A method, system, device and medium for predicting an electric wave propagation path based on a wavelet neural network to establish a troposphere three-dimensional atmospheric model

Abstract

The application discloses a method, system, device and medium for establishing a troposphere three-dimensional atmospheric model-based radio wave propagation path prediction method; the method comprises the following steps: collecting and establishing historical sounding data sets of different latitudes, longitudes and altitude information, to obtain preprocessed historical sounding data sets; constructing a troposphere three-dimensional atmospheric model, to obtain troposphere atmospheric refractive index distribution; setting a transmitting and receiving antenna coordinate, a transmitting antenna elevation angle, a working frequency, reading a space terrain, and respectively calculating the electrical conductivity and dielectric constant of the ground and sea surface based on the working frequency, to obtain distance error and elevation angle error caused by troposphere atmospheric refraction; and optimizing an electromagnetic wave ray propagation path by using the distance error and the elevation angle error, to obtain a radio wave propagation path prediction result; the system, the device and the medium are used for realizing the method for establishing the troposphere three-dimensional atmospheric model-based radio wave propagation path prediction method; and the application has the characteristics of small error, accurate prediction and good annual periodicity.

Description

Technical Field

[0001] This invention relates to the field of radio wave propagation path prediction technology, specifically to a method, system, device, and medium for predicting radio wave propagation paths based on a three-dimensional atmospheric model of the troposphere established using a wavelet neural network. Background Technology

[0002] Currently, in the field of deterministic electromagnetic wave propagation computation, the forward ray tracing algorithm is a commonly used simulation method to model the propagation and interaction of electromagnetic waves in complex environments. It treats electromagnetic waves as composed of many infinitely small rays or beams, and simulates the behavior of electromagnetic waves in the environment by tracing the propagation paths and interactions of these rays. The core idea of ​​the algorithm is to start from the emission source, emit rays along a specific direction, and determine the propagation path of the electromagnetic wave in the environment by calculating the intersection points between the rays and objects, as well as phenomena such as reflection and refraction.

[0003] Existing radio wave propagation path prediction uses forward ray tracing algorithms. When calculating the propagation path of radio waves over large areas or long distances, the non-uniform distribution of the refractive index of the troposphere and atmospheric refraction effects cause light to bend and diffuse, resulting in changes in the path of electromagnetic waves during propagation. The propagation distance and angle errors caused by atmospheric refraction effects are large, leading to large errors in the prediction results.

[0004] Patent application [CN103217177A] discloses "a method, device, and system for correcting radio wave refraction." The method includes the following steps: 1. Obtaining the tropospheric refractive index profile in real time based on historical meteorological radiosonde data and surface temperature, humidity, and pressure data; 2. Based on GNSS signals collected by a ground-based single-station GNSS receiver, obtaining the measured value of the ionospheric vertical total electron content in real time, and then using a genetic nonlinear optimization algorithm to invert the ionospheric electron density profile in real time; 3. Calculating the tropospheric refractive index profile and ionospheric electron density profile according to a segmented tropospheric refractive index model, and using a ray tracing method to calculate the radio wave environment refraction error in real time; 4. Correcting the target detection data based on the target radio wave environment refraction error. Because the tropospheric segmented model used in this invention only takes surface temperature, humidity, and pressure as input, and does not consider the variation of atmospheric refractive index at different latitudes, longitudes, and altitudes, there may be certain errors when simulating the refractive index of regions with different climatic environments.

[0005] Patent application [CN109164439A] discloses a method for calculating the atmospheric refractive index on radio waves. The method includes the following steps: 1. Based on my country's geographical location and the characteristics of radio meteorological environment changes, the entire country is divided into 1840 grids using atmospheric environmental grid technology, establishing a national atmospheric refractive index profile database; 2. For regions with uniform atmospheric structure, only the variation of atmospheric refractive index with altitude is considered, treating the atmosphere as a spherical layered atmosphere. The atmospheric profile of the radar area is obtained using a direct detection method, and the refractive index at any position of the radio wave is obtained by shifting the measured refractive index at the same altitude; 3. For regions with non-uniform atmospheric structure and complex underlying surfaces, a three-dimensional atmospheric profile is used. When the starting position of the radio wave is within the grid where the radar is located, the position of the ray point is first calculated, and then the atmospheric refractive index profile of that ray point is obtained using the established national atmospheric profile data. This refractive index profile is then used to calculate the atmospheric refractive index on the radio wave. Because this invention does not further quantify the changes in the atmosphere under different seasons, it cannot effectively simulate the impact of different seasons on the tropospheric environment of a fixed region, and has the disadvantage of poor annual periodicity. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, the present invention aims to provide a method, system, device, and medium for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model established using a wavelet neural network. By considering historical radiosonde datasets distributed across various latitudes and longitudes, a three-dimensional tropospheric atmospheric model considering different latitudes, longitudes, altitudes, and dates is established based on a wavelet neural network. The refractive index distribution of the troposphere is calculated using the three-dimensional tropospheric atmospheric model, and the propagation path of radio waves is calculated using a forward ray tracing algorithm. The resulting spatial refractive index distribution data exhibits characteristics of variation with latitude and longitude and seasonality. It also considers the influence of atmospheric refractive index on rays, resulting in more accurate prediction of radio wave propagation paths. Therefore, it features small errors, accurate predictions, and good annual periodicity.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model using wavelet neural networks includes the following steps:

[0009] Step 1: Collect and build a historical radiosonde dataset with different latitude, longitude, and altitude information;

[0010] Step 2: Preprocess the historical radiosonde dataset obtained in Step 1 to obtain the preprocessed historical radiosonde dataset;

[0011] Step 3: Input the preprocessed historical sounding dataset obtained in Step 2 into a wavelet neural network to construct a three-dimensional atmospheric model of the troposphere, and calculate the refractive index distribution of the troposphere based on the three-dimensional atmospheric model of the troposphere;

[0012] Step 4: Set the coordinates of the transmitting and receiving antennas within the simulation range, the elevation angle of the transmitting antenna to 0°~90°, and the operating frequency to 30MHz or higher. Read the spatial terrain and calculate the conductivity and dielectric constant of the ground and sea surface based on the operating frequency.

