Railway mobile communication antenna remote control and adaptive optimization system and method

Abstract

The invention discloses a remote control and adaptive optimization system and method for a railway mobile communication antenna, and the system comprises a data collection module which is used for collecting the multi-source heterogeneous data of the physical attitude parameters of the antenna, the wireless link quality index parameters and the surrounding environment information in real time, the collected multi-source heterogeneous data is transmitted to the central control platform; the central control platform is used for generating an optimal antenna adjustment strategy and outputting a control instruction of antenna adjustment based on a built-in intelligent decision-making module by adopting a reinforcement learning algorithm and fusing historical operation data with current multi-dimensional perception input; and the antenna control module is used for analyzing and executing the control instruction and adjusting the attitude of the antenna through an executing mechanism so as to realize high-precision dynamic adjustment of the antenna. By fusing environment perception, intelligent decision and a high-reliability communication mechanism, the purposes of staff reduction, efficiency improvement and dynamic real-time high-precision adjustment are achieved, and intelligent railway development is promoted.

Description

Technical Field

[0001] This invention relates to the field of antenna control technology, specifically to a remote control and adaptive optimization system and method for railway mobile communication antennas. Background Technology

[0002] In railway train control systems, the GSM-R (Global System for Mobile Communications – Railway) mobile communication network undertakes the critical tasks of train operation control and train-to-ground information transmission, and its network coverage quality is directly related to train operation safety. However, because railway mobile communication systems typically use strip-shaped base stations and directional antennas for coverage, these antennas are affected by factors such as changes in the line alignment, additions or subtractions of buildings along the line, terrain evolution, and wind loads. This causes the original installation angle to gradually deviate from the design state. The signals emitted by the mobile communication system antennas, which have only been optimized once during the engineering construction phase, cannot achieve linear and uniform coverage of the line for a long period of time, which can seriously affect the normal operation of trains.

[0003] To ensure train communication quality, existing technologies typically rely on periodic testing and network optimization to calibrate base station antenna parameters (including horizontal azimuth and vertical elevation angles). However, the current network has a massive antenna network. For example, on the CTC-3 (China Train Control System Level 3) line, the distance between base stations is approximately 3–3.5 km, and each base station is generally equipped with two antennas, resulting in a huge cumulative number across the country. At the same time, existing optimization methods mainly rely on maintenance personnel climbing towers to manually adjust the antennas on-site. This not only poses risks associated with working at heights but also results in long optimization cycles, high labor costs, and difficulty in guaranteeing adjustment accuracy due to subjective experience.

[0004] In addition, existing antenna control methods are mostly periodic static optimizations, which cannot respond in real time to changes in the dynamic electromagnetic environment. For example, when the channel degrades, multipath interference increases, or obstacles are generated on the trackside, the antenna coverage is difficult to adjust and compensate in a timely manner, resulting in frequent fluctuations in communication quality, which is not conducive to the reliable operation of the train control system.

[0005] In summary, existing railway mobile communication antenna control technology has several shortcomings: First, optimization methods heavily rely on manual operation, which is not only inefficient but also poses significant personal safety risks. Second, parameter adjustment lacks a refined control mechanism, leading to excessive reliance on the operator's subjective experience for optimization results. Third, the system struggles to continuously and effectively compensate for dynamically changing electromagnetic environments and coverage performance degradation that occurs during long-term operation. Finally, in a large-scale railway network environment, existing technologies cannot meet the practical needs of rapid iterative updates and intelligent operation and maintenance.

[0006] To address the aforementioned technical shortcomings, there is currently a lack of a technical solution for remote, precise, and efficient adjustment of railway mobile communication antenna parameters to support the continuous network optimization needs in the context of intelligent railways. Summary of the Invention

[0007] This application provides a remote control and adaptive optimization system and method for railway mobile communication antennas to solve the problems of heavy manual high-risk operation, parameter adjustment accuracy limited by human experience, and lack of real-time dynamic adaptive capability in the prior art.

[0008] According to a first aspect, one embodiment provides a railway mobile communication antenna remote control and adaptive optimization system, the system comprising:

[0009] The data acquisition module is used to collect multi-source heterogeneous data in real time, including antenna physical attitude parameters, wireless link quality index parameters and surrounding environmental information, and transmit the collected multi-source heterogeneous data to the central control platform.

[0010] The central control platform, based on the built-in intelligent decision-making module, uses reinforcement learning algorithms to integrate historical operating data with current multi-dimensional sensing inputs, generate the optimal antenna adjustment strategy, and output control commands for antenna adjustment.

[0011] The antenna control module is used to parse and execute the control commands, and adjust the antenna attitude through the actuator to achieve high-precision dynamic antenna adjustment.

[0012] Furthermore, the data acquisition module is specifically used for:

[0013] Multi-source heterogeneous data acquisition is achieved by deploying attitude measurement devices and radio frequency performance monitoring devices inside the antenna body, as well as environmental sensing devices installed on the external structure of the antenna and the base station area.

