Optimization method and system for high-speed railway wireless access point deployment

Abstract

The application provides an optimization method and system for high-speed railway wireless access point deployment, which is used to solve the problem that the existing high-speed railway wireless access point deployment method is prone to data island and insufficient dynamic environment adaptation capability. Through collecting high-speed railway wireless communication state data of the location of the railway area node, training a wireless access point deployment optimization model based on the data, and synchronizing the model parameters obtained by the training to a federal server for aggregation to output a global wireless access point deployment optimization model, the global model issued by the federal server is received, and the global model is updated online based on a continuous learning mechanism. Real-time wireless communication state data of the railway area node is collected and input into the updated global model for optimization to obtain a wireless access point deployment optimization result of the location of the railway area node, which includes at least one of the deployment position, deployment interval, transmission power and wireless switching threshold of the wireless access point.

Description

Technical Field

[0001] This application relates to the field of railway wireless communication technology, and in particular to an optimization method and system for deploying wireless access points on high-speed railways. Background Technology

[0002] High-speed railway communication systems place extremely high demands on the continuity of wireless network coverage, real-time communication, and handover stability. As a critical infrastructure element, the deployment strategy of wireless access points directly impacts communication quality.

[0003] Existing methods for deploying wireless access points on high-speed railways primarily employ fixed parameter configuration or centralized optimization, achieving initial deployment through static network coverage analysis along the railway line. However, the high-speed railway operating environment is characterized by high-speed movement, complex scenarios, and dynamic changes, making traditional static deployment methods difficult to adapt to in the long term. When train speed, user access density, weather, or electromagnetic environment changes, existing methods are prone to increased wireless coverage blind spots and communication latency. Furthermore, data sharing between nodes in different railway areas is difficult, creating data silos and preventing the model from effectively adapting to dynamic environmental changes. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide an optimized method and system for the deployment of wireless access points on high-speed railways, so as to solve the technical problems in the prior art that the deployment scheme of wireless access points is not adaptable due to dynamic changes in the environment and data silos, and that it is prone to coverage blind spots and high communication latency.

[0005] In a first aspect, embodiments of this application provide an optimized method for deploying wireless access points on high-speed railways, comprising: Collect wireless communication status data of the high-speed railway where the railway area node is located; The wireless access point deployment optimization model is trained based on the wireless communication status data, and the parameters of the trained wireless access point deployment optimization model are synchronized to the federated server to realize the aggregation processing and output of the global wireless access point deployment optimization model for high-speed railway. The system receives the global wireless access point deployment optimization model issued by the federation server and performs online incremental updates on the global wireless access point deployment optimization model based on a continuous learning mechanism. The online incremental update involves introducing historical scene feature memory constraints into the global wireless access point deployment optimization model. Real-time wireless communication status data of the railway area nodes are collected and input into the updated global wireless access point deployment optimization model for optimization, so as to obtain the wireless access point deployment optimization result of the railway area node location. The deployment optimization result includes at least one of the following wireless access point deployment parameters: deployment location of wireless access points, deployment spacing, transmission power and wireless handover threshold.

[0006] Optionally, training the wireless access point deployment optimization model based on the wireless communication state data includes: Based on the wireless communication status data, a set of wireless communication status features for high-speed railways is constructed, wherein the set of wireless communication status features includes at least one of the following features: received signal strength, signal-to-interference-plus-noise ratio, train speed, number of users accessing the network, and load status of wireless access points. The wireless communication state feature set is input into a deep neural network for training to obtain a wireless access point deployment optimization model.

[0007] Optionally, the step of inputting the wireless communication state feature set into a deep neural network for training to obtain a wireless access point deployment optimization model includes: The feature set of each wireless access point in the wireless communication state feature set is input into a deep neural network for training to obtain multiple optimized models. The association calculation is performed based on the feature values ​​corresponding to each wireless access point, and the dynamic evaluation is performed based on the calculation results. Based on the evaluation results, one of the multiple optimization models is selected as the wireless access point deployment optimization model corresponding to the location of the railway area node.

[0008] Optionally, the calculation formula for associating features based on the feature values ​​corresponding to each wireless access point is as follows: ,in, Indicates the target wireless access point currently being accessed by the user terminal. Indicates the first The signal-to-interference-plus-noise ratio (SIR) for each wireless access point Indicates the first Current load status of each wireless access point Indicates the cost of wireless access point handover. This represents the weighting coefficient of the corresponding indicator.

[0009] Optionally, the expression for calculating the signal-to-interference-plus-noise ratio is: ,in, Indicates the transmit power of the target wireless access point. This indicates the channel gain corresponding to the target wireless access point. This indicates the power of interference signals generated by nearby wireless access points or the external electromagnetic environment. This indicates the system noise power.

[0010] Optionally, the online incremental update of the global wireless access point deployment optimization model based on the continuous learning mechanism includes: When a change in the operating environment of a railway area node is detected relative to the previous detection time, new wireless communication status data of the location of the railway area node is acquired. By utilizing a continuous learning mechanism, the new wireless communication state data is matched with similar historical wireless communication state data, and incremental training is performed on the global wireless access point deployment optimization model.

[0011] Optionally, the step of utilizing a continuous learning mechanism to match similar historical wireless communication state data based on the new wireless communication state data and incrementally training the global wireless access point deployment optimization model includes: Based on the new wireless communication state data, similar historical wireless communication state data are matched from the historical database, and a historical memory constraint loss function is constructed based on the historical wireless communication state data. The global wireless access point deployment optimization model is subjected to constrained incremental training based on the historical memory constraint loss function and the new wireless communication state data. The expression for the historical memory constraint loss function is: ,in, This represents the total loss function of the model. This represents the task loss function in the current high-speed railway communication scenario. Represents the importance coefficient of the model parameters. Indicates the current model parameters. This represents the key model parameters corresponding to the historical scenario. This represents the set of historical scene features that are remembered at the current moment. This represents a set of memories of historical scene features from the previous moment. This represents the weighting coefficient of historical parameter constraints. This represents the weighting coefficient for preserving historical scene memories; The update expression for the historical scene feature memory set is: ,in, Indicates the historical memory retention coefficient. This indicates the characteristics of the current high-speed railway wireless communication scenario.

[0012] Optionally, after synchronizing the parameters of the trained wireless access point deployment optimization model to the federated server, the method further includes: Receive parameters of the wireless access point deployment optimization model uploaded by multiple railway area nodes; The model parameters uploaded by each railway area node are weighted and aggregated to obtain a global wireless access point deployment optimization model; The global wireless access point deployment optimization model is distributed to each railway area node.