[0013] Step 5: Based on the set transceiver antenna coordinates and transmit antenna elevation angle obtained in Step 4, use the forward ray tracing algorithm combined with the tropospheric atmospheric refractive index distribution data obtained in Step 3 to obtain the distance error and elevation angle error caused by tropospheric atmospheric refraction.

[0014] Step 6: Optimize the electromagnetic wave propagation path using the distance error and elevation angle error obtained in Step 5, and obtain the electromagnetic wave propagation path prediction results based on the three-dimensional atmospheric model of the troposphere established by the wavelet neural network.

[0015] The specific process of step 2 is as follows: Check the historical radiosonde dataset obtained in step 1 for outliers and radiosonde data with insufficient single-measurement data. Outliers refer to data falling outside the range [μ-a*σ, μ+a*σ], where μ represents the median of the current sequence, σ represents the median absolute deviation (MAD) of the current sequence, and a is 2-6. Insufficient single-measurement data refers to radiosonde data with fewer than 30 samples. These data are discarded or supplemented using surrounding data via cubic spline interpolation to obtain processed data. The dew point temperature of the processed data and the data that meets the requirements is converted into air humidity to obtain the preprocessed historical radiosonde dataset. The formula for converting dew point temperature into air humidity is:

[0016]

[0017] Convert to:

[0018]

[0019] In the formula, e represents water vapor pressure, with the unit being hectopascals (hPa); E0 represents the saturated water vapor pressure at 0℃, with a value of 6.1078 hPa; a is a coefficient, taken as 7.69; and b is a coefficient, taken as 243.92.

[0020] The structure of the three-dimensional atmospheric model of the troposphere in step 3 is as follows:

[0021] Wavelet neural network module: used to fit the preprocessed historical sounding dataset to obtain the tropospheric atmospheric calculation weight coefficients;

[0022] Empirical model: used to optimize the tropospheric atmospheric calculation weight coefficients output by the wavelet neural network module to obtain the optimized tropospheric atmospheric calculation weight coefficients, and to process the coefficients of the annual cycle variation curve to obtain the point distribution data of temperature, humidity and pressure points on the grid plane;

[0023] Mesh plane generation module: used to divide the simulation area into N*N mesh planes; and to spherically divide the area below the tropopause into vertical layers, each containing N*N mesh planes;

[0024] Tropospheric temperature, humidity and pressure calculation module: used to calculate the temperature, humidity and pressure data at each grid plane point, and obtain the point distribution data of temperature, humidity and pressure points on the grid plane;

[0025] Atmospheric refractive index calculation module: This module is used to calculate the point distribution data of temperature, humidity, and pressure points on the grid plane obtained from the tropospheric temperature, humidity, and pressure calculation module, and thus obtain the tropospheric atmospheric refractive index distribution data.

[0026] The specific process of step 3 is as follows:

[0027] Step 3.1: Use the wavelet neural network module to fit the preprocessed historical sounding dataset obtained in Step 2, and input the fitted data into the empirical model to obtain the optimized tropospheric atmospheric calculation weight coefficients;

[0028] Step 3.2: Input the latitude and longitude range and date information of the simulation area, read the radiosonde station data within the simulation area, and set the horizontal and vertical step size of the simulation;

[0029] Step 3.3: Using the mesh plane generation module, divide the simulation area into N*N mesh planes according to the simulation horizontal and vertical step sizes set in Step 3.2;

[0030] Step 3.4: Based on the radiosonde station data read in Step 3.2, the grid plane division module determines the lowest altitude corresponding to the atmospheric structure with an atmospheric pressure in the range of 500hPa to 70hPa, a temperature lapse rate of less than or equal to 2℃ / km, and an average temperature lapse rate within 2km above this altitude of no more than 2℃ / km. The corresponding lowest altitude is taken as the tropopause height. According to the simulation vertical step size set in Step 3.2, the region below the tropopause height is spherically divided to obtain the number of vertical division layers, each containing N*N grid planes.

[0031] Step 3.5: For the grid planes on different vertical layers generated in Step 3.3 and Step 3.4, the temperature, humidity and pressure data at each grid point are calculated using the tropospheric temperature, humidity and pressure calculation module to obtain the point distribution data of temperature, humidity and pressure points on the grid plane;

[0032] Step 3.6: Using the atmospheric refractive index calculation module, the point distribution data of temperature, humidity, and pressure points obtained in Step 3.5 are used to calculate the surface distribution data of temperature, humidity, and pressure points using the cubic spline interpolation method. The obtained temperature, humidity, and pressure point data are then used to calculate the tropospheric atmospheric refractive index distribution data using the refractive index calculation formula.

[0033] The specific process of step 3.1 is as follows:

[0034] Step 3.1.1: Use the temperature, humidity, and pressure data in the preprocessed historical sounding dataset obtained in Step 2 as input parameters for the input layer to obtain the temperature weighting coefficient, humidity weighting coefficient, and pressure weighting coefficient that change over time;

[0035] Step 3.1.2: Use the temperature weighting coefficient, humidity weighting coefficient, and pressure weighting coefficient obtained in Step 3.1.1 as input parameters for the hidden layer, and input them into the hidden layer to obtain the corrected temperature weighting coefficient, humidity weighting coefficient, and pressure weighting coefficient; the hidden layer output calculation formula is:

[0036]

[0037] In the formula, h(j) is the output value of the j-th node in the hidden layer; ω ij b represents the connection weights between the input layer and the hidden layer. j h is the translation factor of the wavelet basis function; j a is the scaling factor of the wavelet basis functions; j These are wavelet basis functions;

[0038] The wavelet basis function a j The Morlet mother wavelet basis function is expressed as:

[0039]

[0040] Step 3.1.3: Input the corrected temperature weight coefficients, humidity weight coefficients, and pressure weight coefficients obtained in Step 3.1.2 into the output layer to obtain the output layer weights. The calculation formula for the wavelet neural network output layer is as follows:

[0041]

[0042] In the formula, ω jkdenoted as the weights from the hidden layer to the output layer; h(i) is the output of the i-th hidden layer node; l is the number of nodes in the hidden layer; m is the number of nodes in the output layer;

[0043] Step 3.1.4: Calculate the output layer weights obtained in Step 3.1.3 according to the time series to be predicted, obtain the prediction curve corresponding to the time series, and then perform least squares fitting on the prediction curve to obtain the optimized tropospheric atmospheric calculation weight coefficients.

