Beam Prediction Method Based on Radar Environment Perception and RIS Power Distribution Pattern

Abstract

The present invention discloses a beam prediction method based on radar environment perception and RIS power distribution pattern, belonging to the field of communication technology, comprising: obtaining radar point cloud data at each time step, and extracting features from the radar point cloud data using a multi-layer perceptron and a maximum pooling layer to obtain radar point cloud features; obtaining a RIS power delay distribution data graph, and extracting RIS power delay distribution features using a CNN convolutional layer and position encoding; fusing the two features using a cross-modal fusion module; encoding the fused features using a Transformer encoder to obtain encoded features containing spatiotemporal dependencies; calculating the dynamic weight of each target, and weightedly aggregating the encoded features using the dynamic weights; decoding the weighted aggregated features using a Transformer decoder to obtain a beam prediction result. Through the above method, the present invention improves the accuracy of beam prediction.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and more particularly to a beam prediction method based on radar environment perception and RIS power distribution mode. Background Art

[0002] Beam prediction refers to the use of algorithms or models to predict the optimal signal transmission direction and beam parameters in wireless communication systems, thereby achieving more efficient signal coverage and improving communication quality.

[0003] Currently, beamforming predictions are often combined with artificial intelligence (AI) and machine learning (ML) to achieve accurate predictions.

[0004] However, existing beam prediction technologies have many shortcomings, such as poor adaptability to dynamic environments and inability to solve the beam splitting problem caused by multipath effects, resulting in beam prediction accuracy that does not meet requirements.

[0005] Therefore, how to improve the accuracy of beam prediction is an urgent problem that those skilled in the art need to solve. Summary of the Invention

[0006] In view of this, the present invention provides a beam prediction method based on radar environment perception and RIS power distribution mode

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] Obtain radar point cloud data for each time step in the radar time series data, and use a multi-layer perceptron and a maximum pooling layer to extract features from the radar point cloud data for each time step to obtain radar point cloud features;

[0009] Obtain the RIS power delay distribution data graph and use the CNN convolution layer and position encoding to extract the RIS power delay distribution features;

[0010] The cross-modal fusion module is used to fuse the extracted radar point cloud features and RIS power delay distribution features to obtain fused features;

[0011] The obtained fusion features are encoded using the Transformer encoder to obtain encoding features containing spatiotemporal dependencies;

[0012] Calculating the dynamic weight of each target, and performing weighted aggregation on the encoding features using the dynamic weight to obtain a weighted aggregate feature;

[0013] The Transformer decoder is used to decode the weighted aggregate features and obtain the beam prediction results.

[0014] Furthermore, the radar point cloud data of each time step in the radar time series data is obtained, including the following radar point cloud data sets:

[0015] P t ={p i |p i =(x i ,y i ,z i ,v y,i ,σ i )∈R 6}, wherein the radar point cloud data set P t For each data point p i Including position (x i ,y i , z i ), speed v y,i and radar cross section σ i ,R 6 Represents the 6 dimensions of radar point cloud data.

[0016] Furthermore, a multi-layer perceptron and a maximum pooling layer are used to extract features from the radar point cloud data at each time step, specifically including:

[0017] For each data point p i , get p through multi-layer perceptron i local features;

[0018] For each data point p i All data points p in the neighborhood N(i) j All are transformed by MLP and the maximum eigenvalue is obtained;

[0019] will p i The local features of and the maximum eigenvalue in the neighborhood N(i) are spliced to obtain the spliced features;

[0020] The obtained splicing features are sequentially input into the multi-layer perceptron and the maximum pooling layer to output the radar point cloud features.