[0013] Optionally, the weighted aggregation of model parameters uploaded by nodes in each railway area to obtain a global wireless access point deployment optimization model includes: The aggregation weight coefficients of each node are dynamically adjusted based on the communication quality, data scale, and training contribution of each railway area node. The federated averaging algorithm is used to weight and aggregate the model parameters uploaded by each railway area node based on the aggregation weight coefficient to obtain a global wireless access point deployment optimization model.

[0014] Secondly, embodiments of this application also provide a high-speed railway wireless access point deployment system, including: A railway area node is used to collect wireless communication status data of the high-speed railway where the railway area node is located; a wireless access point deployment optimization model is trained based on the wireless communication status data, and the parameters of the trained wireless access point deployment optimization model are synchronized to the federated server to achieve aggregated processing and output of a global wireless access point deployment optimization model for the high-speed railway; the global wireless access point deployment optimization model is received from the federated server, and online incremental updates are performed on the global wireless access point deployment optimization model based on a continuous learning mechanism, wherein the online incremental update introduces historical scene feature memory constraints into the global wireless access point deployment optimization model; real-time wireless communication status data of the railway area node is collected and input into the updated global wireless access point deployment optimization model for optimization to obtain the wireless access point deployment optimization result where the railway area node is located, and the deployment optimization result includes at least one of the following wireless access point deployment parameters: deployment location of the wireless access point, deployment spacing, transmission power, and wireless handover threshold. The federated server receives parameters of the wireless access point deployment optimization model uploaded by multiple railway area nodes; performs weighted aggregation on the model parameters uploaded by each railway area node to obtain a global wireless access point deployment optimization model; and distributes the global wireless access point deployment optimization model to each railway area node.

[0015] The embodiments of this application bring the following beneficial effects: This application provides an optimization method and system for high-speed railway wireless access point deployment, which collaboratively optimizes wireless access point deployment by combining federated learning and continuous learning mechanisms. First, each railway node trains a model locally based on wireless communication state data, uploading only the model parameters to the federated server for aggregation. This process avoids sharing raw data, directly alleviating the problem of insufficient model generalization ability caused by data silos, and helps improve the robustness of the global deployment scheme. Second, by introducing a continuous learning mechanism that incorporates historical scene feature memory constraints, the aggregated global model is incrementally updated online. This allows the model to continuously learn new knowledge while retaining historical scene memory when facing dynamic environmental changes such as train speed and user access density. This directly overcomes the shortcomings of traditional static deployment methods, which struggle to adapt to complex dynamic environmental changes over long periods, effectively reducing wireless coverage blind spots and communication latency. Furthermore, by constructing a set of wireless communication state features that includes characteristics such as received signal strength, signal-to-interference-plus-noise ratio, and number of users accessing the network, and training it with a deep neural network, it is possible to more accurately model complex wireless environments, thereby outputting more optimized parameters such as wireless access point deployment location, deployment spacing, transmission power, and handover threshold, ultimately achieving a comprehensive improvement in the continuity of high-speed railway wireless network coverage, handover stability, and resource utilization.

[0016] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the application. The objectives and other advantages of this application are realized and obtained through the structures particularly pointed out in the description, claims and drawings.

[0017] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of this application or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating an optimized method for deploying wireless access points on high-speed railways, provided in an embodiment of this application. Figure 2 A schematic diagram of the high-speed railway wireless communication status data acquisition and local model training process provided in this application embodiment; Figure 3Another flowchart illustrating an optimized method for deploying wireless access points on high-speed railways, provided in an embodiment of this application; Figure 4 A schematic diagram of the federated learning collaborative training and model aggregation framework provided in the embodiments of this application; Figure 5 A schematic diagram illustrating the continuous learning online incremental update and historical scene memory mechanism provided in this application embodiment; Figure 6 This is a schematic diagram illustrating the optimized dynamic deployment of high-speed railway wireless access points according to an embodiment of this application. Figure 7 This is a schematic diagram of a high-speed railway wireless access point deployment system provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0022] The technical solution provided in this application can be applied to the planning and dynamic optimization of wireless communication networks along high-speed railways, and is particularly suitable for multi-line collaborative optimization scenarios that span different geographical environments and administrative regions and have high requirements for data privacy and security. This solution allows railway area nodes distributed in different locations to collaboratively train a highly adaptable wireless access point deployment optimization model without sharing original communication data, and can also complete real-time self-updating of the model in response to environmental changes.

[0023] Before introducing the specific implementation details of this application, let's first explain some key terms that run throughout the text: Railway regional node: refers to an edge computing entity or logical functional unit responsible for wireless communication data collection, model training, and deployment optimization of a specific high-speed railway line or section.

[0024] Wireless communication status data: refers to a series of measurement parameters that can reflect the quality of wireless channels, network load and train operation conditions along high-speed railways.

[0025] Federation server: refers to the central functional entity that coordinates federated learning among various railway regional nodes, and is responsible for aggregating model parameters and distributing the global model.

[0026] Continuous learning mechanism: refers to a training paradigm that enables the model to continuously learn from new data while overcoming the problem of forgetting old knowledge through specific constraints.

[0027] Historical scene feature memory constraint: refers to the constraint term applied in the loss function or model update stage to associate historical scene state features in order to maintain the model's ability to recognize and process previously learned wireless environment scenes during continuous learning.

[0028] Figure 1 A flowchart illustrating an optimized method for deploying wireless access points on high-speed railways is shown. Figure 1 As shown in the figure, this application embodiment provides an optimized method for deploying wireless access points on high-speed railways, the method comprising: S101, collects wireless communication status data of the high-speed railway where the railway area node is located; The wireless communication status data is collected in real time by wireless access points, edge computing nodes and vehicle communication terminals deployed along the high-speed railway. The wireless communication status data includes one or more of the following: train speed, train location information, received signal strength index (RSSI), signal-to-interference-plus-noise ratio (SINR), path loss, communication delay, data throughput, wireless access point load, number of users accessing the railway, handover success rate and packet loss rate.

