Railway communication digital twin monitoring method based on space-time prediction and active avoidance network

Abstract

The application provides a mobile communication digital twin monitoring method based on space-time prediction and active avoidance network. The method comprises the following steps: collecting real-time state data of the train through a data acquisition probe, constructing a three-dimensional channel feature digital base containing a historical signal strength heat map based on a three-dimensional rendering engine and macroscopic geographic terrain data; deducing the network environment quality change trend of the train in the three-dimensional channel feature digital base through a front space-time prediction algorithm; when it is predicted that the train will enter a communication attenuation area within a preset distance threshold, issuing a cross-layer pre-scheduling instruction from a cloud decision hub to a vehicle-mounted gateway; the vehicle-mounted gateway executes the cross-layer pre-scheduling instruction, activates multi-link aggregation, transfers non-urgent business data into an edge memory, and forcibly locks a priority transmission channel for core control signaling. The application realizes cross-dimension early intervention of geographic space data on the network bottom protocol stack, and eliminates the inevitable communication lag and packet loss when the train enters the blind area.

Description

Technical Field

[0001] This invention relates to the field of integrated air-ground-space vehicle-to-ground communication and air-to-ground communication technologies, and particularly to a mobile communication digital twin monitoring method based on spatiotemporal prediction and active network avoidance. Background Technology

[0002] The vehicle-to-ground communication network protocol stack refers to the underlying data transmission link standard upon which information exchange between the train and the ground control center relies. Connection-oriented data flows are traditionally established using transmission control protocols and the Internet protocol suite. However, in the complex terrain conditions of ordinary railways, this protocol stack is susceptible to signal fading, making head-of-train congestion particularly severe.

[0003] To ensure the quality of service for message queue telemetry transmission, a confirmation mechanism is established within the lightweight IoT protocol to ensure that messages are delivered at least once. Messages will be continuously retransmitted until the sender receives an acknowledgment from the receiver. The lower limit of the transmission of core control signaling in railway communications is constrained by this mechanism.

[0004] 3. The QUIC protocol's zero-round-trip time handshake mechanism is a next-generation transport layer protocol built on UDP (User Datagram Protocol). Server session parameters are cached by the client after the initial complete handshake. During subsequent reconnections, encrypted data can be combined with the handshake request and sent, eliminating the need for additional round-trip communication. This significantly improves link reconstruction efficiency in scenarios with frequent disconnections in weak networks, reducing reconnection latency from seconds in traditional TCP to milliseconds.

[0005] The concept of macro-micro digital twin coordinate mapping refers to the extraction of geographic elevation data and vector coordinates from the real physical world to construct a macro-level virtual environment. Simultaneously, a micro-level, finely detailed 3D mesh model of the physical equipment is also established. Real-time latitude and longitude coordinates and attitude parameters are injected into the micro-model via algorithms, driving it to perform second-level synchronous displacement within the macro-level virtual terrain, thus forming a strictly physically constrained mirror space.

[0006] The spatiotemporal prediction algorithm is based on a joint probability distribution model constructed from historical time-series data and three-dimensional geospatial attributes. Using the train's current vector speed and location coordinates as input, the algorithm extrapolates and outputs the network channel degradation trend of its specific geographical location within a future time window. This enables proactive network environment awareness.

[0007] Currently, the shortcomings of existing railway communication digital twin monitoring methods include:

[0008] 1. Feedback lag caused by lack of prior information. In current vehicle-to-ground communication systems, network scheduling and link switching mechanisms strictly follow the passive response mode defined by the OSI model. The instantaneous signal-to-noise ratio, RSRP, and other low-level parameters reported by the physical layer become the sole basis for the network layer to trigger link switching decisions; the transport layer relies on the TCP retransmission timeout mechanism to determine whether the connection is still valid. Once the system detects a decline in network quality, the link has usually already suffered significant attenuation, or even been completely interrupted.

[0009] 2. Significant technical bottlenecks exist. When trains enter areas with strong electromagnetic shielding, such as tunnels, at high speeds, the physical layer can no longer support maintaining a valid connection, and the underlying protocol stack only then triggers link switching and retransmission mechanisms. This lack of proactive intervention inevitably leads to severe backlog and packet loss of critical signaling within the onboard buffer. This deterministic event of "connection failure upon entering a blind spot" is classified as an anomaly in traditional architectures, rather than a predictable norm, thus creating a systemic monitoring blind spot that is difficult to avoid through conventional means, directly threatening the integrity and reliability of train safety data.

[0010] 3. The current state of data disorder and lack of verification between sessions. In the extremely weak network environment of railways, frequent intermittent outages and heterogeneous switching of multiple links coexist. At this time, the reliable transmission mechanism on which TCP relies constitutes a structural obstacle to the consistency of timing. The phenomenon that early data blocks arrive at the receiving end later than later ones due to retransmission delay is caused by its sliding window algorithm and retransmission mechanism, resulting in the timestamp being reversed with the actual arrival order. The lack of a global timing verification mechanism across TCP session cycles is particularly prominent. After the network is restored, historical cached data at the receiving end is misaligned or overwritten with real-time newly transmitted data, causing an irreversible offset on the timeline between the physical parameters obtained by the ground center at the moment of the fault and the actual state. This timing distortion makes the root cause analysis of the fault based on timing relationships completely ineffective, and it is difficult to fully restore the true physical state at the moment the fault occurs. The fault tracing chain is systematically destroyed from the source of data collection.

[0011] 4. Description of the current state of resource mismatch caused by inter-layer semantic isolation. In weak network environments, non-core high-bandwidth services often compete with core control signaling for limited uplink channel resources, easily inducing head-of-line congestion and timeout packet loss in core signaling. The predicament of critical services being squeezed out by non-critical services is actually a systemic resource mismatch caused by the lack of cross-layer collaboration capabilities in the layered architecture. At best, this causes monitoring screen lag; at worst, it leads to delays or even loss of control commands, directly compromising train operation safety.

