A method for selecting electromagnetic wave rain attenuation model based on upper bound of confidence interval

Abstract

The present invention discloses a method for selecting an electromagnetic wave rain attenuation model based on a confidence interval upper bound, and belongs to the technical field of electromagnetic wave propagation loss. The present invention adopts an improved confidence interval upper bound reinforcement learning algorithm, defines the action space as a plurality of different rain attenuation models, and by inputting rainfall rate information, the system can dynamically select the optimal model from a plurality of candidate models; the reward function is based on the absolute error between the electromagnetic wave rain attenuation value per unit distance predicted by each model and the actual observed value, the smaller the error, the higher the reward, and vice versa. Through continuous iterative training, the present invention can adaptively select the rain attenuation model with the highest prediction accuracy in different environments, thereby improving the accuracy of electromagnetic wave propagation prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of electromagnetic wave propagation loss, and in particular to a method for selecting an electromagnetic wave rain attenuation model based on an upper bound of a confidence interval. Background Art

[0002] In the process of electromagnetic wave propagation, rain attenuation is an important factor affecting the quality of signal transmission, especially in the microwave and millimeter wave frequency bands. The propagation of electromagnetic waves in a rainy environment will be scattered and absorbed by raindrops, resulting in signal attenuation. The calculation of rain attenuation is of great significance to wireless communications, radar systems and other fields.

[0003] There are many rain attenuation models in the prior art: In the "Research on Rain Attenuation of Sanya Ka-band Satellite Remote Sensing Data Signals" published by Xin Jin et al. of Xi'an Satellite Tracking and Control Center in China in the journal "Radio Communication Technology", rain attenuation modeling was performed based on the measured data of satellite remote sensing data signals. In the "Modeling and Analysis of Scattering Characteristics of Rainfall Environment in Autonomous Driving Radar Band" published by Zou Mengfan et al. of Beijing University of Aeronautics and Astronautics in the journal "Journal of System Simulation", a scattering characteristic model of rainfall environment was derived based on assumptions such as rainfall physical characteristics. The national invention patent "A Method and Device for Reliability Assurance of Satellite Broadband Communication Links" applied by Liu Zhixin et al. of the Institute of Systems Engineering of the Chinese People's Liberation Army Academy of Military Sciences, with the authorization announcement number CN117294341B, directly adopted the ITU-R model. In addition, there are also K / N models, Simple models, etc. Each model performs differently under different rainfall conditions, making it difficult to determine the optimal model in practical applications. Specifically, due to different parameter settings of different models, the prediction accuracy of some models is better than that of other models in specific application environments. Therefore, how to intelligently select the optimal rain attenuation model according to the actual environment has become an urgent problem to be solved.

[0004] Current model selection methods usually rely on fixed model parameters or manual selection, lacking flexibility and adaptability. With the development of reinforcement learning technology, strategies based on the balance between exploration and exploitation can be used to dynamically select the best model and improve prediction accuracy. Summary of the invention

[0005] The purpose of the present invention is to provide a method for selecting an electromagnetic wave rain attenuation model based on an upper bound of a confidence interval. By adopting an improved upper bound algorithm of the confidence interval, the optimal model is dynamically selected from multiple rain attenuation models according to the actual regional environmental conditions, thereby accurately predicting the rain attenuation of electromagnetic waves under different rainfall conditions.

[0006] The technical solution adopted by the present invention is:

[0007] A method for selecting an electromagnetic wave rain attenuation model based on an upper bound of a confidence interval comprises the following steps:

[0008] Step 1: Construct alternative model space;

[0009] Step 2: Set the total number of rounds T and initialize the number of times each model is selected in the candidate model space =0, total reward =0;

[0010] Step 3: Select each model once and update the number of times each model is selected and total reward ;

[0011] Step 4: Calculate the upper bound of the confidence interval for each model and select the model with the largest upper bound of the confidence interval ;

[0012] Step 5: Update the model The number of times selected and total reward ;

[0013] Step 6: Repeat steps 4 and 5 until the total round T is reached, and then execute step 4 again to output the model with the largest upper bound of the selected confidence interval to complete the selection of the electromagnetic wave rain attenuation model.

