Rail transit dynamic channel prediction method and system based on communication task performance optimization

Abstract

The invention provides a rail transit dynamic channel prediction method and system based on communication task performance optimization, and the method comprises the steps: obtaining historical CSI data and related spatial-temporal characteristic data in a rail transit system, and carrying out the preprocessing, and obtaining model input data; performing deep spatial-temporal feature extraction on the model input data based on a deep learning model to obtain joint spatial-temporal feature representation; constructing a channel state prediction model, and obtaining a channel state prediction value based on the channel state prediction model and the joint spatial-temporal feature representation; training and optimizing the channel state prediction model by adopting a task-oriented loss function based on the channel state prediction value; and deploying the trained and optimized channel state prediction model to a 5G base station side or a vehicle-mounted terminal to realize real-time channel prediction in a high-speed rail transit environment. According to the invention, through task-oriented channel prediction, high quality and reliability of communication in a high-speed rail transit environment are ensured.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, specifically relating to a method and system for dynamic channel prediction in rail transit based on communication task performance optimization. Background Technology

[0002] With the continuous evolution of high-speed rail towards 5G and even future 6G communication technologies, the signal stability and reliability of rail transit communication systems face unprecedented challenges in high-speed movement and complex, ever-changing environments. The high-speed operation of trains, rapid changes in terrain along the line (such as tunnels, viaducts, and open areas), and variable weather conditions all have a drastic and non-stationary impact on the propagation of wireless signals, leading to rapid and unpredictable changes in channel quality. This renders traditional channel modeling and prediction methods (usually based on channel stationarity assumptions or static models) inadequate for meeting the stringent communication quality requirements of high-speed rail transit.

[0003] In advanced communication systems such as 5G massive multiple-input multiple-output (MIMO), accurate channel state information (CSI) prediction is the cornerstone for achieving key technologies such as efficient beamforming, precoding, resource scheduling, and fast handover. However, existing channel prediction methods mainly focus on the mathematical error (e.g., mean square error, MSE) between the predicted and actual values. While these methods, based on minimizing prediction error, can provide high prediction accuracy under ideal conditions, their limitations become increasingly apparent in scenarios like high-speed rail transit where the channel is highly non-stationary and dynamically changing. First, even with small mathematical errors in the predictions, these predictions may still fail to effectively support performance optimization for real-world communication tasks. For example, a channel prediction that is "accurate" in the MSE sense may still lead to a significant decrease in actual system performance (such as system throughput, signal-to-noise ratio, or handover success rate) if its error happens to occur on the channel components that have the most critical impact on beam direction, signal power allocation, or handover decisions. Traditional methods do not directly optimize the contribution of channel prediction to actual communication benefits.

[0004] Secondly, in the rail transit environment, channel variations are influenced by a variety of complex factors, including real-time train speed and location, obstruction from surrounding buildings, tunnel effects, and atmospheric conditions. Existing methods often struggle to comprehensively and intelligently capture these multi-dimensional spatiotemporal characteristics, resulting in insufficient adaptability of prediction models to real-world channel changes. More importantly, these models lack awareness of communication task requirements and cannot selectively predict channels based on the actual needs of different communication scenarios, such as beamforming, resource scheduling, or handover decisions.

[0005] Therefore, there is an urgent need for a channel prediction method that breaks through the traditional paradigm of minimizing prediction errors. This method should not only be able to accurately capture the spatiotemporal dynamic evolution characteristics of rail transit channels, but also be directly guided by improving the key performance indicators of communication systems in actual operation. By optimizing the channel prediction model, it should be able to provide channel state information with greater practical application value and performance assurance for 5G / 6G rail transit communication systems. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a dynamic channel prediction method and system for rail transit based on communication task performance optimization. It aims to break through the traditional channel prediction method that minimizes prediction error, and is guided by improving the key performance indicators of the communication system in actual operation. By optimizing the channel state prediction model, it can provide channel state information with more practical application value and performance assurance for 5G / 6G rail transit communication systems.

[0007] To achieve the above objectives, the present invention provides the following solution: A dynamic channel prediction method for rail transit based on communication task performance optimization, the method comprising: S1. Acquire historical CSI data and relevant spatiotemporal feature data from the rail transit system and preprocess them to obtain model input data; S2. Based on the deep learning model, perform deep spatiotemporal feature extraction on the input data of the model to obtain a joint spatiotemporal feature representation; S3. Construct a channel state prediction model, and obtain the channel state prediction value based on the channel state prediction model and the joint spatiotemporal feature representation; S4. Based on the predicted channel state values, the channel state prediction model is trained and optimized using a task-oriented loss function. S5. Deploy the trained and optimized channel state prediction model to the 5G base station or vehicle terminal to achieve real-time channel prediction in high-speed rail transit environments.