[0021] Furthermore, the RIS power delay distribution data graph is obtained, and the RIS power delay distribution features are extracted using the CNN convolution layer and position encoding, which specifically includes the following steps:

[0022] Get RIS power delay distribution data graph P r ;

[0023] The power delay distribution data graph P r Input the 2D CNN convolution layer and flatten the convolution result;

[0024] The flattened convolution result and the power delay distribution data graph Pr The RIS power delay distribution features are obtained by splicing and combining the position coding features:

[0025]

[0026] in, Represents the RIS power delay distribution characteristics; Flatten represents the flattening operation; Conv2D represents the two-dimensional convolution layer; E pos Represents a positional encoding operation.

[0027] Furthermore, the position encoding operation is implemented by the following formula:

[0028]

[0029]

[0030] Among them, m∈{0,1,2,....,M-1} represents the spatial position index value, where M is the maximum spatial position index, i∈{0,1,…,⌊D c / 2⌋-1} represents the feature dimension index, D c Indicates the number of convolution output channels.

[0031] Furthermore, the cross-modal fusion module is used to fuse the extracted radar point cloud features and RIS power delay distribution features, specifically including:

[0032] Use the attention mechanism to construct the target correlation feature matrix:

[0033]

[0034] in, represents the radar point cloud features at time t, represents the RIS power delay distribution characteristics at time t, Represents the query space matrix obtained after linear transformation of radar features; represents the key space matrix obtained after the linear change of the RIS power delay distribution characteristics; d k Denotes the dimension compression factor of the attention mechanism, D r represents the feature dimension of radar point cloud, Represents the RIS power delay distribution characteristic dimension;

[0035] Based on the target correlation feature matrix, the RIS power delay distribution feature is converted to the radar point cloud dimension, and then fused with the radar point cloud feature to obtain the fusion feature. :

[0036]

[0037] Among them, W v It represents the value space matrix obtained after the linear change of the RIS power delay distribution characteristics, and LayerNorm represents the layer normalization function.

[0038] Furthermore, the obtained fusion features are encoded using the Transformer encoder to obtain encoding features containing spatiotemporal dependencies, specifically including the following expressions:

[0039] H enc =TransformerEncoder(H fuse )

[0040] Among them, TransformerEncoder represents the Transformer encoder, H enc Represents the obtained fusion coding features.

[0041] Furthermore, the dynamic weight of each target is calculated, and the coding features are weightedly aggregated using the dynamic weight to obtain weighted aggregate features, which specifically includes the following steps:

[0042] Calculate the weight of each objective:

[0043]

[0044] in, Represents the data value of the i-th target point in the fused coding feature at time t, represents the data value of the i-th target point in the RIS power delay distribution data sequence at time t, tanh is the hyperbolic tangent function, v is a learnable parameter, and W w is a learnable matrix;

[0045] The target weight Perform weighted summation with the corresponding data points in the fusion coding feature sequence to obtain the weighted aggregate feature:

[0046]

[0047] Among them, F H represents the weighted aggregate feature, H i Represents the corresponding i-th data point in the fused encoding feature sequence.

[0048] Furthermore, the Transformer decoder is used to decode the weighted aggregate features and obtain the beam prediction results, which specifically include:

[0049] Use Transformer decoder to decode the weighted aggregate features;

[0050] The output of the decoder is passed through a linear layer and an activation function to obtain the amplitude and phase prediction results of the RIS:

[0051]

[0052] Among them, W out is the output of the Transformer decoder;

[0053] Amplitude and phase can be output separately:

[0054]

[0055] Among them, A pred Indicates the amplitude, Indicates the phase output.

[0056] The present invention also discloses a beam prediction system based on radar environment perception and RIS power distribution mode, including a computer program, which can implement the beam prediction method described in any one of the present invention when running.

[0057] It can be seen from the above technical solution that, compared with the prior art, the present invention discloses a beam prediction method based on radar environment perception and RIS power distribution mode, which has the following beneficial effects:

[0058] 1. Improve beam prediction accuracy

[0059] By combining radar data with RIS PDP optimization, this invention can accurately predict beam direction in dynamic environments, significantly improving beam prediction accuracy. Compared with existing technologies (such as traditional CSI-based methods), this invention improves prediction accuracy, especially in complex mobile scenarios and highly dynamic environments.

