Signal takeover control system for emergency working condition of vehicle base

Abstract

The invention relates to the technical field of rail transit signal control, and discloses a signal takeover control system for emergency working conditions of a vehicle base. The system is deployed between an existing signal control system and outdoor signal equipment of a vehicle base, a station graph model is constructed by collecting multi-source state information of a turnout, a signal machine, a track section, an axle counter and the like, and station structure reconstruction is achieved by introducing a structure maintaining rectification current matching graph generation mechanism; on the basis, a risk assessment result and a control boundary are generated through graph reasoning, an equipment control permission matrix and a control script are formed in combination with a safety rule base, and equipment operation and state backward reading are achieved through a command execution closed loop. Therefore, the safe takeover control of the vehicle base outdoor signal equipment is realized under the condition that the signal interlocking system is abnormal or the communication is interrupted.

Description

Technical Field

[0001] This invention relates to the field of rail transit signal control technology, and in particular to a signal takeover control system for emergency situations in vehicle depots. Background Technology

[0002] With the continuous expansion of urban rail transit systems, depots typically house a large number of outdoor signaling devices, including switches, signals, track sections, axle counting equipment, and track circuits. These devices are controlled by a signal interlocking system and a centralized control system to ensure the safe organization of train shunting and depot operations. Under normal operating conditions, the interlocking system performs interlocking verification and executes equipment control based on information such as section occupancy status, switch position status, and signal opening conditions. However, in actual operation, when the signal interlocking system experiences communication anomalies, control system failures, or control link interruptions, the original control system may be unable to continue effectively controlling the outdoor signaling devices, affecting critical operations such as switch switching, signal opening, and section closure. Although existing technologies include emergency control devices, they typically only possess basic control functions and lack the overall modeling and analysis capabilities for the status, topology, and interlocking constraints of station equipment. With the development of intelligent rail transit signaling systems, some technologies have introduced data analysis and intelligent algorithms for equipment status assessment. However, they lack support for station structure reconstruction, control boundary reasoning, and equipment operation constraint generation under interlocking anomalies, making it difficult to meet the safety control requirements in emergency takeover scenarios at vehicle depots. Summary of the Invention

[0003] This invention addresses the problem of difficulty in safely taking over control of outdoor signaling equipment under abnormal or malfunctioning conditions in vehicle depot signaling interlocking systems. It proposes a signal takeover control system for emergency situations in vehicle depots. This system establishes an independent emergency takeover control link between the existing signaling control system and outdoor equipment in the vehicle depot. By uniformly collecting and structuring multi-source status information from switches, signals, track sections, axle counters, and power supply status, a station map model reflecting the station topology, equipment control dependencies, and interlocking constraints is constructed. A structure-preserving rectifier flow matching graph generation mechanism is introduced to jointly model the station topology and equipment status, achieving stable reconstruction and consistent representation of the station structure. Based on this, a graph reasoning mechanism performs structured calculations on the station map, generating risk assessment results and control boundaries by integrating equipment status, section occupancy relationships, and interlocking constraints. This results in an equipment control permission matrix and operational constraints, enabling automatic identification of the controllable equipment range and key interlocking relationships. Meanwhile, by integrating the graph reasoning results with the pre-set safety rule base, control scripts that meet the interlocking rule constraints are generated. An execution closed loop including precondition verification, command issuance, status readback, and timeout handling is constructed at the command execution layer, enabling the equipment operation process to have real-time feedback and traceability capabilities. This allows for the safe takeover and stable control of outdoor signal equipment in the vehicle base under conditions of signal interlocking system abnormality or communication interruption, thereby improving the safety and reliability of station control in emergency situations.

[0004] This invention provides a signal takeover control system for emergency conditions in vehicle depots. The system includes: an emergency judgment module, an isolation switching module, a station map construction module, a graph reasoning boundary module, a control permission generation module, a command execution closed-loop module, an emergency takeover host, an interface isolation and safety relay unit, a field signal acquisition and protocol adaptation unit, a manual control console and human-machine interaction terminal, and a command issuance and execution closed-loop unit.

[0005] The emergency judgment module constructs multi-source status data; the emergency takeover host performs consistency verification, timeout judgment, and abnormal threshold comparison on the multi-source status data to form an emergency condition judgment result; the judgment is made based on the emergency condition judgment result and the judgment result is obtained; when the judgment result meets the preset takeover trigger condition, a takeover request signal is generated and the system state is switched from monitoring state to takeover preparation state.

