Low earth orbit satellite channel prediction method based on hierarchical collaborative optimization-random vector functional link network

Abstract

The low-orbit satellite channel prediction method based on hierarchical collaborative optimization-random vector function link network belongs to the technical field of wireless communication.The present application solves the problem of low prediction accuracy of channel state information due to long transmission distance between satellite and ground.The hierarchical collaborative optimization algorithm is introduced into the parameter optimization process of the random vector function link network to realize the global optimization configuration of the weights and biases between the input layer and the hidden layer of the random vector function link network;then the trained random vector function link network is used to predict the channel state information at the future time with high precision.The method can improve the effectiveness and reliability of channel state information prediction while keeping the model structure relatively simple, and provide more accurate channel prior information for resource allocation and beam scheduling of low-orbit satellite communication system.The method can be applied to low-orbit satellite channel prediction.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, specifically relating to a method for predicting low Earth Orbit (LEO) satellite channels based on hierarchical collaborative optimization-random vector function linking networks. Background Technology

[0002] Low Earth Orbit (LEO) satellites are widely considered a crucial infrastructure supporting high-speed global connectivity and large-scale Internet of Things (IoT) services. With the continuous increase in user terminal scale and service traffic, the deep integration of massive MIMO (Multiple Input Multiple Output) technology with LEO satellite communication systems has become a key approach to improving system capacity and achieving seamless global high-speed coverage. In this context, high-precision and real-time acquisition of satellite-to-ground downlink channel state information is essential for ensuring link reliability and spectrum utilization efficiency. However, in LEO satellite scenarios, the high-speed motion of satellites leads to highly dynamic channels, and the long satellite-to-ground transmission distance introduces significant round-trip time delays. Furthermore, the onboard platform is subject to strict constraints in terms of computing power and payload, making traditional channel state information acquisition and update mechanisms ineffective in addressing these challenges.

[0003] In summary, in order to address the problem of low channel state information prediction accuracy caused by long transmission distances between satellites and ground, constructing an intelligent channel prediction method with nonlinear modeling and time-series prediction capabilities has become a key issue in current research on channel state information prediction for low-Earth orbit satellites. Summary of the Invention

[0004] The purpose of this invention is to solve the problem of low channel state information prediction accuracy caused by long transmission distances between satellites and ground, and to propose a low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-random vector function linking network.

[0005] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0006] A low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network, the method specifically includes the following steps:

[0007] Step 1: Establish a low-Earth orbit (LEO) satellite channel model and obtain the channel state information of the LEO satellite channel model at consecutive P+L time points. Use the obtained channel state information as a training sample in the supervised learning dataset of the random vector function linking network (RVC). Use the channel information of the first P time points in the training sample as the input of the RVC and the channel information of the last L time points in the training sample as the output of the RVC.

[0008] Where P and L are both positive integers;

[0009] Step 2: Train the random vector function link network using a supervised learning dataset. During the training process, optimize the weights and bias parameters of the random vector function link network based on the hierarchical collaborative optimization algorithm to obtain the trained random vector function link network.

[0010] Step 3: Use the trained random vector function to link the network to predict channel state information.

[0011] Furthermore, in each training sample, the channel state information for the first P time steps is estimated using the minimum mean square error channel estimation algorithm.

[0012] Furthermore, for each training sample, the channel information for the last L time steps is as follows:

[0013]

[0014] in, Indicates the first One user, Indicates the first The user in the future The actual channel information at any given moment Indicates total path fading. Indicates the first The user in the future Small-scale decay at a given moment .

[0015] Furthermore, the total path fading for:

[0016]

[0017] in, Indicates path loss. Indicates atmospheric decay, This indicates scintillation attenuation in the ionosphere or troposphere. This indicates the penetration loss of the building.

[0018] Furthermore, the small-scale fading for:

[0019]

[0020] in, Represents the line-of-sight component. This represents the non-line-of-sight component.