[0012] 5. Spatial amnesia-induced experience accumulation and knowledge silos. Experience cannot be strongly linked to specific physical locations, leading to organizational knowledge "amnesia." When subsequent trains run through the same physical blind spot, they cannot automatically recall similar troubleshooting experiences or proactively push them to maintenance personnel. On-site maintenance personnel must relive the process of fault occurrence, troubleshooting, location, and resolution. The same type of network fault occurring in the same spatial location will recur, requiring further investigation. Organizational knowledge within the same spatial area cannot be digitally learned or passed down through generations, further reducing maintenance efficiency to a "person-to-person" level. Problems such as scattered stations, delayed responses on remote lines, and wasted resources also persist. Summary of the Invention

[0013] This invention provides a railway communication digital twin monitoring method based on spatiotemporal prediction and active network avoidance, so as to effectively improve the safety of train operation.

[0014] To achieve the above objectives, the present invention adopts the following technical solution.

[0015] A mobile communication digital twin monitoring method based on spatiotemporal prediction and active network avoidance includes:

[0016] By deploying lightweight data acquisition probes on trains and ground equipment, real-time status data of trains and environmental data along the line are collected at preset intervals. Based on a 3D rendering engine and macro-geographic terrain data, a 3D digital base of channel characteristics containing historical signal strength heatmaps is constructed.

[0017] The real-time status data and the along-line environment data are input into the pre-spatiotemporal prediction algorithm to deduce the trend of network environment quality changes of the train in the future time window in the three-dimensional channel feature digital base; when the deduction result predicts that the train is about to enter the communication attenuation zone within the preset distance threshold, the cloud decision center sends a cross-layer pre-scheduling instruction to the vehicle gateway.

[0018] In response to the cross-layer pre-scheduling instruction, the vehicle gateway drives the underlying lightweight transmission protocol stack to perform the following actions: activate multi-link aggregation to widen redundant channels, transfer non-urgent service data to edge storage for buffering and suppression, and forcibly lock the priority transmission channel for core control signaling.

[0019] Preferably, the method of collecting real-time status data of the train and environmental data along the line at preset intervals using lightweight data acquisition probes deployed on the train and ground equipment includes:

[0020] Lightweight, multi-source heterogeneous data acquisition probe arrays are deployed in base station equipment rooms along the railway line and at the onboard router nodes of ordinary railway trains. When the train is in normal operation, the data acquisition probes collect real-time status data of the train and environmental interference data along the line at a fixed frequency. The real-time status data of the train includes: longitude, latitude, altitude and speed data of the train; temperature, load status and memory usage of the central processing unit of the hardware; latency, jitter, packet loss rate, throughput data and wireless channel reference signal received power of the onboard router; the environmental data along the line includes geographic elevation benchmark data based on the national unified coordinate system, GIS vector road network data of the entire ordinary railway line and environmental interference data along the line.

[0021] The heterogeneous device adaptive gateway calls the protocol conversion mapping table to uniformly parse the non-standard data of various industrial buses collected by the data acquisition probe into standard application layer payloads. Based on the built-in auto-incrementing timing algorithm, it assigns a unique 32-bit unsigned integer sequential index parameter to each independent data stream captured in the current cycle.

[0022] Preferably, the construction of a three-dimensional channel feature digital base containing historical signal strength heatmaps based on a three-dimensional rendering engine and macro-geographic terrain data includes:

[0023] By using a 3D rendering engine to load macroscopic geographic terrain elevation data and vector road network data from the entire line's geographic information system, a spatial absolute coordinate reference system is established. The historical signal strength heat map of the entire line is spatially aligned and fused with the spatial absolute coordinate reference system. Computer-aided design grid drawings of trains and communication equipment along the line are imported. With the help of a microscopic rendering engine, virtual models of physical equipment are instantiated, and a one-to-one entity binding operation is carried out between the virtual models of physical equipment and the corresponding physical media access control addresses in the database, thus constructing a 3D physical digital twin base.

[0024] Preferably, the real-time status data and the along-line environmental data are input into a pre-spatiotemporal prediction algorithm to extrapolate the network environment quality change trend of the train within a future time window in the three-dimensional channel feature digital base; when the extrapolation result predicts that the train is about to enter the communication attenuation zone within a preset distance threshold, the cloud decision center issues a cross-layer pre-scheduling instruction to the vehicle gateway, including:

[0025] The aforementioned pre-train spatiotemporal prediction algorithm extracts the three-axis coordinate data and running speed scalar currently reported by the train. Using a three-dimensional physical digital twin as the coordinate reference system, it aligns the train's real-time coordinates and speed with the three-dimensional channel characteristic digital base coordinate system. It reads the historical signal strength thermal attenuation distribution matrix of a preset section ahead of the train's forward vector direction. Utilizing the environmental data along the line, the algorithm extrapolates the probability of network fading ahead at the current train speed. Based on the historical central processing unit status 1200 meters ahead of the train, the historical network performance of the onboard router, and a pre-set threshold for determining strong attenuation of radio electromagnetic signals in the geographical blind zone 1200 meters ahead, it extrapolates the probability and degree of network channel degradation within a future time window. When it is predicted that the train is about to enter a communication attenuation zone, a cross-layer pre-scheduling instruction is generated. This instruction includes a service priority label and a link aggregation strategy, and is sent from the cloud decision center to the onboard gateway.

[0026] Preferably, the vehicle gateway, in response to the cross-layer pre-scheduling instruction, drives the underlying lightweight transmission protocol stack to perform the following actions: activating multi-link aggregation to widen redundant channels, transferring non-urgent service data to edge storage for buffering and suppression, and forcibly locking the priority transmission channel for core control signaling, including:

[0027] After receiving the cross-layer pre-scheduling instruction, the pre-scheduling execution unit at the bottom layer of the vehicle gateway activates the emergency communication state machine before the physical train enters the blind zone. The emergency communication state machine forces the vehicle gateway to complete three underlying technical actions before the physical train enters the blind zone: activate the multi-link aggregation engine, transfer non-emergency business data to the edge storage for caching and suppression, lock the MQTT QoS1 channel as the only transmission path for core operation signaling, and lock the highest quality of service channel of the message queue telemetry transmission protocol in the gateway network layer protocol stack, setting the highest quality of service channel as the absolutely exclusive uplink for core train control data and key train operation status parameters.

[0028] When the train completely enters the blind zone, the underlying QUIC protocol stack of the on-board gateway continuously detects the round-trip delay jitter variance and packet loss rate threshold. The cloud-based environmental compensation controller automatically adjusts the packet segmentation volume parameters and timeout retransmission backoff time window of the current QUIC connection based on the detection feedback. In addition, combined with the MQTT QoS1 automatic retransmission mechanism, the extreme value of the end-to-end communication signaling transmission delay is hard-constrained to within the limit range of 500 milliseconds.