[0014] Furthermore, the specific method of step 1 is:

[0015] Set multiple different rain attenuation models as the candidate model space corresponding to the reinforcement learning action space. Each action corresponds to a model, and the total number of models is I. Then the candidate model space It is expressed as:

[0016] M = [ M 1 , ⋯ , M i , ⋯ , M I ] .

[0017] Furthermore, in step 3 and step 5, the number of times selected and total reward The update method is:

[0018] Step S1: Update the model The number of times selected :

[0019] ;

[0020] Step S2: Measuring rainfall and the rain attenuation of electromagnetic waves per unit distance , the rainfall Input Model , get the predicted rain attenuation value of electromagnetic waves per unit distance , the reward value is obtained through the reward function :

[0021] ;

[0022] Step S3: Update the model Total Rewards :

[0023] .

[0024] Furthermore, the specific method of step 4 is:

[0025] Step 4-1: Calculate the upper bound of the confidence interval for each model :

[0026] ,

[0027] in, is the current round number;

[0028] Step 4-2: Select the model with the largest upper bound of the confidence interval ; If there are multiple models with the largest upper bound of the confidence interval, select the model with the serial number The smallest model.

[0029] The beneficial effects of the present invention are:

[0030] 1. This method selects models based on rainfall rate input information by setting the action dimension to the number of optional rain attenuation models. The electromagnetic wave rain attenuation per unit distance predicted by each model based on the rainfall rate is calculated, and the difference between these values ​​and the actual observed values ​​is used as the reward function. The larger the difference, the smaller the reward; the smaller the difference, the greater the reward. Through multiple iterative training, the system can gradually converge to the optimal model selection strategy.

[0031] 2. Compared with the prior art, the present invention can dynamically select the optimal model in a specific area by introducing a reinforcement learning mechanism, has strong adaptability and high application value.

[0032] 3. The present invention adopts an improved confidence interval upper bound algorithm, which retains the exploration and utilization balance characteristics of the original confidence interval upper bound algorithm while enhancing the ability to suppress models with poor performance, so that the weights of models with poor performance are dynamically reduced during the model selection process.

[0033] 4. The method of the present invention has strong adaptability and high accuracy and can be widely used in the field of electromagnetic wave propagation prediction.

[0034] In summary, the present invention can adaptively select the rain attenuation model with the highest prediction accuracy in different environments, thereby improving the accuracy of electromagnetic wave propagation prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is the rain attenuation value of electromagnetic waves per unit distance obtained by measurement. DETAILED DESCRIPTION

[0036] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0037] A method for selecting an electromagnetic wave rain attenuation model based on an upper bound of a confidence interval specifically comprises the following steps:

[0038] Step 1: Construct alternative model space;

[0039] Set multiple different rain attenuation models as candidate model spaces corresponding to the reinforcement learning action space. Each action corresponds to a model. The three models are model1, model2 and model3. The total number of models is I=3, so the candidate model space is It can be expressed as

[0040] M = [ M 1 , M 2 , M 3 ] ;

[0041] Correspondingly, a function is constructed to record the total rewards obtained by each model. and the number of selections variable .

[0042] Step 2: Set the total number of rounds T and initialize the number of times each model is selected in the candidate model space =0, total reward =0;

[0043] Step 3: Select each model once and update the number of times each model is selected and total reward ; Among them, the reward function is calculated based on the absolute difference between the predicted value and the actual value:

[0044] ;

[0045] In the formula, is the reward value, is the rain attenuation value of electromagnetic waves per unit distance predicted by the i-th model according to the rainfall rate x, is the rain attenuation value of electromagnetic waves per unit distance measured by the equipment;

[0046] Step 4: Calculate the upper bound of the confidence interval for each model:

[0047] ;

[0048] in, is the upper bound of the confidence interval, For the current round, is the dynamic exploration coefficient and The decay weight is used to reduce the probability of selecting a model with poor prediction performance;

[0049] Select the model with the largest upper bound of the confidence interval ;

[0050] Step 5: Update the model The number of times selected and total reward ;

[0051] Step 6: Repeat steps 4 and 5 until the total round T is reached, and then execute step 4 again to output the model with the largest upper bound of the selected confidence interval to complete the selection of the electromagnetic wave rain attenuation model.