[0029] S102, train the wireless access point deployment optimization model based on wireless communication status data, and synchronize the parameters of the trained wireless access point deployment optimization model to the federated server to realize the aggregation processing and output of the global wireless access point deployment optimization model for high-speed railway. S103, receive the global wireless access point deployment optimization model issued by the federation server, and perform online incremental updates on the global wireless access point deployment optimization model based on the continuous learning mechanism. The online incremental update introduces historical scene feature memory constraints into the global wireless access point deployment optimization model. S104. Collect real-time wireless communication status data of railway area nodes and input it into the updated global wireless access point deployment optimization model for optimization to obtain the wireless access point deployment optimization result of the railway area node location. The deployment optimization result includes at least one of the following wireless access point deployment parameters: wireless access point deployment location, deployment spacing, transmission power, and wireless handover threshold.

[0030] In this embodiment, the above steps work in synergy: Step S101 acquires basic data describing the current wireless environment, which forms the basis for all subsequent intelligent analysis and decision-making; Step S102 uses this data to train a model locally that adapts to the wireless characteristics of the local area, and then synchronizes the parameters to the federated server, taking a key step in breaking down data silos and aggregating multi-party experience; Step S103 transmits the global model obtained by the federated server from multiple parties back to the local machine, and immediately performs online incremental updates through a continuous learning mechanism that introduces historical scene feature memory constraints—this operation allows the newly acquired global model to quickly adapt to the new, dynamically changing local environment, while ensuring that the model remains compatible with optimized historical scenes, effectively avoiding catastrophic forgetting problems; Step S104 applies the model optimized by both federated aggregation and continuous learning to real-time data, and finally outputs decision parameters such as deployment location, deployment spacing, transmission power, and wireless handover threshold that can directly guide engineering implementation, forming a complete closed loop from perception, learning, collaboration to decision-making. The entire process directly solves the problems of increased coverage blind spots and communication latency caused by the inability to share data and adapt to dynamic environmental changes in traditional static deployment methods.

[0031] Taking a specific application scenario as an example: Suppose there is a long-distance high-speed railway that traverses mountains, plains, and tunnels, with three railway area nodes, A, B, and C, deployed along the line. First, in step S101, nodes A, B, and C respectively collect wireless communication status data such as received signal strength, signal-to-interference-plus-noise ratio, and train speed within their respective jurisdictions. Next, in step S102, each node uses its local data to train a wireless access point deployment optimization model that reflects the characteristics of its own section of the line, and then encrypts and uploads its model parameters to the federated server. The server then aggregates these parameters to generate a global wireless access point deployment optimization model that integrates experience from multiple scenarios, including mountains, plains, and tunnels. In step S103, when the user access density in the plain section where node A is located changes drastically due to the addition of a new station, node A receives the global model and immediately initiates an online incremental update based on a continuous learning mechanism. During the update process, by introducing historical scene feature memory constraints, the model can learn the high-density access characteristics of the new station while retaining the previously learned optimization strategies for high-speed traffic scenarios in open plain areas. Finally, in step S104, node A applies the updated model to the real-time data, accurately outputting the new deployment location of the wireless access point near the new station, the adjusted transmission power, and the wireless handover threshold adapted to high-speed handover, ensuring the continuity and stability of communication after the new station is opened.

[0032] In one embodiment, step S102, training the wireless access point deployment optimization model based on the wireless communication state data, includes: Based on the wireless communication status data, a set of wireless communication status features for high-speed railways is constructed, wherein the set of wireless communication status features includes at least one of the following features: received signal strength, signal-to-interference-plus-noise ratio, train speed, number of users accessing the network, and load status of wireless access points. The wireless communication state feature set is input into a deep neural network for training to obtain a wireless access point deployment optimization model.

[0033] Specifically, extracting a set of features representing the core attributes of the wireless environment from raw wireless communication state data can significantly reduce data dimensionality and remove redundant information. Among these features, received signal strength and signal-to-interference-plus-noise ratio directly quantify channel quality; train speed is a key factor triggering handover and dynamic coverage adjustment; and the number of users accessing the network and the load status of wireless access points reflect the pressure on network resources. By fusing these multi-dimensional key features and training them with a deep neural network possessing strong nonlinear fitting capabilities, the model learns the deep mapping between complex wireless environment states and optimal deployment parameters, rather than relying solely on empirical rules based on a single threshold. This directly improves the accuracy of the final output wireless access point deployment optimization model in perceiving and understanding complex scenarios.

[0034] In one embodiment, the step of inputting the wireless communication state feature set into a deep neural network for training to obtain a wireless access point deployment optimization model includes: The feature set of each wireless access point in the wireless communication state feature set is input into a deep neural network for training to obtain multiple optimized models. The association calculation is performed based on the feature values ​​corresponding to each wireless access point, and the dynamic evaluation is performed based on the calculation results. Based on the evaluation results, one of the multiple optimization models is selected as the wireless access point deployment optimization model corresponding to the location of the railway area node.

[0035] It's important to note that a railway regional node typically contains multiple wireless access points, each with vastly different local environments (e.g., near mountains, tunnels, or densely populated areas). Training a separate model for each access point's feature set captures extremely fine-grained local features, creating multiple "local expert models." Subsequently, correlation calculations and dynamic evaluations are performed based on these feature values, essentially establishing a model selection or arbitration mechanism. This mechanism comprehensively weighs the current performance and scenario adaptability of each model, ultimately selecting the optimal model as the primary model for that node. This optimization strategy avoids using a single set of average parameters to fit all scenarios, allowing model parameters to better match the overall optimal communication strategy for the current node's region, ultimately achieving a superior deployment optimization effect compared to a single global model.

[0036] In another embodiment, the calculation formula for the association calculation based on the feature values ​​corresponding to each wireless access point is as follows: ,in, Indicates the target wireless access point currently being accessed by the user terminal. Indicates the first The signal-to-interference-plus-noise ratio (SIR) for each wireless access point Indicates the first Current load status of each wireless access point Indicates the cost of wireless access point handover. This represents the weighting coefficient of the corresponding indicator.

[0037] In practical applications, by linearly weighting and combining three key indicators—signal-to-interference-plus-noise ratio (SNR), load status, and handover cost—and determining the wireless access point that maximizes its overall benefit, the ideal decision-making process for terminal access selection or network-side load balancing is simulated. A higher SNR indicates better signal quality, which is a positive factor; higher load and greater handover cost indicate poorer post-access experience and stability, which are negative factors. Weighting coefficients... The introduction of this feature allows for flexible adjustment of the relative importance of various factors based on different line operation strategies (such as prioritizing throughput or ensuring uninterrupted connection). This formula, as an evaluation function, ensures that the selected model's internal characteristics tend to favor wireless access points with good signal strength, light load, and smooth handover, directly leading to better network coverage continuity and user communication experience.