[0006] After receiving the takeover request signal, the isolation switching module, interface isolation and safety relay unit performs electrical isolation and logic interlocking operations to switch the system state from takeover preparation state to takeover operation state.

[0007] After entering the takeover operation state, the station map construction module drives the field signal acquisition and protocol adaptation unit to perform a full refresh acquisition of the current station status, obtaining refreshed multi-source status data. Based on the refreshed multi-source status data, a traceable station status observation set is constructed and consistency verification and anomaly cleaning are performed. Then, according to equipment code, location mileage, turnout number, signal number, section number, and interlocking area number, a structured mapping is performed to generate a set of node feature vectors and a set of edge relationships. The structure-preserving rectifier flow matching graph generation method is introduced to perform graph generation processing. The noise prior station map is constructed through the graphette prior generation mechanism. By integrating the Gromov-Wasserstein structure optimal transmission mechanism, the structural migration and endpoint generation are completed under the constraints of structural coupling and node alignment and continuous-time rectifier flow matching reconstruction, resulting in a vehicle base station map that meets the interlocking rules.

[0008] The graph reasoning boundary module completes the construction of the vehicle base station map, performs multi-round message passing calculations on the vehicle base station map, and performs joint reasoning at the graph level and subgraph level to generate station map reasoning results corresponding to the current takeover operation state;

[0009] The control permit generation module, based on the station map reasoning results, integrates the key interlock constraint set with the vehicle base's pre-set safety rule base to generate a manual control permit matrix and step constraints corresponding to the current takeover operation state, and forms a control script;

[0010] The command execution closed-loop module displays controllable devices, current status, interlock verification items, risk scores, and suggested operation sequences after receiving the control script from the manual control console and human-machine interface terminal. When the dispatcher selects the target device and initiates an operation, the system performs identity authentication and dual-person confirmation procedures, and submits the confirmation results and operation parameters to the command issuance and execution closed-loop unit. The command issuance and execution closed-loop unit executes the actions of "precondition verification - command encoding - issuance - execution readback - timeout handling" in sequence according to the control script. After the command is issued, the field signal acquisition and protocol adaptation unit reads back the representation quantity, acknowledgment quantity, and related segment occupancy changes of the target device, and feeds back the readback results to the emergency takeover host. When the readback result meets the success criteria, the status database is updated and a closed loop is completed. When the readback fails or a timeout occurs, a rollback strategy and alarm prompts are triggered, and the reason for the failure and the explanation of the related nodes are written to the log.

[0011] Furthermore, a structure-preserving rectified flow matching graph generation method is introduced to perform graph generation processing to obtain the vehicle base station map. The specific steps include:

[0012] Step S1: Add a collection timestamp, data source identifier, and quality flag to each piece of status data of the refreshed multi-source status data to form a traceable station status observation set;

[0013] Step S2: Perform consistency verification and anomaly cleaning on the station status observation set, and establish a structured mapping index based on equipment code, location mileage, turnout number, signal number, section number and interlocking area number; during the mapping process, normalize the observation features according to equipment type, generate a set of node feature vectors and a set of edge relationships, and obtain the observation station map;

[0014] Step S3: Invoke the graphette prior generation mechanism. The graphette prior in the graphette prior generation mechanism consists of a graph function, a sparse control sequence, and a graph editing function. The graph function is used to generate an initial topological skeleton that matches the scale of the vehicle base station. The sparse control sequence is used to constrain the connectivity density of the initial topological skeleton to match the typical density of the vehicle base tracks and interlocking areas. The graph editing function is used to perform structural editing on the initial topological skeleton to inject repeating structural phantoms into the station. Based on the graphette prior, sampling is performed to obtain a noisy prior station map.

[0015] Step S4: The emergency takeover host invokes the pre-trained edge-aware graph encoder to encode the noise prior station map and the observation station map respectively, and extracts the node-level structure embedding representations of the two. Based on the node-level structure embedding representations, the intra-graph structure cost matrix of the noise prior station map and the observation station map is calculated respectively to characterize the structural differences between different nodes within the same station map. At the same time, the cross-graph feature cost matrix is ​​calculated based on the node features of the noise prior station map and the node observation features of the observation station map to characterize the differences in node attributes between the two maps.