[0021] Furthermore, the optimization of the weights and bias parameters of the stochastic vector function link network based on the hierarchical collaborative optimization algorithm is carried out as follows:

[0022] Step 3. Initialize the parameters of the hierarchical collaborative optimization algorithm: population size is... The number of individuals in each subgroup is The maximum number of iterations is The operation selection parameters are as follows: And the fitness function is ;

[0023] Each individual represents a vector consisting of the weights and biases from the input layer to the hidden layer;

[0024] The first initialization vector The upper bound of the elements is , No. The lower bound of the elements is , order the The first in the subpopulation Upper bound of elements , No. The first in the subpopulation lower bound of elements , , , This represents the dimension of the vector corresponding to each individual.

[0025] Step 3.2: Initialize the first The first subpopulation The vector corresponding to the individual is the th _ ... The elements are:

[0026]

[0027] in, Indicates the first The first subpopulation The vector corresponding to the individual is the th _ ... One element, This indicates that a random number between 0 and 1 will be returned.

[0028] Step 33: Initialize the number of iterations ;

[0029] Steps three and four, for the first Individual populations, according to Select the first The individual with the lowest fitness function value in the subpopulation. , order the The individual with the lowest fitness function value in the subpopulation is the [number]th [individual]. The current best individual in the subpopulation, the th The best individual in the current subpopulation ;

[0030] in, They represent the first After the nth iteration, the th The first, second, ..., the second best individuals in the subpopulation. One element;

[0031] Then from the first After each iteration, the individual with the lowest fitness function value is selected from the best individuals in each subpopulation, and this selected individual is taken as the global best individual. ;

[0032] in, They represent the 1st, 2nd, ..., 1st individuals in the globally optimal group. One element;

[0033] Step 35: Generate random numbers , ,judge and The size relationship between them;

[0034] like Then proceed to step three six;

[0035] like Then proceed to step three seven;

[0036] Step 36: Perform a global bootstrap update, updating the position, upper bound, and lower bound of all individuals; specifically:

[0037] Step 361: If the globally optimal individual is at the th... Within a subpopulation,

[0038]

[0039] If the globally optimal individual is not in the _ _ Within a subpopulation,

[0040]

[0041] Step 362: Update the upper and lower bounds to obtain the updated upper bound. and the updated Nether ;

[0042] Step 363, jump to step 38;

[0043] Step 37: Perform a local perturbation update to update the positions of all individuals; specifically:

[0044]

[0045] In the formula, It is from 1 to Random integers generated between [a certain range] Indicates a Gaussian distribution;

[0046] like The corresponding fitness function value is greater than The corresponding fitness function value is then set to ;

[0047] Otherwise, let ;

[0048] make Then jump to step 38;

[0049] Step 38: Determine whether the fitness function value of the best individual in each subpopulation has decreased;

[0050] If the fitness function value of the best individual in at least one subpopulation decreases, then proceed directly to step 39;

[0051] Otherwise, perform cross-recombination on individuals in each subpopulation before proceeding to step 39;

[0052] Step 39: Determine if the condition is met. ;

[0053] If satisfied Then the individual with the smallest fitness function value obtained from the last iteration update will be output as the optimal parameter of the random vector function linking network.

[0054] If not satisfied Then let Return to steps three and four.

[0055] Furthermore, in step 362, the specific process of updating the upper and lower bounds is as follows:

[0056]

[0057] in, Indicates the updated upper bound. Indicates the updated lower bound; This represents the equilibrium parameter.

[0058] Furthermore, the fitness function for:

[0059]

[0060] in, , express The channel matrix between the satellite and the user terminal at any given time. express Time Satellite and the Channel vectors between user terminals , This represents the predictions made by the random vector function link network. The channel matrix between the satellite and the user terminal at any given time. , This indicates finding the Frobenius norm of a matrix. This indicates the operation of seeking the expected value.

[0061] Furthermore, the balance parameters for:

[0062]

[0063] in, This indicates the calculation of absolute value.