[0038] In one embodiment, the expression for calculating the signal-to-interference-plus-noise ratio is: ,in, Indicates the transmit power of the target wireless access point. This indicates the channel gain corresponding to the target wireless access point. This indicates the power of interference signals generated by nearby wireless access points or the external electromagnetic environment. This indicates the system noise power.

[0039] Understandably, this expression precisely defines the core metric for communication quality. The signal-to-interference-plus-noise ratio (SINR) depends not only on the signal strength itself, but more importantly on the ratio of the signal to the interference plus noise. The numerator... The useful signal energy received from the target wireless access point is quantified. The denominator comprehensively considers the total interference power from neighboring wireless access points, as well as the system's inherent noise related to the operating environment. In scenarios with dense high-speed rail deployments, improper deployment locations and transmit power settings for wireless access points can lead to a sharp increase in mutual interference between neighboring access points, resulting in an increased denominator and a deterioration in the signal-to-interference-plus-noise ratio (SNR). The model incorporates this precise calculation during training and evaluation, enabling it to understand that increasing transmit power is not always beneficial, thus outputting deployment optimization results that achieve the best balance between coverage and interference.

[0040] Furthermore, railway regional nodes dynamically construct a set of wireless communication state features based on path loss values ​​and signal-to-interference-plus-noise ratio (SINNR), and combine this with train operation status and user access status to form a high-speed railway wireless communication state matrix, which is used for subsequent training of wireless access point deployment optimization models.

[0041] To evaluate the coverage effect of the high-speed railway wireless network, this embodiment further establishes a wireless coverage model, the expression of which is: ,in, Indicates wireless network coverage. This indicates the total number of high-speed rail zones. This represents the decision function, which determines the communication quality threshold when the signal-to-interference-plus-noise ratio (SIR) of the corresponding region meets the threshold. The value is 1 if the condition is met, and 0 otherwise. This represents the wireless communication quality threshold.

[0042] To evaluate the stability of wireless communication handover on high-speed railways, this embodiment establishes a wireless handover success rate model, the expression of which is: ,in, Indicates the success rate of wireless handover. Indicates the total number of switches. Indicates the first Secondary wireless handover latency This indicates the maximum allowable handover delay threshold of the system. This represents the decision function.

[0043] The deployment optimization model dynamically optimizes and adjusts the deployment location, spacing, and transmission power of high-speed railway wireless access points based on wireless coverage, wireless handover success rate, and wireless communication quality, in order to improve the continuity of high-speed railway wireless network coverage and the stability of wireless communication.

[0044] In one embodiment, before constructing the wireless communication status feature set of the high-speed railway based on the wireless communication status data in step S102, in order to ensure the stability and accuracy of subsequent model training, the method further includes: normalizing, removing outliers, and encoding features on the collected wireless communication status data to reduce the data distribution differences between nodes in different railway areas.

[0045] This data preprocessing is crucial for ensuring successful convergence of federated learning and effective generalization of the aggregated global model. Normalization eliminates the magnitude differences caused by varying units between different data dimensions, preventing gradient bias towards features with large values ​​(such as transmit power) during model training. Outlier removal filters out outliers caused by momentary equipment failures or sudden interference, as such noisy data can mislead the model's learning direction. Feature encoding converts non-numerical state information into numerical forms that the model can understand. This process eliminates the inconsistencies in the distribution of raw data caused by differences in hardware and calibration among nodes in different railway areas, ensuring that the model parameters uploaded by each node are within a similar semantic space. This creates a favorable foundation for weighted aggregation by the federated server, directly improving the aggregation efficiency and generalization ability of the final global wireless access point deployment optimization model.

[0046] In one embodiment, such as Figure 5 As shown, for step S103, this embodiment further introduces a continuous learning mechanism to improve the long-term adaptability of the wireless access point deployment model to changes in the dynamic environment of high-speed railways.

[0047] When the operating environment of high-speed railways changes, including changes in train speed, user access density, weather, electromagnetic environment, or the addition of new railway lines, the railway regional nodes automatically trigger the online incremental update process of the model.

[0048] In this embodiment, the continuous learning mechanism stores historical wireless communication scene features and historical key model parameters through a historical scene feature memory.

[0049] The historical scene feature memory includes information such as historical wireless coverage status, historical path loss distribution, historical user access characteristics, and historical wireless handover status.

[0050] During the incremental update of the model, railway area nodes first match similar historical scenarios based on the current wireless communication status, and call the corresponding historical scenario features to participate in model training, thereby improving the model's adaptability to complex dynamic environments.

[0051] In this embodiment, the online incremental update of the global wireless access point deployment optimization model based on the continuous learning mechanism includes: When a change in the operating environment of a railway area node is detected relative to the previous detection time, new wireless communication status data of the location of the railway area node is acquired. By utilizing a continuous learning mechanism, the new wireless communication state data is matched with similar historical wireless communication state data, and incremental training is performed on the global wireless access point deployment optimization model.

[0052] This section essentially explains when to trigger continuous learning and an effective method for incremental training. Continuous learning doesn't occur constantly; rather, it's driven by changes in the operating environment, significantly saving computational resources and energy. When events such as changes in train speed or user access density are detected, the model begins acquiring new data. Compared to direct full-scale fine-tuning, this approach uses similar historical wireless communication state data for incremental training—a more ingenious design. It not only allows the model to learn new features but also enables it to adapt to new environments more quickly and reduce the forgetting of historical knowledge by finding and learning from experiences in similar historical scenarios. This incremental training mechanism, triggered by environmental changes and incorporating the comparison of old and new knowledge, greatly enhances the model's response speed and adaptability to complex dynamic environments.

[0053] In one embodiment, the step of utilizing a continuous learning mechanism to match similar historical wireless communication state data based on the new wireless communication state data and incrementally training the global wireless access point deployment optimization model includes: Based on the new wireless communication state data, similar historical wireless communication state data are matched from the historical database, and a historical memory constraint loss function is constructed based on the historical wireless communication state data. The global wireless access point deployment optimization model is subjected to constrained incremental training based on the historical memory constraint loss function and the new wireless communication state data. The expression for the historical memory constraint loss function is: ,in, This represents the total loss function of the model. This represents the task loss function in the current high-speed railway communication scenario. Represents the importance coefficient of the model parameters. Indicates the current model parameters. This represents the key model parameters corresponding to the historical scenario. This represents the set of historical scene features that are remembered at the current moment. This represents a set of memories of historical scene features from the previous moment. This represents the weighting coefficient of historical parameter constraints. This represents the weighting coefficient for preserving historical scene memories; The update expression for the historical scene feature memory set is: ,in, Indicates the historical memory retention coefficient. This indicates the characteristics of the current high-speed railway wireless communication scenario.

