A nonlinear regression method for predicting electromagnetic wave rain attenuation

Abstract

The present invention discloses a nonlinear regression electromagnetic wave rain attenuation prediction method, which belongs to the field of communication technology. The method is based on the classic ITU-R rain attenuation model, and by collecting rainfall rate and attenuation data, constructs training and test data sets for fitting model parameters k and b; adopts the nonlinear regression method in deep learning, designs and trains a model containing k and b parameters, uses mean square error as the loss function, and solves the parameters through a gradient descent optimization algorithm; when the similarity between the fitting result on the test set and the true attenuation value reaches a specified threshold value, stops training and outputs the optimal parameters. The present invention realizes the automatic solution of rain attenuation model parameters, reduces human intervention, improves calculation accuracy and the generalization ability of the model, and is suitable for electromagnetic wave propagation research under various rain attenuation scenarios.

Description

Technical Field

[0001] The invention relates to the field of communication technology, and in particular to an electromagnetic wave rain attenuation prediction method based on nonlinear regression. Background Art

[0002] The characteristics of electromagnetic wave propagation play a vital role in the estimation and demonstration of equipment capabilities. In the fields of communications, radar, navigation, etc., the performance of equipment is often strongly affected by the electromagnetic wave propagation environment. Electromagnetic waves will be affected by rainfall during propagation, which will cause attenuation. How to accurately estimate the electromagnetic wave rain attenuation in the current environment has become a research focus.

[0003] In the article "Research on Rain Attenuation of Sanya Ka-band Satellite Remote Sensing Data Signals" published in the journal "Radio Communication Technology" by Xin Jin and others from Xi'an Satellite Control Center in China, the factors that cause attenuation of remote sensing signals were analyzed in detail, and the electromagnetic wave rain attenuation model was accurately modeled by combining interpretable theory and measured data. However, the process of obtaining model parameters in this method is very complicated.

[0004] In the paper "Modeling and Analysis of Scattering Characteristics of Rainfall Environment in Autonomous Driving Radar Frequency Band" published in the Journal of System Simulation by Zou Mengfan and others from the Beijing University of Aeronautics and Astronautics, the impact of different rainfall environments on electromagnetic signals was analyzed in detail. However, the modeling in this paper is based on too specific assumptions and has limited scope of application.

[0005] In the national invention patent "A satellite broadband communication link reliability assurance method and device" applied by Liu Zhixin and others from the Institute of Systems Engineering of the Chinese People's Liberation Army Academy of Military Sciences, with the authorization announcement number CN117294341B, a typical rain attenuation estimation model is used to achieve reliable assurance of the communication link. However, there is no detailed description of how to set the model parameters to obtain the rain attenuation value.

[0006] Although the typical numerical values ​​of the rain attenuation model are highly universal, they lack the ability to characterize specific areas. Traditional parameter solution methods usually rely on a large amount of experimental measurement data and require manual adjustment, which is complicated and inefficient. Summary of the invention

[0007] In view of this, the present invention proposes a method for predicting electromagnetic wave rain attenuation based on nonlinear regression. The present invention can automatically and efficiently solve the 𝑘 and 𝑏 parameters of the rain attenuation formula model in the ITU-R P.838 standard recommended by the radiocommunication department under the International Telecommunication Union, thereby accurately describing the rain attenuation characteristics of specific frequency-using equipment under different rainfall rate conditions.

[0008] The technical solution adopted by the present invention is:

[0009] A nonlinear regression electromagnetic wave rain attenuation prediction method comprises the following steps:

[0010] Step 1: Collect rainfall rate data and electromagnetic wave rain attenuation data to construct training data sets and test data sets;

[0011] Step 2: Construct an electromagnetic wave rain attenuation prediction model;

[0012] Step 3: Design loss function;

[0013] Step 4: Optimize and verify the electromagnetic wave rain attenuation prediction model to obtain a trained electromagnetic wave rain attenuation prediction model;

[0014] Step 5: Input the rainfall rate into the trained electromagnetic wave rain attenuation prediction model to predict the rain attenuation per unit path length.

[0015] Furthermore, the specific method of step 1 is:

[0016] Step 1-1: Under the specified electromagnetic equipment operating frequency f and rainfall rate R, use the electromagnetic wave transmitter and receiver to measure the electromagnetic wave rain attenuation α per unit path length, repeat the measurement T times, and construct a rainfall rate data set and electromagnetic wave rain attenuation dataset ,in R = [ R 1 , ⋯ , R t , ⋯ R T ] T and α = [ α 1 , ⋯ , α t , ⋯ α T ] ;

[0017] Step 1-2: According to the proportion , the rainfall rate dataset Randomly split into rainfall rate training data set And rainfall rate test dataset , the electromagnetic wave rain attenuation dataset Randomly split into electromagnetic wave rain attenuation training data set and electromagnetic wave rain attenuation test dataset , and The number of data in is , and The number of data in is , Indicates rounding down.