[0060] 2. Effectively suppress multipath interference and beam splitting

[0061] This invention leverages RIS's PDP optimization to dynamically adjust the phase, amplitude, and polarization of the reflector unit, enhancing the energy of the main path signal and reducing interference caused by multipath propagation, particularly beam splitting. Experiments have shown that sidelobe interference is reduced by 15dB, signal attenuation is minimized, and system stability and communication quality are effectively improved.

[0062] 3. Improve computing efficiency and reduce overhead

[0063] By compressing and optimizing deep learning models (e.g., knowledge distillation techniques), the amount of training data and inference time are reduced, lowering computational complexity by 40%. Compared to existing technologies, particularly traditional CSI-based deep learning models, this invention significantly improves computational efficiency without sacrificing accuracy, making it suitable for high-speed mobile scenarios.

[0064] 4. Enhance adaptability to dynamic environments

[0065] The system of this invention can perceive environmental changes (such as target movement and obstacle relocation) in real time and adaptively adjust beam prediction strategies through reinforcement learning (RL), ensuring sustained high-quality signal transmission in diverse dynamic environments. Compared to traditional static models, this system offers greater robustness and adaptability, making it particularly well-suited for rapidly evolving applications such as 6G connected vehicles and the Industrial Internet of Things.

[0066] 5. Reduce costs and energy consumption

[0067] By optimizing the system architecture and deep learning model, this invention reduces hardware requirements and computing resource consumption, thereby reducing overall system energy consumption and costs. This advantage makes the technology more economical and scalable when deployed on a large scale, especially in high-density network environments.

[0068] 6. Optimize beam steering strategy

[0069] This invention incorporates a reinforcement learning algorithm, enabling the system to optimize beam steering strategies based on real-time feedback, offering significant advantages in ensuring system stability and efficiency. Compared to traditional fixed or manually adjusted beam steering methods, this invention can adapt to environmental changes, improving the system's ability to operate stably over extended periods of time. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0071] Figure 1 A schematic diagram of the overall flow of a beam prediction method based on radar environment perception and RIS power distribution mode provided in an embodiment of the present invention.

[0072] Figure 2 Schematic diagram of the specific steps for obtaining radar point cloud features using a multi-layer perceptron and a maximum pooling layer, provided in an embodiment of the present invention.

[0073] Figure 3 A schematic diagram of feature fusion using a cross-modal fusion module provided in an embodiment of the present invention.

[0074] Figure 4 A comparison diagram of the robustness effects of the method of the present invention and the existing method under different noise environments is provided in the embodiment of the present invention.

[0075] Figure 5 A comparison chart of the inference prediction speed effects of the method of the present invention and the existing methods provided in the embodiment of the present invention.

[0076] Figure 6 A comparison chart of the prediction accuracy of the method provided in the embodiment of the present invention and the existing method. DETAILED DESCRIPTION

[0077] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0078] refer to Figure 1-Figure 3 The embodiment of the present invention discloses a beam prediction method based on radar environment perception and RIS power distribution mode, comprising the following steps:

[0079] Obtain radar point cloud data for each time step in the radar time series data, and use a multi-layer perceptron and a maximum pooling layer to extract features from the radar point cloud data for each time step to obtain radar point cloud features;

[0080] Obtain the RIS power delay distribution data graph and use the CNN convolution layer and position encoding to extract the RIS power delay distribution features;

[0081] The cross-modal fusion module is used to fuse the extracted radar point cloud features and RIS power delay distribution features to obtain fused features;

[0082] The obtained fusion features are encoded using the Transformer encoder to obtain encoding features containing spatiotemporal dependencies;

[0083] Calculating the dynamic weight of each target, and performing weighted aggregation on the encoding features using the dynamic weight to obtain a weighted aggregate feature;

[0084] The Transformer decoder is used to decode the weighted aggregate features and obtain the beam prediction results.