[0008] Preferably, the relevant spatiotemporal feature data includes: the real-time location of the train, the speed of the train, and the geographical environmental markers of the train's location; the geographical environmental markers include tunnels, bridges, or open areas.

[0009] Preferably, the method in S2 for extracting deep spatiotemporal features from the model input data based on a deep learning model to obtain a joint spatiotemporal feature representation includes: Temporal dependency features are extracted from the model input data based on a recurrent neural network structure; Spatial correlation features are extracted from the model input data based on a graph neural network structure; The temporal dependency features and spatial correlation features are fused and weighted based on an attention mechanism to obtain a joint spatiotemporal feature representation.

[0010] Preferably, the method in S4 for training and optimizing the channel state prediction model based on the predicted channel state value using a task-oriented loss function includes: Based on the predicted channel state values, the strategy adopted by the train communication system when performing preset communication tasks is simulated or calculated. Obtain the actual communication task performance corresponding to the strategy under real channel conditions; A task-oriented loss function is constructed with the actual communication task performance as the optimization objective. Based on the task-oriented loss function, the channel state prediction model is trained and optimized so that the channel state prediction model generates the channel state prediction value that maximizes the performance of the actual communication task.

[0011] Preferably, the task-oriented loss function is: ; in, This represents a task-oriented loss function. Indicates the actual channel state Below, based on the channel state prediction value The actual communication task performance achieved by the obtained strategy, Represents the policy function. It represents the mathematical expectation.

[0012] The present invention also provides a dynamic channel prediction system for rail transit based on communication task performance optimization. The system is used to implement the aforementioned method and includes: a data acquisition and processing module, a feature extraction module, a first prediction module, a model optimization module, and a second prediction module. The data acquisition and processing module is used to acquire historical CSI data and related spatiotemporal feature data from the rail transit system and preprocess them to obtain model input data. The feature extraction module is used to perform deep spatiotemporal feature extraction on the input data of the model based on the deep learning model to obtain a joint spatiotemporal feature representation; The first prediction module is used to construct a channel state prediction model and obtain channel state prediction values ​​based on the channel state prediction model and the joint spatiotemporal feature representation. The model optimization module is used to train and optimize the channel state prediction model based on the channel state prediction value using a task-oriented loss function; The second prediction module is used to deploy the trained and optimized channel state prediction model to the 5G base station or vehicle terminal to achieve real-time channel prediction in high-speed rail transit environments.

[0013] Preferably, the relevant spatiotemporal feature data includes: the real-time location of the train, the speed of the train, and the geographical environmental markers of the train's location; the geographical environmental markers include tunnels, bridges, or open areas.

[0014] Preferably, the model optimization module includes: a policy acquisition unit, an actual communication task performance acquisition unit, a loss function construction unit, and a training optimization unit; The strategy acquisition unit is used to simulate or calculate the strategy adopted by the train communication system when performing a preset communication task based on the channel state prediction value. The actual communication task performance acquisition unit is used to acquire the actual communication task performance corresponding to the strategy under real channel conditions. The loss function construction unit is used to construct a task-oriented loss function with the actual communication task performance as the optimization objective. The training and optimization unit is used to train and optimize the channel state prediction model based on the task-oriented loss function, so that the channel state prediction model generates the channel state prediction value when the performance of the actual communication task is maximized.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned method.

[0016] The present invention also provides a computer-readable storage medium storing a computer program that, when executed, implements the aforementioned method.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a dynamic channel prediction method and system for rail transit based on communication task performance optimization. Through this task-oriented channel prediction, the channel state prediction model can more effectively and accurately support 5G communication systems in beamforming, resource scheduling, and handover decisions, directly improving the capacity, throughput, and stability of the communication system, and ensuring high quality and reliability of communication in high-speed rail transit environments. Attached Figure Description