[0054] It should be noted that this total loss function... It actually consists of three parts, with a very clear direction: the first item To ensure the model can fit the wireless communication state data well in the new environment and complete the learning of the new task; the second item It is a historical parameter constraint term, which calculates the importance coefficient of each parameter in the model to solving old tasks. And constrain the key parameters not to deviate from their historical optimal values ​​during the current training. Too far away, thus locking in historical knowledge at the microscopic parameter level; the third item It is a historical scene memory retention term, which forces the model to represent the set of memories in the feature space after learning new scene features. Unable to connect with the memory of the previous moment Despite significant shifts, knowledge continuity was maintained at the macroscopic feature representation level. Simultaneously, the update formula of the memory set itself... It is an exponential moving average, which ensures that the memory evolves smoothly. In this way, under the joint constraints of both, the model can still retain its optimization ability for old tunnel and bridge scenarios when learning new wireless communication scenarios such as train speed or weather conditions, which greatly improves the robustness and generalization of the model in long-term operation.

[0055] When the addition of new high-speed railway communication scenarios causes changes in key model parameters to exceed preset thresholds, the system automatically triggers a historical knowledge playback mechanism to retrain the model on historical scenario data, thereby reducing the probability of catastrophic forgetting in the model.

[0056] Through the aforementioned continuous learning mechanism, this embodiment can improve the long-term stability and generalization ability of the wireless access point deployment optimization model in complex dynamic environments.

[0057] In one embodiment, the aforementioned historical scene feature memory set Specifically, it includes at least one type of historical information used to comprehensively depict the wireless environment: historical wireless coverage status; historical path loss distribution; historical user access characteristics; and historical wireless handover status.

[0058] This historical information is a crucial piece of the puzzle for reconstructing the full picture of historical scenarios. Historical wireless coverage records the distribution of signal strength at different locations; historical path loss distribution reflects the channel attenuation patterns under specific terrain features; historical user access characteristics characterize traffic volume and user mobility patterns; and historical wireless handover status summarizes the triggering patterns and success rates of handovers. During incremental training, constraints are constructed by extracting these multi-dimensional historical scenario data from the feature memory set, enabling the model to recall the optimal decision logic made when facing these comprehensive states. This ensures the comprehensiveness and accuracy of the constraints, making the model's retention of historical knowledge more robust.

[0059] In one embodiment, after completing the above-described constraint-based incremental training, to further enhance model stability, after performing constraint-based incremental training on the global wireless access point deployment optimization model based on the historical memory constraint loss function and the new wireless communication state data, the method further includes: Detect the changes in key model parameters caused by the addition of new high-speed railway communication scenarios; Determine whether the change exceeds a preset threshold; When the preset threshold is exceeded, the historical knowledge playback mechanism is automatically triggered to retrain the historical scene data.

[0060] When a model experiences drastic changes in key parameters due to learning a completely new high-speed rail communication scenario that differs significantly from previous ones (e.g., a newly opened line using a signaling system drastically different from the past), it means that conventional constraints may not be sufficient to completely prevent the disruption of old knowledge. This step captures this risk by setting a threshold; once the threshold is exceeded, a historical knowledge replay mechanism is automatically triggered. This mechanism remixes stored historical scenario data into the new data for training, pulling the model back to a stable state that balances old and new knowledge in a more direct and forceful way, effectively preventing a precipitous drop in model performance caused by rare new scenario shocks. This mechanism directly ensures the smooth and controlled evolution of the model's performance when encountering various complex and extreme changes in operating environments.

[0061] For the process of training the wireless access point deployment optimization model, such as Figure 2 As shown, after completing the collection of wireless communication status data, the railway area node preprocesses the wireless communication status data and establishes a local wireless access point deployment optimization model based on the processed wireless communication status feature set.

[0062] In this embodiment, the local wireless access point deployment optimization model is constructed using a deep neural network. The input is a set of wireless communication state features, and the output is wireless access point deployment parameters.

[0063] The set of wireless communication state features includes:

[0064] in, Indicates the received signal strength. Indicates the signal-to-interference-plus-noise ratio. Indicates the train's speed. Indicates the number of users connected. This indicates the load status of the wireless access point.

[0065] In this embodiment, the deep neural network includes an input layer, a hidden layer, and an output layer; the input layer is used to receive a set of wireless communication state features, the hidden layer is used to extract the wireless communication feature relationships in the high-speed railway operating environment, and the output layer is used to output wireless access point deployment optimization parameters.

[0066] The wireless access point deployment optimization parameters include wireless access point deployment location, deployment spacing, transmission power, coverage area, and wireless handover threshold.

[0067] In this embodiment, the railway user terminal performs wireless access point association selection based on the wireless communication quality, load status, and wireless handover cost of the wireless access point, so as to improve the continuity of high-speed railway wireless network coverage and the stability of wireless handover.

[0068] In this embodiment, the railway user terminal dynamically evaluates the comprehensive correlation benefit value corresponding to different wireless access points and selects the target wireless access point with the best comprehensive communication performance to reduce the probability of communication interruption caused by frequent wireless handover in high-speed railway high-speed movement scenarios and improve the utilization efficiency of wireless network resources.

[0069] The railway area nodes iteratively train a deep neural network based on local wireless communication status data and update the network parameters using the gradient descent method, thereby obtaining the local model parameters of the corresponding railway area nodes.

[0070] In this embodiment, the original wireless communication data is not shared between the nodes in each railway area. Instead, local model parameters are uploaded to the federated server for collaborative training, thereby reducing the risk of railway communication data leakage.

[0071] In this embodiment, as Figure 6 As shown, for step S104, the railway area node outputs the dynamic optimization results of the wireless access point based on the continuously updated global model.

[0072] The optimization results include parameters such as wireless access point deployment location, deployment spacing, transmission power, coverage area, and wireless handover threshold.

[0073] In this embodiment, the system dynamically adjusts the wireless access point deployment strategy based on the high-speed railway operating status and changes in wireless communication quality.