[0016] Step S5: Invoke the Gromov-Wasserstein structure-optimal transport mechanism, construct a fusion cost function based on the intra-graph structure cost matrix and the cross-graph feature cost matrix, solve the structure-preserving coupling relationship between the noisy prior station map and the observation station map, and obtain the structure-aligned paired station map pair; the structure-aligned paired station map pair includes the structure-aligned source station map and the target station map;

[0017] Step S6: Construct a continuous-time rectified flow matching trajectory on the structure-aligned paired station map pair to generate a linear interpolated station representation; specifically: select a time parameter in a continuous time interval from zero to one, and use the time parameter as a weight to perform linear interpolation on the source and target station maps after structure alignment at three levels: adjacency relationship, node features, and edge features, respectively, to generate the corresponding linear interpolated station representation;

[0018] Step S7: Construct a permutation-equivariant graph neural network Transformer velocity field predictor. Input the linear interpolation station field representation and the corresponding time parameters into the velocity field predictor. Output the adjacent velocity, which characterizes the rate of change of adjacency relationships; the node feature velocity, which characterizes the rate of change of node features; and the edge feature velocity, which characterizes the rate of change of edge features. Construct a local velocity matching loss function. By minimizing the local velocity matching loss function, ensure that the adjacent velocity, node feature velocity, and edge feature velocity are consistent with their corresponding changes. This constrains the velocity field predictor to output an instantaneous transmission direction that conforms to the rectified flow matching trajectory at any intermediate time.

[0019] Step S8: Introduce endpoint consistency supervision terms and interlocking feasibility regularization constraints, and combine them with the local velocity matching loss function for weighted fusion to form the total loss function used to train the velocity field predictor;

[0020] Step S9: Iteratively optimize the velocity field predictor based on the total loss function to obtain the station structure-preserving velocity field; after obtaining the station structure-preserving velocity field, use the source station map after structure alignment as the initial input, and perform finite-step rectification integral update according to the preset time step to obtain the endpoint station representation at the termination time; perform discretization projection and interlocking rule correction on the endpoint station representation to make the adjacency relationship, node state and edge feature satisfy the equipment type value range and interlocking constraints to obtain the vehicle base station map.

[0021] Furthermore, the edge relationship set of the vehicle depot station map consists of topological connection edges, control dependency edges, interlocking constraint edges, and fault propagation edges. Among them: topological connection edges are used to represent the spatial connectivity between turnouts and sections, and between sections and signals; control dependency edges are used to represent the influence of equipment control commands on equipment representation quantities and interlocking zone states; interlocking constraint edges are used to represent the constraint relationships between turnout positions, section occupancy, and signal opening conditions; and fault propagation edges are used to represent the influence of power supply abnormalities, communication abnormalities, and control cabinet abnormalities on adjacent equipment and equipment within the same interlocking zone.

[0022] By adopting the above solution, the beneficial effects achieved by the present invention are as follows:

[0023] This invention establishes an independent emergency takeover control link between the existing signal control system and outdoor signal equipment in the vehicle depot, and performs unified collection and structured processing of multi-source status information such as turnouts, signals, track sections, axle count status, and power supply status. This enables centralized acquisition and unified modeling of the status of equipment in the vehicle depot, improves the integrity and consistency of equipment status information under emergency conditions, and solves the problem of difficulty in unified acquisition and comprehensive analysis of outdoor equipment status in the event of signal interlocking system anomalies or communication link interruptions. The system can quickly generate basic station status data that can be used for control decision-making in complex station environments, thereby enhancing the vehicle depot's equipment takeover capability and system operation continuity under abnormal conditions.

[0024] This invention constructs a station map model that includes equipment node features and multiple types of relational edges, and introduces a structure-preserving rectifier flow matching graph generation mechanism to jointly model the station topology and equipment status. This achieves stable reconstruction and consistent expression of the vehicle depot station structure, improves the system's ability to express complex station topologies and equipment dependencies, and solves the problem in existing emergency control technologies of the difficulty in comprehensively depicting the control dependencies, interlocking constraints, and state propagation relationships between turnouts, track sections, signals, and interlocking areas. This enables the system to accurately reconstruct the station structure and maintain the consistency of interlocking rules in emergency takeover scenarios, thereby enhancing the structural reliability and operational safety of the vehicle depot emergency takeover control process.