[0018] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the process of the dynamic channel prediction method for rail transit based on communication task performance optimization according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the process for obtaining model input data in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the working principle of the task-oriented loss function in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the training and optimization process of the channel state prediction model according to an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] Example 1 This invention provides a dynamic channel prediction method for rail transit based on communication task performance optimization, comprising: S1. Acquire historical CSI data and relevant spatiotemporal feature data from the rail transit system and preprocess them to obtain model input data; S2. Based on the deep learning model, perform deep spatiotemporal feature extraction on the input data of the model to obtain a joint spatiotemporal feature representation; S3. Construct a channel state prediction model, and obtain the channel state prediction value based on the channel state prediction model and the joint spatiotemporal feature representation; S4. Based on the predicted channel state values, the channel state prediction model is trained and optimized using a task-oriented loss function. S5. Deploy the trained and optimized channel state prediction model to the 5G base station or vehicle terminal to achieve real-time channel prediction in high-speed rail transit environments.

[0023] Figure 1 This is a flowchart illustrating the dynamic channel prediction method for rail transit based on communication task performance optimization according to an embodiment of the present invention. It specifically demonstrates how channel data input, deep spatiotemporal feature extraction, task-oriented channel prediction, and the crucial task-oriented loss function work together to optimize the performance of the actual communication system. The specific implementation process of the present invention is as follows: S1. Methods for obtaining model input data include acquiring historical CSI data and relevant spatiotemporal feature data from the rail transit system and preprocessing them: Figure 2 This is a flowchart illustrating the process of obtaining model input data in an embodiment of the present invention. It specifically demonstrates the preprocessing of multi-source channel input data, the integration of spatiotemporal correlation features, and the unification process before inputting channel state prediction models in a rail transit communication system.

[0024] The input data mainly includes: Historical Channel State Information (CSI) data: acquired through onboard equipment or base stations in the rail transit system, including characteristics reflecting channel quality and transmission capacity such as path loss, fading, and delay.

[0025] Relevant spatiotemporal characteristic data includes the train's real-time location, speed (obtained via GPS and sensors), and markers indicating the train's entry into and exit from different geographical environments (such as tunnels, viaducts, and open areas). These features comprehensively characterize the channel state, dynamic behavior, and environmental information of the train's environment.

[0026] Based on the requirements of rail transit communication systems, this invention sequentially performs data cleaning, data alignment, data standardization, and data formatting on the input data. The specific methods are as follows: First, data cleaning is performed on the input data. Since CSI data, train operation parameters, and environmental sign data may be acquired from different devices, outliers, missing values, and duplicate data may exist. Therefore, data cleaning methods are needed to remove invalid data. For example, methods such as sliding window mean filtering, IQR (Inter-Quartile Range) outlier detection, and threshold removal can be used to remove data points with abrupt changes or that clearly do not conform to physical laws. For a small number of missing items, linear interpolation, time series interpolation, or interpolation algorithms based on neighborhood features can be used to fill in the missing data.

[0027] Secondly, this invention performs data alignment between historical CSI data and spatiotemporal feature data. Since train speed, location, and geographic environmental markers originate from different sensors, their sampling frequencies and timestamps may differ. To ensure consistency in subsequent model inputs, all features need to be aligned to a unified timeline. Data alignment can include timestamp-based synchronization, downsampling or upsampling, and interpolation-based sequence reconstruction, ensuring all features are arranged at the same time step, forming a temporally consistent dataset.

[0028] After data cleaning and alignment, this invention performs data standardization preprocessing on the input CSI data and related spatiotemporal feature data. All data are normalized through standardization to eliminate dimensional differences between different data sets, ensuring data consistency and model stability. The standardization method can employ the commonly used Standard Scaler, whose formula is: ; in, For standardized data, This is the original data. The mean, The standard deviation is denoted as . The standardized data, which serves as the model input for subsequent channel state prediction models, is uniformly represented as . .

[0029] Finally, the standardized data undergoes formatting. Formatting includes encoding the raw CSI data (such as amplitude, phase, signal-to-noise ratio, path loss, etc.) into fixed-dimensional feature vectors; and converting train operating parameters (such as speed, latitude and longitude) and geographical landmarks (such as tunnels, viaducts, or open areas) into numerical forms recognizable by the model. For example, one-hot encoding can be used to represent environmental categories, and two-dimensional coordinate encoding can be used to represent the train's spatial location. Furthermore, depending on the model's requirements, different data sources can be combined into a unified spatiotemporal feature matrix to provide structured input for the deep learning model.

