Plane shunting service use scene self-sensing self-repairing AI large model and method

Abstract

The application discloses a kind of plane shunting service use scene self-sensing self-repairing AI big model and method, AI big model includes multimodal perception input layer, big model inference core layer, self-repairing execution layer and knowledge and memory support layer, by fusing text shunting signaling, voice instruction, video and ranging data, generate structured job state representation, and combine standard shunting job rules and historical shunting signaling sequence, to shunting job process violation and communication environment anomaly are judged;When detecting job process violation, automatically generate and send repair shunting signaling, when detecting LTE private network communication anomaly, control terminal switches to 5G public network and re-registers service, while feedback repair result and continuously monitor repair state, form perception, research, repair and feedback closed loop.The application can realize the perception of plane shunting service use scene, abnormality research and automatic repair, improve the safety, continuity and intelligent level of shunting operation.

Description

Technical Field

[0001] This invention relates to the field of railway intelligent shunting and artificial intelligence application technology, specifically to a self-sensing and self-repairing AI large model and method for flat shunting operations. Background Technology

[0002] Shunting is a crucial link in railway transportation organization, undertaking tasks such as coupling, uncoupling, marshalling, and transporting vehicles within stations. During shunting operations, shunting operators use terminal devices such as locomotive controllers and handheld radios to send shunting signals for starting, shunting, deceleration, stopping, and reporting vehicle distances, directing the locomotive to complete the corresponding actions. The accurate transmission of shunting signals and the compliant execution of operational procedures directly affect operational safety and transportation efficiency.

[0003] The existing shunting system mainly relies on manual monitoring and a single network transmission. After the shunting operator sends a signal, the system lacks the ability to automatically assess the compliance of the operational procedures in real time. When high-risk operations occur, such as sending acceleration commands while the distance between trains is decreasing continuously, or failing to stop in time within a safe distance, the system often relies on manual experience to detect these issues afterward, making it difficult to intervene in real time during the operation. At the same time, shunting communication mostly uses a dedicated LTE network. When the dedicated network signal is abnormal or the communication link is interrupted, the operation is forced to stop, requiring manual troubleshooting and network re-establishment, resulting in poor operational continuity and long fault recovery time.

[0004] Furthermore, existing systems only perceive shunting signaling and network status at a simple status acquisition level, typically using preset rule-based judgment methods. This approach can only handle single, specific anomalies and cannot dynamically infer based on signaling trends during continuous operations, nor can it generate corresponding repair strategies for complex scenarios. It lacks multimodal information fusion capabilities, failing to comprehensively utilize text signaling, voice commands, onboard video, and ranging data for integrated judgment; it also lacks a logical reasoning mechanism based on historical signaling sequences and standardized operating rules, making it difficult to form a closed-loop automated "perception-diagnosis-repair" capability. As the requirements for safety, continuity, and automation in railway shunting operations continue to increase, traditional methods relying on manual monitoring and single-network communication are no longer sufficient to meet practical needs.

[0005] Therefore, there is an urgent need for an intelligent method and system that can autonomously perceive the shunting operation process and network environment status, rely on large models for semantic understanding and scenario reasoning, and automatically perform repair actions for illegal signaling and network anomalies, in order to overcome the shortcomings of existing technologies and improve the safety, continuity and automation level of horizontal shunting operations. Summary of the Invention

[0006] To address the aforementioned technical issues, this invention aims to provide a self-perceiving and self-repairing AI model and method for flat shunting operations based on a multimodal artificial intelligence model. By fusing and perceiving multi-source data such as shunting text signaling, voice commands, video, and ranging information, and combining historical shunting signaling sequences and standardized operating rules for intelligent analysis, it achieves automatic identification of abnormal shunting operation states and abnormal communication environment states, and performs shunting signaling repair or communication network switching based on the identification results.