[0014] The physical attitude parameters include azimuth, elevation, polarization direction and their rotational trends, which are used to accurately describe the current pointing state of the antenna; the wireless link quality indicators include RSSI, SINR, RSRP, CQI, VSWR, return loss, neighboring cell interference intensity, handover success rate and link interruption frequency, which are used to dynamically evaluate the stability of the communication link and signal integrity; the surrounding environmental information includes temperature, humidity, wind speed, wind direction, thunderstorm conditions, electromagnetic interference level, changes in the shape of buildings around the base station, terrain slope, curve curvature, real-time position and speed of trains, which are used to identify external interference factors that may cause signal fading.

[0015] Furthermore, the intelligent decision-making module is specifically used for:

[0016] An antenna attitude optimization model is constructed based on a deep reinforcement learning framework, and the learning process is driven by the environmental state vector and the reward function. The antenna attitude optimization model takes multi-source sensing state information, including antenna physical attitude parameters, wireless link quality indicators and surrounding environment information, as input and takes the optimal antenna physical attitude information as output.

[0017] Furthermore, the intelligent decision-making module is specifically used for:

[0018] A deep reinforcement learning framework based on Actor-Critic is adopted, including:

[0019] The action generation unit is responsible for generating continuous antenna attitude control action commands based on the input state information.

[0020] The performance evaluation unit is responsible for evaluating the degree of improvement in link quality after antenna adjustment. Based on the reward function, it comprehensively considers various factors including communication link stability, signal strength, energy consumption and control smoothness, generates a reward signal through weighted calculation, and feeds the corresponding signal back to the action generation unit, thereby achieving end-to-end strategy optimization.

[0021] Furthermore, the intelligent decision-making module is specifically used for:

[0022] Supports online transfer learning: Based on the differences in railway lines, climate conditions, or terrain environments, when the detected changes in the line environment characteristics exceed the set threshold, the transfer learning process is automatically triggered to fine-tune the model parameters without the need for manual retraining.

[0023] Furthermore, the intelligent decision-making module is specifically used for:

[0024] Supports multi-antenna collaborative optimization strategy: When the train runs to the overlapping area of ​​base station coverage, the antennas of multiple adjacent base stations are synchronously scheduled. Based on the real-time position, speed and track curvature of the train, the main lobe pointing, beamwidth and downtilt angle of each antenna are dynamically coordinated to achieve smooth connection of coverage area and seamless transition of signal strength.

[0025] Furthermore, the antenna control module is specifically used for:

[0026] It integrates a high-precision angle adjustment device based on a worm gear transmission mechanism and has a self-locking function, allowing only the worm to drive the worm wheel, while reverse transmission is mechanically locked.

[0027] The built-in attitude real-time feedback unit is used for closed-loop verification of the consistency between the actual adjustment results and the target command, ensuring that the antenna is accurately pointed to the train's running trajectory;

[0028] It has a fault self-diagnosis mechanism that issues an alarm when an abnormal situation is detected, and automatically triggers a safety policy to instruct the antenna to return to the preset default posture, ensuring basic coverage and preventing equipment damage and safety accidents.

[0029] Furthermore, the system also includes:

[0030] The communication and networking module is used for data interaction between modules through the railway dedicated optical transmission backbone network or 5G private network. It has the ability to automatically switch between wired / wireless dual channels and monitor link status, and provides a standardized API interface for seamless integration with the existing GSM-R network management platform.

[0031] Furthermore, the central control platform is also specifically used for:

[0032] As the core of human-computer interaction and system management, it provides a graphical monitoring interface to realize real-time display of antenna operation status, historical data backtracking query, performance trend prediction and analysis, as well as centralized scheduling of antennas at multiple sites across regions, anomaly detection and remote system version updates.

[0033] According to a second aspect, one embodiment provides a method for remote control and adaptive optimization of a railway mobile communication antenna, the method comprising:

[0034] The system collects multi-source heterogeneous data in real time, including antenna physical attitude parameters, wireless link quality index parameters, and surrounding environmental information, and transmits the collected multi-source heterogeneous data to the central control platform.

[0035] The intelligent decision-making module built into the central control platform uses reinforcement learning algorithms to integrate historical operating data with current multi-dimensional sensing inputs, generate the optimal antenna adjustment strategy, and output control commands for antenna adjustment.

[0036] The control commands are parsed and executed, and the antenna attitude is adjusted through the actuator to achieve high-precision dynamic antenna adjustment.

[0037] This application provides a remote control and adaptive optimization system and method for railway mobile communication antennas, specifically including: a data acquisition module for real-time acquisition of multi-source heterogeneous data, including antenna physical attitude parameters, wireless link quality index parameters, and surrounding environmental information, and transmitting the acquired multi-source heterogeneous data to a central control platform; a central control platform for generating an optimal antenna adjustment strategy based on a built-in intelligent decision-making module using a reinforcement learning algorithm, fusing historical operating data with current multi-dimensional sensing input, and outputting control commands for antenna adjustment; and an antenna control module for parsing and executing the control commands, adjusting the antenna attitude through an actuator to achieve high-precision dynamic antenna adjustment. Through this invention, railway mobile communication antennas can adaptively adjust their attitude according to the dynamic operating status of trains and complex environmental changes, effectively improving the continuity of railway communication network coverage, anti-interference capability, and overall service quality, providing reliable communication guarantees for train safety and railway transportation efficiency. Attached Figure Description

[0038] Figure 1 A flowchart illustrating a remote control and adaptive optimization method for railway mobile communication antennas, provided in one embodiment of the present invention;

[0039] Figure 2 A flowchart illustrating the generation of target antenna control strategies using a deep reinforcement learning algorithm based on an Actor-Critic architecture in a railway mobile communication antenna remote control and adaptive optimization system, as provided in an embodiment of the present invention.