[0025] This invention generates an equipment control permission matrix and control script by performing structured graph reasoning calculations on the station map and integrating the reasoning results with a pre-set safety rule base. Simultaneously, it constructs a command execution closed loop that includes precondition verification, command issuance, status readback, and timeout handling. This enables automatic identification of equipment control boundaries and closed-loop management of the control process in vehicle depot emergency takeover scenarios. It enhances the ability to generate safe operation sequences under complex equipment states and interlocking constraints, solving the problems of difficult-to-determine control boundaries, lack of unified constraints on equipment operation, and lack of feedback verification of execution results in existing emergency control methods. Therefore, it enhances the safety, controllability, and traceability of the vehicle depot emergency control process. Attached Figure Description

[0026] Figure 1 A schematic diagram of a signal takeover control system for emergency conditions at a vehicle depot, provided by the present invention;

[0027] Figure 2 This is a schematic diagram of the station topology provided by the station map construction module in Example 1. Detailed Implementation

[0028] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0029] Example 1, according to Figure 1 This invention provides a signal takeover control system for emergency situations in vehicle depots. It is applied to emergency control scenarios of outdoor signal equipment in rail transit vehicle depots. The system is deployed between the existing signal control system and outdoor equipment in the vehicle depot and is connected to the on-site communication network of the vehicle depot. The system includes: an emergency judgment module, an isolation switching module, a station map construction module, a graph reasoning boundary module, a control permission generation module, a command execution closed-loop module, an emergency takeover host, an interface isolation and safety relay unit, an on-site signal acquisition and protocol adaptation unit, a manual control console and human-machine interaction terminal, and a command issuance and execution closed-loop unit.

[0030] The emergency judgment module continuously receives operating heartbeat signals, interlocking communication status information, station centralized monitoring status, and alarm information from the vehicle base signal interlocking and centralized control system. The field signal acquisition and protocol adaptation unit collects and structures turnout indications, signal indications, track section occupancy, axle count status, track circuit occupancy status, power supply status, and access control status, generating corresponding auxiliary status data. The operating heartbeat signals, interlocking communication status information, station centralized monitoring status and alarm information, and auxiliary status data together constitute multi-source status data. The emergency takeover host performs consistency verification, timeout judgment, and anomaly threshold comparison on the multi-source status data to form an emergency condition judgment result. Based on the emergency condition judgment result, a judgment is made and the judgment result is obtained. When the judgment result meets the preset takeover trigger conditions, a takeover request signal is generated, and the system status is switched from monitoring state to takeover preparation state.

[0031] In this embodiment, when the system is in monitoring mode, the emergency judgment module receives the operating heartbeat signal of the existing signal interlocking system at a period of 500ms, receives the interlocking communication status and centralized monitoring status at a period of 1s, receives the field discrete quantity acquisition data at a period of 20ms, and receives the analog quantity acquisition data at a period of 200ms. The emergency judgment module performs consistency verification, timeout judgment and abnormal threshold comparison on the multi-source status data.

[0032] 1) The timeout thresholds are set as follows: Heartbeat timeout threshold: 3s (6 consecutive missing heartbeat cycles are judged as timeout); Interlocking communication timeout threshold: 2s (communication status is interrupted for 2 consecutive seconds and is judged as timeout); Key equipment refresh timeout threshold: 1s (any type of section occupancy, turnout indication, or signal indication is not updated for 1 consecutive second and is judged as timeout).

[0033] 2) Consistency Verification Rule Examples: Section Occupancy and Axle Counting Consistency: If the same detection object shows a mismatch of "the axle count has increased but the section is still continuously idle" ≥3 times within a 10s sliding window, it is judged as a mismatch anomaly; Turnout Position and Locking Consistency: If the locking indicator is not in place within 500ms after the turnout position change is completed, it is judged as a locking anomaly; Signal Opening Condition Consistency: If the signal shows open and its protected section occupancy is "occupied" for ≥200ms, it is judged as an opening condition conflict.

[0034] 3) Alarm density threshold: If the number of alarms accumulated within 60 seconds is ≥8 and includes any of the critical alarms such as "communication interruption / interlock host abnormality / critical acquisition mismatch", the takeover priority will be increased.