[0085] Furthermore, the radar point cloud data of each time step in the radar time series data is obtained, including the following radar point cloud data sets:

[0086] P t ={p i |p i =(x i ,yi ,z i ,v y,i ,σ i )∈R 6}, wherein the radar point cloud data set P t For each data point p i Including position (x i ,y i , z i ), speed v y,i and radar cross section σ i ,R 6 Represents the 6 dimensions of radar point cloud data.

[0087] Furthermore, a multi-layer perceptron and a maximum pooling layer are used to extract features from the radar point cloud data at each time step, specifically including:

[0088] For each data point p i , get p through multi-layer perceptron i local features;

[0089] For each data point p i All data points p in the neighborhood N(i) j All are transformed by MLP and the maximum eigenvalue is obtained;

[0090] will p i The local features of and the maximum eigenvalue in the neighborhood N(i) are spliced to obtain the spliced features;

[0091] The obtained splicing features are input into the multi-layer perceptron and the maximum pooling layer in sequence to output the radar point cloud features. , where N is the number of targets, D r is the feature dimension, which is expressed as: .

[0092] In this step, the neighborhood N(i) is selected using the fixed variable diameter method, and all data points within a certain radius r of the current data point are regarded as the neighborhood range of the data point.

[0093] Furthermore, the RIS power delay distribution data graph is obtained, and the RIS power delay distribution features are extracted using the CNN convolution layer and position encoding, which specifically includes the following steps:

[0094] Get RIS power delay distribution data graph P r ;

[0095] The power delay distribution data graph P r Input the 2D CNN convolution layer and flatten the convolution result;

[0096] The flattened convolution result and the power delay distribution data graph P r The RIS power delay distribution features are obtained by splicing and combining the position coding features:

[0097] ;

[0098] in, Represents the RIS power delay distribution characteristics; Flatten represents the flattening operation; Conv2D represents the two-dimensional convolution layer; E pos Represents a positional encoding operation.

[0099] Specifically, in the embodiment of the present invention, position encoding is performed in the following manner:

[0100] ;

[0101] ;

[0102] Among them, m∈{0,1,2,....,M-1} represents the spatial position index value, where M is the maximum spatial position index, i∈{0,1,…,⌊D c / 2⌋-1} represents the feature dimension index, D c Indicates the number of convolution output channels.

[0103] Furthermore, the cross-modal fusion module is used to fuse the extracted radar point cloud features and RIS power delay distribution features, specifically including:

[0104] Use the attention mechanism to construct the target correlation feature matrix:

[0105] ;

[0106] in, represents the radar point cloud features at time t, represents the RIS power delay distribution characteristics at time t, Represents the query space matrix obtained after linear transformation of radar features; represents the key space matrix obtained after the linear change of the RIS power delay distribution characteristics; d k Denotes the dimension compression factor of the attention mechanism, D r represents the feature dimension of radar point cloud, represents the feature dimension of RIS power delay distribution; in this step, the purpose of the correlation feature matrix is to establish the data association between the radar point cloud features and the RIS power delay distribution, so that the data of two different modalities can be fused.

[0107] Then, based on the target correlation feature matrix in the previous step, the RIS power delay distribution feature is converted to the radar point cloud dimension, and fused with the radar point cloud feature to obtain the fusion feature. :

[0108]

[0109] Among them, W v It represents the value space matrix obtained after the linear change of the RIS power delay distribution characteristics, and LayerNorm represents the layer normalization function.

[0110] Furthermore, the obtained fusion features are encoded using the Transformer encoder to obtain encoding features containing spatiotemporal dependencies, specifically including the following expressions:

[0111] H enc =TransformerEncoder(H fuse )

[0112] Among them, TransformerEncoder represents the Transformer encoder, H enc Represents the obtained fusion coding features.