[0007] To achieve the above objectives, the technical solution of this invention is: a self-perceiving and self-repairing AI large-scale model for shunting operations, characterized by comprising a multimodal perception input layer, a large-scale model reasoning core layer, a self-repairing execution layer, and a knowledge and memory support layer; the multimodal perception input layer is used to receive and preprocess multi-source information from the shunting operation site; the large-scale model reasoning core layer is used to perform semantic understanding and scenario reasoning on the multi-source information, determine whether the operation process is in violation of regulations or whether the operation environment is abnormal, and generate self-repairing decision instructions; the self-repairing execution layer is used to execute repair actions according to the self-repairing decision instructions and feed back the execution results to the large-scale model reasoning core layer; the knowledge and memory support layer is used to provide rule and historical data support for the large-scale model reasoning core layer.

[0008] A self-awareness and self-repair method for a flat-plane shunting operation scenario, implemented based on the aforementioned large AI model, includes the following steps: A. Equipment Registration and Network Access: After the terminal equipment completes LTE private network attachment and registration with the equipment management platform and obtains the local dispatch service port number, it enters the shunting signaling service processing state; the shunting operator sends shunting signaling according to the work plan, and the locomotive executes the corresponding operation according to the received shunting signaling; B. Shunting Operation Judgment: The multimodal perception input layer receives text shunting signals, voice commands, and video and ranging data from the train leader platform. The semantic understanding and state representation module generates a structured operation state representation. The scenario reasoning and violation judgment module combines the standardized shunting operation rule library and the historical shunting signal sequence memory library to determine whether the operation process is in violation and whether the operation environment is abnormal. C. Initiate self-repair process: If the operation process is found to be in violation or the operation network environment is abnormal, the self-repair process is initiated; otherwise, the shunting operation continues. The decision generation module generates a repair decision, which is executed by the self-repair execution layer. When the operation process is in violation, the signaling repair execution module sends a repair shunting signaling. When the operation network environment is abnormal, the network switching control module switches the terminal from the LTE private network to the 5G public network and re-registers it with the equipment management platform. D. Repairing and terminating the normal shunting operation: The scenario reasoning and violation judgment module continuously monitors the operation status and network status after the repair. If it is determined that the operation process after the repair is compliant and the network environment is normal, the self-repair process ends and the normal shunting operation is restored.

[0009] Compared with the prior art, the present invention has the following technical effects: First, relying on the semantic understanding and logical reasoning capabilities of large models, this invention enables automated parsing and scenario judgment of shunting signals. It can automatically identify violations and send repair signals based on historical signal sequences, significantly improving the safety of shunting operations and reducing the risk of accidents caused by human error.

[0010] Secondly, this invention automatically switches to the 5G public network when the LTE dedicated network signal is abnormal, and sends a fault stop prompt to the locomotive during the switching process, ensuring the continuity of shunting operations and reducing operation interruptions caused by network failures.

[0011] Furthermore, this invention constructs a closed-loop process encompassing perception, analysis, judgment, repair, and feedback, reducing the need for manual inspection and intervention and improving the automation and intelligence levels of horizontal shunting operations. Finally, through multimodal perception and knowledge memory support, this invention enables the system to comprehensively utilize text, voice, video, and ranging information, and combine standardized rules and historical data for reasoning, enhancing its adaptability and decision reliability in complex operational scenarios. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the overall architecture of the application system of this invention; Figure 2 This is a schematic diagram of the topology of the dual-network converged communication system for planar shunting operations according to the present invention; Figure 3 This is a schematic diagram of the large-scale model architecture of the present invention; Figure 4 This is a schematic diagram of the method flow of the present invention; Figure 5 This is a schematic diagram of the timing sequence of terminal device registration and network access interaction according to the present invention. Detailed Implementation

[0013] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings.

[0014] like Figure 1 As shown, the self-sensing and self-repairing AI big model for the use scenario of flat shunting operations of the present invention is deployed in the shunting terminal equipment in the form of software modules. The shunting terminal equipment includes a machine controller, a handheld console, and a train leader console, which are used to perceive, analyze and judge the signaling, voice, video and distance measurement data in real time during the shunting operation, and execute corresponding self-repair operations according to the diagnostic results.