[0035] The system detected the following during the period from 09:17:23 to 09:17:28: 7 consecutive missing heartbeats of the existing interlocking system, with each missing heartbeat lasting 3.5 seconds; the interlocking communication status changed from "normal" to "interrupted" for 2.4 seconds; 11 alarms were accumulated on the centralized monitoring platform within 60 seconds, including "interlocking communication interrupted" and "station display refresh failed"; and the axle counting object AX-03 experienced 4 mismatch events of "entry segment count changed but the segment is still idle" within a 10-second sliding window.

[0036] Based on this, the emergency judgment module outputs the emergency condition judgment result as "the takeover trigger condition is met", generates a takeover request signal, and switches the system status from monitoring state to takeover preparation state.

[0037] After receiving the takeover request signal, the isolation and switching module, interface isolation and safety relay unit, performs electrical isolation and logic interlocking operations on the control link between the vehicle base signal interlocking and centralized control system, disconnecting the original control system's drive path to the outdoor equipment control circuit, making the outdoor equipment a state with only one controllable source; after completing the isolation, it switches the outdoor equipment control circuit to the emergency control link of the command issuance and execution closed-loop unit, and returns isolation completion signal and switching completion signal to the emergency takeover host; after receiving the isolation completion signal and switching completion signal simultaneously, the emergency takeover host switches the system state from takeover preparation state to takeover operation state;

[0038] After entering the takeover operation state, the station map construction module drives the field signal acquisition and protocol adaptation unit to perform a full refresh acquisition of the current station status, obtaining refreshed multi-source status data. Based on the refreshed multi-source status data, a traceable station status observation set is constructed and consistency verification and anomaly cleaning are performed. Then, according to equipment code, location mileage, turnout number, signal number, section number, and interlocking area number, a structured mapping is performed to generate a set of node feature vectors and a set of edge relationships. The structure-preserving rectifier flow matching graph generation method is introduced to perform graph generation processing. The noise prior station map is constructed through the graphette prior generation mechanism. By integrating the Gromov-Wasserstein structure optimal transmission mechanism, the structural migration and endpoint generation are completed under the constraints of structural coupling and node alignment and continuous-time rectifier flow matching reconstruction, resulting in a vehicle base station map that meets the interlocking rules.

[0039] The station map construction module generates a station topology diagram based on the set of edge relationships. Figure 2 Specifically, the location mileage, equipment type, and equipment number in the node feature vector set are parsed into visual coordinates and graphic element styles, and the connection relationship, constraint relationship, and propagation relationship in the edge relationship set are parsed into line and arrow graphic elements. During the generation process, the node graphic elements are laid out according to the preset layout rules to form the station topology skeleton, and the status mark and timestamp corresponding to the currently refreshed multi-source status data are superimposed on the node graphic elements to obtain the station topology map corresponding to the takeover operation state. Figure 2 In the diagram, black straight lines represent connection relationships, blue dashed lines represent constraint relationships, and red dashed lines represent propagation relationships; rectangles represent sections, rhombuses represent turnouts, circles represent signals, hexagons represent areas, and triangles represent facilities.

[0040] The graph reasoning boundary module, after completing the construction of the vehicle base station map, performs multi-round message passing calculations on the station map and conducts joint reasoning at the graph level and subgraph level to generate station map reasoning results corresponding to the current takeover operation state. The station map reasoning results include risk assessment results and control boundaries. Among them, the risk assessment results include risk scores and risk source explanation information, and the control boundaries include controllable equipment sets, prohibited equipment sets, suggested operation sequences, and key interlock constraint sets. The controllable equipment sets and prohibited equipment sets are used to limit the range of equipment that are allowed or prohibited from manual control in the current state, and the key interlock constraint set is used to characterize the interlocking restriction relationship between equipment under the current position and occupancy conditions. The risk score is calculated comprehensively based on the current track section occupancy status, turnout position status, signal opening conditions, power supply status, and communication status to quantify the conflict intensity and safety risk level that may occur when executing a specific manual control command. The risk source explanation information is used to identify key nodes and key edges that contribute significantly to the risk score in the station map, providing a basis for subsequent control permission generation and manual confirmation.

[0041] The graph reasoning boundary module performs multi-round message passing reasoning on the vehicle base station map, with the number of message passing rounds set to 4, and outputs the station map reasoning results: Risk score: R=0.78 (0–1 normalized), risk level mapped to "high"; Explanation of risk sources: Key nodes are section T07, turnout P03, and signal S04; Key edges are "T07—P03 interlock constraint edge" and "P03—S04 open condition edge"; Set of controllable devices: {P01, P02, P04, P06, S01, S02, T02 blocked, T03 blocked}; Set of prohibited devices: {P03, S04, T07 related operations}; Suggested operation sequence: First execute "T07 block confirmation", then execute "P03 related route disabling", then execute "P04 switch to positioning".