[0044] The specific process of step 3.5 is as follows: First, determine whether the grid point in the grid plane belongs to the area included by the radiosonde station in the simulation area. If so, calculate the temperature, humidity, and pressure point distribution data on the grid point using the optimized tropospheric atmospheric calculation weight coefficient obtained in step 3.1 based on the latitude, longitude, and altitude information of the grid point. If not, calculate the Julian day based on the simulation year, month, and day to obtain the corresponding Julian day information, which is used to obtain the coefficient of the annual cycle variation curve. Based on this coefficient, apply Hopfield, GPT3, and the piecewise model to obtain the point distribution data of temperature, humidity, and pressure points on the grid plane.

[0045] The specific process of step 5 is as follows: Based on the coordinates of the transmitting and receiving antennas and the elevation angle of the transmitting antenna set in step 4, the forward ray tracing algorithm is used. When the ray passes through grid blocks with different refractive indices, the incident angle and exit angle of the ray on the block interface are calculated based on the tropospheric atmospheric refractive index distribution data obtained in step 3. The propagation path of the radio wave is optimized to obtain the distance error and elevation angle error caused by refraction.

[0046] The distance error caused by refraction is:

[0047] ΔR=R g -R0

[0048] The elevation angle error caused by refraction is:

[0049] ε=θ0-a0

[0050] In the formula, R0 represents the straight-line distance from the station to the target; R g θ0 represents the spatial ray trajectory (actual geometric distance) from the station to the target; θ0 represents the apparent elevation angle; a0 represents the target's true elevation angle.

[0051] A radio wave propagation path prediction system based on a wavelet neural network to establish a three-dimensional tropospheric atmospheric model includes:

[0052] Data preprocessing module: Collects and builds historical radiosonde datasets, preprocesses the collected and built radiosonde datasets to obtain preprocessed historical radiosonde datasets;

[0053] Three-dimensional tropospheric atmospheric model: The preprocessed historical sounding dataset is input into the three-dimensional tropospheric atmospheric model to obtain the tropospheric atmospheric refractive index distribution;

[0054] Simulation parameter setting module: used to set the coordinates of the transmitting and receiving antennas, the elevation angle of the transmitting antenna, the operating frequency, read the spatial terrain, and calculate the conductivity and dielectric constant of the ground and sea surface based on the operating frequency;

[0055] Tropospheric radio wave propagation path calculation module: Based on the set coordinates of the transmitting and receiving antennas and the elevation angle of the transmitting antenna, the forward ray tracing algorithm is used in conjunction with the refractive index distribution data of the troposphere to obtain the distance error and elevation angle error caused by refraction. Based on the obtained distance error and elevation angle error, the electromagnetic wave ray propagation path is optimized to obtain the radio wave propagation path prediction result based on the three-dimensional atmospheric model of the troposphere established by the wavelet neural network.

[0056] A device for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model using wavelet neural networks includes:

[0057] Memory: Used to store the computer program that implements the method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model using a wavelet neural network;

[0058] A processor is used to execute the computer program to implement the method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model established by a wavelet neural network.

[0059] A computer-readable storage medium storing a computer program that, when executed by a processor, enables the method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model established using a wavelet neural network.

[0060] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0061] 1. Taking into account tropospheric temperature, humidity, and pressure data with varying latitudes, longitudes, and time, the tropospheric atmosphere is spatially stratified and partitioned. Wavelet neural networks are used to fit the regional temperature, humidity, and pressure data. Combined with existing empirical models, such as the Hopfield model, GPT3 model, and piecewise model, the refractive index of the tropospheric atmosphere is simulated. The output tropospheric atmospheric model incorporates latitude, longitude, and annual cycle variation factors, thus exhibiting accurate prediction, small error, and good annual cycle characteristics.

[0062] 2. Using the forward ray tracing method, based on the tropospheric refractive index data output by the constructed tropospheric atmospheric model, the propagation path of radio waves can be predicted. Considering the distance error and elevation angle error caused by atmospheric refraction, the influence of atmospheric refraction on radio wave propagation can be taken into account, which can more accurately estimate the signal transmission performance, optimize antenna configuration and wireless network planning, and improve the accuracy of prediction.

[0063] In summary, this invention, by considering historical radiosonde datasets from radiosonde stations distributed across various latitudes and longitudes, establishes a three-dimensional tropospheric atmospheric model based on wavelet neural networks, taking into account different latitudes, longitudes, altitudes, and date information. The tropospheric atmospheric refractive index distribution is calculated using this model, and the propagation path of radio waves is calculated using a forward ray tracing algorithm. The resulting spatial refractive index distribution data exhibits variations with latitude and longitude and seasonality. Furthermore, it considers the influence of atmospheric refractive index on radio waves, leading to more accurate predictions of radio wave propagation paths. Therefore, it features small errors, accurate predictions, and good annual periodicity. Attached Figure Description

[0064] Figure 1 This is a flowchart of a method for predicting radio wave propagation paths based on a wavelet neural network to establish a three-dimensional atmospheric model of the troposphere, provided in an embodiment of the present invention.

[0065] Figure 2 This is a schematic diagram of the locations of domestic sounding stations provided in an embodiment of the present invention.

[0066] Figure 3 This is a schematic diagram of the wavelet neural network topology in step 2 of the present invention.

[0067] Figure 4 This is a schematic diagram showing the comparison error between the optimized tropospheric atmospheric calculation model and the data from the Hanzhong sounding station in step 3 of the present invention.

[0068] Figure 5 The figures provided in this embodiment of the invention are root mean square error (RMSE) comparisons of January data from 85 radiosonde stations across the country, using a wavelet neural network-optimized model. Figure (a) is the RMSE comparison of January air pressure deviation data, Figure (b) is the RMSE comparison of January temperature deviation data, and Figure (c) is the RMSE comparison of January water vapor deviation data.