[0015] This system uses an LTE private network to transmit the main shunting signaling services, while using a 5G public network as a backup communication guarantee, thus constructing a dual-network communication architecture that integrates the LTE private network and the 5G public network to improve communication reliability and service continuity during shunting operations.

[0016] The system comprises a 5G network component, a shunting operation site component, and an LTE network component. The 5G network component includes a video management platform, firewall, and 5G base stations, used to transmit onboard video surveillance data and provide backup public network communication links for terminal equipment. The shunting operation site component includes one or more shunting teams, each consisting of a machine controller, a lead-up station, and a handheld radio, used to send, receive, and display shunting signaling and collect on-site status data. The LTE network component includes LTE base stations, an LTE core network, firewall, equipment management platform, local area network, area manager's station, and shunting order editing terminal, used to realize shunting signaling forwarding, equipment registration management, dispatching command, and business data interaction.

[0017] The terminal device integrates an LTE private network module and a 5G public network module. It runs an AI large-scale model software program through a high-performance processor and processes collected multi-source data in real time. The terminal device uses a Linux operating system, supports multi-tasking and multi-threading, and assigns the highest priority to LTE shunting service processing tasks to ensure the real-time performance of shunting signaling services. Under normal circumstances, the terminal device completes shunting signaling transmission through the LTE private network. When an LTE private network communication anomaly is detected, the AI ​​large-scale model controls the terminal to activate the 5G public network module for network switching based on network status diagnostic results. This enables the integrated operation of dual-network services on the same device, providing dual safety guarantees for shunting operations.

[0018] As a preferred option, the LTE private network module uses the MAG SLM792A smart module, and the 5G public network module uses the MAG SRM813A smart module.

[0019] like Figure 2 As shown, the information center deploys an equipment management platform and a video management platform, which are connected to the station's shunting teams and dispatch hall via 5G public network, LTE private network, and local area network, respectively. Handheld radios, train dispatching stations, and machine controllers access the LTE private network wirelessly to interact with the equipment management platform; the district manager's station connects to the equipment management platform within the local area network via wired connection to receive and broadcast shunting signals. When the LTE private network malfunctions, the terminal devices can automatically switch to the 5G public network to ensure service continuity. Handheld radios, train dispatching stations, and machine controllers access the 5G public network wirelessly to interact with the video management platform for transmitting vehicle-mounted video surveillance data.

[0020] The core equipment functions of the system are as follows: the ward master station is responsible for receiving and broadcasting shunting signals; the machine controller is responsible for sending, receiving and displaying shunting signals; the handheld station is responsible for displaying shunting signals; the lead station is responsible for real-time on-site monitoring (including video and distance measurement); the equipment management platform software provides functions such as interface processing with the core network, data forwarding and storage, and equipment management.

[0021] like Figure 3 As shown, a self-perceiving and self-repairing AI big model for a flat shunting operation scenario includes a multimodal perception input layer, a big model reasoning core layer, a self-repairing execution layer, and a knowledge and memory support layer. Each layer works together to form a closed loop of perception-reasoning-decision-execution.

[0022] The multimodal perception input layer is used to receive and preprocess multi-source information from the shunting operation site, including a text signaling parsing module, a speech recognition and command conversion module, and a video and distance measurement feature extraction module. Among them, the text signaling parsing module is used to receive and parse the text-based shunting signals sent by the shunting operator; the speech recognition and command conversion module is used to receive the shunting operator's voice commands and convert them into structured text commands through speech recognition; the video and distance measurement feature extraction module is used to receive the onboard video monitoring screen and distance measurement data provided by the train leader station and extract key visual features such as train distance and coupling status.