[0040] Figure 3 The flowchart illustrates an adaptive dynamic update algorithm in a railway mobile communication antenna remote control and adaptive optimization system, as provided in one embodiment of the present invention. Detailed Implementation

[0041] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of this application. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to this application are not shown or described in the specification. This is to avoid obscuring the core parts of this application with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0042] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0043] The first embodiment of this invention provides a remote control and adaptive optimization system for railway mobile communication antennas, deployed at communication base stations along the railway line. It is collaboratively constructed by a data acquisition module, a communication and networking module, a central control platform, an intelligent decision-making module, and an antenna control module. Figure 1 As shown, the system utilizes sensing, deep learning inference, remote closed-loop control, status feedback verification, and strategy iterative optimization as its core technologies to achieve autonomous adjustment of railway mobile communication antennas.

[0044] The data acquisition module is configured on the antenna body and its surrounding environment to acquire in real time the antenna's physical attitude parameters (including horizontal azimuth, elevation angle and polarization direction), wireless link quality indicators (Received Signal Strength Indicator (RSSI) and Signal-to-Interference-Ratio (SINR)), electromagnetic interference level, meteorological parameters (including wind speed, wind direction and ambient temperature) and equipment operating status information, forming a multi-source heterogeneous environmental and operating condition dataset.

[0045] The intelligent decision-making module, deployed on the central server, uses reinforcement learning algorithms to integrate historical operating data with current multi-dimensional sensing input, continuously learns online and generates the optimal antenna adjustment strategy, and outputs control commands for antenna adjustment to maximize communication link stability and coverage efficiency.

[0046] The antenna control module is used to parse and execute control commands generated by the intelligent decision module. It adjusts the antenna elevation and azimuth angles through a worm gear transmission mechanism to achieve high-precision dynamic antenna adjustment. At the same time, the worm gear transmission mechanism has a self-locking characteristic, which only allows the worm to drive the worm wheel. The reverse transmission is mechanically locked to prevent the antenna angle from changing under conditions such as gravity, wind, and power failure.

[0047] The communication and networking module is built on the standard TCP / IP protocol stack. It supports bidirectional data interaction through railway dedicated optical transmission backbone network or 5G private network. It has the ability to automatically switch between wired / wireless dual channels and link status monitoring, and provides standardized API interfaces to seamlessly connect to the existing GSM-R network management platform.

[0048] The central control platform, as the core of human-computer interaction and system management, provides a graphical monitoring interface, supports real-time display of antenna operating status, historical data backtracking and query, performance trend prediction and analysis, as well as centralized scheduling of antennas at multiple sites across regions, anomaly detection, and remote system version updates.

[0049] In this embodiment, the data acquisition module specifically includes:

[0050] Embedded sensing nodes communicate with the antenna control module via CAN bus or Ethernet to acquire attitude and environmental parameters in real time.

[0051] The data fusion unit employs a weighted Kalman filter algorithm to denoise and dynamically weight the data from multiple sensor nodes, thereby reducing the impact of time drift and measurement noise and outputting high-confidence environmental state estimation results.

[0052] The intelligent decision-making module specifically includes:

[0053] An adaptive control unit is used to establish an antenna control model based on a deep reinforcement learning algorithm, and the learning process is driven by the environmental state vector and the reward function.

[0054] The strategy optimization unit is used to comprehensively evaluate the stability of the communication link, signal strength, energy consumption and control smoothness based on the reward function, and dynamically generate the optimal adjustment strategy for the antenna elevation angle and azimuth angle.

[0055] In this embodiment, the adaptive control unit adopts an Actor-Critic structure, including:

[0056] The Actor unit is responsible for generating continuous angle control commands based on the current environmental conditions. This unit takes a state vector containing the antenna's current elevation and azimuth angles, electromagnetic interference level, wind speed, wind direction, ambient temperature, and polarization direction as input, and outputs adjustments to the elevation, azimuth, and polarization directions. The azimuth adjustment range is limited to (-15°, +15°), the elevation adjustment range is (-60°, 0°), and the minimum adjustment step for all control commands is (0.5^circ) to match the drive accuracy requirements of the worm gear transmission mechanism.

[0057] The performance evaluation unit (Critic) is used to evaluate the performance of the control strategy output by the action generation unit. This unit takes link quality indicators as input and comprehensively considers three factors: signal-to-noise ratio improvement, signal coverage improvement, and the degree of wear and tear on the mechanical structure caused by the operation. It generates a reward signal through weighted calculation and feeds this signal back to the action generation unit, thereby achieving end-to-end strategy optimization.

[0058] In this embodiment, the intelligent decision-making module also has online transfer learning capabilities, including:

[0059] The online transfer learning module is used to automatically trigger the transfer learning process when changes in the line environment characteristics (such as changes in surrounding buildings, strong electromagnetic interference events, extreme weather, etc.) exceed a set threshold, based on differences in different railway lines, climate conditions, or terrain environments. This allows for fine-tuning of the model parameters without the need for manual retraining.