[0052] Specifically, the following steps are included:

[0053] S1, initialize the total reward obtained by each model =0, selected times variable =0 and current round =0, set the total optimization rounds to T=1000;

[0054] S2, round Add 1 and calculate the upper bound of the confidence interval for each model , get the model with the largest upper bound of the confidence interval If multiple models have the same upper confidence interval and the largest upper confidence interval, select the serial number Small model. Update the number of times the model is selected :

[0055] ;

[0056] S3, set the sensor frequency to 5 GHz, and obtain the rainfall in the current area by measuring =10 mm / h and rain attenuation of electromagnetic waves per unit distance , according to the model Input rainfall Predict the rain attenuation value of electromagnetic waves per unit distance The measured rain attenuation value of electromagnetic waves per unit distance is as follows: Figure 1 As shown. The reward value is calculated by the reward function , update total reward :

[0057] ;

[0058] S4, when the round When , return to step S2; when the round Calculate the upper bound of the confidence interval for each model when , and obtain the model with the largest upper bound of the confidence interval As the optimal model. If multiple models have the same upper confidence interval and the largest upper confidence interval, select the model with the same upper confidence interval. The smaller model is taken as the optimal model.

[0059] After 1000 rounds of optimization, the number of times the three models were selected and the average reward values ​​are shown in Table 1:

[0060]

[0061] It can be seen from the observation table that model1 is selected the most times and has the largest average reward value, making it the optimal unit distance electromagnetic wave model.

[0062] In summary, the present invention adopts an improved confidence interval upper bound reinforcement learning algorithm, defines the action space as multiple different rain attenuation models, and by inputting rainfall rate information, the system can dynamically select the optimal model from multiple candidate models; the reward function is based on the absolute error between the electromagnetic wave rain attenuation value per unit distance predicted by each model and the actual observed value. The smaller the error, the higher the reward, and vice versa. Through continuous iterative training, the present invention can adaptively select the rain attenuation model with the highest prediction accuracy in different environments, thereby improving the accuracy of electromagnetic wave propagation prediction.

Claims

1. A method for selecting an electromagnetic wave rain attenuation model based on an upper bound of a confidence interval, characterized in that: The following steps are involved: Step 1: Construct an alternative model space; the specific method is: Set multiple different rain attenuation models as the candidate model space corresponding to the reinforcement learning action space. Each action corresponds to a model, and the total number of models is I. Then the candidate model space It is expressed as: ； Step 2: Set the total number of rounds T and initialize the number of times each model is selected in the candidate model space =0, total reward =0; Step 3: Select each model once and update the number of times each model is selected and total reward ; Step 4: Calculate the upper bound of the confidence interval for each model and select the model with the largest upper bound of the confidence interval ; The specific method is: Step 4-1: Calculate the upper bound of the confidence interval for each model : ， in, is the current round number; Step 4-2: Select the model with the largest upper bound of the confidence interval ; If there are multiple models with the largest upper bound of the confidence interval, select the model with the serial number The smallest model; Step 5: Update the model The number of times selected and total reward ; Step 6: Repeat steps 4 and 5 until the total round T is reached, and then execute step 4 again to output the model with the largest upper bound of the selected confidence interval, thus completing the selection of the electromagnetic wave rain attenuation model; In steps 3 and 5, the number of selections and total reward The update method is: Step S1: Update the model The number of times selected : ； Step S2: Using sensors to measure rainfall , using equipment to measure the rain attenuation value of electromagnetic waves per unit distance , the rainfall Input Model , get the predicted rain attenuation value of electromagnetic waves per unit distance , the reward value is obtained through the reward function : ； Step S3: Update the model Total Rewards : 。

Images