[0023] The core layer of the large-scale model inference is the core decision-making part of this AI large-scale model. It is used to perform semantic understanding and scenario reasoning on multi-source information, determine whether the operation process is in violation of regulations or whether the operation environment is abnormal, and generate self-repair decision instructions. The core layer of the large-scale model inference includes a semantic understanding and state representation module, a scenario reasoning and violation judgment module, and a decision generation module. Among them, the semantic understanding and state representation module is used to perform unified semantic parsing on multimodal information such as text signaling, voice commands, and visual features output from the multimodal perception input layer, and combine it with historical shunting signaling sequences to generate a structured real-time operation state representation (including current train distance level, locomotive operating status, coupling trend, network communication quality, etc.). The scenario reasoning and violation judgment module is used to perform logical reasoning based on the structured state representation, combined with the standardized shunting operation rule base in the knowledge and memory support layer, and combined with historical shunting signaling sequences to determine whether the current operation process is in violation of regulations or whether the network environment is abnormal. The decision generation module is used to generate specific self-repair decision instructions based on the judgment results, that is, repair shunting signaling or network switching instructions.

[0024] The self-repair execution layer is used to execute repair actions according to self-repair decision instructions and feed the execution results back to the large model inference core layer. The self-repair execution layer includes a signaling repair execution module, a network switching control module, and a human-machine interaction prompt and closed-loop feedback module. Among them, the signaling repair execution module is used to automatically generate and send corresponding repair shunting signaling (such as deceleration, emergency stop, acceleration, etc.) according to the decision instructions. The network switching control module is used to control the terminal to reattach to the network using the built-in 5G module when the LTE private network environment is determined to be abnormal, switch to the 5G public network, and re-register with the equipment management platform. The human-machine interaction prompt and closed-loop feedback module is used to display the status through the equipment human-machine interaction interface during network switching or critical repair actions, send prompt shunting signaling (such as "fault stop") to the locomotive, and feed the execution results back to the large model inference core layer to form a closed loop.

[0025] The knowledge and memory support layer provides rules and historical data support for the reasoning core layer, including a standardized shunting operation rule base, a historical shunting signal sequence memory, and a real-time operation status cache. The standardized shunting operation rule base stores the standard operating rules and safety constraints for railway horizontal shunting; the historical shunting signal sequence memory caches recent shunting signal sequences and their corresponding operation statuses, supporting trend analysis; and the real-time operation status cache temporarily stores the structured state representation of the current operation for use in multiple rounds of reasoning.

[0026] like Figure 4 As shown, the overall process of a self-sensing and self-repairing method for a flat-plane shunting operation includes: equipment registration and network access, shunting operation determination, initiation of the self-repair process, and termination of repair to restore normal shunting operation. The specific process is as follows: A. Device registration and network access All terminal devices are managed and activated uniformly. Each terminal device is configured with its local ID, identity, dispatch number, and alias, as well as the device management platform's IP address and registration port number. After powering on, the terminal device attaches to the LTE private network. Once network attachment is successful, it requests to log in to the LTE network and obtains an IP address, subsequently requesting login from the device management platform. The system performs permission checks on the devices to prevent unauthorized devices from entering the shunting operation group.

[0027] like Figure 5 As shown, a terminal device already connected to the network sends a service registration request to the device management platform. The request includes the device's ID, identity, IP address, port number, local dispatch code, and dispatch code alias. The device management platform verifies the device's identity and permissions and maintains a service mapping table for identity, ID, dispatch code, IP address, and port number. Upon successful verification, a service registration response is returned to the terminal device, containing a registration success identifier and the local dispatch service port number. After obtaining the local dispatch service port number, the terminal device enters the dispatch signaling service processing state.

[0028] If no response is received during the registration process or the response indicates temporary registration failure, the service registration will be attempted periodically; if the response indicates permanent registration failure, the LTE network access registration will be terminated and no further attempts will be made.

[0029] After successful registration, the shunting operator, according to the work plan, sends shunting signals to command the locomotive to start, shunt, decelerate, stop, and report the distance between locomotives. The locomotive then executes the corresponding operations based on the received shunting signals. This step provides the equipment and operational foundation for subsequent sensing and repair.