[0064] Furthermore, the specific process of step three is as follows:

[0065] No. The channel state information at each time point is predicted in the following way. It is a natural number;

[0066] like Then the first The channel state information at each time point is estimated using the minimum mean square error method;

[0067] like Then the first The channel state information at each time step is used to predict the output by linking the network with a trained random vector function.

[0068] The beneficial effects of this invention are:

[0069] This invention introduces a hierarchical collaborative optimization algorithm into the parameter optimization process of a stochastic vector function linking network (SVR), achieving global optimization of the weights and biases between the input and hidden layers of the SVR. The trained SVR is then used to perform high-precision prediction of channel state information at future time points. This method improves the effectiveness and reliability of channel state information prediction while maintaining a relatively simple model structure, providing more accurate prior channel information for resource allocation and beam scheduling in low-Earth orbit satellite communication systems. Attached Figure Description

[0070] Figure 1 This is a model of the LEO satellite communication system in this invention;

[0071] The system model includes two parts: a low-Earth orbit (LEO) satellite and a ground user terminal. This invention considers the downlink channel for LEO satellite communication, and the LEO satellite side is equipped with... A uniform planar array composed of three antennas. , Each user terminal (UT) has a single antenna, and the system model operates in time division duplex (TDD) mode, dividing time resources into multiple frames;

[0072] Figure 2 This is a flowchart of the channel prediction process based on RVFL networks according to the present invention;

[0073] In the first P steps, the conventional minimum mean square error channel estimation algorithm is used to obtain the satellite-to-ground channel state. Then, the RVFL network and the historical channel estimation results of the first P steps are used to predict the channel state information for the next L steps. In the next L steps, no channel estimation or channel prediction operation is required.

[0074] Figure 3 This is a flowchart of the training process for the RVFL network in this invention. Detailed Implementation

[0075] Specific Implementation Method 1: This implementation method describes a low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-random vector function linking network. The method specifically includes the following steps:

[0076] Step 1, such as Figure 1 As shown, a low-Earth orbit (LEO) satellite channel model is established, and the channel state information of the LEO satellite channel model at consecutive P+L time points is obtained. The obtained channel state information is used as a training sample in the supervised learning dataset of the random vector function linking network (RVFL). The channel information of the first P time points in the training sample is used as the input of the random vector function linking network, and the channel information of the last L time points in the training sample is used as the output of the random vector function linking network.

[0077] Where P and L are both positive integers;

[0078] Step one specifically involves setting the specific parameters in the channel model according to 3GPP TR 38.811, including setting the total path fading and the initial satellite positions. Set the initial position of the kth user as... The satellite's orbital altitude is set to 1000km. Assuming the satellite and user move in the same direction, the satellite's speed is 7.45km / s, and the user's speed is 2m / s. Based on these parameters, the satellite's azimuth, elevation, transmission delay, and transmission distance can be directly calculated, thus obtaining the true channel state information. The input time step P for the historical estimated channel state information is set to 15 frames, and the output time step L for the predicted true channel information is set to 5 frames.

[0079] Furthermore, by changing distance-related parameters (such as satellite position and user position), multiple sets of low-Earth orbit satellite channel models can be established. Based on the established models, multiple sets of training sample data can be generated. The obtained data is divided into three parts: training set, validation set, and test set in a ratio of 8:1:1.

[0080] Step 2: Train the stochastic vector function link network using a supervised learning dataset. During training, optimize the weights and bias parameters of the stochastic vector function link network using the hierarchical collaborative optimization algorithm (JOA). (It should be noted that the weights optimized using the hierarchical collaborative optimization algorithm only include the connection weights between the input layer and the hidden layer.) This will yield the trained stochastic vector function link network.

[0081] Step 3: Use the trained random vector function to link the network to predict channel state information.

[0082] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that the channel state information for the first P time steps in each training sample is estimated using the minimum mean square error channel estimation algorithm.

[0083] The other steps and parameters are the same as in Specific Implementation Method 1.