[0074] In this embodiment, to avoid frequent switching of wireless access points during high-speed train operation, the system adopts a dynamic handover decision mechanism, the expression of which is:

[0075] in, Indicates the signal strength of the target wireless access point. Indicates the current signal strength of the wireless access point. This indicates the wireless handover threshold.

[0076] When the system detects an expansion of wireless coverage blind spots, an increase in wireless handover failure rate, or abnormal load on wireless access points, it automatically re-optimizes the deployment parameters of wireless access points, thereby improving the continuity of high-speed railway wireless network coverage, the stability of wireless handover, and the efficiency of communication resource utilization.

[0077] By implementing the method described above, wireless communication status data during high-speed railway operation is first collected, including train speed, received signal strength, signal-to-noise ratio, user density, and wireless access point load. Then, each railway node trains a wireless access point deployment model based on local data, and uploads the model parameters to a central server via a federated learning framework for global aggregation, achieving multi-line collaborative optimization without sharing original communication data. Furthermore, a continuous learning mechanism is introduced, employing a strategy of maintaining historical key parameters and incremental updates to enable the model to adaptively learn online based on changes in the communication environment of the high-speed railway scenario, effectively mitigating the catastrophic forgetting problem in traditional incremental learning. Finally, the optimal deployment location, transmission power, and handover strategy for wireless access points are output. This method can improve the continuity of high-speed railway wireless network coverage, handover stability, and resource utilization efficiency, while reducing the cost of manual deployment and repetitive optimization, demonstrating good real-time performance, adaptability, and engineering application value.

[0078] like Figure 3 and 4 The diagram illustrates another implementation of the optimized method for deploying high-speed railway wireless access points provided in this application. This method is provided from a system perspective, comprising multiple railway area nodes and a federated server. The method includes steps S201-S207, wherein steps S201-S202 and S206-S207 are the same as steps S101-S104, and steps S203-S205 are as follows: S203 receives parameters from multiple railway area nodes regarding the wireless access point deployment optimization model.

[0079] S204. The model parameters uploaded by each railway area node are weighted and aggregated to obtain a global wireless access point deployment optimization model.

[0080] S205 distributes the global wireless access point deployment optimization model to each railway area node.

[0081] It should be noted that the execution entity for steps S203-S205 is the federation server, while the execution entity for steps S201-S202 and S206-S207 is the railway area node. Each railway area node will independently execute steps S201-S202 and S206-S207, and after completing S201-S202 and S206-S207, it will send the obtained model parameters to the federation server. The federation server will then perform parameter fusion by executing steps S203-S205 to output a global wireless access point deployment optimization model, which will then be distributed to each railway area node in sequence.

[0082] In other words, the federated server does not access any raw data; it merely acts as a central hub for aggregating and distributing model parameters. By receiving optimized model parameters uploaded from geographically dispersed railway regional nodes that possess data on different railway line environments, the federated server builds a distributed knowledge base. The weighted aggregation process is a form of knowledge fusion, extracting general wireless network optimization rules across lines and scenarios. This results in a globally comprehensive wireless access point deployment optimization model that is more knowledgeable and generalizable than any local model. This model is then distributed, forming a closed-loop knowledge transfer mechanism. This allows each railway regional node to benefit from the indirect experience of other nodes, directly solving the data silo problem caused by the inability to share data.

[0083] In another embodiment, the weighted aggregation of model parameters uploaded by each railway area node to obtain a global wireless access point deployment optimization model includes: The aggregation weight coefficients of each node are dynamically adjusted based on the communication quality, data scale, and training contribution of each railway area node. The federated averaging algorithm is used to weight and aggregate the model parameters uploaded by each railway area node based on the aggregation weight coefficient to obtain a global wireless access point deployment optimization model.

[0084] This approach introduces dynamically adjusted aggregation weight coefficients based on multiple factors, making the federated aggregation process more intelligent and fair. Nodes with poor communication quality may upload parameters containing significant noise, and their weights should be reduced. Nodes with large data volumes typically have more reliable local models, and their contributions should be higher. Training contribution can be understood as the model gain achieved by a node on its data; better gains indicate more effective local learning and should be given higher weights. A federated averaging algorithm is used, employing these dynamically adjusted weight coefficients to perform a weighted average of the local model parameters of each node, replacing a simple average. This allows the final aggregation result to automatically favor nodes with higher-quality data, more stable communication, and more effective models, significantly improving the performance ceiling and robustness of the global wireless access point deployment optimization model.

[0085] The expression for the federated averaging algorithm is as follows: ,in, Indicates the first The global model parameters after round aggregation Indicates the first Local model parameters for each railway regional node. Indicates the first Number of data samples per railway regional node This represents the total number of data samples across all nodes participating in the training. This represents the total number of rail region nodes participating in federal training. Indicates the first The dynamic aggregation weight coefficients corresponding to each railway regional node.

[0086] Each local model parameter The contribution is determined by the proportion of its data sample. and dynamic aggregation weight coefficient A joint decision. The proportion of data samples reflects the basic information content of a node, while This is a dynamic correction factor that integrates communication quality and training contribution. For example, if node A has a large amount of data but extremely poor communication quality, leading to distorted uploaded parameters, then its... The size of the aggregate will be small, thus weakening its influence in the aggregation. This refined aggregation calculation is the core guarantee that the final output global model can perform well in various complex cross-scenarios.

[0087] In this embodiment, the federated server performs weighted aggregation of the model parameters uploaded by each node based on the data scale, communication quality, and training contribution of nodes in different railway regions.

[0088] After the federated server completes model aggregation, it distributes the aggregated global model to each railway area node to achieve collaborative optimization of the model deployment across multiple wireless access points.

[0089] As a preferred implementation, after the federated server receives the parameters of the wireless access point deployment optimization model uploaded by multiple railway area nodes, in order to further improve the efficiency of federated training, it may also include: compressing the received model parameters and using an asynchronous update mechanism to aggregate the parameters, so as to reduce the communication load of the federated training process.

[0090] It's worth noting that parameter compression, such as gradient sparsity or quantization, can significantly reduce the size of parameters that need to be uploaded and distributed, directly reducing network bandwidth consumption. The asynchronous update mechanism allows the federated server to begin aggregation without waiting for all nodes to have finished uploading. Instead, it can set a trigger condition (such as a time window or after receiving a certain number of parameters) to immediately begin aggregation and distribute the updated model to ready nodes. This non-blocking, asynchronous parallel approach avoids the problem of individual node communication delays or slow computation slowing down the entire federated training iteration process. The combined use of these two measures greatly improves the real-time performance and system throughput of federated training in large-scale, cross-regional deployments.