[0030] Furthermore, S2, the method for extracting deep spatiotemporal features from the model input data based on a deep learning model to obtain a joint spatiotemporal feature representation includes: Temporal dependency features are extracted from the model input data based on a recurrent neural network structure; Spatial correlation features are extracted from the model input data based on a graph neural network structure; The temporal dependency features and spatial correlation features are fused and weighted based on an attention mechanism to obtain a joint spatiotemporal feature representation.

[0031] Specifically, this invention employs a deep learning model to extract deep spatiotemporal features from the model input data, aiming to extract potential features from complex and ever-changing rail transit channel data that are crucial for predicting future channel states and optimizing the performance of subsequent communication tasks.

[0032] Methods for extracting deep spatiotemporal features from model input data using deep learning models may include, but are not limited to, using Long Short-Term Memory (LSTM) networks to capture the temporal dependencies of channels, using Graph Neural Networks (GCN) to model the spatial correlations of channels, and using spatiotemporal attention mechanisms to adaptively fuse and weight features from different spatiotemporal dimensions.

[0033] S201, Feature Mapping: Input data into the model (For example, channel gain, signal-to-noise ratio, train speed, train position, geographical landmarks, etc.) are mapped to a high-dimensional feature space through a linear projection (fully connected layer) to enhance the expressive power of the features and provide a unified input for subsequent temporal feature extraction modeling and spatial feature extraction modeling. The calculation method of this process is as follows: ; in, This is the weight matrix. For bias terms, For activation function, and These are the high-dimensional features after mapping.

[0034] S202, Temporal Feature Extraction: By utilizing recurrent neural network structures such as LSTM, deep learning is performed on the time-series information in historical CSI data to capture the long-term dependence and dynamic evolution patterns of channel states during train operation. Its core computational model can be summarized as follows: ; in, Input the mapping features at the current time. and These represent the hidden state and cell state from the previous time step, respectively. Output the hidden state at the current moment.

[0035] Finally, the time-series feature representation output by the LSTM module is denoted as... .

[0036] S203, Spatial Feature Extraction: Using graph neural network structures such as GCN, a graph model is constructed based on the physical connections of the track network and the train's environment. Spatial dependency features of the channel are then extracted to reflect the influence of the train's position and surrounding environment on the channel. Its core graph convolution operation can be summarized as follows: ; in, It is the first The node feature matrix of the layer It is an activation function, and It is the first The learnable weight matrix of the layer. It is an adjacency matrix containing self-loops, which is derived from the original adjacency matrix. Add the identity matrix To obtain, that is Original adjacency matrix Describes the connections between nodes in a graph model (e.g., connections between adjacent positions on a track), the identity matrix. This adds a self-connection (self-loop) to each node, ensuring that the node includes its own information when aggregating neighbor information. yes The degree matrix, a diagonal matrix whose diagonal elements are The sum of the corresponding rows (i.e., the degree of the node).

[0037] Finally, the spatial feature representation output by the GCN module is denoted as: .

[0038] S204, Spatiotemporal Feature Fusion: Temporal features are adaptively fused through modules such as spatiotemporal attention mechanisms. and spatial features This mechanism generates a unified, weighted joint spatiotemporal feature representation. Based on the dynamic changes in spatiotemporal features, it adjusts the degree of attention given to different time periods and spatial regions, ensuring that the extracted features more accurately reflect the dynamic evolution of the channel state.

[0039] The fused joint spatiotemporal features are represented as follows: ; in, For joint spatiotemporal feature representation, Indicates feature splicing, This indicates the weighting of the attention mechanism.

[0040] Furthermore, S3, the method for constructing a channel state prediction model and obtaining channel state prediction values ​​based on the channel state prediction model and the joint spatiotemporal feature representation includes: After deep spatiotemporal feature extraction and fusion, the joint spatiotemporal feature representation will enter the channel state prediction module to generate predicted channel state values ​​for a future period. The prediction result of this step serves as an intermediate representation for subsequent optimization of specific communication tasks, and the training of the channel state prediction model will be optimized through a customized task loss function.

[0041] Channel state prediction models can be constructed using fully connected layers or other regression networks to jointly represent spatiotemporal features. Mapped to channel state prediction values : ; in, It is the weight matrix of the prediction layer. It is a bias term.