[0029] Preferably, the method further includes:

[0030] Based on the received real-time data and the underlying link status, the cloud monitoring platform triggers an alarm when an anomaly is detected, obtains the three-dimensional spatial coordinates of the fault point, and uses the lightweight transmission protocol stack link statistics instantaneous communication quality table and similar historical maintenance work orders to reconstruct the three-dimensional spatial coordinates of the fault point, the instantaneous communication quality table, and similar historical maintenance work orders into dynamic context instructions carrying spatiotemporal physical constraints.

[0031] The dynamic context instructions are input into a large language model fine-tuned by railway vertical domain data to generate a multi-level interactive fault diagnosis decision tree with physical constraints.

[0032] Based on the three-dimensional spatial coordinates of the fault point, the corresponding physical point's live video stream is retrieved for simultaneous verification. After the verification is completed, the operation and maintenance personnel perform spatial annotation operations on the fault point in the three-dimensional virtual space, bind the spatial coordinates of the annotated point with the troubleshooting log, generate a spatial structured tag, and feed it back to the cloud knowledge base.

[0033] Preferably, the construction of the dynamic context instruction is performed by a structured dynamic prompt word assembly engine, which concurrently performs the following data extraction operations:

[0034] Extract the three-dimensional absolute geographic spatial coordinates bound to the faulty device from the three-dimensional channel feature digital base;

[0035] Extract message delivery rate, packet loss rate, delay jitter and reference signal receiving power within a preset time period before and after the fault from the time series database;

[0036] Extract historical fault repair and closure reports of equipment of the same physical section and model from the relational database;

[0037] The three types of heterogeneous data mentioned above are serialized into an instruction text format with strong geospatial feature constraints and instantaneous communication physical state constraints.

[0038] Preferably, the lightweight transport protocol stack is built on the Fast UDP network connection protocol, and utilizes the zero round-trip time handshake and headless blocking characteristics of the Fast UDP network connection protocol and the characteristics of the upper layer carrying message queue telemetry transport protocol, and is configured with the highest quality of service channel.

[0039] As can be seen from the technical solutions provided by the embodiments of the present invention above, the method of the present invention defines the specific input structure and operation mechanism of the large language model in the railway operation and maintenance scenario. It utilizes real physical parameters to curb the illusionary generalization of the model, ensuring that diagnostic conclusions are no longer general statements, but rather precise measures with strong on-site physical constraints, significantly reducing the average repair time for complex faults. The method of the present invention creates a complete human-computer interaction and knowledge evolution closed loop, from situational awareness to one-click real-scene verification and then to the spatial annotation and accumulation of hidden dangers. This enables the digital inheritance of network hidden dangers with strong geospatial attributes along railway lines, fundamentally solving the problems of organizational knowledge amnesia and experience reuse.

[0040] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of the invention. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A schematic diagram illustrating the implementation principle of a railway communication digital twin monitoring method based on spatiotemporal prediction and active network avoidance, provided in an embodiment of the present invention.

[0043] Figure 2 This is a schematic diagram of a vehicle-to-ground communication active network avoidance and pre-scheduling mechanism based on three-dimensional spatiotemporal prior knowledge, provided for an embodiment of the present invention.

[0044] Figure 3 This is a schematic diagram of an edge dwelling and temporal anti-out-of-order self-healing method for extremely weak networks, provided by an embodiment of the present invention.

[0045] Figure 4 A schematic diagram of a dynamic generation method for a large model obstacle removal decision tree that integrates three-dimensional spatial coordinates and a multi-dimensional communication scale, provided in an embodiment of the present invention;

[0046] Figure 5 A schematic diagram of a digital twin virtual-real mapping and spatial knowledge accumulation mechanism based on a macro-micro dual-engine provided in this embodiment of the invention;

[0047] Figure 6 A timing diagram of base construction and data acquisition provided in an embodiment of the present invention;

[0048] Figure 7 A time-series diagram for spatiotemporal prediction and active network avoidance scheduling provided in an embodiment of the present invention;

[0049] Figure 8 A timing diagram for network outage self-healing and data retransmission provided in an embodiment of the present invention;

[0050] Figure 9 This is a timing diagram for alarm triggering and intelligent troubleshooting provided in an embodiment of the present invention. Detailed Implementation

[0051] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0052] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0053] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0054] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0055] The embodiments of the present invention fundamentally reconstruct the control logic of traditional vehicle-to-ground communication systems, break through the physical isolation boundary between the network layer and the application layer of the OSI model, and create a cross-layer control flow architecture with "spatiotemporal prior perception driving the active regulation of the underlying protocol" as the core concept.

[0056] The implementation principle diagram of a railway communication digital twin monitoring method based on spatiotemporal prediction and active network avoidance provided in this embodiment of the invention is shown below. Figure 1 As shown. This embodiment of the invention takes three-dimensional geospatial space as the basis for perception, uses the MQTT (Message Queuing Telemetry Transport) + QUIC (Quick UDP Internet Connections) fusion transmission protocol as the underlying execution key, uses temporal consistency as data protection, and uses spatial physical constraints as the basis for intelligent diagnosis, ultimately forming a knowledge closed loop of experience feedback.

[0057] This invention first utilizes macroscopic geographic terrain data and historical signal strength heatmaps to construct a three-dimensional physical digital twin base with physical attributes. This three-dimensional physical digital twin base not only carries spatial perception but also serves as the coordinate reference mode for all subsequent predictions and decisions. Based on this digital base, the system extracts the train's real-time position and speed, then inputs these data into a pre-emptive spatiotemporal prediction algorithm to predict network attenuation trends ahead of the train in digital space. If it is predicted that the train is about to enter a communication attenuation zone, the cloud will proactively send cross-layer control signaling to the onboard gateway, prompting the underlying MQTT+QUIC fusion protocol stack to perform actions such as multi-link aggregation, non-urgent service caching, and core signaling channel locking before the train physically enters the blind zone, achieving "dimensionality reduction intervention" from geospatial data to the network protocol stack. This protocol stack leverages QUIC's 0-RTT fast handshake and headless blocking features of multiplexing, combined with MQTT's QoS1 quality level, to strictly limit end-to-end latency to within 500 milliseconds, thus providing programmable transmission execution capabilities for proactive network avoidance. While this proactive network avoidance mechanism solves the communication interruption problem at the blind spot entry point, it still cannot completely prevent complete network outages under extreme conditions.