[0030] B. Shunting Operation Judgment The AI-powered large-scale model performs checks for operational process violations and network environment anomalies. Dispatch signaling is forwarded from the equipment management platform connected to the core network to the terminal equipment connected to the LTE private network. This step is completed collaboratively by the multimodal perception input layer and the large-scale model inference core layer, with the knowledge and memory support layer providing rule and historical data support.

[0031] The text signaling parsing module, speech recognition and command conversion module, and video and ranging feature extraction module in the multimodal perception input layer receive text shunting signals, shunting operator voice commands, and video / ranging data from the lead platform, respectively. The semantic understanding and state representation module performs unified semantic parsing on the multimodal information to generate a structured operation state representation, including the current distance level, locomotive operating status, coupling trend, and network communication quality.

[0032] The scenario reasoning and violation judgment module combines the standardized shunting operation rule library and the historical shunting signal sequence memory library to perform logical reasoning on the structured state representation, and respectively judge whether the operation process is in violation of regulations and whether the operation environment is abnormal.

[0033] The determination of whether the operation process is in violation is based on the violation of shunting signaling. Specifically, when the historical shunting signaling sequence shows a continuous decrease in the distance between cars (e.g., consecutive occurrences of "ten cars, five cars, three cars"), the large model performs joint reasoning based on the decreasing distance trend in the continuous car distance signaling sequence, the vehicle's motion status, and the current control command. When it is determined that there is a conflict between the current control command and the vehicle status (if an "accelerate" shunting signaling or voice acceleration command appears, it is determined to be a high-risk violation), a high-risk violation judgment result is output. When the vehicle has entered a close-range state (e.g., when a "one car" shunting signaling appears, it is determined that the current car distance is below the safety threshold; continuing operation poses a collision risk), a judgment of violation is output. When a "start" or "release" shunting signaling appears, it is determined that the locomotive is in a low-speed state after just starting.

[0034] Whether the operating environment is abnormal is determined by the network communication status: when the terminal does not receive a reply confirmation message from the device management platform, or receives a reply but the timeout exceeds the preset threshold (preferably 500ms), the current LTE dedicated network environment is determined to be abnormal.

[0035] C. Initiate the self-repair process If the work process is determined to be in violation of regulations or the network environment is abnormal, a self-repair process is initiated; otherwise, the shunting operation continues. This step is completed collaboratively by the decision generation module and the self-repair execution layer in the core layer of the large model inference. If the scenario inference and violation assessment module determines that the work process is compliant and the work environment is normal, no intervention is taken, and the normal shunting operation continues.

[0036] If the work process is determined to be in violation of regulations or the work network environment is abnormal, the decision generation module generates a repair decision, and the self-repair execution layer executes the corresponding actions: 1. Shunting Signal Repair: When a violation of the operational process is determined (i.e., a shunting signal violation), the signal repair execution module automatically generates and sends a repair shunting signal based on the decision. For example, when it is determined that the distance between trains is continuously decreasing and the control command conflicts with the vehicle status, the equipment management platform automatically sends a repair shunting signal to reduce the operating speed (e.g., a "decelerate" shunting signal) to repair the shunting operation process; when it is determined that the current distance between trains is less than the safe operating distance (e.g., when it is determined that the distance between trains is "one car" and the locomotive continues to run), a repair shunting signal to stop the locomotive is sent, such as an "emergency stop" shunting signal, to stop the locomotive; when it is determined that the locomotive has just started, an "accelerate" shunting signal is sent to accelerate the locomotive's operating speed and improve the efficiency of the shunting operation.

[0037] 2. Network Environment Repair: When the network environment is determined to be abnormal, the network switching control module controls the terminal to reattach to the network using the built-in 5G IoT SIM card and 5G smart module, switching to the 5G public network and resending the service registration request to the equipment management platform. During the switching process, the human-machine interaction prompt and closed-loop feedback module displays the network switching status on the equipment's human-machine interface and simultaneously sends a message to the locomotive to prompt for a suspension of operations (e.g., a "fault stop" message).