[0042] Furthermore, S4, the method for training and optimizing the channel state prediction model using a task-oriented loss function based on the predicted channel state values ​​includes: Based on the predicted channel state values, the strategy adopted by the train communication system when performing preset communication tasks is simulated or calculated. Obtain the actual communication task performance corresponding to the strategy under real channel conditions; A task-oriented loss function is constructed with the actual communication task performance as the optimization objective. Based on the task-oriented loss function, the channel state prediction model is trained and optimized so that the channel state prediction model generates the channel state prediction value that maximizes the performance of the actual communication task.

[0043] S401. Construction of Task-Oriented Loss Function: The core of the training phase is to optimize network weights by maximizing the actual performance metrics of the channel state prediction model on specific communication tasks (preset communication tasks, including but not limited to: received signal-to-noise ratio (SNR) or signal-to-interference-plus-noise ratio (SINR); system throughput; spectral efficiency; beamforming gain; resource allocation or handover decisions), rather than simply minimizing the mathematical error between the predicted channel state and the actual channel state. Therefore, this invention employs a task-oriented loss function.

[0044] Figure 3 This paper details how to calculate the corresponding communication task performance (e.g., beamforming gain) based on the predicted channel state and compare it with the task performance under the actual channel state, thereby constructing a differentiable loss function to guide model training, rather than simply calculating the prediction error. The working principle of the task-oriented loss function is as follows: Calculate the task performance based on the predicted channel: use the predicted channel state values ​​predicted by the channel state prediction model ( This is used to simulate or calculate the performance metrics that a communication system can achieve when performing a specific communication task. For example, if the task is to maximize beamforming gain, then according to... The optimal beamforming vector is calculated and its performance in real channel conditions is evaluated. H The actual received signal strength or signal-to-noise ratio (SNR) under these conditions SNR ).

[0045] Define the optimization objective: The task-oriented loss function aims to ensure that the predictions generated by the channel state prediction model are able to guide the communication system in the actual channel state. H To achieve optimal task performance, the task-oriented loss function measures the difference between the actual and expected results of decisions made based on the predicted channel (e.g., beamforming direction, MCS selection, etc.) on the real physical channel.

[0046] Differentiability handling: To ensure that the task-oriented loss function can be used for gradient descent optimization, for communication system operations involving discontinuous or non-differentiable operations, techniques such as differentiable simulators, surrogate loss functions, or reinforcement learning can be used for approximation or transformation.

[0047] For example, a task-oriented loss function for beamforming gain optimization can be expressed as: ; in, This represents a task-oriented loss function. Indicates the actual channel state Below, based on the channel state prediction value The actual communication task performance (e.g., SINR, throughput, or spectral efficiency) achieved by the resulting strategy (such as beamforming vectors or precoding matrices). Represents the policy function. (Symbol) This represents the mathematical expectation, used to average the system performance under different time slices, samples, or channel conditions. Introducing the mathematical expectation allows for a more stable reflection of the overall statistical performance of the prediction model, rather than optimization based solely on a single channel sample. Since the goal of the task-oriented loss function is to maximize the actual performance of the communication system under real-world channel conditions, it is negative, making it easier to minimize the expected performance during training. It can maximize system performance.

[0048] S402, Channel State Prediction Model Optimization: Figure 4This demonstrates how to use a task-oriented loss function to optimize the weights of the entire channel state prediction model, ensuring that the prediction results can directly improve the actual performance indicators of the communication system.

[0049] like Figure 4 As shown, this invention uses a common gradient descent optimization algorithm (such as the Adam optimizer) to continuously adjust the weights of the channel state prediction model according to the task-oriented loss function described above.

[0050] During training, the preprocessed input feature data (historical CSI data and relevant spatiotemporal feature data) undergoes deep spatiotemporal feature extraction and is then used to output channel state prediction values ​​through the channel state prediction model. These channel state prediction values ​​are subsequently used to calculate the task-oriented loss function, and then backpropagate the gradient to optimize the channel state prediction model parameters.

[0051] S403, Regularization: To avoid overfitting, techniques such as Dropout and L2 regularization can be integrated during training to enhance the generalization ability of the channel state prediction model and ensure that the channel state prediction model performs stably and reliably under new data and unseen channel environments.

[0052] Furthermore, S5, methods for deploying the trained and optimized channel state prediction model to 5G base stations or vehicle-mounted terminals to achieve real-time channel prediction in high-speed rail transit environments include: The trained and optimized channel state prediction model will be deployed to 5G base stations or vehicle terminals for real-time channel prediction in high-speed rail transit environments.