[0060] In this embodiment, the communication and networking module specifically includes:

[0061] The link monitoring unit is used to monitor the signal strength, bit error rate and delay indicators of the communication link in real time, and is compatible with the signaling interface specifications of the railway communication system (GSM-R).

[0062] The health assessment unit is used to automatically trigger backup channel switching and adjust data transmission priority when a link quality degradation exceeds a set threshold, ensuring communication link continuity.

[0063] In this embodiment, the central control platform specifically includes:

[0064] The timing prediction unit is used to predict communication quality trends based on the Long Short-Term Memory (LSTM) neural network model, identify potential signal attenuation or environmental risks in advance, and guide the pre-adjustment actions of the antenna control module to avoid link interruption.

[0065] The data analysis unit is used to retrospectively analyze and statistically process historical operating data, and output performance trend reports and optimization suggestions.

[0066] The access control unit is used to implement multi-level access control for operations and maintenance personnel, regional managers and the headquarters dispatch center, so as to realize hierarchical authorization and cross-departmental collaborative management.

[0067] The offline caching unit is used to temporarily store locally acquired data and control command queues when communication is interrupted, and automatically synchronize them after communication is restored to ensure data continuity.

[0068] The self-diagnostic unit is used to detect hardware sensor malfunctions and software operating status in real time.

[0069] The remote upgrade unit automatically reports to the central control platform and performs remote maintenance or online model update operations when it detects that the system version or intelligent decision-making module is lagging behind.

[0070] The following is a detailed explanation of the content of each module:

[0071] During continuous wireless communication operation, the data acquisition module, as the system's front-end sensing unit, integrates attitude measurement devices and RF performance monitoring devices deployed inside the antenna body, as well as environmental sensing devices installed on the external antenna structure and base station area. These components form a high-precision multi-source fusion measurement system via a unified bus. This module can simultaneously acquire real-time operating parameters such as antenna physical attitude, wireless link quality, and surrounding environmental conditions. Physical attitude parameters include azimuth, elevation, polarization direction, and their rotational trends, accurately describing the antenna's current pointing state. Wireless link quality indicators include RSSI, SINR, RSRP, CQI, VSWR, and return loss, used to dynamically assess communication link stability and signal integrity. Environmental operating data covers temperature, humidity, wind speed, wind direction, thunderstorm conditions, electromagnetic interference levels, and changes in the morphology of buildings surrounding the base station, used to identify external interference factors that may cause signal fading. Combined with real-time train position, speed, curve curvature, and gradient information returned by the railway train control system, the antenna coverage area can proactively adjust to the train's movement.

[0072] It should be noted that the environmental data around the antenna is obtained through a variety of specialized equipment. The above information is collected at high frequency by hardware equipment such as inertial measurement unit, tilt sensor, electronic compass, RF monitoring module and micro weather station, with a sampling frequency of up to 200 Hz. This allows for timely response even in scenarios where the signal changes rapidly, such as severe weather, strong wind load, and high-speed driving.

[0073] The collected data is encapsulated and transmitted back through the communication and networking module. This module implements security strategies such as CRC check and AES encryption on the data stream, and simultaneously achieves high-speed and reliable transmission based on the railway optical transmission network. To reduce the burden on the communication link and the processing pressure on the server, this module has edge prediction capabilities. When the system detects that the antenna attitude or link quality is stabilizing, it automatically reduces the data reporting frequency; when it identifies sudden environmental changes, link degradation trends, or trains entering obstructed sections at high speed, it proactively increases the sampling and upload rates, realizing data enhancement and response acceleration strategies under key operating conditions.

[0074] The central control platform, built on a server cluster, undertakes core functions such as data aggregation, remote control, operation monitoring, and strategy optimization. Employing a distributed architecture and multi-layered disaster recovery design, the platform enables centralized management and cross-regional collaborative control of multiple base station antennas, possessing high availability and elastic scalability. The intelligent decision-making module integrated within the platform is the core of system control, and its control strategy originates from an antenna attitude optimization model built on a deep reinforcement learning framework. This model takes antenna attitude parameters, wireless link performance indicators, and environmental background information as input state variables, and antenna pointing and gain distribution adjustment as action outputs, using maximizing train-side communication quality and link reliability as the optimization objective function.

[0075] During the offline phase, the model is pre-trained based on a typical railway environment database. The data covers actual operating conditions such as multiple terrains, climates, and high and low traffic density. The model network structure and strategy parameters are continuously iterated by combining simulation and experimental feedback. During the online operation phase, the model receives status information from each base station in real time, predicts the communication degradation trend in the future, and performs attitude adjustments in advance.