[0058] This invention deeply integrates the MQTT + QUIC underlying transport protocol with the business layer time-series index, generating a unique sequential index (record_id) for each frame of data at the data acquisition end. In the event of a sudden network outage, the edge computing unit will prioritize data and store it locally. Once the network is restored, leveraging QUIC's 0-RTT fast reconnection feature, it will strictly adhere to the timestamp and index to perform precise application-layer breakpoint resumption, ensuring that the physical parameters at the moment of the failure can be completely restored. When network degradation or device failure ultimately triggers a system alarm, this invention does not employ the conventional simple calling mode of a large language model, but instead designs a structured dynamic prompt word assembly engine.

[0059] The aforementioned structured dynamic prompt word assembly engine simultaneously extracts the three-dimensional spatial coordinates of the fault point, the instantaneous communication quality table of the underlying MQTT+QUIC link statistics, including packet loss rate, latency, RSRP, etc., and also extracts similar historical maintenance work orders. Then, it reconstructs these three types of heterogeneous physical parameters into contextual instructions with strict spatiotemporal constraints, and inputs them into a large language model that has been fine-tuned in the railway vertical domain, thereby dynamically generating interactive fault diagnosis results with physical constraints.

[0060] The output results from the aforementioned intelligent fault diagnosis will ultimately form a closed-loop model with the front-end visualization mechanism. The system employs a macro-micro dual-engine rendering architecture, using real-time train attitude data to drive the micro-model, thereby achieving second-level virtual-real synchronous mapping in the macro-terrain. Once an alarm is triggered, the 3D model will not only highlight the alarm but also directly acquire the live video stream of the corresponding physical point for simultaneous review. After the review is completed, maintenance engineers can directly mark specific fault points in the 3D coordinate system. The system will automatically associate the spatial coordinates of the point with the troubleshooting log, generate a spatially structured label, and feed it back to the cloud knowledge base, thus completing the spatialization of experience.

[0061] Based on the aforementioned closely linked technological chain, this invention establishes a closed-loop model that extends from proactive network avoidance at the beginning, self-healing and fault resistance during the process, to intelligent fault clearing and experience accumulation at the end, thus achieving a transformation of railway vehicle-to-ground communication from "passive response" to "proactive prevention".

[0062] This invention provides a vehicle-to-ground communication active network avoidance and pre-scheduling mechanism based on three-dimensional spatiotemporal prior knowledge, as shown in the embodiments of the present invention. Figure 2 As shown. First, a 3D rendering engine is used to load macroscopic geographic terrain elevation data and historical signal strength heatmaps along the entire line to construct a 3D physical digital twin base with physical attributes. The system extracts the absolute positioning coordinates and vector velocity reported by the train in real time, inputs them into a pre-prepared spatiotemporal prediction algorithm, and predicts the network fading probability along the train's future displacement path in digital virtual space.

[0063] When the algorithm predicts that the train will enter a strong attenuation zone within a preset distance threshold, the cloud-based decision-making center breaks through the isolation barrier between the application layer and the network layer, directly issuing low-level cross-layer control signaling to the onboard edge gateway. This signaling forces the onboard gateway to complete three low-level technical actions before the train physically enters the blind zone: activating the multi-link aggregation engine, transferring non-urgent business data to the edge storage for caching and suppression, and forcibly locking the MQTT QoS1 channel as the sole transmission path for core operational signaling.

[0064] A schematic diagram of an edge dwell and temporal anti-out-of-order self-healing method for extremely weak networks provided in this embodiment of the invention is shown below. Figure 3 As shown in the figure, this embodiment of the invention uses a lightweight transport protocol stack that deeply integrates MQTT and QUIC as the underlying transport base, and performs cross-layer collaborative encapsulation with business layer timing index and edge caching strategy to build a data self-healing channel that is resistant to high-frequency interruptions and enforces timing consistency.

[0065] Firstly, the protocol stack offers robust protection against weak network conditions and enables rapid reconnection. The vehicle gateway employs an MQTT+QUIC converged protocol stack to replace traditional TCP / IP. Leveraging QUIC's 0-RTT fast handshake feature, data transmission can be restored without rebuilding a complete connection after a network outage, reducing reconnection latency from seconds to milliseconds. Furthermore, QUIC's multiplexing and headless blocking features fundamentally eliminate timing inversions caused by TCP head-of-line blocking. Simultaneously, MQTT is configured with QoS1 service quality to ensure that core control signaling is delivered at least once in unstable links.

[0066] Secondly, global temporal consistency is guaranteed. The physical acquisition end collects heterogeneous data at fixed intervals. Based on the unique constraint field in the database, a unique 32-bit unsigned integer absolute index is generated for each frame of data, and a microsecond-level timestamp and check bit are added to form a standardized data frame. This index is independent of the underlying transmission protocol and maintains global increment across session cycles, fundamentally solving the problem of data order loss after TCP reconnection.

[0067] Third, tiered persistence and precise retransmission. When the system receives proactive network avoidance pre-scheduling signaling or encounters a sudden physical link disruption, the edge computing unit triggers a tiered persistence strategy, forcibly persisting all unconfirmed data to local solid-state storage based on preset service priority tags. Upon network recovery, the underlying QUIC protocol utilizes a 0-RTT fast reconnection mechanism to restore the data channel first. Subsequently, the breakpoint resumption daemon starts, strictly relying on timestamps and unique sequence indexes to perform precise breakpoint resumption at the application layer. This eliminates the disordered retransmission behavior of traditional protocol stacks, forcibly queuing and retransmitting edge-resident data to the cloud in a unidirectional ascending time sequence.