[0069] Figure 6 These are simulation area refractive index distribution and refractive index profile between two points provided in the embodiments of the present invention. Figure (a) is an atmospheric refractive index distribution diagram, and Figure (b) is a vertical refractive index profile between two points.

[0070] Figure 7 This is a map showing the distribution of atmospheric refractive index across the country in January, provided in an embodiment of the present invention.

[0071] Figure 8 These are distance and elevation errors caused by atmospheric refraction provided in an embodiment of the present invention. Figure (a) shows the distance error caused by atmospheric refraction, and Figure (b) shows the elevation error caused by atmospheric refraction.

[0072] Figure 9 This is a schematic diagram comparing the radio wave propagation paths calculated using tropospheric atmospheric data constructed using this invention and tropospheric atmospheric data provided in ITU-RP453 Recommendation. Detailed Implementation

[0073] The present invention will now be described in detail with reference to the embodiments.

[0074] See Figure 1 A method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model using wavelet neural networks includes the following steps:

[0075] Step 1: Collect and establish historical radiosonde datasets with different latitude, longitude, and altitude information; the historical radiosonde datasets include data from domestic radiosonde stations and global radiosonde data. Global radiosonde data can be obtained by downloading upper-air data from the website of the Department of Meteorological Sciences at the University of Wyoming and the China Meteorological Data Network; the data from domestic radiosonde stations includes historical data from 88 domestic radiosonde stations.

[0076] See Figure 2 Using historical data from 88 domestic radiosonde stations as the primary dataset, the empirical models were first preliminarily validated. Figure 2 The dots in the diagram represent the distribution of various radiosonde stations within China, and the shade of color of each dot represents the altitude of that station. These radiosonde stations are an important source of real training data for analyzing the tropospheric atmospheric environment.

[0077] Step 2: Preprocess the historical radiosonde dataset obtained in Step 1 to obtain a preprocessed historical radiosonde dataset. The specific process of Step 2 is as follows: Check the historical radiosonde dataset obtained in Step 1 for outliers and radiosonde data with insufficient data from a single measurement. Outliers refer to values ​​falling outside the range [μ-a*σ, μ+a*σ], where μ represents the median of the current sequence, σ represents the median absolute deviation (MAD) of the current sequence, and a is 2-6. Insufficient data from a single measurement refers to radiosonde data with fewer than 30 samples. These data are discarded or supplemented using surrounding data via cubic spline interpolation to obtain processed data. The dew point temperature of the processed data and the data that meets the requirements is converted into air humidity to obtain the preprocessed historical radiosonde dataset. The formula for converting dew point temperature into air humidity is:

[0078]

[0079] Convert to:

[0080]

[0081] In the formula, e represents the water vapor pressure, with the unit being hectopascals (hPa); E0 represents the saturated water vapor pressure at 0℃, with a value of 6.1078 hPa; a is a coefficient, taken as 7.69; b is a coefficient, taken as 243.92;

[0082] Simulations were conducted in an area with a relatively high density of radiosonde stations (latitude: 30°N~37°N / longitude: 106.6°E~117.6°E). Atmospheric parameters within the edge contours of the radiosonde stations were derived using cubic spline interpolation, while atmospheric parameters outside the edge contours were supplemented using empirical models. Two arbitrary points were selected within the modeling area, with the starting point coordinates at (30.98°N, 107.08°E) and the ending point coordinates at (33.63°N, 113.51°E), spanning a distance of approximately 700 kilometers. The mesh generation accuracy was 0.1°.

[0083] Step 3: Input the preprocessed historical sounding dataset obtained in Step 2 into a wavelet neural network to construct a three-dimensional tropospheric atmospheric model, and calculate the tropospheric atmospheric refractive index distribution based on the three-dimensional tropospheric atmospheric model; the structure of the three-dimensional tropospheric atmospheric model in Step 3 is as follows:

[0084] Wavelet neural network module: used to fit the preprocessed historical sounding dataset to obtain the tropospheric atmospheric calculation weight coefficients;

[0085] Empirical model: used to optimize the tropospheric atmospheric calculation weight coefficients output by the wavelet neural network module to obtain the optimized tropospheric atmospheric calculation weight coefficients, and to process the coefficients of the annual cycle variation curve to obtain the point distribution data of temperature, humidity and pressure points on the grid plane;

[0086] Mesh plane generation module: used to divide the simulation area into N*N mesh planes; and to spherically divide the area below the tropopause into vertical layers, each containing N*N mesh planes;

[0087] Tropospheric temperature, humidity and pressure calculation module: used to calculate the temperature, humidity and pressure data at each grid plane point, and obtain the point distribution data of temperature, humidity and pressure points on the grid plane;

[0088] Atmospheric refractive index calculation module: used to calculate the point distribution data of temperature, humidity, and pressure points on the grid plane obtained by the tropospheric temperature, humidity, and pressure calculation module, and obtain the tropospheric atmospheric refractive index distribution data;

[0089] The Wavelet Neural Network (WNN) algorithm was used to construct the three-dimensional tropospheric atmospheric model. WNN is a hybrid network model combining wavelet analysis and backpropagation (BP) neural networks. Wavelet analysis addresses the limitation of Fourier transform in obtaining time-domain information. Its basic principle is to generate a waveform of finite length with an average value of 0, called the wavelet basis function. This waveform is then shifted and multiplied with the input signal, decomposing the signal into a superposition of wavelet functions. This allows the time series to be decomposed into multiple subsequences of different frequencies, each corresponding to a different time scale. Wavelet analysis can extract local features of the time series, providing information for predicting its future trends.

[0090] The specific process of step 3 is as follows:

[0091] Step 3.1: Use the wavelet neural network module to fit the preprocessed historical sounding dataset obtained in Step 2, and input the fitted data into the empirical model to obtain the optimized tropospheric atmospheric calculation weight coefficients;

[0092] The specific process of step 3.1 is as follows:

[0093] See Figure 3 Step 3.1.1: Use the temperature, humidity, and pressure data from the preprocessed historical radiosonde dataset obtained in Step 2 from each radiosonde station from 2017 to 2019 as input parameters for the input layer to obtain the temperature weighting coefficient, humidity weighting coefficient, and pressure weighting coefficient that change over time.