[0038] D. Repair and terminate the operation to resume normal shunting operations. This step involves the scenario reasoning and violation assessment modules in the core layer of the large model inference continuously monitoring the post-repair operation status and network status. If it is determined that the post-repair operation process is compliant and the network environment is normal, the self-repair process ends and normal shunting operations resume.

[0039] To further illustrate how this AI model operates in actual shunting operations, the detailed execution flow for two typical application scenarios is given below.

[0040] Example 1: Repair of Shunting Signal Violations The shunting operator sends shunting commands in text or voice format via a machine controller or handheld radio. These commands are then forwarded to the terminal equipment via the equipment management platform. The multimodal perception input layer receives the shunting commands and video / distance measurement data from the lead-up platform. The semantic understanding and status representation module performs unified semantic parsing to generate a structured operational status representation, which is then compared with recent sequences in the historical shunting command sequence memory. The scenario reasoning and violation judgment module performs logical reasoning based on the standardized shunting operation rule base. If the shunting command is determined to be compliant, no intervention is taken; if it is determined to be non-compliant, the decision generation module generates a repair decision, and the signal repair execution module automatically sends the corresponding repair shunting command.

[0041] Specific scenarios include: (1) When the shunting operator continuously sends shunting signals for "ten cars", "five cars", and "three cars", the historical shunting signal sequence memory records the decreasing trend of car spacing. The scenario reasoning and violation judgment module determines that the locomotive spacing is decreasing. If the shunting operator sends an "accelerate" shunting signal or a voice acceleration command at this time, it is judged as a high-risk operation, which is very likely to cause an operational accident. The shunting signal repair execution module immediately and automatically sends a "decelerate" repair shunting signal to repair the shunting operation process.

[0042] (2) When the shunting operator sends the shunting signal “one car”, the scenario reasoning and violation judgment module determines that the current car distance is less than the safe operating distance. The shunting signal repair and execution module immediately and automatically sends the “emergency stop” shunting signal to stop the locomotive operation.

[0043] (3) When the shunting operator sends a “start” or “release” shunting signal, the scenario reasoning and violation judgment module determines that the locomotive is in a low-speed state just started. The shunting signal repair and execution module automatically sends an “accelerate” shunting signal to improve the efficiency of shunting operations.

[0044] The human-computer interaction prompt and closed-loop feedback module feeds the execution results back to the inference core layer. The scenario inference and violation judgment module reassesses the post-repair status. If the workflow is compliant, the current round of repair is terminated; otherwise, monitoring continues.

[0045] Example 2: Automatic switching in case of network signal abnormalities After the shunting operator sends the shunting command, the terminal and the equipment management platform use a message reply confirmation mechanism to monitor the communication status. The scenario reasoning and violation judgment module monitors the arrival time of the reply confirmation message in real time. If the reply message is received normally and within the timeout period, no intervention is taken; if no reply confirmation message is received, or if a reply confirmation message is received but the timeout period is greater than 500ms, the LTE dedicated network where the shunting operation equipment is located is determined to be abnormal. The decision generation module generates a network switching decision and automatically switches the shunting operation equipment on the original network to the 5G public network.

[0046] The network switching control module's control terminal uses the device's built-in 5G IoT SIM card and 5G smart module to re-register network services. During the switching process, the device's human-machine interface (HMI) displays the complete network status on the HMI and sends a "fault stop" message to the locomotive, prompting it to suspend operations. After the device successfully registers and joins the 5G public network, the HMI notifies the shunting operator that operations can continue, ensuring the continuity of shunting operations. Simultaneously, the execution results are fed back to the core layer of the large model inference, completing the closed loop. Once the scenario inference and violation assessment module confirms that the network environment has returned to normal, it ends the current network repair process, and the system resumes processing shunting signaling services using the currently available network.