[0053] S501, Deployment of Channel State Prediction Model: Once trained, the channel state prediction model will be integrated into the real-time processing link of the communication system, such as in the digital baseband unit of a base station or the communication module on a train.

[0054] S502, Real-time Data Input: During real-time prediction, the channel state prediction model receives real-time historical CSI data and train spatiotemporal characteristics (such as location and speed). This data serves as input to the model, driving it to generate future channel state predictions.

[0055] S503, Task-Oriented Prediction Output: Based on its capabilities gained from task-oriented training, the channel state prediction model outputs predicted channel states for a future period of time. These channel state predictions directly serve subsequent communication system decisions, for example: Beamforming and precoding: Provide more accurate channel information, guide the base station to form the optimal beam, maximize signal gain, and suppress interference.

[0056] Resource scheduling: Based on predicted channel quality, subcarriers and modulation and coding schemes (MCS) are dynamically allocated to improve spectrum efficiency.

[0057] Handover decision: Predict future channel quality changes to achieve smoother and more reliable base station handover, reducing call drop rate and handover latency.

[0058] Through this task-oriented channel prediction method, the present invention can significantly improve the capacity, throughput, connection stability and overall reliability of 5G / 6G rail transit communication systems, effectively address the communication challenges in high-speed mobile scenarios, and provide users with high-quality, uninterrupted communication services.

[0059] In summary, this invention provides a dynamic channel prediction method for rail transit based on communication task performance optimization. Through this task-oriented channel prediction, the channel state prediction model can more effectively and accurately support 5G communication systems in beamforming, resource scheduling, and handover decisions, directly improving the capacity, throughput, and stability of the communication system, and ensuring high-quality and reliable communication in high-speed rail transit environments.

[0060] Example 2 Based on the same inventive concept, the present invention also provides a dynamic channel prediction system for rail transit based on communication task performance optimization, for implementing the method described in the foregoing embodiments. The system includes: a data acquisition and processing module, a feature extraction module, a first prediction module, a model optimization module, and a second prediction module. The data acquisition and processing module is used to acquire historical CSI data and related spatiotemporal feature data from the rail transit system and preprocess them to obtain model input data. The feature extraction module is used to perform deep spatiotemporal feature extraction on the input data of the model based on the deep learning model to obtain a joint spatiotemporal feature representation; The first prediction module is used to construct a channel state prediction model and obtain channel state prediction values ​​based on the channel state prediction model and the joint spatiotemporal feature representation. The model optimization module is used to train and optimize the channel state prediction model based on the channel state prediction value using a task-oriented loss function; The second prediction module is used to deploy the trained and optimized channel state prediction model to the 5G base station or vehicle terminal to achieve real-time channel prediction in high-speed rail transit environments.

[0061] Furthermore, the relevant spatiotemporal feature data includes: the train's real-time location, the train's speed, and the geographical environmental markers where the train is located; the geographical environmental markers include tunnels, bridges, or open areas.

[0062] Furthermore, the feature extraction module includes: a temporal feature extraction unit, a spatial feature extraction unit, and a feature fusion unit; The temporal feature extraction unit is used to extract temporal dependency features from the model input data based on the recurrent neural network structure. A spatial feature extraction unit is used to extract spatial correlation features from the model input data based on a graph neural network structure. The feature fusion unit is used to fuse and weight the temporal dependency features and the spatial correlation features based on an attention mechanism to obtain a joint spatiotemporal feature representation.

[0063] Furthermore, the model optimization module includes: a policy acquisition unit, an actual communication task performance acquisition unit, a loss function construction unit, and a training optimization unit; The strategy acquisition unit is used to simulate or calculate the strategy adopted by the train communication system when performing a preset communication task based on the channel state prediction value. The actual communication task performance acquisition unit is used to acquire the actual communication task performance corresponding to the strategy under real channel conditions. The loss function construction unit is used to construct a task-oriented loss function with the actual communication task performance as the optimization objective. The training and optimization unit is used to train and optimize the channel state prediction model based on the task-oriented loss function, so that the channel state prediction model generates the channel state prediction value when the performance of the actual communication task is maximized.

[0064] Furthermore, the task-oriented loss function is: ; in, This represents a task-oriented loss function. Indicates the actual channel state Below, based on the channel state prediction value The actual communication task performance achieved by the obtained strategy, Represents the policy function. It represents the mathematical expectation.

[0065] Example 3 Based on the same inventive concept, corresponding to the methods of any of the above embodiments, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the methods described in any of the above embodiments.