[0094] Step 3.1.2: Use the temperature weighting coefficient, humidity weighting coefficient, and pressure weighting coefficient obtained in Step 3.1.1 as input parameters for the hidden layer, and input them into the hidden layer to obtain the corrected temperature weighting coefficient, humidity weighting coefficient, and pressure weighting coefficient; the hidden layer output calculation formula is:

[0095]

[0096] In the formula, h(j) is the output value of the j-th node in the hidden layer; ω ij b represents the connection weights between the input layer and the hidden layer. j h is the translation factor of the wavelet basis function; j a is the scaling factor of the wavelet basis functions; j These are wavelet basis functions;

[0097] The wavelet basis function a j The Morlet mother wavelet basis function is expressed as:

[0098]

[0099] Step 3.1.3: Input the corrected temperature weight coefficients, humidity weight coefficients, and pressure weight coefficients obtained in Step 3.1.2 into the output layer to obtain the output layer weights. The calculation formula for the wavelet neural network output layer is as follows:

[0100]

[0101] In the formula, ω jk denoted as the weights from the hidden layer to the output layer; h(i) is the output of the i-th hidden layer node; l is the number of nodes in the hidden layer; m is the number of nodes in the output layer;

[0102] Step 3.1.4: Calculate the output layer weights obtained in Step 3.1.3 according to the time series to be predicted, obtain the prediction curve corresponding to the time series, and then perform least squares fitting on the prediction curve to obtain the optimized tropospheric atmospheric calculation weight coefficients.

[0103] Step 3.2: Input the latitude and longitude range and date information of the simulation area, read the radiosonde station data within the simulation area, and set the horizontal and vertical step size of the simulation;

[0104] Step 3.3: Using the mesh plane generation module, divide the simulation area into N*N mesh planes according to the simulation horizontal and vertical step sizes set in Step 3.2;

[0105] Step 3.4: Based on the radiosonde station data read in Step 3.2, the grid plane division module determines the lowest altitude corresponding to the atmospheric structure with an atmospheric pressure in the range of 500hPa to 70hPa, a temperature lapse rate of less than or equal to 2℃ / km, and an average temperature lapse rate within 2km above this altitude of no more than 2℃ / km. The corresponding lowest altitude is taken as the tropopause height. According to the simulation vertical step size set in Step 3.2, the region below the tropopause height is spherically divided to obtain the number of vertical division layers, each containing N*N grid planes.

[0106] Step 3.5: For the grid planes on different vertical layers generated in Steps 3.3 and 3.4, the temperature, humidity, and pressure data at each grid point are calculated using the tropospheric temperature, humidity, and pressure calculation module to obtain the point distribution data of temperature, humidity, and pressure points on the grid plane. The specific process of Step 3.5 is as follows: First, it is determined whether the grid point in the grid plane belongs to the area included by the radiosonde station in the simulation area. If so, the temperature, humidity, and pressure point distribution data on the grid point are calculated using the optimized tropospheric atmospheric calculation weight coefficients obtained in Step 3.1 based on the latitude, longitude, and altitude information of the grid point. If not, the Julian day is calculated based on the simulation year, month, and day to obtain the corresponding Julian day information, which is used to obtain the coefficients of the annual cycle variation curve. Based on these coefficients, Hopfield, GPT3, and the piecewise model are applied to obtain the point distribution data of temperature, humidity, and pressure points on the grid plane.

[0107] Step 3.6: Using the atmospheric refractive index calculation module, the point distribution data of temperature, humidity, and pressure obtained in Step 3.5 are used to calculate the area distribution data of temperature, humidity, and pressure using the cubic spline interpolation method. The obtained temperature, humidity, and pressure point data are then used to calculate the tropospheric atmospheric refractive index distribution data using the refractive index calculation formula. For microwave and lower frequency bands, the atmospheric refractive index n and atmospheric refractive index N can be determined by meteorological parameters such as temperature, humidity, and pressure. The formula for calculating the atmospheric refractive index N is:

[0108]

[0109] Calculate the atmospheric refractive index:

[0110] n = N × 10 -6 +1

[0111] In the formula, p is atmospheric pressure, in hPa; T is atmospheric thermodynamic temperature, in K; and e is the partial pressure of water vapor, in hPa.

[0112]

[0113]

[0114] In the formula, R h The relative humidity of the atmosphere is generally between 250 and 400 N.

[0115] To treat the Earth's surface as a flat plane, making it easier to evaluate the atmospheric refractive index gradient and its impact on radio wave propagation, a corrected refractive index quantity is defined as a substitute for the atmospheric refractive index, commonly used for correcting for the Earth's curvature. Its relationship with the atmospheric refractive index N is as follows:

[0116]

[0117] In the formula, r e is the average Earth radius; h is the altitude; M is in M ​​units.

[0118] Step 4: Set the coordinates of the transmitting and receiving antennas within the simulation range, the elevation angle of the transmitting antenna to 0°~90°, and the operating frequency to above 30MHz. Read the spatial terrain and calculate the conductivity and dielectric constant of the ground and sea surface based on the operating frequency.

[0119] Step 5: Based on the transmit and receive antenna coordinates and transmit antenna elevation angle set in Step 4, the forward ray tracing algorithm is used in conjunction with the tropospheric atmospheric refractive index distribution data obtained in Step 3 to obtain the distance error and elevation angle error caused by tropospheric atmospheric refraction. The specific process of Step 5 is as follows: Based on the transmit and receive antenna coordinates and transmit antenna elevation angle set in Step 4, the forward ray tracing algorithm is used. The forward ray tracing algorithm starts from the transmit antenna coordinates and faces the receive antenna. The electromagnetic wave ray is emitted at the transmit antenna elevation angle, and the path points (x, y, z) of the ray are recorded. When the ray passes through grid blocks with different refractive indices, the incident angle and exit angle of the ray at the block interface are calculated based on the tropospheric atmospheric refractive index distribution data obtained in Step 3. The propagation path of the radio wave is optimized, and the distance error and elevation angle error caused by tropospheric atmospheric refraction are obtained.