[0068] A schematic diagram of a dynamic generation method for a large-scale obstacle-clearing decision tree that integrates three-dimensional spatial coordinates and a multi-dimensional communication scale provided by an embodiment of the present invention is shown below. Figure 4 As shown in the figure, this embodiment of the invention designs a structured dynamic prompt word assembly engine. When the underlying link monitoring module captures a high-level anomaly and triggers a system alarm, the engine starts a concurrent multi-threaded data capture mechanism to accurately extract three types of heterogeneous parameters: the three-dimensional absolute geospatial coordinate parameters mapped from the damaged physical hardware in the three-dimensional physical digital twin base; the instantaneous communication quality table based on the message queue telemetry transmission protocol and the fast UDP network connection protocol link statistics, including packet loss rate, latency, and reference signal received power; and the historical maintenance and repair work order record with the highest matching degree in the knowledge base.

[0069] The system reconstructs the aforementioned multi-source heterogeneous environmental characteristic parameters into contextual instructions carrying strict spatiotemporal physical constraints, and inputs them into a large language model that has undergone secondary incremental fine-tuning based on railway vertical domain communication and maintenance data. The model performs inference based on the execution of strong physical constraint instructions and dynamically outputs a multi-level interactive fault diagnosis decision tree containing a strict handling sequence.

[0070] A schematic diagram of a digital twin virtual-real mapping and spatial knowledge accumulation mechanism based on a macro-micro dual-engine provided by an embodiment of the present invention is shown below. Figure 5 As shown, this invention employs a data-driven dual-rendering engine architecture to construct the front-end visualization mechanism. The first rendering engine parses the macroscopic geographic elevation data of the entire line to establish the terrain base, while the second rendering engine loads a high-precision communication base station and a microscopic mesh physical model of the train. The system uses a coordinate matrix transformation algorithm and real-time captured train latitude, longitude, pitch, and roll attitude data to drive the microscopic model to perform second-level virtual-real synchronous mapping within the macroscopic terrain.

[0071] When the system triggers an alarm, the platform not only highlights the corresponding 3D model, but also directly uses the globally unique identification code of the device bound to the model to trigger the live video protocol direct signaling, establish a point-to-point real-time communication channel, and pull the monitoring video stream of the corresponding physical location to perform cross-verification and review on the same screen.

[0072] After verification, the system provides a spatial collaborative annotation tool, allowing maintenance engineers to perform spatial marking operations at specific fault points in a virtual 3D spatial environment. The system automatically binds the 3D absolute coordinates of the point to the troubleshooting log, generates a spatially structured tag with geographic attributes, and feeds it back into the cloud knowledge base.

[0073] The underlying execution flow, cross-layer scheduling algorithm control mechanism, and closed-loop self-healing process of the system of this invention are progressively developed according to strict temporal logic. The processing flow of the railway communication digital twin monitoring method based on spatiotemporal prediction and active network avoidance includes the following processing steps:

[0074] Step S10: Collect geographic elevation benchmark data based on the national unified coordinate system and GIS vector road network data of the entire conventional railway line to construct a three-dimensional physical digital twin base.

[0075] This step first requires acquiring geographic elevation benchmark data based on the national unified coordinate system (geographic graphics in the unified coordinate system) and GIS vector road network data for the entire conventional railway line (inputting a kml route map). After acquiring the data and filtering out valid data, the train's longitude, latitude, altitude, and speed data will be output. Hardware data such as CPU temperature and memory usage (stored as historical data), and onboard router latency, jitter, packet loss rate, and throughput data (stored as historical data) will also be included.

[0076] On-site engineers deployed lightweight, multi-source heterogeneous data acquisition probe arrays at base station equipment rooms along the railway line and at onboard router nodes on ordinary railway trains. At the kernel layer of the onboard computing gateway, a lightweight transmission control stack deeply integrating MQTT and QUIC was installed. Its underlying layer is based on the QUIC protocol, leveraging QUIC's 0-RTT fast handshake and head-of-line blocking multiplexing characteristics. The upper layer carries the MQTT protocol, configured with a QoS1 service quality level to ensure reliable delivery of core signaling in weak network environments. At the control center's private cloud server cluster, a high-concurrency message distribution middleware, namely the EMQX cluster, was built. This cluster supports MQTT over QUIC. A dual-engine database cluster was also built, with a time-series database specifically configured for persistent high-frequency network quality parameters and a relational database specifically configured for recording structured equipment ledgers.

[0077] When the train is in normal operation, the multi-source data acquisition probes strictly adhere to a fixed frequency of 100 milliseconds to collect the train's absolute coordinates, real-time vector velocity, hardware central processing unit load status, and wireless channel reference signal reception power. The heterogeneous device adaptive gateway calls the protocol conversion mapping table to uniformly parse the non-standard data from various industrial buses into standard application layer payloads. The sequential data transmission engine intervenes to process the data, assigning a unique 32-bit unsigned integer sequence index parameter to each independent data stream captured in the current cycle based on the built-in auto-incrementing timing algorithm. It then serializes and assembles the microsecond-level system timestamp, cyclic redundancy check code, and payload to generate standardized data frames with strong timing consistency and anti-out-of-order properties, which are then queued for transmission.

[0078] Figure 6 This invention provides a sequence diagram for the construction and data acquisition of a three-dimensional physical digital twin base. Cloud-based rendering computing nodes load macroscopic geographic topographic elevation benchmark data based on the national unified coordinate system and vector road network data from the geographic information system of the entire conventional railway line in parallel, thereby establishing a spatial absolute coordinate reference system. Then, computer-aided design mesh drawings of the train and its communication equipment along the line are imported. A micro-rendering engine is used to instantiate virtual models of the physical equipment, and a one-to-one entity binding operation is performed between these models and the corresponding physical media access control addresses in the database, thus constructing the three-dimensional physical digital twin base.

[0079] The aforementioned three-dimensional physical digital twin base is directly used in step S20: as a coordinate reference system for the spatiotemporal prediction algorithm; and to provide a historical signal intensity thermal attenuation distribution matrix for a preset section ahead of the train's forward vector direction, in preparation for subsequent prediction.

[0080] Step S20: Based on the current location and speed data of the train, the probability of network fading ahead of the train at the current speed is inferred by the three-dimensional physical digital twin base and spatiotemporal algorithm. When it is predicted that the train is about to enter the communication attenuation zone, a cross-layer pre-scheduling instruction is generated to drive the train's underlying MQTT+QUIC fusion protocol stack to execute the above-mentioned cross-layer pre-scheduling instruction.