[0047] Through the collaboration of the above system architecture and AI large model software module, this invention realizes the automatic identification and repair of shunting signal violations, as well as the automatic switching of communication network anomalies, which significantly reduces the need for manual intervention and improves the safety and continuity of horizontal shunting operations.

Claims

1. A self-perceiving and self-repairing AI large-scale model for a flat-plane shunting operation scenario, characterized in that: The large model is deployed as a software module in the terminal device software system, including a multimodal perception input layer, a large model inference core layer, a self-repair execution layer, and a knowledge and memory support layer. The multimodal perception input layer is used to receive and preprocess multi-source information from the shunting operation site. The large model inference core layer is used to perform semantic understanding and scenario reasoning on the multi-source information, determine whether the operation process is in violation of regulations or whether the operation environment is abnormal, and generate self-repair decision instructions. The self-repair execution layer is used to execute repair actions according to the self-repair decision instructions and feed back the execution results to the large model inference core layer. The knowledge and memory support layer is used to provide rule and historical data support for the large model inference core layer.

2. The self-perceiving and self-repairing AI large-scale model for flatbed shunting operations as described in claim 1, characterized in that, The multimodal perception input layer includes a text signaling parsing module, a speech recognition and command conversion module, and a video and ranging feature extraction module. The text signaling parsing module receives and parses text-based shunting signals sent by the shunting operator. The speech recognition and command conversion module receives the shunting operator's voice commands and converts them into structured text commands. The video and ranging feature extraction module receives onboard video monitoring footage and ranging data provided by the lead vehicle station and extracts key visual features such as vehicle distance and coupling status.

3. The self-perceiving and self-repairing AI large-scale model for flatbed shunting operations as described in claim 1, characterized in that, The core layer of the large model inference includes a semantic understanding and state representation module, a scene reasoning and violation judgment module, and a decision generation module. The semantic understanding and state representation module performs unified semantic parsing on the text signaling, voice commands, and visual features output from the multimodal perception input layer, and generates a structured real-time operation state representation by combining it with historical shunting signaling sequences. The scene reasoning and violation judgment module performs logical reasoning based on the structured state representation, combined with the standardized shunting operation rule base and historical shunting signaling sequences in the knowledge and memory support layer, to determine whether the current operation process is in violation or whether the network environment is abnormal. The decision generation module generates self-repair decision instructions based on the judgment results, including repair shunting signaling and network switching instructions.

4. The self-perceiving and self-repairing AI large-scale model for flatbed shunting operations as described in claim 1, characterized in that, The self-healing execution layer includes a signaling repair execution module, a network switching control module, and a human-machine interaction prompt and closed-loop feedback module. The signaling repair execution module automatically generates and sends corresponding repair and shunting signaling based on decision instructions. The network switching control module, when determining an abnormal LTE private network environment, controls the terminal to reattach to the network using its built-in 5G module, switch to the 5G public network, and re-register with the equipment management platform. The human-machine interaction prompt and closed-loop feedback module displays the status through the equipment's human-machine interface during network switching or critical repair actions, sends prompt shunting signaling to the locomotive, and simultaneously feeds back the execution results to the large model inference core layer, forming a closed loop.

5. The self-perceiving and self-repairing AI large-scale model for flatbed shunting operations as described in claim 1, characterized in that, The knowledge and memory support layer includes a standardized shunting operation rule base, a historical shunting signal sequence memory, and a real-time operation status cache. The standardized shunting operation rule base stores the standard operating rules and safety constraints for railway horizontal shunting. The historical shunting signal sequence memory caches recent shunting signal sequences and their corresponding operation statuses, supporting trend analysis. The real-time operation status cache temporarily stores the structured state representation of the current operation for use in multi-round reasoning.