[0120] The distance error caused by refraction is:

[0121] ΔR=R g -R0

[0122] The elevation angle error caused by refraction is:

[0123] ε=θ0-a0

[0124] In the formula, R0 represents the straight-line distance from the station to the target; R g θ0 represents the spatial ray trajectory (actual geometric distance) from the station to the target; θ0 represents the apparent elevation angle; a0 represents the target's true elevation angle.

[0125] Step 6: Optimize the electromagnetic wave propagation path using the distance error and elevation angle error obtained in Step 5, and obtain the electromagnetic wave propagation path prediction results based on the three-dimensional atmospheric model of the troposphere established by the wavelet neural network.

[0126] A radio wave propagation path prediction system based on a wavelet neural network to establish a three-dimensional tropospheric atmospheric model includes:

[0127] Data preprocessing module: Collects and builds historical radiosonde datasets, preprocesses the collected and built radiosonde datasets to obtain preprocessed historical radiosonde datasets;

[0128] Three-dimensional tropospheric atmospheric model: The preprocessed historical sounding dataset is input into the three-dimensional tropospheric atmospheric model to obtain the tropospheric atmospheric refractive index distribution;

[0129] Simulation parameter setting module: used to set the coordinates of the transmitting and receiving antennas, the elevation angle of the transmitting antenna, the operating frequency, read the spatial terrain, and calculate the conductivity and dielectric constant of the ground and sea surface based on the operating frequency;

[0130] Tropospheric radio wave propagation path calculation module: Based on the set coordinates of the transmitting and receiving antennas and the elevation angle of the transmitting antenna, the forward ray tracing algorithm is used in conjunction with the refractive index distribution data of the troposphere to obtain the distance error and elevation angle error caused by refraction. Based on the obtained distance error and elevation angle error, the electromagnetic wave ray propagation path is optimized to obtain the radio wave propagation path prediction result based on the three-dimensional atmospheric model of the troposphere established by the wavelet neural network.

[0131] A device for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model using wavelet neural networks includes:

[0132] Memory: Used to store the computer program that implements the method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model built using a wavelet neural network;

[0133] A processor is used to execute the computer program to implement the method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model established by a wavelet neural network.

[0134] A computer-readable storage medium storing a computer program that, when executed by a processor, enables the method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model established using a wavelet neural network.

[0135] See Figure 4 The horizontal axis represents different vertical altitudes, the vertical axis represents the temperature data at that altitude, the black triangular curve represents the measured data of the radiosonde station, the red circle curve represents the temperature vertical variation curve at the latitude and longitude of the station simulated by the GPT3 model, and the blue asterisk curve represents the temperature vertical variation curve at the latitude and longitude of the station simulated by the tropospheric three-dimensional atmospheric model constructed in this invention. The calculated RMSE between the GPT3 model and the measured data is 7.4768, and the RMSE between the optimized model and the measured data is 2.9564, indicating that the accuracy of the tropospheric three-dimensional atmospheric model constructed in this invention is higher than that of the existing GPT3 model.

[0136]

[0137] As shown in the table above, it can be seen that in the unoptimized model, the average RMSE of atmospheric pressure / hPa is 8.6197, while in the optimized model it is 7.3081. In the unoptimized model, the average RMSE of temperature / ℃ is 8.4265, while in the optimized model it is 3.8788. In the unoptimized model, the average RMSE of water vapor pressure / hPa is 1.7068, while in the optimized model it is 1.3822. Therefore, it can be seen that the accuracy of the optimized model, which is the tropospheric three-dimensional atmospheric model constructed in this invention, is significantly improved.

[0138] See Figure 5 Figure (a) shows the pressure deviation data for January, from which the root mean square error (RMSE) can be obtained. The RMSE for January pressure decreased from 8.6197 to 7.3081, improving accuracy by 15%. Figure (b) shows the pressure deviation data for January temperature, from which the RMSE for January temperature decreased from 8.4263 to 3.8788, improving accuracy by 54%. Figure (c) shows the water vapor deviation data for January, from which the RMSE for January water vapor decreased from 1.7068 to 1.3822, improving accuracy by 19%.

[0139] Figure 6 Figure 1 shows the refractive index distribution of the simulated region and the refractive index profile between two points. Figure 2 shows the atmospheric refractive index distribution, with the horizontal axis representing the longitude range of the simulated region and the vertical axis representing the latitude range. Different colors represent the tropospheric refractive index distribution at the current altitude. A black horizontal line in the figure represents the refractive index profile between these two points, which is shown in Figure 3 on the right. Figure 4 shows the vertical refractive index profile between two points, with the horizontal axis representing the distance between the two points and the vertical axis representing the altitude. It can be seen that the atmospheric refractive index changes drastically below 2000m, gradually stabilizing and decreasing with increasing altitude. Figure 5 (a) Figure 5 (b) As can be seen, the three-dimensional tropospheric atmospheric model constructed by the present invention can fully construct the tropospheric atmospheric distribution data in the three-dimensional space of the simulation area.

[0140] The influence of atmospheric refraction was determined by using monthly average atmospheric parameter data from 86 radiosonde stations across the country in January 2019 as input, and interpolating with an empirical tropospheric model to obtain the surface atmospheric refractive index distribution for the entire January range.

[0141] See Figure 7The blue area in the image (between 26° and 39° north latitude and 73° and 104° east longitude) represents the Qinghai-Tibet Plateau region. The Qinghai-Tibet Plateau region has a high altitude and is characterized by low air pressure, dry air, and low average temperature. The atmospheric refractive index in the region is also the lowest, at approximately 230 N units. The red area in the image (between 23° and 26° north latitude and 119° and 113° east longitude) represents the coastal areas of Guangdong Province and Hong Kong Special Administrative Region in my country. The coastal areas of Guangdong Province and Hong Kong Special Administrative Region have a lower average altitude, higher temperature, and more frequent precipitation, resulting in the highest refractive index, at approximately 300 N units.

[0142] For beyond-line-of-sight radio wave propagation, the effect of the Earth's curvature on the propagation process needs to be considered; therefore, spherical layering theory is required for atmospheric stratification. The distance and elevation angle errors caused by atmospheric exhaust are simulated using a ray tracing algorithm. Figure 8 As shown.