[0076] Crucially, the intelligent decision-making module supports multi-antenna collaborative optimization strategies: when the train reaches areas where base station coverage overlaps (such as curves, tunnel entrances, or mountain valleys), the platform can simultaneously schedule the antennas of 2-3 adjacent base stations. Based on the train's real-time position, speed, and track curvature, it dynamically coordinates the main lobe pointing, beamwidth, and downtilt angle of each antenna to achieve smooth coverage transitions and seamless signal strength transitions. For example, 500 meters before the train enters a tunnel, the system enhances the horizontal beam focusing capability of the entrance base station antenna and appropriately raises the elevation angle to compensate for diffraction loss at the tunnel entrance; simultaneously, the rear base stations gradually narrow their beams and reduce their gains to avoid over-coverage interference, forming a "relay" dynamic coverage chain.

[0077] Once the intelligent decision-making module generates the optimal attitude adjustment strategy, the control command is quickly sent to the antenna control module through the communication and networking module, where it is executed.

[0078] The antenna control module integrates a high-precision angle adjustment device based on a worm gear transmission mechanism. It features a self-locking function, maintaining antenna stability under conditions such as power outages, strong winds, or icing, preventing angle shifts due to external forces. The module supports a horizontal adjustment range of ±45° and a vertical pitch range of 0° to 30°, with an angle control accuracy of ±0.1°. It also includes a built-in real-time attitude feedback unit for closed-loop verification of the consistency between the actual adjustment results and the target command, ensuring the antenna accurately points along the train's trajectory and enabling dynamic optimization of the communication link. The adjustment process follows a closed-loop feedback mechanism, specifically:

[0079] During execution, the system performs status verification on the adjustment results, that is, it re-collects antenna attitude and link quality data and compares the difference with the target. If the deviation still exceeds the threshold, the system immediately initiates a secondary refinement and correction process to ensure that the optimal coverage state is finally achieved. In addition, this module has a fault self-checking mechanism, which can issue alarms when abnormal conditions such as motor overload, mechanical jamming, power fluctuations, and communication interruption are detected. At the same time, it automatically triggers safety policies, instructing the antenna to return to the preset default attitude (such as 0° horizontal and -5° pitch) to ensure basic coverage and prevent equipment damage and safety accidents.

[0080] The entire system's control process is continuously recorded and monitored by the central control platform, which is used for subsequent performance evaluation, operating condition analysis, maintenance auditing, and continuous strategy optimization to achieve intelligent and stable operation of the system throughout its entire life cycle.

[0081] Through the design of this invention, the railway mobile communication antenna can adaptively adjust its posture according to the dynamic operating status of the train and changes in complex environment, effectively improving the continuity of railway communication network coverage, anti-interference capability and overall service quality, and providing reliable communication guarantee for train safety and railway transportation efficiency.

[0082] like Figure 2 As shown, the intelligent decision-making module employs the Deep Deterministic Policy Gradient (DDPG) algorithm based on the Actor-Critic architecture, which enables efficient optimization of antenna attitude within a continuous action space. This method enhances dynamic communication performance in complex orbital environments through the collaborative training of the policy network (Actor) and the value network (Critic).

[0083] Specifically, the policy network Actor outputs continuous action instructions a(t) based on the real-time environmental state s(t), aiming to optimize antenna attitude parameters and improve communication link quality.

[0084] This environmental state consists of a series of detailed parameter data, including but not limited to:

[0085] Antenna attitude parameters: current antenna horizontal angle θ, elevation angle φ, and polarization direction;

[0086] Environmental parameters: temperature, humidity, wind speed, wind direction, rainfall intensity, height and distribution of buildings along the track, terrain slope, curve curvature, real-time train location and speed;

[0087] Communication performance parameters: Received Signal Strength Indicator (RSSI), Received Reference Signal Power (RSRP), Signal-to-Noise Ratio and Interference Ratio (SINR), Channel Quality Indicator (CQI), Standing Wave Ratio (VSWR), Return Loss, Neighboring Cell Interference Intensity, Handover Success Rate, and Link Outage Frequency.

[0088] After standardization and feature encoding, the aforementioned multidimensional state variables together constitute the input state space of the reinforcement learning model, thereby comprehensively depicting the physical environment of the current antenna and the quality of its wireless communication link, providing data support for the policy network to generate high-precision, low-latency antenna control actions.

[0089] Meanwhile, the Critic network is responsible for evaluating the improvement in link quality after antenna adjustments. By estimating the state-action value of the action selected by the Actor in the current state, the immediate reward function r(t) is calculated. This reward function needs to comprehensively consider factors such as improved communication performance, optimized coverage, and the cost of control actions; its exemplary definition is as follows:

[0090]

[0091] Where α, β, and γ are weighting coefficients used to balance the priorities of different optimization objectives;

[0092] This indicates the improvement in signal-to-noise ratio and interference ratio; It reflects changes in the effective coverage area; the operating cost takes into account factors such as the energy consumption of the actuator, motor wear and mechanical fatigue, aiming to suppress unnecessary frequent adjustments and improve system stability and equipment life.

[0093] Through this reward mechanism, the system can maximize communication performance while taking into account control efficiency and hardware reliability, effectively avoiding control oscillations caused by over-optimization.

[0094] To achieve the above functions, the Actor-Critic architecture of this system is trained using the Deep Deterministic Policy Gradient Algorithm (DDPG), utilizing historical data to train our intelligent decision-making model. Specifically:

[0095] First, sufficient historical operation and maintenance data is collected using the acquisition module as an initial experience pool. This data includes, but is not limited to, multi-dimensional information such as antenna attitude parameters, environmental parameters, and communication index parameters, as well as corresponding adjustment actions and result feedback (such as changes in signal strength after adjustment). Finally, it is stored in the form of a four-tuple: state, action, reward, and next state (s, a, r, s'). This data provides basic data support for subsequent offline training.