[0113] Furthermore, the dynamic weight of each target is calculated, and the coding features are weightedly aggregated using the dynamic weight to obtain weighted aggregate features, which specifically includes the following steps:

[0114] Calculate the weight of each objective:

[0115] ;

[0116] in, Represents the data value of the i-th target point in the fused coding feature at time t, represents the data value of the i-th target point in the RIS power delay distribution data sequence at time t, tanh is the hyperbolic tangent function, v is a learnable parameter, and W w is a learnable matrix;

[0117] The target weight Perform weighted summation with the corresponding data points in the fusion coding feature sequence to obtain the weighted aggregate feature:

[0118] ;

[0119] Among them, F H represents the weighted aggregate feature, H i Represents the corresponding i-th data point in the fused encoding feature sequence.

[0120] Furthermore, the Transformer decoder is used to decode the weighted aggregate features and obtain the beam prediction results, which specifically include:

[0121] Use Transformer decoder to decode the weighted aggregate features;

[0122] The output of the decoder is passed through a linear layer and an activation function to obtain the amplitude and phase prediction results of the RIS: Among them, W out is the output of the Transformer decoder;

[0123] Amplitude and phase can be output separately:

[0124]

[0125] Among them, A pred Indicates the amplitude, Indicates the phase output.

[0126] The overall loss function of the model in this invention is as follows:

[0127] ;

[0128] Among them, λ1, λ2, and λ3 represent weight coefficients. The first term of the loss function represents the beam reconstruction loss, the second term represents the power distribution similarity loss, and the third term represents the final loss of the key target. represents the beam prediction value, Y represents the beam true value, pr and They represent the similarity of the real and predicted power distributions, KL represents the KL divergence, and p i and denote the true and predicted target positions, respectively.

[0129] Furthermore, physical constraint optimization is added after the prediction output layer of the present invention to further optimize the obtained prediction results, specifically including:

[0130] Amplitude constraint: A opt =min(max(A pred ,0),A max );

[0131] Phase Constraint: .

[0132] In the physical constraint process, the projected gradient descent method is used to iterate so that the final constraint conditions are met:

[0133] ;

[0134] .

[0135] In order to verify the effect of the present invention, the method of the present invention is compared with the existing similar methods. Figure 4 、 Figure 5 、 Figure 6 The beam prediction method combining radar environment perception and RIS power distribution pattern disclosed in the present invention is superior to existing similar methods in terms of robustness, inference speed and inference accuracy.

[0136] The present invention also discloses a beam prediction system based on radar environment perception and RIS power distribution mode, including a computer program, which can implement the beam prediction method described in any one of the present invention when running.

[0137] The present invention may specifically include the following steps when running the above-mentioned steps including the above-mentioned method on a device terminal (such as a computer terminal).

[0138] 1. Initialize the environment

[0139] Global variable declaration: At the beginning of the script, the global variable SELECTED_DATA_PATH is declared to store the selected dataset path. The code uses global SELECTED_DATA_PATH for global variable declaration.

[0140] Clean the workspace: The clear; clc; command clears all variables in the workspace and clears the command window, ensuring that the script runs in a clean environment.

[0141] Add subdirectories to the path: Use addpath(genpath(pwd)) to add the current directory and all its subdirectories to the MATLAB search path to ensure that subsequent functions can be called smoothly.

[0142] 2. Create the directory structure

[0143] Check and create directories: The code checks and creates the following directories through mkdir:

[0144] oconfigs

[0145] oresults

[0146] oresults / models

[0147] oresults / figures If these directories do not exist, the script will create them to store configuration files, experimental results, trained models, and figure files.

[0148] 3. Scan available datasets

[0149] Scan the dataset directory: The code scans the current working directory and its subdirectories through scan_data_files(pwd) to find available datasets. data_paths = scan_data_files(pwd); will return the paths and file counts of all available datasets.