[0081] This step first obtains the train's current location and speed data. Using the three-dimensional physical digital twin as the coordinate reference system for the spatiotemporal prediction algorithm, the train's real-time coordinates and speed are aligned with the three-dimensional channel characteristic digital base coordinate system. The spatiotemporal algorithm is used to deduce the probability of network fading ahead of the train at the current speed. Based on the historical CPU status of 1200 meters ahead of the train (leaving distance for braking or deceleration), the historical network performance of the onboard router, and the pre-set threshold for the strong attenuation of radio electromagnetic signals in the geographical blind zone 1200 meters ahead, when it is predicted that the train is about to enter the communication attenuation zone, a cross-layer pre-scheduling instruction is generated to drive the underlying MQTT+QUIC converged protocol stack to perform multi-link aggregation, service cache suppression, and core channel locking.

[0082] Figure 7 This invention provides a spatiotemporal prediction and active network avoidance scheduling timing diagram. The cloud-based predictive network scheduling algorithm module continuously extracts the currently reported three-axis coordinate data and running speed scalar from the train according to a predetermined calculation cycle. The algorithm engine aligns this data with the three-dimensional channel characteristic digital base coordinate system to read the historical signal strength thermal attenuation distribution matrix of a preset section ahead of the train's forward vector direction. If the spatiotemporal algorithm determines that the train, at its current speed, will enter a geographically blind zone with strong radio electromagnetic signal attenuation 1200 meters ahead, the cloud-based decision center immediately establishes a cross-layer pre-scheduling instruction with the highest scheduling priority and sends it to the designated train's onboard gateway via the downlink control channel.

[0083] Edge-end response and proactive network avoidance execution. Upon receiving a cross-layer pre-scheduling instruction, the pre-scheduling execution unit at the bottom layer of the onboard gateway immediately activates the emergency communication state machine before the train's physical body enters the blind zone. The gateway's bottom-layer driver interface forcibly initiates a multi-physical link broadband aggregation mechanism and modifies the internal network routing policy to redirect non-emergency priority service traffic, such as ordinary video surveillance streams, to the local solid-state storage of the edge node for caching and suppression. Simultaneously, the gateway's network layer protocol stack forcibly locks the QoS level one channel of the message queue telemetry transmission protocol, setting it as an absolutely dedicated uplink for core train control data and critical train operation status parameters, thereby ensuring that critical application layer messages can be delivered to the minimum limit in weak network environments.

[0084] The protocol stack undergoes adaptive parameter tuning. When the train enters a high-attenuation zone, such as deep mountains or long tunnels, the underlying QUIC protocol stack continuously detects the round-trip delay jitter variance and packet loss rate threshold. The cloud-based environmental compensation controller automatically adjusts the packet segmentation volume parameters and timeout retransmission backoff window of the current QUIC connection based on the detection feedback. It leverages QUIC's 0-RTT fast handshake feature to combat frequent wireless channel interruptions. Simultaneously, combined with the MQTT QoS1 automatic retransmission mechanism, it strictly limits the extreme value of end-to-end communication signaling transmission delay to within a 500-millisecond limit.

[0085] Step S30: Network Disconnection Self-Healing and Data Retransmission

[0086] This step first requires acquiring the remote message data for each heartbeat cycle of the train, and then using the heartbeat detection program to perform network disconnection detection on the message data, triggering a network connection failure interruption event.

[0087] The edge computing control unit acquires the data frames carrying the service priority tag of the train, the system timestamp when the last data frame was successfully saved, the 32-bit unsigned integer sequential index, and ensures global timing consistency across session periods. Based on the service priority tag, the edge computing control unit forces the persistence of all unconfirmed data to ensure global timing consistency across session periods.

[0088] During the execution of the aforementioned cross-layer pre-scheduling instructions by the underlying MQTT+QUIC fusion protocol stack of the train, when a proactive network avoidance pre-scheduling signaling is received, hierarchical resident is triggered. The QUIC protocol stack triggers a 0-RTT fast reconnection process based on the encrypted session cache. The breakpoint resume daemon strictly compares the timestamp and index to block disordered retransmission behavior.

[0089] The final output of this step includes hierarchical persistent data, precise retransmission sequences, and restored physical parameter slices. The hierarchical persistent data includes data written to the vehicle-mounted drop-proof storage array according to control signaling, status parameters, and multimedia stream hierarchical rules. The precise retransmission sequences include retransmission data streams strictly queued according to timestamps and unique order indexes. The restored physical parameter slices include continuous physical state trajectories within the blind zone restored after cloud parsing.

[0090] Figure 8This invention provides a timing diagram for network outage self-healing and data retransmission in an embodiment of the invention. In the extremely harsh geographical environment of a conventional railway, a train encounters a vacuum in base station signal coverage, resulting in a complete physical layer blockage. After failing to receive a remote acknowledgment message for three consecutive heartbeat cycles, the onboard network heartbeat detection program triggers a network connection failure interruption event. The edge computing control unit takes over the data flow, blocking data distribution to the external network interface and calling the business rule engine to parse the priority tags of data frames remaining in the local queue. All data is persistently written to the onboard drop-proof storage array according to the hierarchical rules of control signaling, status parameters, and multimedia streams, thereby ensuring the complete preservation of the instantaneous underlying characteristic data of the physical nodes during the fault.

[0091] Rapid reconnection and precise retransmission. The moment the train leaves the blind spot and captures a weak base station signal, the underlying QUIC protocol stack triggers a 0-RTT rapid reconnection process based on encrypted session caching. This means that the upper-layer MQTT data channel can be restored in a very short time without another handshake operation. After network recovery, the breakpoint resume daemon starts, and the system extracts the system timestamp and 32-bit unsigned integer sequential index from the cloud database when the last successfully saved data frame. The on-board gateway strictly compares the above two parameters, accurately extracts the first frame data at the moment of network failure from the local solid-state storage, and blocks the disordered retransmission behavior of the traditional protocol stack, forcing the edge-resident data pool to be queued and retransmitted to the cloud through the MQTT QoS1 channel in a unidirectional ascending order of time sequence. After the retransmission sequence is completed, the cloud will parse and restore the continuous physical parameter slices within the blind spot, thereby achieving lossless closed-loop self-healing of the state trajectory during the fault occurrence period.

[0092] Step S40: Alarm Triggering and Intelligent Troubleshooting.