[0042] The control permission generation module, based on the station map reasoning results, integrates the key interlock constraint set with the vehicle base's pre-set safety rule base to generate a manual control permission matrix and step constraints corresponding to the current takeover operation state. The safety rule base includes rules for prohibiting switching due to occupancy, prohibiting the opening of signals when the turnout is not locked, prioritizing section blocking, and restricting concurrent operations in the same interlocking area. The manual control permission matrix is ​​used to determine whether turnout switching operations, turnout single locking and unlocking operations, signal opening and closing operations, section blocking and unlocking operations, and guidance and shunting permission operations are allowed to execute, prohibited to execute, or require secondary confirmation. For operations determined to be allowed to execute, a corresponding control script is generated. The control script includes a set of preconditions, interlock verification items, and execution timeout threshold parameters, which are used to constrain the execution process of the command issuance and execution closed-loop unit.

[0043] The command execution closed-loop module displays controllable devices, current status, interlock verification items, risk scores, and suggested operation sequences after receiving the control script from the manual control console and human-machine interface terminal. When the dispatcher selects the target device and initiates an operation, the system performs identity authentication and dual-person confirmation procedures, and submits the confirmation results and operation parameters to the command issuance and execution closed-loop unit. The command issuance and execution closed-loop unit executes the actions of "precondition verification - command encoding - issuance - execution readback - timeout handling" in sequence according to the control script. After the command is issued, the field signal acquisition and protocol adaptation unit reads back the representation quantity, acknowledgment quantity, and related segment occupancy changes of the target device, and feeds back the readback results to the emergency takeover host. When the readback result meets the success criteria, the status database is updated and a closed loop is completed. When the readback fails or a timeout occurs, a rollback strategy and alarm prompts are triggered, and the reason for the failure and the explanation of the related nodes are written to the log.

[0044] After the dispatcher selects "P04 switch to location" on the manual console and completes double confirmation, the command issuance and execution closed-loop unit is executed as follows: precondition verification passed → command encoding → issuance → readback;

[0045] The data readback for this operation is as follows: Command issuance time: 09:18:06.120; Conversion start readback: 09:18:06.260 (Conversion in progress = 1); Lock in place readback: 09:18:07.540 (Locked in = 1); Position in place readback: 09:18:07.620 (Position in place = 1); Changes in associated segment occupancy: No new occupancy;

[0046] The system determines success and updates the status database, recording a successful closed-loop log.

[0047] Example 2, based on Example 1, introduces a structure-preserving rectifier flow matching diagram generation method to perform diagram generation processing and obtain the vehicle base station diagram. The specific steps include:

[0048] Step S1: Add a collection timestamp, data source identifier, and quality flag to each piece of status data of the refreshed multi-source status data to form a traceable station status observation set;

[0049] Step S2: Perform consistency verification and anomaly cleaning on the station status observation set, and establish a structured mapping index based on equipment codes, location mileage, turnout numbers, signal numbers, section numbers, and interlocking area numbers; during the mapping process, normalize the observation features according to equipment type to generate a set of node feature vectors, and construct a set of edge relationships according to the hierarchy of "physical topology connection relationship - control dependency relationship - interlocking constraint relationship - fault propagation relationship" to obtain the observation station map;

[0050] Step S3: Invoke the graphette prior generation mechanism. The graphette prior in the graphette prior generation mechanism consists of a graph function, a sparse control sequence, and a graph editing function. The graph function is used to generate an initial topological skeleton that matches the scale of the vehicle base station. The sparse control sequence is used to constrain the connectivity density of the initial topological skeleton to match the typical density of the vehicle base tracks and interlocking areas. The graph editing function is used to perform structural editing on the initial topological skeleton to inject repeating structural phantoms into the station. Based on the graphette prior, sampling is performed to obtain a noisy prior station map.

[0051] The definition of a graphette:

[0052] ;

[0053] in, This represents the graphette prior, which is the overall probabilistic model used to generate the noise / prior graph. Represents graph functions (graphon kernel). Represents a number of nodes A varying sparse control sequence is used to control the overall edge density of the graph; The graph editing function represents the function used to perform structural editing on the base sampled graph for injection.