[0150] Selecting a dataset: If multiple datasets are scanned, the script asks the user to enter the dataset number and provides an option to select all datasets. If the user does not enter a selection, the first dataset is selected by default. The code obtains user input through input and selects the corresponding dataset path.

[0151] 4. Create a label file

[0152] Label file check: The script will check whether the dataset selected by the user contains a label file (test_labels.mat). If the file does not exist, the system will prompt the user whether to create a label file.

[0153] Generate label file: Users can choose from three label generation methods: exhaustive search, hierarchical search, or optimal method. The code will call create_labels_file(selected_path, methods{method_choice}); to create the label file.

[0154] 5. Execution method comparison experiment

[0155] Experimental comparison selection: In this step, the system will ask the user whether to perform a method comparison experiment. If the user selects "yes", the script will call the compare_methods function to compare different beam prediction methods and save the results to the results / figures directory.

[0156] 6. Operation mode selection

[0157] Mode selection: The user selects the operating mode in this step, providing the following four options:

[0158] ①Complete experimental process

[0159] ② Comparison of beam prediction performance

[0160] ③Training deep learning predictors

[0161] ④Visualize the dataset

[0162] The user enters the mode number through input, and the script performs the corresponding task according to the selected mode.

[0163] 7. Perform the corresponding tasks

[0164] Complete experimental process: If the user selects mode 1, the script will call the run_all_experiments() function to execute the complete experimental process including data loading, preprocessing, training, evaluation, and result saving.

[0165] Beam prediction performance comparison: If the user selects mode 2, the script will execute compare_methods('results / figures') to compare different beam prediction methods and save the results in the results / figures directory.

[0166] Train the deep learning predictor: If the user selects mode 3, the script will call train_dl_predictor() to train the deep learning prediction model and optimize the beam prediction.

[0167] Visualize the dataset: If the user selects mode 4, the script will call visualize_dataset(data_config.radar_train_dir) to visualize the radar data in the dataset and generate power maps, obstacle maps, and PDP maps.

[0168] 8. End execution

[0169] Execution completed: After executing all tasks, the script displays the prompt "Run completed!", indicating that the script has ended successfully.

[0170] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0171] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A beam prediction method based on radar environment perception and RIS power distribution pattern, characterized in that: The following steps are involved: Obtain radar point cloud data for each time step in the radar time series data, and use a multi-layer perceptron and a maximum pooling layer to extract features from the radar point cloud data for each time step to obtain radar point cloud features; Obtain the RIS power delay distribution data graph and use the CNN convolution layer and position encoding to extract the RIS power delay distribution features; The cross-modal fusion module is used to fuse the extracted radar point cloud features and RIS power delay distribution features to obtain fused features. Specifically, the following steps are performed: Use the attention mechanism to construct the target correlation feature matrix: ； in, represents the radar point cloud features at time t, represents the RIS power delay distribution characteristics at time t, Represents the query space matrix obtained after linear transformation of radar features; represents the key space matrix obtained after the linear change of the RIS power delay distribution characteristics; d k Denotes the dimension compression factor of the attention mechanism, D r represents the feature dimension of radar point cloud, Represents the RIS power delay distribution characteristic dimension; Based on the target correlation feature matrix, the RIS power delay distribution feature is converted to the radar point cloud dimension, and then fused with the radar point cloud feature to obtain the fusion feature. : ； Among them, W v It represents the value space matrix obtained after the linear change of the RIS power delay distribution characteristics, and LayerNorm represents the layer normalization function; The obtained fusion features are encoded using the Transformer encoder to obtain encoding features containing spatiotemporal dependencies; Calculating the dynamic weight of each target, and performing weighted aggregation on the encoding features using the dynamic weight to obtain weighted aggregate features; The Transformer decoder is used to decode the weighted aggregate features and obtain the beam prediction results.