[0096] Next, two identical neural network structures are created, called the target Q-network and the reference Q-network, respectively, to learn the policy network. Specifically, the target Q-network is used to learn the state-action value function Q(s,a), while the reference Q-network is used to provide a stable reference value and reduce fluctuations during training. Initially, the weights of the two can be the same or similar.

[0097] Next, during each training iteration, a batch of samples is randomly selected from the experience replay pool, and for each sample, its Temporal Difference Error (TD) is calculated: , where a' is the action predicted by the Actor network.

[0098] Where γ is the discount factor, Q' is the value of the action predicted by the reference Q network in the current environment, and Q is the value predicted by the current target Q network.

[0099] Furthermore, the weights of the target Q network are updated using the TD error, typically employing gradient descent. Simultaneously, the weights of the reference Q network are periodically updated (e.g., every N iterations) to reflect the latest weights of the target Q network.

[0100] Furthermore, during offline training, the model should be validated regularly to ensure it can effectively learn from historical data and generalize to unseen scenarios. Model performance can be further optimized by adjusting hyperparameters (such as learning rate, batch size, and network architecture). Once the model reaches a satisfactory performance level, it should be exported into a form suitable for online inference, ready for deployment in a real-world environment for real-time decision support.

[0101] In actual deployment, the antenna's current state vector at time t is obtained through the acquisition module and the communication and networking module. Input to the Actor network, and the Actor network directly outputs the control actions. This information is then passed to the antenna control module for execution.

[0102] In this way, the entire control system not only achieves high-precision, low-latency antenna control, but also balances system stability and equipment lifespan, avoiding control oscillations caused by over-optimization. This solution, through data-driven closed-loop optimization, significantly improves the response speed and robustness of antenna control in high-speed rail scenarios, while reducing the frequency of ineffective mechanical component actions. Actual deployment shows that the system can stably maintain high SINR and low switching failure rate in complex track environments, verifying the engineering applicability of DDPG in continuous control tasks.

[0103] like Figure 3 As shown, the adaptive dynamic update algorithm module proposed in this embodiment of the invention serves as a key component in the system's intelligent decision-making closed loop and is deployed within the central control platform, thereby effectively addressing the complex and ever-changing extreme operating environment along the railway line.

[0104] This module uses environmental disturbance perception as the core triggering mechanism, integrates transfer learning and online fine-tuning strategies, and realizes the dynamic evolution and rapid adaptation of the reinforcement learning control model, thereby ensuring that the antenna attitude optimization strategy still has high robustness and real-time response capability under non-steady-state conditions.

[0105] During actual operation, the system continuously acquires the physical status of the antenna and the performance indicators of the communication link through the data acquisition module, and transmits the data to the central control platform for aggregation and analysis via the communication and networking module. The platform's built-in environmental change discrimination unit monitors the input data stream in real time, focusing on evaluating the magnitude of changes in external environmental parameters such as sudden increases in wind speed, new building obstructions, and enhanced electromagnetic interference. Once a change in one or more environmental parameters is detected to exceed a preset threshold (e.g., a continuous drop in SINR of more than 3 dB for more than 10 seconds, or a sudden increase in wind speed to more than 80% of the design limit), it is judged as a significant environmental disturbance event, and the system automatically triggers the model retraining process.

[0106] To avoid the problems of high computational resource consumption, slow convergence speed, and policy fluctuations caused by traditional full-parameter retraining, this invention adopts a policy fine-tuning mechanism based on transfer learning. Specifically, the system obtains high-quality sample data from the collected data within a short window (usually 5-30 minutes) before and after the disturbance occurs. While retaining the backbone structure of the original reinforcement learning neural network and most of its parameters, gradient updates are performed only on the output layer or some intermediate layer parameters of the policy network. This process, aided by a mini-batch online backpropagation algorithm, can complete the rapid adjustment of model parameters within minutes, thereby efficiently achieving the transfer and accurate matching of the control policy to the new environment.

[0107] During fine-tuning, the traditional training process only requires replacing the step of "creating two identical neural network structures (i.e., the target Q-network and the reference Q-network)" with using a pre-trained neural network model, and then continuing with subsequent training steps based on this model. Since the original network already possesses basic antenna control capabilities, fine-tuning on this basis not only significantly improves the convergence speed of the algorithm but also ensures the stability of policy updates.

[0108] The entire adaptive update process forms a closed-loop feedback mechanism: from environmental disturbance identification, model fine-tuning, policy distribution to execution effect verification, all are uniformly scheduled and recorded by the central control platform, providing data support for subsequent policy effectiveness evaluation, anomaly tracing analysis and continuous model iteration optimization.

[0109] Through the above design, this invention realizes the intelligent evolution of the entire chain of railway mobile communication antenna control strategy from "perception to decision-making to adaptation to evolution", endowing the system with the ability to learn autonomously and improve itself in long-term operation, and providing a solid technical guarantee for the stable, efficient and safe operation of railway communication systems in complex environments.