6. A self-sensing and self-repairing method for a flatbed shunting operation scenario, implemented based on the self-sensing and self-repairing AI large-scale model for flatbed shunting operation scenarios as described in any one of claims 1 to 5, characterized in that, Includes the following steps: A. Equipment Registration and Network Access: After the terminal equipment completes LTE private network attachment and registration with the equipment management platform and obtains the local dispatch service port number, it enters the shunting signaling service processing state; the shunting operator sends shunting signaling according to the work plan, and the locomotive executes the corresponding operation according to the received shunting signaling; B. Shunting Operation Judgment: The multimodal perception input layer receives text shunting signals, voice commands, and video and ranging data from the train leader platform. The semantic understanding and state representation module generates a structured operation state representation. The scenario reasoning and violation judgment module combines the standardized shunting operation rule library and the historical shunting signal sequence memory library to determine whether the operation process is in violation and whether the operation environment is abnormal. C. Initiate self-repair process: If the operation process is found to be in violation or the operation network environment is abnormal, the self-repair process is initiated; otherwise, the shunting operation continues. The decision generation module generates a repair decision, which is executed by the self-repair execution layer. When the operation process is in violation, the signaling repair execution module sends a repair shunting signaling. When the operation network environment is abnormal, the network switching control module switches the terminal from the LTE private network to the 5G public network and re-registers it with the equipment management platform. D. Repairing and terminating the normal shunting operation: The scenario reasoning and violation judgment module continuously monitors the operation status and network status after the repair. If it is determined that the operation process after the repair is compliant and the network environment is normal, the self-repair process ends and the normal shunting operation is restored.

7. The self-sensing and self-repairing method for flatbed shunting operations according to claim 6, characterized in that, In step A, the terminal device already connected to the network sends a service registration request to the device management platform. The request includes the device ID, identity, IP address and port, local access code and access code alias. The device management platform verifies the device's identity and permissions and maintains a service mapping table for identity, ID, access code, IP address and port. After successful verification, the platform returns a service registration response to the terminal device, which includes a registration success identifier and the local access code service port number. During the registration process, if no response is received or the response indicates temporary registration failure, the platform periodically attempts to register the service. If the response indicates permanent registration failure, the current LTE network access registration is terminated and no further attempts are made.

8. The self-sensing and self-repairing method for flatbed shunting operations according to claim 6, characterized in that, In step B, whether the operation process is in violation is determined by whether the shunting signaling is in violation: when the historical shunting signaling sequence shows that the distance between vehicles is continuously decreasing, the big model performs joint reasoning based on the distance decreasing trend in the continuous distance signaling sequence, the vehicle movement status and the current control command. When it is determined that there is a conflict between the current control command and the vehicle status, and an acceleration shunting signaling occurs again, it is determined to be a high-risk violation operation. If the current distance to the vehicle is determined to be less than the safe working distance and the vehicle does not stop, it is considered a violation of operating procedures. When a shunting start signal is received, it is determined that the locomotive is in a low-speed state just starting.

9. The self-sensing and self-repairing method for flatbed shunting operations according to claim 6, characterized in that, In step B, whether the operating environment is abnormal is determined by the network communication status: when the terminal and the device management platform do not receive a reply confirmation message, or receive a reply but the timeout exceeds a preset threshold, the current LTE dedicated network environment status is determined to be abnormal.

10. The self-sensing and self-repairing method for flatbed shunting operations according to claim 6, characterized in that, In step C, when a violation of the work process is determined, the signaling repair execution module automatically generates and sends a repair shunting signal based on the decision: when it is determined that the train distance is continuously decreasing and the control command conflicts with the vehicle status, a repair shunting signal to reduce the operating speed is sent; when it is determined that the current train distance is less than the safe working distance, a repair shunting signal to stop the locomotive is sent; when it is determined that the locomotive is in a low-speed state after starting, a repair shunting signal to increase the operating speed is sent; when it is determined that the work network environment is abnormal, the network switching control module controls the terminal to activate the built-in 5G module to reattach to the network, switch to the 5G public network, and resend the service registration request to the equipment management platform. During the switching process, the human-machine interaction prompt and closed-loop feedback module displays the network switching status on the equipment human-machine interaction interface, and at the same time sends a message to the locomotive to prompt the suspension of work.

Images