[0018] Furthermore, the electromagnetic wave rain attenuation prediction model constructed in step 2 is expressed as the following function:

[0019]

[0020] Among them, 𝑘 and 𝑏 are the parameters to be solved.

[0021] Furthermore, in step 3, the mean square error is used as the loss function, and its formula is as follows:

[0022] Loss = 1 ⌊ μ T ⌋ ∑ t = 1 ⌊ μ T ⌋ [ α ′ t − f k , b ( R ′ t ) ] 2 .

[0023] Furthermore, the specific method of step 4 is:

[0024] Step 4-1: Initialize the parameters 𝑘 and 𝑏 to be solved, and ensure that their gradients are calculated during back propagation;

[0025] Step 4-2: Use the Adam optimizer, pass k and b as optimizer parameters, and set the learning rate lr;

[0026] Step 4-3: Based on the training data set and Calculate the loss function, clear the gradient values ​​stored in the optimizer, calculate the gradient of the loss function with respect to k and b, and then update the values ​​of k and b in the model;

[0027] Step 4-4: Set the model to evaluation mode and use the rainfall rate test dataset Estimated electromagnetic wave rain attenuation output from computational model , calculate and estimate the electromagnetic wave rain attenuation Electromagnetic wave rain attenuation test data set Correlation coefficient :

[0028] ;

[0029] Step 4-5: Repeat steps 4-2 to 4-4. When the threshold value is reached, the values ​​of k and b are saved to obtain the trained electromagnetic wave rain attenuation prediction model.

[0030] The beneficial effects of the present invention are:

[0031] 1. The present invention can solve the k and 𝑏 parameters in the rain attenuation model quickly and efficiently through automated model training without manual intervention.

[0032] 2. The present invention can adapt to various rainfall conditions and has strong generalization ability and practicality.

[0033] 3. The present invention uses nonlinear regression technology to achieve accurate fitting of rain attenuation model parameters and significantly improves the solution accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The figure is a flow chart of a method for predicting electromagnetic wave rain attenuation based on nonlinear regression in an embodiment of the present invention.

[0035] Figure 2 This is a schematic diagram of data collected in an embodiment of the present invention.

[0036] Figure 3 Schematic diagram of the relationship between the number of iterations and the correlation coefficient in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0038] A nonlinear regression electromagnetic wave rain attenuation prediction method, the overall process is as follows Figure 1 As shown, the specific steps include:

[0039] Step 1: Data collection and dataset construction;

[0040] Step 1-1: Data collection and construction

[0041] In the experimental area, the electromagnetic equipment is specified to use the frequency f=1GHz, and the electromagnetic wave transmitter and receiver are used to measure the electromagnetic wave rain attenuation per unit path length α at the rainfall rate R. The above process is repeated T=200 times to construct the rainfall rate data set and electromagnetic wave rain attenuation dataset ,in R = [ R 1 , ⋯ , R t , ⋯ R T ] T and α = [ α 1 , ⋯ , α t , ⋯ α T ] The collected rainfall rate data and electromagnetic wave rain attenuation data are as follows: Figure 2 shown.

[0042] Step 1-2: Dataset segmentation

[0043] According to the proportion = 0.7, and randomly divide the data set obtained from T tests into a training data set and a test data set, the numbers of which are and ,in represents rounding down; the rainfall rate training data set is , the dimension is ; Electromagnetic wave rain attenuation training data set , the dimension is ; The rainfall rate test data set is , the dimension is ;Electromagnetic wave rain attenuation test data set , the dimension is .

[0044] Step 2: Design model function;

[0045]

[0046] Where 𝑘 and 𝑏 are the parameters to be solved.

[0047] Step 3: Design loss function;

[0048] The mean square error is used as the loss function, and its formula is as follows:

[0049] Loss = 1 ⌊ μ T ⌋ ∑ t = 1 ⌊ μ T ⌋ [ α ′ t − f k , b ( R ′ t ) ] 2

[0050] Step 4: Model optimization and verification;

[0051] Step 4-1: Initialize the parameters 𝑘 and 𝑏 to be solved, both with a value of 0.1, and ensure that their gradients are calculated during back propagation.

[0052] Step 4-2: Define the optimizer. Use the Adam optimizer, pass k and b as the optimized parameters, and set the learning rate lr=0.0001.