2. The beam prediction method based on radar environment perception and RIS power distribution mode according to claim 1 is characterized in that: Get the radar point cloud data for each time step in the radar time series data, including the following radar point cloud data sets: P t ={p i |p i =(x i ,y i ,z i ,v y,i ,σ i )∈R 6 }, wherein the radar point cloud data set P t For each data point p i Including position (x i ,y i , z i ), speed v y,i and radar cross section σ i ,R 6 Represents the 6 dimensions of radar point cloud data.

3. The beam prediction method based on radar environment perception and RIS power distribution mode according to claim 2 is characterized in that: The multi-layer perceptron and maximum pooling layer are used to extract features from the radar point cloud data at each time step, specifically including: For each data point p i , get p through multi-layer perceptron i local features of For each data point p i All data points p in the neighborhood N(i) j All are transformed by MLP and the maximum eigenvalue is obtained; will p i The local features of and the maximum eigenvalue in the neighborhood N(i) are spliced to obtain the spliced features; The obtained splicing features are sequentially input into the multi-layer perceptron and the maximum pooling layer to output the radar point cloud features.

4. The beam prediction method based on radar environment perception and RIS power distribution mode according to claim 1 is characterized in that: Obtain the RIS power delay distribution data graph and use the CNN convolution layer and position encoding to extract the RIS power delay distribution features. The specific steps include: Get RIS power delay distribution data graph P r ; The power delay distribution data graph P r Input the 2D CNN convolution layer and flatten the convolution result; The flattened convolution result and the power delay distribution data graph P r The RIS power delay distribution features are obtained by splicing and combining the position coding features: ； in, Represents the RIS power delay distribution characteristics; Flatten represents the flattening operation; Conv2D represents the two-dimensional convolution layer; E pos Represents a positional encoding operation.

5. The beam prediction method based on radar environment perception and RIS power distribution mode according to claim 4 is characterized in that: The position encoding operation is implemented by the following formula: ； ； Among them, m∈{0,1,2,....,M-1} represents the spatial position index value, where M is the maximum spatial position index, i∈{0,1,…,D c / 2-1} represents the feature dimension index, D c Indicates the number of convolution output channels.

6. According to the beam prediction method based on radar environment perception and RIS power distribution pattern in claim 1, the obtained fusion features are encoded using a Transformer encoder to obtain encoding features containing spatiotemporal dependencies, specifically including the following expression: H enc =TransformerEncoder(H fuse ) in, TransformerEncoder represents the Transformer encoder, H enc Represents the obtained fusion coding features.

7. The beam prediction method based on radar environment perception and RIS power distribution mode according to claim 6, characterized in that: Calculate the dynamic weight of each target and use the dynamic weight to perform weighted aggregation on the encoding features to obtain weighted aggregate features, specifically including the following steps: Calculate the weight of each objective: ； in, Represents the data value of the i-th target point in the fused coding feature at time t, represents the data value of the i-th target point in the RIS power delay distribution data sequence at time t, tanh is the hyperbolic tangent function, v is a learnable parameter, and W w is a learnable matrix; The target weight Perform weighted summation with the corresponding data points in the fusion coding feature sequence to obtain the weighted aggregate feature: ； Among them, F H represents the weighted aggregate feature, H i Represents the corresponding i-th data point in the fused encoding feature sequence.

8. The beam prediction method based on radar environment perception and RIS power distribution pattern according to claim 7, wherein the method comprises decoding the weighted aggregate features using a Transformer decoder and obtaining a beam prediction result, specifically comprising: Use Transformer decoder to decode the weighted aggregate features; The output of the decoder is passed through a linear layer and an activation function to obtain the amplitude and phase prediction results of the RIS: ； Among them, W out is the output of the Transformer decoder; Amplitude and phase can be output separately: ； Among them, A pred Indicates the amplitude, Indicates the phase output.

9. The beam prediction system based on radar environment perception and RIS power distribution mode is characterized by: It includes a computer program, which can implement the beam prediction method according to any one of claims 1 to 8 when running.

Images