[0143] See Figure 8 Figure (a) shows the distance error caused by atmospheric refraction. The horizontal axis represents the altitude distance in the vertical direction of electromagnetic wave propagation, and the vertical axis represents the distance error caused by atmospheric refraction. Each curve represents a different antenna transmission angle (simulated angles are 1°, 2°, 3°, 5°, 10°, 15°, 20°, 40°, and 80°). It can be seen that different antenna elevation angles have a significant impact on the distance error. Figure (b) shows the elevation angle error caused by atmospheric refraction. The horizontal axis represents the altitude distance in the vertical direction of electromagnetic wave propagation, and the vertical axis represents the distance error caused by atmospheric refraction. The simulated angle of each curve is the same as that in Figure (a). It can be seen that the elevation angle error caused by atmospheric refraction increases continuously with the increase of propagation distance.

[0144] The path of the ray in the simulation area is simulated using the forward ray tracing method, with the simulation parameters set as follows:

[0145]

[0146]

[0147] Using a digital elevation map of Yan'an City as ground input, the radio wave propagation paths obtained from the simulated tropospheric atmospheric 3D model are compared with those obtained using data from ITU-R Recommendation P453. The results are as follows: Figure 9 As shown.

[0148] See Figure 9In the figure, the x-axis represents the longitude range of Yan'an City, the y-axis represents the latitude range of Yan'an City, and the z-axis represents the vertical altitude. The black dashed line is the ray propagation path calculated by the tropospheric three-dimensional atmospheric model built based on this model, and the red solid line is the ray propagation path calculated by the tropospheric atmospheric model based on the ITU-R Recommendation P453. It can be seen that different atmospheric models cause different refractive index distributions, which have a significant impact on the propagation path of radio waves.

[0149] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model established using wavelet neural networks, characterized in that... Includes the following steps: Step 1: Collect and build a historical radiosonde dataset with different latitude, longitude, and altitude information; Step 2: Preprocess the historical radiosonde dataset obtained in Step 1 to obtain the preprocessed historical radiosonde dataset; Step 3: Input the preprocessed historical sounding dataset obtained in Step 2 into a wavelet neural network to construct a three-dimensional atmospheric model of the troposphere, and calculate the refractive index distribution of the troposphere based on the three-dimensional atmospheric model of the troposphere; Step 4: Set the coordinates of the transmitting and receiving antennas within the simulation range, the elevation angle of the transmitting antenna to 0°~90°, and the operating frequency to 30MHz or higher. Read the spatial terrain and calculate the conductivity and dielectric constant of the ground and sea surface based on the operating frequency. Step 5: Based on the set transceiver antenna coordinates and transmit antenna elevation angle obtained in Step 4, use the forward ray tracing algorithm combined with the tropospheric atmospheric refractive index distribution data obtained in Step 3 to obtain the distance error and elevation angle error caused by tropospheric atmospheric refraction. Step 6: Optimize the electromagnetic wave propagation path using the distance error and elevation angle error obtained in Step 5, and obtain the electromagnetic wave propagation path prediction results based on the three-dimensional atmospheric model of the troposphere established by the wavelet neural network; The structure of the three-dimensional atmospheric model of the troposphere in step 3 is as follows: Wavelet neural network module: used to fit the preprocessed historical sounding dataset to obtain the tropospheric atmospheric calculation weight coefficients; Empirical model: used to optimize the tropospheric atmospheric calculation weight coefficients output by the wavelet neural network module, to obtain the optimized tropospheric atmospheric calculation weight coefficients, and to process the coefficients of the annual cycle variation curve to obtain the point distribution data of temperature, humidity, and pressure points in the wavelet neural network plane; Mesh plane generation module: used to divide the simulation area into N*N mesh planes; and to spherically divide the area below the tropopause into vertical layers, each containing N*N mesh planes; Tropospheric temperature, humidity and pressure calculation module: used to calculate the temperature, humidity and pressure data at each grid plane point, and obtain the point distribution data of temperature, humidity and pressure points on the grid plane; Atmospheric refractive index calculation module: This module is used to calculate the point distribution data of temperature, humidity, and pressure points on the grid plane obtained from the tropospheric temperature, humidity, and pressure calculation module, and thus obtain the tropospheric atmospheric refractive index distribution data.

2. The method for predicting radio wave propagation paths based on a wavelet neural network to establish a three-dimensional tropospheric atmospheric model according to claim 1, characterized in that, The specific process of step 2 is as follows: check whether the historical radiosonde dataset obtained in step 1 contains outliers or insufficient data from a single measurement; outliers refer to data falling within... Values ​​outside the range, μ represents the median of the current sequence, σ represents the median absolute deviation (MAD) of the current sequence, and a is 2-6; insufficient data from a single measurement refers to radiosonde data with fewer than 30 samples. These data are discarded or supplemented using surrounding data through cubic spline interpolation to obtain processed data. The dew point temperature of the processed data and the data that meets the requirements are converted into air humidity to obtain the preprocessed historical radiosonde dataset; the formula for converting dew point temperature into air humidity is: Convert to: In the formula, e represents the water vapor pressure, and the unit is hectopascals; This represents the saturated water vapor pressure at 0℃, with a value of 6.1078 hPa. a The coefficient is set to 7.69; b The coefficient is 243.

92.

3. The method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model established using a wavelet neural network according to claim 1, characterized in that, The specific process of step 3 is as follows: Step 3.1: Use the wavelet neural network module to fit the preprocessed historical sounding dataset obtained in Step 2, and input the fitted data into the empirical model to obtain the optimized tropospheric atmospheric calculation weight coefficients; Step 3.2: Input the latitude and longitude range and date information of the simulation area, read the radiosonde station data within the simulation area, and set the horizontal and vertical step size of the simulation; Step 3.3: Using the mesh plane generation module, divide the simulation area into N*N mesh planes according to the simulation horizontal and vertical step sizes set in Step 3.2; Step 3.4: Based on the radiosonde station data read in Step 3.2, the grid plane division module determines the lowest altitude corresponding to the atmospheric structure with an atmospheric pressure between 500 hPa and 70 hPa, a temperature lapse rate less than or equal to 2℃ / km, and an average temperature lapse rate within 2km above the location not greater than 2℃ / km. The corresponding lowest altitude is taken as the tropopause height. According to the simulation vertical step size set in Step 3.2, the region below the tropopause height is spherically divided to obtain the number of vertical division layers, each containing N*N grid planes. Step 3.5: For the grid planes on different vertical layers generated in Step 3.3 and Step 3.4, the temperature, humidity and pressure data at each grid point are calculated using the tropospheric temperature, humidity and pressure calculation module to obtain the point distribution data of temperature, humidity and pressure points on the grid plane; Step 3.6: Using the atmospheric refractive index calculation module, the point distribution data of temperature, humidity, and pressure points obtained in Step 3.5 are used to calculate the surface distribution data of temperature, humidity, and pressure points using the cubic spline interpolation method. The obtained temperature, humidity, and pressure point data are then used to calculate the tropospheric atmospheric refractive index distribution data using the refractive index calculation formula.