[0084] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that the low-Earth orbit satellite channel includes path fading and small-scale fading. In each training sample, the true channel information for the last L time moments is as follows:

[0085]

[0086] in, Indicates the first One user, Indicates the first The user in the future The actual channel information at any given moment Indicates total path fading. Indicates the first The user in the future Small-scale decay at a given moment .

[0087] Other steps and parameters are the same as in specific implementation method one or two.

[0088] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that the total path fading... for:

[0089]

[0090] in, Indicates the basic path loss. Indicates atmospheric decay, This indicates scintillation attenuation in the ionosphere or troposphere. This indicates the penetration loss of the building.

[0091] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0092] Specific Implementation Method Five: This implementation method differs from Specific Implementation Methods One to Four in that the small-scale fading... for:

[0093]

[0094] in, Represents the line-of-sight (LoS) component. This represents the non-line-of-sight (NLoS) component.

[0095] The other steps and parameters are the same as those in one of the specific implementation methods one to four.

[0096]

[0097] in, Indicates the carrier frequency. Represents Rice factor; This indicates the total number of antennas equipped on the LEO satellite side; The base of the natural logarithm; Represents the imaginary unit; This indicates Doppler frequency shift caused by the satellite. Indicating the line-of-sight situation, the user terminal side's first... Doppler shift for individual users; This indicates that in non-line-of-sight situations, the user terminal side... Doppler frequency shift along the path; Indicates the propagation delay along the line-of-sight path. Represents the array response vector. Indicates the first The azimuth angle of each user relative to the satellite. Indicates the first The elevation angle of each user relative to the satellite; Indicates the first The first user's Latency spread under a single path Indicates the relationship with the first The number of scattering paths associated with each user Indicates the first The first user's The multiple channel gain of each path.

[0098] Specific Implementation Method Six: Combination Figure 3 This implementation method is described below. The difference between this implementation method and one of the specific implementation methods one through five is that the optimization of the weights and bias parameters of the random vector function link network based on the hierarchical collaborative optimization algorithm is as follows:

[0099] Step 3. Initialize the parameters of the hierarchical collaborative optimization algorithm: the number of subpopulations is... The number of individuals in each subpopulation is The maximum number of iterations is The operation selection parameters are as follows: (This is crucial for both global boot updates and protection actions) and the fitness function is... ;

[0100] Each individual represents a vector consisting of the weights and biases from the input layer to the hidden layer; that is, the encoding method for each individual is:

[0101]

[0102] in, This represents the dimension of the input weights in a random vector function linking network. This represents the dimension of the input bias in a random vector function link network;

[0103] The first initialization vector The upper bound of the elements is , No. The lower bound of the elements is , order the The first in the subpopulation Upper bound of elements , No. The first in the subpopulation lower bound of elements , , , This represents the dimension of the vector corresponding to each individual, i.e. ;

[0104] Step 3.2: Initialize the first The first subpopulation The vector corresponding to the individual is the th _ ... The elements are:

[0105]

[0106] in, Indicates the first The first subpopulation The vector corresponding to the individual is the th _ ... One element, This indicates that a random number between 0 and 1 will be returned.

[0107] Step 33: Initialize the number of iterations ;

[0108] Steps three and four, for the first Individual populations, according to Select the first The individual with the lowest fitness function value in the subpopulation. , order the The individual with the lowest fitness function value in the subpopulation is the [number]th [individual]. The current best individual in the subpopulation, the th The best individual in the current subpopulation ;

[0109] in, They represent the first After the nth iteration, the th The first, second, ..., the second best individuals in the subpopulation. One element;

[0110] Then from the first After each iteration, the individual with the lowest fitness function value is selected from the best individuals in each subpopulation, and this selected individual is taken as the global best individual. ;

[0111] in, They represent the 1st, 2nd, ..., 1st individuals in the globally optimal group. One element;

[0112] Step 35: Generate random numbers , ,judge and The size relationship between them;