[0093] This step requires obtaining the precise latitude, longitude, and elevation coordinates of the faulty device from the three-dimensional physical digital twin, the MQTT+QUIC link statistics (message delivery rate, packet loss rate, latency jitter, RSRP) within 5 minutes before and after the fault occurred from the time-series database of normal transmission and retransmission, and the historical fault repair and maintenance reports of the same physical segment and the same model of equipment extracted from the relational database.

[0094] The data obtained above is subjected to multi-source parameter capture: the structured dynamic prompt word assembly engine concurrently performs three-dimensional coordinate, communication quality and historical work order data extraction and processing.

[0095] The acquired data is then context-constrained: the heterogeneous data is serialized into a command text format with strong geospatial feature constraints and instantaneous communication physical state constraints.

[0096] The acquired data is used for intelligent reasoning: the constraint instructions are input into a large language model that has been fine-tuned in the railway vertical domain to generate diagnostic decision processing for physical constraints.

[0097] Spatial linkage verification is performed on the data obtained above: the twin rendering control center triggers point-to-point communication of live video to realize the verification process.

[0098] Output dynamic context instructions, including structured prompts carrying strict spatiotemporal physical constraints (used for inputting ai).

[0099] Output an interactive troubleshooting decision tree, including troubleshooting logic guidance with conditional branching structure (used for troubleshooting).

[0100] Output spatially structured labels, including empirical data of geographic attributes bound to the three-dimensional absolute coordinates of the fault point and troubleshooting logs (used for subsequent fault prediction and repair).

[0101] Output the knowledge base update results, including data that is synchronously fed back to the large model training corpus and the troubleshooting work order knowledge management pool.

[0102] Figure 9 This invention provides an alarm triggering and intelligent troubleshooting timing diagram. It involves multi-source parameter capture and contextual constraints. When the cloud-based end-to-end control center detects that the continuous latency jitter variance exceeds the limit or the lower limit of a specific hardware module, it sends a high-level warning event to the system service bus. This warning event immediately wakes up the structured dynamic prompt word assembly engine to perform multi-directional data extraction. The first direction of data extraction is to extract the precise latitude, longitude, and elevation triaxial coordinate data of the faulty hardware registration and binding from the twin base. The second direction is to obtain the instantaneous communication quality table of MQTT+QUIC link statistics within five minutes before and after the network anomaly from the time-series database. This table includes data such as message delivery rate, packet loss rate, latency jitter, and RSRP. The third direction is to extract historical fault repair and closure reports of the same physical segment and the same model of equipment from the relational database. Afterwards, the assembly engine serializes the above heterogeneous data into a command text format with strong geospatial feature constraints and instantaneous communication physical state constraints.

[0103] Decision tree generation and spatial linkage verification. The system inputs constrained instruction text into the computational graph of a large language model that has undergone secondary fine-tuning using railway professional corpus. The model engine can suppress the divergence of irrelevant parameters, dynamically infer based on the characteristics of the current absolute physical environment, generate a troubleshooting logic interactive decision tree with conditional branching structure, and then push it to the console terminal. The twin rendering control center receives alarm binding information, issues rendering layer material change instructions, causes the 3D mesh model of the damaged component to flash as an alarm, and simultaneously sends a direct pull signal for live video point-to-point communication to the media server. The dispatch terminal directly retrieves camera footage from the bound coordinate points to perform cross-verification and verification of the actual on-site environment.

[0104] Spatial experience accumulation and knowledge evolution. After the on-site maintenance team performs physical repairs and restarts the equipment according to the interactive decision tree, they log into the remote collaborative annotation management module. Engineers use mouse ray picking at specific fault trigger points in the 3D virtual coordinate system to obtain the 3D absolute coordinates and input the actual disaster-causing interference source judgment results and repair parameters. The system serializes and loads coordinate labels, synchronously feeding this spatial structured experience data back to the large model training corpus and the troubleshooting work order knowledge management pool, promoting a closed-loop improvement in the system's cognitive judgment capabilities.

[0105] In summary, this invention deeply integrates and encapsulates QUIC's 0-RTT fast reconnection and headless blocking characteristics of multiplexing, MQTT's QoS1 reliable delivery mechanism, and a globally unique sequential index at the service layer. From the physical acquisition end to the cloud receiving pool, a self-healing data channel is established that resists high-frequency interruptions, enforces timing consistency, and ensures no packet loss during network outages. After network recovery, the cloud can completely reconstruct continuous physical parameter slices within the blind zone, ensuring accurate reproduction of the true physical state at the moment of the fault, providing a reliable data foundation for root cause analysis, and completely solving the inherent defects of the traditional TCP / IP protocol stack in extremely weak railway network scenarios.

[0106] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0107] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0108] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0109] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A mobile communication digital twin monitoring method based on spatiotemporal prediction and active network avoidance, characterized in that, include: By deploying lightweight data acquisition probes on trains and ground equipment, real-time status data of trains and environmental data along the line are collected at preset intervals. Based on a 3D rendering engine and macro-geographic terrain data, a 3D digital base of channel characteristics containing historical signal strength heatmaps is constructed. The real-time status data and the along-line environment data are input into the pre-spatiotemporal prediction algorithm to deduce the trend of network environment quality changes of the train in the future time window in the three-dimensional channel feature digital base; when the deduction result predicts that the train is about to enter the communication attenuation zone within the preset distance threshold, the cloud decision center sends a cross-layer pre-scheduling instruction to the vehicle gateway. In response to the cross-layer pre-scheduling instruction, the vehicle gateway drives the underlying lightweight transmission protocol stack to perform the following actions: activate multi-link aggregation to widen redundant channels, transfer non-urgent service data to edge storage for buffering and suppression, and forcibly lock the priority transmission channel for core control signaling.