[0054] Step S4: The emergency takeover host invokes the pre-trained edge-aware graph encoder to encode the noise prior station map and the observation station map respectively, extracting the node-level structural embedding representations of both. Based on the node-level structural embedding representations, the intra-graph structural cost matrix of each of the noise prior station map and the observation station map is calculated to characterize the structural differences between different nodes within the same station map. Simultaneously, based on the node features of the noise prior station map and the node observation features of the observation station map, a cross-graph feature cost matrix is ​​calculated to characterize the differences in node attributes between the two maps. Through the joint construction of the intra-graph structural cost matrix and the cross-graph feature cost matrix, a unified measurement of the topological structural differences and node attribute differences of the station map is achieved.

[0055] Step S5: Invoke the Gromov-Wasserstein optimal transport mechanism and construct a fusion cost function based on the intra-graph structural cost matrix and the cross-graph feature cost matrix to solve for the structurally preserving coupling relationship between the noisy prior station map and the observation station map. Specifically, under the constraint of uniform weighting of nodes in both graphs, the optimal transport plan is obtained. The intra-graph structural cost and the cross-graph feature cost are weighted and fused by setting a tradeoff coefficient so that the obtained optimal transport plan satisfies both the similarity of node attributes and the compatibility of station topology. Based on the optimal transport plan, node correspondence and consistent node budget constraints are generated, and permutation-consistent node alignment processing is performed on the noisy prior station map and the observation station map accordingly to obtain structurally aligned paired station map pairs, which provide structurally consistent supervised pairing inputs for subsequent continuous time-stream matching generation. The structurally aligned paired station map pairs include the structurally aligned source station map and the target station map.

[0056] The optimization objective function for the optimal transport mechanism of the Gromov-Wasserstein structure is defined as follows:

[0057] ;

[0058] in, This represents the noise priori site map. This represents a map of the observation station. The coefficient represents the trade-off factor. The larger the coefficient, the more it emphasizes the consistency of the station topology, such as the connection relationship between turnouts, sections, and signals, the dependency structure of interlocking zones, and the dependency edges of control cabinets. The smaller the coefficient, the more it emphasizes the similarity of node attributes, such as the closer the observed characteristics of turnout position status, signal open / closed status, section occupancy status, communication status, and power supply status. This represents the minimum value of the optimal transmission target for fusion; Indicates the transmission plan, Represents the cross-graph feature cost matrix; Indicates feature matching terms; This represents the weight distribution of each node in the noise prior station map. This represents the weight distribution of each node in the observation station field map; This represents the set of transmission plans that satisfy the marginal distribution constraint. , This indicates node-level structure embedding. Represents the cost function. Represents the Gromov-Wasserstein structure term;

[0059] Step S6: Construct a continuous-time rectified flow matching trajectory on the structure-aligned paired station map pair to generate a linear interpolated station representation; specifically: select a time parameter within a continuous time interval from zero to one, and use the time parameter as weight to perform linear interpolation on the source and target station maps after structure alignment at the three levels of adjacency relationship, node features, and edge features, respectively, to generate the corresponding linear interpolated station representation; define the sequence of linear interpolated station representations obtained by linear interpolating the source and target station maps after structure alignment according to the time parameter within a continuous time interval from zero to one as the linear interpolation trajectory; and determine the constant velocity displacement vector from the source and target station maps after structure alignment based on the linear interpolation trajectory. The constant velocity displacement vector corresponds to the change in adjacency relationship, the change in node features, and the change in edge features, respectively, and is used as the supervision target for continuous-time velocity field learning to constrain the subsequent velocity field predictor to output an instantaneous transmission direction consistent with the constant velocity displacement vector at any intermediate time.

[0060] Step S7: Construct a permutation-equivariant graph neural network Transformer velocity field predictor. Input the linear interpolation station field representation and the corresponding time parameters into the velocity field predictor. Output the adjacent velocity, which characterizes the rate of change of adjacency relationship; the node feature velocity, which characterizes the rate of change of node features; and the edge feature velocity, which characterizes the rate of change of edge features. The velocity field predictor maintains equivariance for node permutations to ensure consistent inference results even under equipment number rearrangement. Construct a local velocity matching loss function to measure the deviation of the adjacent velocity, node feature velocity, and edge feature velocity from the changes in adjacency relationship, node features, and edge features from the source station field map to the target station field map after structural alignment, respectively. By minimizing the local velocity matching loss function, the adjacent velocity, node feature velocity, and edge feature velocity are made consistent with their corresponding changes, thereby constraining the velocity field predictor to output an instantaneous transmission direction that conforms to the rectified flow matching trajectory at any intermediate time.