[0110] The antenna control method proposed in this invention employs multiple high-precision data acquisition units deployed around the antenna to collect real-time data on the antenna's own operating parameters (such as azimuth, elevation, gain, and VSWR) and surrounding environmental information (including electromagnetic interference intensity, weather conditions, train speed and position, and track terrain features). This multi-source heterogeneous data is then efficiently transmitted to a central control console. Within the central control console, an intelligent decision-making module constructs an antenna control action generation model based on a deep reinforcement learning Actor-Critic architecture. This model is trained offline using historical operating data to optimize model parameters and improve decision accuracy. During actual operation, this action generation unit performs multi-level feature extraction on the input real-time parameter data, dynamically generating the optimal antenna control decision and issuing control commands to the antenna control module. The antenna control module then precisely adjusts the antenna's attitude parameters (such as azimuth, elevation, and polarization) according to the received commands, thereby achieving continuous optimization and assurance of communication link quality.

[0111] The central control platform provided in this embodiment of the invention offers an intuitive and highly interactive antenna status visualization interface, enabling maintenance personnel to monitor the real-time operating status, performance indicators, and alarm information of individual or multiple antennas. It also integrates long-term storage and management functions for historical antenna parameters, supporting multi-dimensional trend analysis, anomaly detection, and performance backtracking of historical data. Furthermore, the platform possesses multi-antenna collaborative control capabilities, enabling batch scheduling and policy linkage of multiple antennas deployed on the same line or adjacent sections, achieving global optimization of regional communication coverage and significantly improving overall system collaborative efficiency and resource utilization.

[0112] The adaptive dynamic optimization algorithm provided in this invention relies on real-time environmental and antenna status data continuously fed back by the data acquisition module. Through a preset multi-dimensional threshold mechanism, it intelligently identifies current line environment characteristics (such as changes in surrounding buildings, strong electromagnetic interference events, extreme weather, etc.). Once a significant environmental change is detected that may affect the effectiveness of existing control strategies, a model parameter update mechanism is automatically triggered. During this process, the system employs transfer learning technology, using the pre-trained basic model as prior knowledge and combining it with a small amount of newly acquired environmental data for rapid fine-tuning. This allows for the efficient generation of optimized control strategies adapted to new scenarios without relying on manual intervention or complete retraining. This mechanism significantly enhances the system's robustness and adaptability in complex and ever-changing railway operating environments, ensuring that the communication link is always in optimal working condition.

[0113] In summary, this invention provides a data-driven intelligent steering control system and method for railway communication antennas. By constructing an intelligent control architecture centered on multi-source sensing fusion and online learning, combined with a high-precision attitude estimation model and an adaptive strategy update mechanism, it achieves a fundamental breakthrough over traditional control methods that rely on human experience, suffer from slow response times, and have weak environmental adaptability. The system introduces online reinforcement learning and strategy self-evolution modules, endowing it with the ability to understand the environment in real time, continuously optimize decisions, and automatically update control strategies. Simultaneously, it integrates an extreme condition adaptive mechanism, ensuring operational stability even under harsh conditions such as strong vibration, high and low temperatures, and complex electromagnetic interference. Furthermore, the fully remote operation and maintenance architecture completely eliminates the need for high-altitude operations, significantly reducing the total lifecycle maintenance cost. This system can be widely applied to railway communication scenarios such as GSM-R, providing high reliability, high precision, and high intelligence for long-term train control communication, strongly supporting the construction and development of next-generation intelligent railway infrastructure.

[0114] Corresponding to the aforementioned railway mobile communication antenna remote control and adaptive optimization system, this invention also discloses a railway mobile communication antenna remote control and adaptive optimization method, which specifically includes:

[0115] The S100 collects multi-source heterogeneous data in real time, including antenna physical attitude parameters, wireless link quality index parameters and surrounding environmental information, and transmits the collected multi-source heterogeneous data to the central control platform.

[0116] The S200, through the intelligent decision-making module built into the central control platform, uses reinforcement learning algorithms to integrate historical operating data with current multi-dimensional sensing inputs, generates the optimal antenna adjustment strategy, and outputs control commands for antenna adjustment.

[0117] S300: The control command is parsed and executed, and the antenna attitude is adjusted through the actuator to achieve high-precision dynamic antenna adjustment.

[0118] It should be noted that for a detailed description of the remote control and adaptive optimization method for railway mobile communication antennas provided in the embodiments of the present invention, please refer to the relevant description of the remote control and adaptive optimization system for railway mobile communication antennas provided in the embodiments of this application, which will not be repeated here.

[0119] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.

[0120] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A remote control and adaptive optimization system for railway mobile communication antennas, characterized in that, The system includes: The data acquisition module is used to collect multi-source heterogeneous data in real time, including antenna physical attitude parameters, wireless link quality index parameters and surrounding environmental information, and transmit the collected multi-source heterogeneous data to the central control platform. The central control platform, based on the built-in intelligent decision-making module, uses reinforcement learning algorithms to integrate historical operating data with current multi-dimensional sensing inputs, generate the optimal antenna adjustment strategy, and output control commands for antenna adjustment. The antenna control module is used to parse and execute the control commands, and adjust the antenna attitude through the actuator to achieve high-precision dynamic antenna adjustment.