[0113] like Then proceed to step three six;

[0114] like Then proceed to step three seven;

[0115] Step 36: Perform a global bootstrap update, updating the position, upper bound, and lower bound of all individuals; specifically:

[0116] Step 361: If the globally optimal individual is at the th... Within a subpopulation,

[0117]

[0118] If the globally optimal individual is not in the _ _ Within a subpopulation,

[0119]

[0120] Step 362: During the update process, due to the dynamic changes in the search space state, the upper and lower bounds of the dimension will also change. Update the upper and lower bounds to obtain the updated upper bound. and the updated Nether ;

[0121] Step 363, jump to step 38;

[0122] Step 37: Perform local perturbation update, using a Gaussian distribution-based update strategy to update the positions of all individuals; specifically:

[0123]

[0124] In the formula, It is from 1 to Random integers generated between [a certain range] Represents a Gaussian distribution, and the Gaussian distribution is... The group works together to dynamically control the scope of individual actions;

[0125] like The corresponding fitness function value is greater than The corresponding fitness function value is then set to ;

[0126] Otherwise, let ;

[0127] make Then jump to step 38;

[0128] Step 38: Determine whether the fitness function value of the best individual in each subpopulation has decreased;

[0129] If the fitness function value of the best individual in at least one subpopulation decreases, then proceed directly to step 39;

[0130] Otherwise, after performing crossover recombination on individuals in each subpopulation (i.e., exchanging some individuals in each subpopulation while keeping the upper and lower bounds of each subpopulation unchanged), proceed to step 39.

[0131] Step 39: Determine if the condition is met. ;

[0132] If satisfied Then the individual with the smallest fitness function value obtained from the last iteration update will be output as the optimal parameter of the random vector function linking network.

[0133] If not satisfied Then let Return to steps three and four.

[0134] The other steps and parameters are the same as those in one of the specific implementation methods one to five.

[0135] After outputting the optimal parameters of the random vector function link network, the trained random vector function link network is obtained. The trained random vector function link network is then used to make predictions on the test set data.

[0136]

[0137] in, This indicates that the trained random vector function connects the network to make predictions on the test set data. This represents the channel information estimated using the least mean square error algorithm. This represents the predicted output of a network linking random vector functions. This represents the parameters of the network linked by the trained random vector function. Indicates the number of predicted times. This represents the number of historical moments entered.

[0138] When the prediction performance of the trained stochastic vector function link network on the test and validation sets fails to meet the requirements, it is necessary to reinitialize the individuals of the hierarchical collaborative optimization algorithm and start training again.

[0139] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One through Six in that the specific process of updating the upper and lower bounds in step 362 is as follows:

[0140]

[0141] in, Indicates the updated upper bound. Indicates the updated lower bound; This represents the equilibrium parameter.

[0142] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0143] Specific Implementation Method Eight: This implementation method differs from Specific Implementation Methods One to Seven in that the fitness function... for:

[0144]

[0145] in, , , For complex fields, express The channel matrix between the satellite and the user terminal at any given time. express Time Satellite and the Channel vectors between user terminals , This represents the predictions made by the random vector function link network. The channel matrix between the satellite and the user terminal at any given time. , This indicates finding the Frobenius norm of a matrix. This indicates the operation of seeking the expected value.

[0146] The other steps and parameters are the same as those in any of the specific implementation methods one to seven.

[0147] By assigning the weights and bias parameters of an individual to the random vector function linking network, the fitness function value on the training set can be obtained, that is, the fitness function value corresponding to the individual.

[0148] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One to Eight in that the balance parameter... for:

[0149]

[0150] in, This indicates the calculation of absolute value.

[0151] The other steps and parameters are the same as those in one of the specific implementation methods one to eight.

[0152] By changing the balancing parameters in stages, a balance can be achieved between the extensibility and exploratory nature of the algorithm.