4. The method for predicting radio wave propagation paths based on a wavelet neural network to establish a three-dimensional tropospheric atmospheric model according to claim 3, characterized in that, The specific process of step 3.1 is as follows: Step 3.1.1: Use the temperature, humidity, and pressure data in the preprocessed historical sounding dataset obtained in Step 2 as input parameters for the input layer to obtain the temperature weighting coefficient, humidity weighting coefficient, and pressure weighting coefficient that change over time; Step 3.1.2: Use the temperature weighting coefficient, humidity weighting coefficient, and pressure weighting coefficient obtained in Step 3.1.1 as input parameters for the hidden layer, and input them into the hidden layer to obtain the corrected temperature weighting coefficient, humidity weighting coefficient, and pressure weighting coefficient; the hidden layer output calculation formula is: In the formula, For the hidden layer j Each node outputs its value; The connection weights between the input layer and the hidden layer; is the translation factor of the wavelet basis function; is the scaling factor of the wavelet basis functions; These are wavelet basis functions; The wavelet basis functions The Morlet mother wavelet basis function is expressed as: Step 3.1.3: Input the corrected temperature weight coefficients, humidity weight coefficients, and pressure weight coefficients obtained in Step 3.1.2 into the output layer to obtain the output layer weights. The calculation formula for the wavelet neural network output layer is as follows: In the formula, Weights from hidden layer to output layer; For the first i The output of each hidden layer node; l This represents the number of nodes in the hidden layer. m This represents the number of nodes in the output layer. Step 3.1.4: Calculate the output layer weights obtained in Step 3.1.3 according to the time series to be predicted, obtain the prediction curve corresponding to the time series, and then perform least squares fitting on the prediction curve to obtain the optimized tropospheric atmospheric calculation weight coefficients.

5. The method for predicting radio wave propagation paths based on a wavelet neural network to establish a three-dimensional tropospheric atmospheric model according to claim 3, characterized in that, The specific process of step 3.5 is as follows: First, determine whether the grid point in the grid plane belongs to the area included by the radiosonde station in the simulation area. If so, calculate the temperature, humidity, and pressure point distribution data of the grid point using the optimized tropospheric atmospheric calculation weight coefficient obtained in step 3.1 based on the latitude, longitude, and altitude information of the grid point. If not, calculate the Julian day based on the simulation year, month, and day to obtain the corresponding Julian day information, which is used to obtain the coefficient of the annual cycle variation curve. Based on this coefficient, apply Hopfield, GPT3, and the piecewise model to obtain the point distribution data of temperature, humidity, and pressure points in the grid plane.

6. The method for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model established using a wavelet neural network according to claim 1, characterized in that, The specific process of step 5 is as follows: Based on the coordinates of the transmitting and receiving antennas and the elevation angle of the transmitting antenna set in step 4, the forward ray tracing algorithm is used. When the ray passes through grid blocks with different refractive indices, the incident angle and exit angle of the ray on the block interface are calculated based on the tropospheric atmospheric refractive index distribution data obtained in step 3. The propagation path of the radio wave is optimized to obtain the distance error and elevation angle error caused by refraction. The distance error caused by refraction is: The elevation angle error caused by refraction is: In the formula, This represents the straight-line distance from the station to the target. Indicates the spatial ray trajectory from the station to the target; Indicates the angle of elevation; Indicates the target's true angle of elevation.

7. A radio wave propagation path prediction system based on a wavelet neural network to establish a three-dimensional tropospheric atmospheric model, used to implement the method described in claim 1, characterized in that, include: Data preprocessing module: Collects and builds historical radiosonde datasets, preprocesses the collected and built radiosonde datasets to obtain preprocessed historical radiosonde datasets; Three-dimensional tropospheric atmospheric model: The preprocessed historical sounding dataset is input into the three-dimensional tropospheric atmospheric model to obtain the tropospheric atmospheric refractive index distribution; Simulation parameter setting module: used to set the coordinates of the transmitting and receiving antennas, the elevation angle of the transmitting antenna, the operating frequency, read the spatial terrain, and calculate the conductivity and dielectric constant of the ground and sea surface based on the operating frequency; Tropospheric radio wave propagation path calculation module: Based on the set coordinates of the transmitting and receiving antennas and the elevation angle of the transmitting antenna, the forward ray tracing algorithm is used in conjunction with the refractive index distribution data of the troposphere to obtain the distance error and elevation angle error caused by refraction. Based on the obtained distance error and elevation angle error, the electromagnetic wave ray propagation path is optimized to obtain the radio wave propagation path prediction result based on the three-dimensional atmospheric model of the troposphere established by the wavelet neural network.

8. A device for predicting radio wave propagation paths based on a three-dimensional tropospheric atmospheric model established using wavelet neural networks, characterized in that, include: Memory: Used to store the computer program for implementing the method for predicting radio wave propagation paths based on a wavelet neural network to establish a three-dimensional atmospheric model of the troposphere as described in any one of claims 1-6; A processor, configured to execute the computer program to implement the radio wave propagation path prediction method based on a wavelet neural network for establishing a three-dimensional atmospheric model of the troposphere as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the radio wave propagation path prediction method based on a wavelet neural network for establishing a three-dimensional atmospheric model of the troposphere, as described in any one of claims 1-6.

Images