2. The method according to claim 1, characterized in that, The method of collecting real-time status data of the train and environmental data along the line at preset intervals using lightweight data acquisition probes deployed on trains and ground equipment includes: Lightweight, multi-source heterogeneous data acquisition probe arrays are deployed in base station equipment rooms along the railway line and at the onboard router nodes of ordinary railway trains. When the train is in normal operation, the data acquisition probes collect real-time status data of the train and environmental interference data along the line at a fixed frequency. The real-time status data of the train includes: longitude, latitude, altitude and speed data of the train; temperature, load status and memory usage of the central processing unit of the hardware; latency, jitter, packet loss rate, throughput data and wireless channel reference signal received power of the onboard router; the environmental data along the line includes geographic elevation benchmark data based on the national unified coordinate system, GIS vector road network data of the entire ordinary railway line and environmental interference data along the line. The heterogeneous device adaptive gateway calls the protocol conversion mapping table to uniformly parse the non-standard data of various industrial buses collected by the data acquisition probe into standard application layer payloads. Based on the built-in auto-incrementing timing algorithm, it assigns a unique 32-bit unsigned integer sequential index parameter to each independent data stream captured in the current cycle.

3. The method according to claim 2, characterized in that, The aforementioned construction of a 3D channel feature digital foundation, including historical signal strength heatmaps, based on a 3D rendering engine and macro-geographic terrain data, includes: By using a 3D rendering engine to load macroscopic geographic terrain elevation data and vector road network data from the entire line's geographic information system, a spatial absolute coordinate reference system is established. The historical signal strength heat map of the entire line is spatially aligned and fused with the spatial absolute coordinate reference system. Computer-aided design grid drawings of trains and communication equipment along the line are imported. With the help of a microscopic rendering engine, virtual models of physical equipment are instantiated, and a one-to-one entity binding operation is carried out between the virtual models of physical equipment and the corresponding physical media access control addresses in the database, thus constructing a 3D physical digital twin base.

4. The method according to claim 3, characterized in that, The method involves inputting the real-time status data and the along-line environmental data into a pre-spatiotemporal prediction algorithm to deduce the trend of network environment quality changes for the train within a future time window using the three-dimensional channel feature digital base. When the simulation results predict that the train is about to enter the communication attenuation zone within a preset distance threshold, the cloud-based decision center issues a cross-layer pre-scheduling instruction to the onboard gateway, including: The aforementioned pre-train spatiotemporal prediction algorithm extracts the three-axis coordinate data and running speed scalar currently reported by the train. Using a three-dimensional physical digital twin as the coordinate reference system, it aligns the train's real-time coordinates and speed with the three-dimensional channel characteristic digital base coordinate system. It reads the historical signal strength thermal attenuation distribution matrix of a preset section ahead of the train's forward vector direction. Utilizing the environmental data along the line, the algorithm extrapolates the probability of network fading ahead at the current train speed. Based on the historical central processing unit status 1200 meters ahead of the train, the historical network performance of the onboard router, and a pre-set threshold for determining strong attenuation of radio electromagnetic signals in the geographical blind zone 1200 meters ahead, it extrapolates the probability and degree of network channel degradation within a future time window. When it is predicted that the train is about to enter a communication attenuation zone, a cross-layer pre-scheduling instruction is generated. This instruction includes a service priority label and a link aggregation strategy, and is sent from the cloud decision center to the onboard gateway.

5. The method according to claim 4, characterized in that, The vehicle gateway, in response to the cross-layer pre-scheduling command, drives the underlying lightweight transmission protocol stack to perform the following actions: activate multi-link aggregation to widen redundant channels, transfer non-urgent service data to edge storage for buffering and suppression, and forcibly lock the priority transmission channel for core control signaling, including: After receiving the cross-layer pre-scheduling instruction, the pre-scheduling execution unit at the bottom layer of the vehicle gateway activates the emergency communication state machine before the physical train enters the blind zone. The emergency communication state machine forces the vehicle gateway to complete three underlying technical actions before the physical train enters the blind zone: activate the multi-link aggregation engine, transfer non-emergency business data to the edge storage for caching and suppression, lock the MQTT QoS1 channel as the only transmission path for core operation signaling, and lock the highest quality of service channel of the message queue telemetry transmission protocol in the gateway network layer protocol stack, setting the highest quality of service channel as the absolutely exclusive uplink for core train control data and key train operation status parameters. When the train completely enters the blind zone, the underlying QUIC protocol stack of the on-board gateway continuously detects the round-trip delay jitter variance and packet loss rate threshold. The cloud-based environmental compensation controller automatically adjusts the packet segmentation volume parameters and timeout retransmission backoff time window of the current QUIC connection based on the detection feedback. In addition, combined with the MQTT QoS1 automatic retransmission mechanism, the extreme value of the end-to-end communication signaling transmission delay is hard-constrained to within the limit range of 500 milliseconds.

6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Based on the received real-time data and the underlying link status, the cloud monitoring platform triggers an alarm when an anomaly is detected, obtains the three-dimensional spatial coordinates of the fault point, and uses the lightweight transmission protocol stack link statistics instantaneous communication quality table and similar historical maintenance work orders to reconstruct the three-dimensional spatial coordinates of the fault point, the instantaneous communication quality table, and similar historical maintenance work orders into dynamic context instructions carrying spatiotemporal physical constraints. The dynamic context instructions are input into a large language model fine-tuned by railway vertical domain data to generate a multi-level interactive fault diagnosis decision tree with physical constraints. Based on the three-dimensional spatial coordinates of the fault point, the corresponding physical point's live video stream is retrieved for simultaneous verification. After the verification is completed, the operation and maintenance personnel perform spatial annotation operations on the fault point in the three-dimensional virtual space, bind the spatial coordinates of the annotated point with the troubleshooting log, generate a spatial structured tag, and feed it back to the cloud knowledge base.

7. The method according to claim 6, characterized in that, The construction of the dynamic context instructions is performed by a structured dynamic prompt word assembly engine, which concurrently performs the following data extraction operations: Extract the three-dimensional absolute geographic spatial coordinates bound to the faulty device from the three-dimensional channel feature digital base; Extract message delivery rate, packet loss rate, delay jitter and reference signal receiving power within a preset time period before and after the fault from the time series database; Extract historical fault repair and closure reports of equipment of the same physical section and model from the relational database; The three types of heterogeneous data mentioned above are serialized into an instruction text format with strong geospatial feature constraints and instantaneous communication physical state constraints.

8. The method according to independent claim 1, characterized in that, The lightweight transport protocol stack is built on the Fast UDP network connection protocol, and utilizes the zero round-trip time handshake and headless blocking characteristics of the Fast UDP network connection protocol, as well as the characteristics of the upper layer carrying message queue telemetry transport protocol, and is configured with the highest quality of service channel.

Images