[0153] Specific Implementation Method Ten: Combining Figure 2 This embodiment is described below. The difference between this embodiment and one of the specific embodiments one through nine is that the specific process of step three is as follows:

[0154] No. The channel state information at each time point is predicted in the following way. It is a natural number;

[0155] like Then the first The channel state information at each time point is estimated using the minimum mean square error method;

[0156] like Then the first The channel state information at each time step is used to predict the output by linking the network with a trained random vector function.

[0157] The other steps and parameters are the same as those in any of the specific implementation methods one to nine.

[0158] No. The moment, the first The moment, ..., the first At the nth time point, this implementation method uses the minimum mean square error method to estimate the channel information, and then uses the estimated nth time point using the minimum mean square error method to estimate the channel information. The moment, the first The moment, ..., the first The channel information at time step is used as the input to the trained random vector function linking network, and the network can predict the _th ... The moment, the first The moment, ..., the first Channel information at each moment.

[0159] Specific Implementation Method Eleven: This implementation method differs from Specific Implementation Methods One through Ten in that the prediction of the random vector function link network is equivalent to the objective function. Minimize:

[0160]

[0161] In the formula, Represents the comprehensive input matrix. , This represents the vector composed of the weights and biases between the input and hidden layers of the trained random vector function linking network (i.e., the optimization result of the hierarchical collaborative optimization algorithm). Represents the history of all users The channel matrix is ​​composed of channel information at each time step. This represents the connection weight matrix between the hidden layer and the output layer. This represents the predicted output matrix. For regularization parameters;

[0162] when At that time, the connection weight matrix The expression is: ,in, Representation matrix The generalized inverse;

[0163] when At that time, the connection weight matrix is ​​solved using the regularized least squares method. :

[0164]

[0165] Where n represents the number of input samples, Indicates the dimension of the input data. This represents the dimension of the hidden layer nodes. express transpose, This represents the identity matrix, with the superscript -1 indicating the inverse of the matrix;

[0166] The output of the trained random vector function link network is:

[0167]

[0168] in, Indicates the future number From the moment to the first Satellite at that moment A matrix composed of channel vectors between user terminals This means converting the network's output vector into the channel matrix form needed to calculate the fitness value.

[0169] The other steps and parameters are the same as those in specific implementation methods one through ten.

[0170] The above examples of the present invention are merely illustrative of the computational model and process of the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is impossible to exhaustively list all possible implementations here. Any obvious variations or modifications derived from the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network, characterized in that, The method specifically includes the following steps: Step 1: Establish a low-Earth orbit (LEO) satellite channel model and obtain the channel state information of the LEO satellite channel model at consecutive P+L time points. Use the obtained channel state information as a training sample in the supervised learning dataset of the random vector function linking network (RVC). Use the channel information of the first P time points in the training sample as the input of the RVC and the channel information of the last L time points in the training sample as the output of the RVC. Where P and L are both positive integers; Step 2: Train the random vector function link network using a supervised learning dataset. During the training process, optimize the weights and bias parameters of the random vector function link network based on the hierarchical collaborative optimization algorithm to obtain the trained random vector function link network. Step 3: Use the trained random vector function to link the network to predict channel state information.

2. The low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network according to claim 1, characterized in that, In each training sample, the channel state information for the first P time steps is estimated using the minimum mean square error channel estimation algorithm.

3. The low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network according to claim 2, characterized in that, For each training sample, the channel information for the last L time steps is as follows: in, Indicates the first One user, Indicates the first The user in the future The actual channel information at each moment Indicates total path fading. Indicates the first The user in the future Small-scale decay at a given moment .

4. The low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network according to claim 3, characterized in that, The path total fading for: in, Indicates path loss. Indicates atmospheric decay, This indicates scintillation attenuation in the ionosphere or troposphere. This indicates the penetration loss of the building.

5. The low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network according to claim 4, characterized in that, The small-scale fading for: in, Represents the line-of-sight component. This represents the non-line-of-sight component.

