intensity data at the corresponding observation time; Constructing a reflectivity factor attenuation correction model set; According to the rainfall intensity data, the horizontal reflectivity factor data is input into the corresponding reflectivity factor attenuation correction model in the reflectivity factor attenuation correction model set to obtain a horizontal reflectivity factor attenuation correction result, wherein the reflectivity factor attenuation correction model is a deep learning model trained based on historical radar horizontal reflectivity factor samples under different rainfall intensity conditions.

2. The artificial intelligence-based Ka-band radar echo attenuation correction method according to claim 1, characterized in that: Also includes: Calculate the corresponding deviation, root mean square error and mean absolute error according to the horizontal reflectivity factor attenuation correction result; The horizontal reflectivity factor attenuation correction result is evaluated according to the deviation, root mean square error and mean absolute error to obtain an evaluation result.

3. The artificial intelligence-based Ka-band radar echo attenuation correction method according to claim 1, characterized in that: The obtaining of horizontal reflectivity factor data detected by radar includes: Determine the horizontal polarization of electromagnetic waves emitted by the target radar; Calculating the backscattered signal intensity of the transmitted horizontally polarized electromagnetic wave; Horizontal reflectivity factor data is determined according to the backscattered signal intensity.

4. The artificial intelligence-based Ka-band radar echo attenuation correction method according to claim 1, characterized in that: The process of constructing the reflectivity factor attenuation correction model set includes: Collect and synchronize the raw horizontal reflectivity factor data of Ka-band radar and C-band radar; Performing quality control on the original horizontal reflectivity factor data and dividing the samples into several categories according to rainfall intensity; A sliding window method is used, using the horizontal reflectivity factor data of C wave as labels and the horizontal reflectivity factor data of the first 5, 10, and 15 distance libraries of Ka band as input features to construct corresponding training samples; Various samples are input into the preset deep network for training to obtain a reflectivity factor attenuation correction model set.

5. The artificial intelligence-based Ka-band radar echo attenuation correction method according to claim 4, characterized in that: The preset deep network includes: Three parallel long short-term memory network layers, each receiving three input sequences; Three independent fully connected layers, corresponding to LSTM output dimensionality reduction; Vector concatenation layer, concatenates the three-way fully connected results into a one-dimensional vector; The linear layer regresses the spliced vector and outputs the horizontal reflectivity factor attenuation correction result.

6. An artificial intelligence-based Ka-band radar echo attenuation correction system, characterized in that: include: An acquisition module is used to obtain the horizontal reflectivity factor data detected by the radar and the rainfall intensity data at the corresponding observation time; A construction module for constructing a reflectivity factor attenuation correction model set; A correction module is used to input the horizontal reflectivity factor data into the corresponding reflectivity factor attenuation correction model in the reflectivity factor attenuation correction model set according to the rainfall intensity data to obtain a horizontal reflectivity factor attenuation correction result, wherein the reflectivity factor attenuation correction model is a deep learning model trained based on historical radar horizontal reflectivity factor samples under different rainfall intensity conditions.

Images