[0053] Step 4-3: Based on the training data set and Calculate the loss function, clear the gradient values ​​stored in the optimizer, calculate the gradient of the loss function with respect to k and b, and then update the values ​​of k and b in the model.

[0054] Step 4-4: Verify the model. Set the model to evaluation mode and use the rainfall rate test data set as Calculate the estimated electromagnetic wave rain attenuation , calculate and estimate the electromagnetic wave rain attenuation Electromagnetic wave rain attenuation test data set Correlation coefficient :

[0055]

[0056] Step 4-5: Iterative optimization. When the correlation coefficient When the specified threshold value of 99.99% is reached, the k and b values ​​are saved; if the specified threshold is not reached, return to step 4-3. The relationship between the number of iterations and the correlation coefficient is shown in the figure below: Figure 3 As shown in Figure 2, the obtained k value is 0.0112 and the b value is 0.7666.

[0057] Step 5: Predict the rain attenuation per unit path length;

[0058] Step 5-1: The model function loads the saved k and b values;

[0059] Step 5-2: Input the rainfall rate into the model to obtain the electromagnetic wave rain attenuation per unit path length. The input rainfall rate value is 49.5051 mm / h, and the attenuation value per unit path length is 0.2238 dB / km.

[0060] In summary, the present invention is based on the classic ITU-R rain attenuation model. By collecting rainfall rate and attenuation data, the training and test data sets for fitting model parameters k and b are constructed. The nonlinear regression method in deep learning is used to design and train the model containing k and b parameters. The mean square error is used as the loss function, and the parameters are solved by the gradient descent optimization algorithm. When the similarity between the fitting result on the test set and the true attenuation value reaches the specified threshold value, the training is stopped and the optimal parameters are output. The present invention can automatically obtain the k and b values ​​in the model, and can use the obtained values ​​in combination with the model to obtain the attenuation value per unit path length corresponding to the rainfall rate, thereby realizing the automatic solution of the rain attenuation model parameters, reducing human intervention, improving the calculation accuracy and the generalization ability of the model, and is suitable for electromagnetic wave propagation research under various rain attenuation scenarios.

Claims

1. A nonlinear regression electromagnetic wave rain attenuation prediction method, characterized in that: The following steps are involved: Step 1: Collect rainfall rate data and electromagnetic wave rain attenuation data to construct training data sets and test data sets; the specific method is: Step 1-1: Under the specified electromagnetic equipment operating frequency f and rainfall rate R, use the electromagnetic wave transmitter and receiver to measure the electromagnetic wave rain attenuation α per unit path length, repeat the measurement T times, and construct a rainfall rate data set and electromagnetic wave rain attenuation dataset ,in and ; Step 1-2: According to the proportion , the rainfall rate dataset Randomly split into rainfall rate training data set And rainfall rate test dataset , the electromagnetic wave rain attenuation dataset Randomly split into electromagnetic wave rain attenuation training data set and electromagnetic wave rain attenuation test dataset , and The number of data in is , and The number of data in is , Indicates rounding down; Step 2: Construct an electromagnetic wave rain attenuation prediction model, expressed as the following function: Among them, 𝑘 and 𝑏 are the parameters to be solved; Step 3: Design loss function; Step 4: Optimize and verify the electromagnetic wave rain attenuation prediction model to obtain a trained electromagnetic wave rain attenuation prediction model; Step 5: Input the rainfall rate into the trained electromagnetic wave rain attenuation prediction model to predict the rain attenuation per unit path length; the specific method is: Step 4-1: Initialize the parameters 𝑘 and 𝑏 to be solved, and ensure that their gradients are calculated during back propagation; Step 4-2: Use the Adam optimizer, pass k and b as optimizer parameters, and set the learning rate lr; Step 4-3: Based on the training data set and Calculate the loss function, clear the gradient values ​​stored in the optimizer, calculate the gradient of the loss function with respect to k and b, and then update the values ​​of k and b in the model; Step 4-4: Set the model to evaluation mode and use the rainfall rate test dataset Estimated electromagnetic wave rain attenuation output from computational model , calculate and estimate the electromagnetic wave rain attenuation Electromagnetic wave rain attenuation test data set Correlation coefficient : ； Step 4-5: Repeat steps 4-2 to 4-4. When the threshold value is reached, the values ​​of k and b are saved to obtain the trained electromagnetic wave rain attenuation prediction model.

2. The electromagnetic wave rain attenuation prediction method based on nonlinear regression according to claim 1 is characterized in that: In step 3, the mean square error is used as the loss function, and its formula is as follows: 。

Images