2. The railway mobile communication antenna remote control and adaptive optimization system as described in claim 1, characterized in that, The data acquisition module is specifically used for: Multi-source heterogeneous data acquisition is achieved by deploying attitude measurement devices and radio frequency performance monitoring devices inside the antenna body, as well as environmental sensing devices installed on the external structure of the antenna and the base station area. The physical attitude parameters include azimuth, elevation, polarization direction and their rotational trends, which are used to accurately describe the current pointing state of the antenna; the wireless link quality indicators include RSSI, SINR, RSRP, CQI, VSWR, return loss, neighboring cell interference intensity, handover success rate and link interruption frequency, which are used to dynamically evaluate the stability of the communication link and signal integrity; the surrounding environmental information includes temperature, humidity, wind speed, wind direction, thunderstorm conditions, electromagnetic interference level, changes in the shape of buildings around the base station, terrain slope, curve curvature, real-time position and speed of trains, which are used to identify external interference factors that may cause signal fading.

3. The railway mobile communication antenna remote control and adaptive optimization system as described in claim 1, characterized in that, The intelligent decision-making module is specifically used for: An antenna attitude optimization model is constructed based on a deep reinforcement learning framework, and the learning process is driven by the environmental state vector and the reward function. The antenna attitude optimization model takes multi-source sensing state information, including antenna physical attitude parameters, wireless link quality indicators and surrounding environment information, as input and takes the optimal antenna physical attitude information as output.

4. The railway mobile communication antenna remote control and adaptive optimization system as described in claim 3, characterized in that, The intelligent decision-making module is further used for: A deep reinforcement learning framework based on Actor-Critic is adopted, including: The action generation unit is responsible for generating continuous antenna attitude control action commands based on the input state information. The performance evaluation unit is responsible for evaluating the degree of improvement in link quality after antenna adjustment. Based on the reward function, it comprehensively considers various factors including communication link stability, signal strength, energy consumption and control smoothness, generates a reward signal through weighted calculation, and feeds the corresponding signal back to the action generation unit, thereby achieving end-to-end strategy optimization.

5. The railway mobile communication antenna remote control and adaptive optimization system as described in claim 1, characterized in that, The intelligent decision-making module is further used for: Supports online transfer learning: Based on the differences in railway lines, climate conditions, or terrain environments, when the detected changes in the line environment characteristics exceed the set threshold, the transfer learning process is automatically triggered to fine-tune the model parameters without the need for manual retraining.

6. The railway mobile communication antenna remote control and adaptive optimization system as described in claim 1, characterized in that, The intelligent decision-making module is further used for: Supports multi-antenna collaborative optimization strategy: When the train runs to the overlapping area of ​​base station coverage, the antennas of multiple adjacent base stations are synchronously scheduled. Based on the real-time position, speed and track curvature of the train, the main lobe pointing, beamwidth and downtilt angle of each antenna are dynamically coordinated to achieve smooth connection of coverage area and seamless transition of signal strength.

7. The railway mobile communication antenna remote control and adaptive optimization system as described in claim 1, characterized in that, The antenna control module is specifically used for: It integrates a high-precision angle adjustment device based on a worm gear transmission mechanism and has a self-locking function, allowing only the worm to drive the worm wheel, while reverse transmission is mechanically locked. The built-in attitude real-time feedback unit is used for closed-loop verification of the consistency between the actual adjustment results and the target command, ensuring that the antenna is accurately pointed to the train's running trajectory; It has a fault self-diagnosis mechanism that issues an alarm when an abnormal situation is detected, and automatically triggers a safety policy to instruct the antenna to return to the preset default posture, ensuring basic coverage and preventing equipment damage and safety accidents.

8. The railway mobile communication antenna remote control and adaptive optimization system as described in claim 1, characterized in that, The system also includes: The communication and networking module is used for data interaction between modules through the railway dedicated optical transmission backbone network or 5G private network. It has the ability to automatically switch between wired / wireless dual channels and monitor link status, and provides a standardized API interface for seamless integration with the existing GSM-R network management platform.

9. The railway mobile communication antenna remote control and adaptive optimization system as described in claim 1, characterized in that, The central control platform is also specifically used for: As the core of human-computer interaction and system management, it provides a graphical monitoring interface to realize real-time display of antenna operation status, historical data backtracking query, performance trend prediction and analysis, as well as centralized scheduling of antennas at multiple sites across regions, anomaly detection and remote system version updates.

10. A method for remote control and adaptive optimization of railway mobile communication antennas, characterized in that, The method includes: The system collects multi-source heterogeneous data in real time, including antenna physical attitude parameters, wireless link quality index parameters, and surrounding environmental information, and transmits the collected multi-source heterogeneous data to the central control platform. The intelligent decision-making module built into the central control platform uses reinforcement learning algorithms to integrate historical operating data with current multi-dimensional sensing inputs, generate the optimal antenna adjustment strategy, and output control commands for antenna adjustment. The control commands are parsed and executed, and the antenna attitude is adjusted through the actuator to achieve high-precision dynamic antenna adjustment.

Images