[0066] Example 4 Based on the same inventive concept, corresponding to the methods of any of the above embodiments, the present invention also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the methods described in any of the above embodiments.

[0067] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A dynamic channel prediction method for rail transit based on communication task performance optimization, characterized in that, The method includes: S1. Acquire historical CSI data and relevant spatiotemporal feature data from the rail transit system and preprocess them to obtain model input data; S2. Based on the deep learning model, perform deep spatiotemporal feature extraction on the input data of the model to obtain a joint spatiotemporal feature representation; S3. Construct a channel state prediction model, and obtain the channel state prediction value based on the channel state prediction model and the joint spatiotemporal feature representation; S4. Based on the predicted channel state values, the channel state prediction model is trained and optimized using a task-oriented loss function. S5. Deploy the trained and optimized channel state prediction model to the 5G base station or vehicle terminal to achieve real-time channel prediction in high-speed rail transit environments.

2. The method according to claim 1, characterized in that, The relevant spatiotemporal feature data includes: the train's real-time location, the train's speed, and the geographical landmarks where the train is located; the geographical landmarks include tunnels, bridges, or open areas.

3. The method according to claim 1, characterized in that, The method in S2 for extracting deep spatiotemporal features from the model input data based on a deep learning model to obtain a joint spatiotemporal feature representation includes: Temporal dependency features are extracted from the model input data based on a recurrent neural network structure; Spatial correlation features are extracted from the model input data based on a graph neural network structure; The temporal dependency features and spatial correlation features are fused and weighted based on an attention mechanism to obtain a joint spatiotemporal feature representation.

4. The method according to claim 1, characterized in that, The method in S4 for training and optimizing the channel state prediction model based on the predicted channel state value using a task-oriented loss function includes: Based on the predicted channel state values, the strategy adopted by the train communication system when performing preset communication tasks is simulated or calculated. Obtain the actual communication task performance corresponding to the strategy under real channel conditions; A task-oriented loss function is constructed with the actual communication task performance as the optimization objective. Based on the task-oriented loss function, the channel state prediction model is trained and optimized so that the channel state prediction model generates the channel state prediction value that maximizes the performance of the actual communication task.

5. The method according to claim 4, characterized in that, The task-oriented loss function is: ； in, This represents a task-oriented loss function. Indicates the actual channel state Below, based on the channel state prediction value The actual communication task performance achieved by the obtained strategy, Represents the policy function. It represents the mathematical expectation.

6. A dynamic channel prediction system for rail transit based on communication task performance optimization, the system being used to implement the method according to any one of claims 1-5, characterized in that, The system includes: a data acquisition and processing module, a feature extraction module, a first prediction module, a model optimization module, and a second prediction module; The data acquisition and processing module is used to acquire historical CSI data and related spatiotemporal feature data from the rail transit system and preprocess them to obtain model input data. The feature extraction module is used to perform deep spatiotemporal feature extraction on the input data of the model based on the deep learning model to obtain a joint spatiotemporal feature representation; The first prediction module is used to construct a channel state prediction model and obtain channel state prediction values ​​based on the channel state prediction model and the joint spatiotemporal feature representation. The model optimization module is used to train and optimize the channel state prediction model based on the channel state prediction value using a task-oriented loss function; The second prediction module is used to deploy the trained and optimized channel state prediction model to the 5G base station or vehicle terminal to achieve real-time channel prediction in high-speed rail transit environments.

7. The system according to claim 6, characterized in that, The relevant spatiotemporal feature data includes: the train's real-time location, the train's speed, and the geographical landmarks where the train is located; the geographical landmarks include tunnels, bridges, or open areas.

8. The system according to claim 6, characterized in that, The model optimization module includes: a policy acquisition unit, an actual communication task performance acquisition unit, a loss function construction unit, and a training optimization unit; The strategy acquisition unit is used to simulate or calculate the strategy adopted by the train communication system when performing a preset communication task based on the channel state prediction value. The actual communication task performance acquisition unit is used to acquire the actual communication task performance corresponding to the strategy under real channel conditions. The loss function construction unit is used to construct a task-oriented loss function with the actual communication task performance as the optimization objective. The training and optimization unit is used to train and optimize the channel state prediction model based on the task-oriented loss function, so that the channel state prediction model generates the channel state prediction value when the performance of the actual communication task is maximized.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1-5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the method as described in any one of claims 1-5.

Images