[0091] In one embodiment, after collecting real-time wireless communication status data of the railway area nodes and inputting it into the updated global wireless access point deployment optimization model for optimization to obtain the wireless access point deployment optimization result for the location of the railway area nodes, the method further includes: Using a pre-constructed objective function for optimizing the deployment of wired access points, the wireless access points of the railway area nodes are deployed based on the optimization results. The expression for the objective function for optimizing the deployment of line access points is as follows: ,in, This represents the overall optimization target value for wireless access point deployment. Indicates wireless coverage. Indicates the success rate of wireless handover. Indicates wireless resource utilization. Indicates system communication power consumption. This represents the weighting coefficient of the corresponding indicator.

[0092] This objective function constitutes the highest-level action plan when the model makes its final decision on the deployment scheme. It does not solely pursue the optimization of a single metric, but rather seeks the optimal balance between wireless coverage, wireless handover success rate, wireless resource utilization, and system communication power consumption through a linearly weighted combination. Weighting coefficients This allows operators to fine-tune the system based on the actual operational needs of different sections of the high-speed railway (for example, in accident-prone sections, handover success rate may be of paramount importance, while in energy-constrained sections, system communication power consumption may be of paramount importance). The model is optimized based on this function, and the series of deployment parameters output ensures that the final deployment scheme is a solution that achieves optimal trade-offs among multiple key performance indicators and is highly consistent with actual needs.

[0093] In one embodiment, after the initial deployment is completed under the guidance of the above objective function, in order to ensure service continuity during train operation, after deploying the wireless access nodes of the railway area nodes using the pre-built wireless access point deployment optimization objective function, the following steps are further included: Obtain the signal strength of the target wireless access point and the current wireless access point; Dynamic handover decisions are made based on preset wireless handover thresholds. When the difference between the signal strength of the target wireless access point and the signal strength of the current wireless access point meets the wireless handover threshold, wireless access point handover is performed.

[0094] It should be noted that this step describes the dynamic handover execution logic based on optimized parameters (especially the wireless handover threshold). Handover is triggered by comparing the signal strength difference between the target wireless access point and the current wireless access point in real time during the post-deployment operational phase, and comparing this difference with a preset wireless handover threshold. This threshold is one of the results of the aforementioned model optimization. A well-chosen threshold can effectively avoid the ping-pong effect (handover back and forth at the critical point due to an excessively small threshold) and prevent dropped calls caused by untimely handover. This dynamic decision ensures that even when the train's high-speed movement causes rapid fluctuations in wireless signal, the terminal can always smoothly and promptly attach to the wireless access point with better signal quality and network load, directly guaranteeing communication continuity and service stability under high-speed movement.

[0095] In this embodiment, after performing the wireless access point handover, the method further includes: Detect whether the wireless coverage blind spot has expanded; Detect whether the wireless handover failure rate is increasing; Check if the load on the wireless access point is abnormal; When any detection result meets the preset abnormal conditions, the deployment location, deployment spacing, transmission power, and wireless handover threshold of the wireless access point are re-optimized.

[0096] In summary, the federated learning mechanism enables collaborative training among multiple railway lines, improving the optimization capability of wireless access point deployment and reducing the risk of railway communication data leakage without sharing original wireless communication data. The continuous learning mechanism enables online incremental updates in the dynamic communication environment of high-speed railways, enhancing the long-term adaptability of the wireless access point deployment model to complex environmental changes. The historical scene feature memory and parameter constraint mechanism reduce catastrophic forgetting during continuous model updates, improving model stability and generalization ability. Dynamic wireless communication state modeling enables real-time optimization of wireless access point deployment parameters, improving the continuity of high-speed railway wireless network coverage, the stability of wireless handover, and the efficiency of communication resource utilization.

[0097] Figure 7 This is a schematic diagram illustrating the composition of a high-speed railway wireless access point deployment system provided in an embodiment of this application. This application also provides a high-speed railway wireless access point deployment system, including: a railway area node 710 and a federated server 720; Railway area node 710 is used to collect wireless communication status data of the high-speed railway where it is located; it trains a wireless access point deployment optimization model based on this wireless communication status data, and synchronizes the parameters of the trained wireless access point deployment optimization model to the federated server 720 to achieve aggregated processing and output of a global wireless access point deployment optimization model for the high-speed railway; it receives the global wireless access point deployment optimization model issued by the federated server 720, and performs online incremental updates to the global wireless access point deployment optimization model based on a continuous learning mechanism, wherein the online incremental update introduces historical scene feature memory constraints into the global wireless access point deployment optimization model; it collects data on the railway area... The real-time wireless communication status data of node 710 is input into the updated global wireless access point deployment optimization model for optimization, resulting in the wireless access point deployment optimization result for the location of node 710 in the railway area. This deployment optimization result includes at least one of the following wireless access point deployment parameters: deployment location, deployment spacing, transmission power, and wireless handover threshold of the wireless access point. A federation server 720 is used to receive the parameters of the wireless access point deployment optimization model uploaded by multiple railway area nodes 710. The model parameters uploaded by each railway area node 710 are weighted and aggregated to obtain the global wireless access point deployment optimization model. The global wireless access point deployment optimization model is then distributed to each railway area node 710.

[0098] This embodiment collects high-speed railway wireless communication status data at the locations of railway regional nodes; trains a wireless access point deployment optimization model based on this data, and synchronizes the trained model parameters to a federated server for aggregation, outputting a global wireless access point deployment optimization model; receives the global model from the federated server and performs online incremental updates to the global model based on a continuous learning mechanism; collects real-time wireless communication status data of railway regional nodes and inputs it into the updated global model for optimization, obtaining the wireless access point deployment optimization results at the locations of railway regional nodes. These results include at least one of the following: deployment location, deployment spacing, transmission power, and wireless handover threshold of the wireless access points.