6. The low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network according to claim 5, characterized in that, The optimization of the weights and bias parameters of the stochastic vector function linking network based on the hierarchical collaborative optimization algorithm is as follows: Step 3. Initialize the parameters of the hierarchical collaborative optimization algorithm: the number of subpopulations is... The number of individuals in each subpopulation is The maximum number of iterations is The operation selection parameters are as follows: And the fitness function is ; Each individual represents a vector consisting of the weights and biases from the input layer to the hidden layer; The first initialization vector The upper bound of the elements is , No. The lower bound of the elements is , order the The first in the subpopulation Upper bound of elements , No. The first in the subpopulation lower bound of elements , , , This represents the dimension of the vector corresponding to each individual. Step 3.2: Initialize the first The first subpopulation The vector corresponding to the individual is the th _ ... The elements are: in, Indicates the first The first subpopulation The vector corresponding to the individual is the th _ ... One element, This indicates that a random number between 0 and 1 will be returned. Step 33: Initialize the number of iterations ; Steps three and four, for the first Individual populations, according to Select the first The individual with the lowest fitness function value in the subpopulation. , order the The individual with the lowest fitness function value in the subpopulation is the [number]th [individual]. The current best individual in the subpopulation, the th The best individual in the current subpopulation ; in, They represent the first After the nth iteration, the th The first, second, ..., the second best individuals in the subpopulation. One element; Then from the first After each iteration, the individual with the lowest fitness function value is selected from the best individuals in each subpopulation, and this selected individual is taken as the global best individual. ; in, They represent the 1st, 2nd, ..., 1st individuals in the globally optimal group. One element; Step 35: Generate random numbers , ,judge and The size relationship between them; like Then proceed to step three six; like Then proceed to step three seven; Step 36: Perform a global bootstrap update, updating the position, upper bound, and lower bound of all individuals; specifically: Step 361: If the globally optimal individual is at the th... Within a subpopulation, If the globally optimal individual is not in the _ _ Within a subpopulation, Step 362: Update the upper and lower bounds to obtain the updated upper bound. and the updated Nether ; Step 363, jump to step 38; Step 37: Perform a local perturbation update to update the positions of all individuals; specifically: In the formula, It is from 1 to Random integers generated between [a certain range] Indicates a Gaussian distribution; like The corresponding fitness function value is greater than The corresponding fitness function value is then set to ; Otherwise, let ; make Then jump to step 38; Step 38: Determine whether the fitness function value of the best individual in each subpopulation has decreased; If the fitness function value of the best individual in at least one subpopulation decreases, then proceed directly to step 39; Otherwise, perform cross-recombination on individuals in each subpopulation before proceeding to step 39; Step 39: Determine if the condition is met. ; If satisfied Then the individual with the smallest fitness function value obtained from the last iteration update will be output as the optimal parameter of the random vector function linking network. If not satisfied Then let Return to steps three and four.

7. The low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network according to claim 6, characterized in that, In step 362, the specific process of updating the upper and lower bounds is as follows: in, Indicates the updated upper bound. Indicates the updated lower bound; This represents the equilibrium parameter.

8. The low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network according to claim 7, characterized in that, The fitness function for: in, , express The channel matrix between the satellite and the user terminal at any given time. express Time Satellite and the Channel vectors between user terminals , This represents the predictions made by the random vector function link network. The channel matrix between the satellite and the user terminal at any given time. , This indicates finding the Frobenius norm of a matrix. This indicates the operation of seeking the expected value.

9. The low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network according to claim 8, characterized in that, The balance parameter for: in, This indicates the calculation of absolute value.

10. The low-Earth orbit satellite channel prediction method based on hierarchical collaborative optimization-stochastic vector function linking network according to claim 9, characterized in that, The specific process of step three is as follows: No. The channel state information at each time point is predicted in the following way. It is a natural number; like Then the first The channel state information at each time point is estimated using the minimum mean square error method; like Then the first The channel state information at each time step is used to predict the output by linking the network with a trained random vector function.

Images