[0099] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0100] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the technical scope disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. An optimized method for deploying wireless access points on high-speed railways, characterized in that, include: Collect wireless communication status data of the high-speed railway where the railway area node is located; The wireless access point deployment optimization model is trained based on the wireless communication status data, and the parameters of the trained wireless access point deployment optimization model are synchronized to the federated server to realize the aggregation processing and output of the global wireless access point deployment optimization model for high-speed railway. The system receives the global wireless access point deployment optimization model issued by the federation server and performs online incremental updates on the global wireless access point deployment optimization model based on a continuous learning mechanism. The online incremental update involves introducing historical scene feature memory constraints into the global wireless access point deployment optimization model. Real-time wireless communication status data of the railway area nodes are collected and input into the updated global wireless access point deployment optimization model for optimization, so as to obtain the wireless access point deployment optimization result of the railway area node location. The deployment optimization result includes at least one of the following wireless access point deployment parameters: deployment location of wireless access points, deployment spacing, transmission power and wireless handover threshold.

2. The optimized method for deploying high-speed railway wireless access points according to claim 1, characterized in that, The training of the wireless access point deployment optimization model based on the wireless communication state data includes: Based on the wireless communication status data, a set of wireless communication status features for high-speed railways is constructed, wherein the set of wireless communication status features includes at least one of the following features: received signal strength, signal-to-interference-plus-noise ratio, train speed, number of users accessing the network, and load status of wireless access points. The wireless communication state feature set is input into a deep neural network for training to obtain a wireless access point deployment optimization model.

3. The optimized method for deploying high-speed railway wireless access points according to claim 2, characterized in that, The step of inputting the wireless communication state feature set into a deep neural network for training to obtain a wireless access point deployment optimization model includes: The feature set of each wireless access point in the wireless communication state feature set is input into a deep neural network for training to obtain multiple optimized models. The association calculation is performed based on the feature values ​​corresponding to each wireless access point, and the dynamic evaluation is performed based on the calculation results. Based on the evaluation results, one of the multiple optimization models is selected as the wireless access point deployment optimization model corresponding to the location of the railway area node.

4. The optimized method for deploying high-speed railway wireless access points according to claim 3, characterized in that, The calculation formula for correlation calculation based on the feature values ​​corresponding to each wireless access point is as follows: ,in, Indicates the target wireless access point currently being accessed by the user terminal. Indicates the first The signal-to-interference-plus-noise ratio (SIR) for each wireless access point Indicates the first Current load status of each wireless access point Indicates the cost of wireless access point handover. This represents the weighting coefficient of the corresponding indicator.

5. The optimized method for deploying high-speed railway wireless access points according to claim 4, characterized in that, The expression for calculating the signal-to-interference-plus-noise ratio is: ,in, Indicates the transmit power of the target wireless access point. This indicates the channel gain corresponding to the target wireless access point. This indicates the power of interference signals generated by nearby wireless access points or the external electromagnetic environment. This indicates the system noise power.

6. The optimized method for deploying high-speed railway wireless access points according to claim 1, characterized in that, The online incremental update of the global wireless access point deployment optimization model based on the continuous learning mechanism includes: When a change in the operating environment of a railway area node is detected relative to the previous detection time, new wireless communication status data of the location of the railway area node is acquired. By utilizing a continuous learning mechanism, the new wireless communication state data is matched with similar historical wireless communication state data, and incremental training is performed on the global wireless access point deployment optimization model.

7. The optimized method for deploying high-speed railway wireless access points according to claim 6, characterized in that, The step of utilizing a continuous learning mechanism to match similar historical wireless communication state data based on the new wireless communication state data and incrementally training the global wireless access point deployment optimization model includes: Based on the new wireless communication state data, similar historical wireless communication state data are matched from the historical database, and a historical memory constraint loss function is constructed based on the historical wireless communication state data. The global wireless access point deployment optimization model is subjected to constrained incremental training based on the historical memory constraint loss function and the new wireless communication state data. The expression for the historical memory constraint loss function is: ,in, This represents the total loss function of the model. This represents the task loss function in the current high-speed railway communication scenario. Represents the importance coefficient of the model parameters. Indicates the current model parameters. This represents the key model parameters corresponding to the historical scenario. This represents the set of historical scene features that are remembered at the current moment. This represents a set of memories of historical scene features from the previous moment. This represents the weighting coefficient of historical parameter constraints. This represents the weighting coefficient for preserving historical scene memories; The update expression for the historical scene feature memory set is: ,in, Indicates the historical memory retention coefficient. This indicates the characteristics of the current high-speed railway wireless communication scenario.

8. The optimized method for deploying high-speed railway wireless access points according to any one of claims 1-7, characterized in that, After synchronizing the parameters of the trained wireless access point deployment optimization model to the federated server, the process also includes: Receive parameters of the wireless access point deployment optimization model uploaded by multiple railway area nodes; The model parameters uploaded by each railway area node are weighted and aggregated to obtain a global wireless access point deployment optimization model; The global wireless access point deployment optimization model is distributed to each railway area node.

9. The optimized method for deploying high-speed railway wireless access points according to claim 8, characterized in that, The weighted aggregation of model parameters uploaded by each railway area node yields a global wireless access point deployment optimization model, including: The aggregation weight coefficients of each node are dynamically adjusted based on the communication quality, data scale, and training contribution of each railway area node. The federated averaging algorithm is used to weight and aggregate the model parameters uploaded by each railway area node based on the aggregation weight coefficient to obtain a global wireless access point deployment optimization model.

10. A high-speed railway wireless access point deployment system, characterized in that, include: Railway area nodes are used to collect wireless communication status data of the high-speed railway where the railway area node is located; The wireless access point deployment optimization model is trained based on the wireless communication status data, and the parameters of the trained wireless access point deployment optimization model are synchronized to the federated server to achieve aggregated processing and output of a global wireless access point deployment optimization model for high-speed railways. The global wireless access point deployment optimization model is received from the federated server, and online incremental updates are performed on the global wireless access point deployment optimization model based on a continuous learning mechanism. The online incremental update introduces historical scene feature memory constraints into the global wireless access point deployment optimization model. Real-time wireless communication status data of the railway area nodes is collected and input into the updated global wireless access point deployment optimization model for optimization, resulting in the wireless access point deployment optimization results for the locations of the railway area nodes. The deployment optimization results include at least one of the following wireless access point deployment parameters: deployment location, deployment spacing, transmission power, and wireless handover threshold of the wireless access points. The federated server receives parameters of the wireless access point deployment optimization model uploaded by multiple railway area nodes; performs weighted aggregation on the model parameters uploaded by each railway area node to obtain a global wireless access point deployment optimization model; and distributes the global wireless access point deployment optimization model to each railway area node.

Images