[0061] Local velocity matching loss function formula:

[0062] ;

[0063] in, This represents the local velocity matching loss function, used to measure the deviation between the instantaneous rate of change of the velocity field predictor output at a certain intermediate moment and the ideal constant velocity displacement. Indicates the adjacency speed. Represents the node characteristic velocity. Indicates the edge feature velocity; Indicates the change in adjacent elements. This represents the amount of change in node characteristics. Indicates the change in edge characteristics; Represents the square of the Frobenius norm; , This represents the weighting coefficient, used to balance the relative contributions of the three types of errors to the total loss;

[0064] Step S8: Introduce the endpoint consistency supervision term and the interlocking feasibility regularization constraint term, and combine them with the local velocity matching loss function for weighted fusion to form the total loss function for training the velocity field predictor; Endpoint consistency supervision term: Based on the linear interpolation station representation, the endpoint predicted station representation at the termination time is derived by single-step rectification integration, and the deviation between the endpoint predicted station representation and the observed endpoint station representation is calculated. Based on the deviation, the endpoint consistency loss is constructed to suppress the global structure drift caused by the accumulation of local errors under the condition of finite-step numerical integration; Interlocking feasibility regularization constraint term: Apply interlocking feasibility regularization constraints to the endpoint predicted station representation. The interlocking feasibility regularization constraints include the constraint of prohibiting conversion under occupancy conditions, the constraint of prohibiting open signals when the turnout is not locked, the constraint of limiting concurrent operation in the same interlocking area, and the constraint of blocking priority. These constraints are used to suppress the combination of topology and equipment state that violates the interlocking rules at the endpoint generation level, thereby improving the rule consistency and executability of the station map reconstruction results;

[0065] Endpoint consistency loss formula:

[0066] ;

[0067] in, This represents the endpoint consistency loss. The adjacency representation of endpoint prediction. The node feature representation representing endpoint prediction. Edge feature representation representing endpoint prediction; This represents the target adjacency representation after structural alignment. This represents the characteristics of the target node after structural alignment. This represents the target edge features after structural alignment;

[0068] Interlocking feasibility regularities include:

[0069] Occupy prohibits conversion constraint:

[0070] ;

[0071] in, This represents the regularized loss value for the occupancy prohibition conversion constraint, used to quantify the degree of violation of "turnout conversion / related operation still occurring under section occupancy conditions" in the endpoint prediction station status; This represents the set of objects to be detected, generally referring to a "segment set". This indicates the section currently being inspected; Indicates and area The relevant "set of prohibited conversion rule instances"; Indicates a section A specific binding clause under the clause; This function represents the degree of violation of prohibited conversions, and outputs a non-negative value. The larger the value, the more severe the violation.

[0072] Signals cannot be opened if the turnout is not locked.

[0073] ;

[0074] in, This represents the regularization loss value for the constraint that the turnout is not locked but the related signal is still open / allowed to be open, and is used to quantify the degree of violation of the "turnout is not locked but the related signal is still open / allowed to be open" rule in the terminal prediction station status. Represents the set of turnouts. Indicates a specific turnout; This function represents the violation degree of the lock-open linkage, and outputs a non-negative value, where a larger value indicates a more severe violation.

[0075] Concurrency limits within the same interlocking zone:

[0076] ;

[0077] in, The regularized loss value represents the restriction on concurrent operations within the same interlocking zone, used to quantify the degree of violation of the rule that "an unacceptable combination of concurrent operations occurs within the same interlocking zone" in the endpoint prediction station status. Represents the set of interlocking zones. Indicates a specific interlocking area; This function represents the violation rate of concurrency limits, outputting a non-negative value; a larger value indicates a violation within the interlocked area. The more severe the internal concurrent conflicts;

[0078] Blocking priority constraints:

[0079] ;

[0080] in, The regularized loss value representing the blockade priority constraint is used to quantify the degree of violation in the endpoint prediction station status of "in the presence of a blockade requirement or under a blockade status, control permissions, access openings or equipment operations that conflict with the blockade are still generated"; Represents the set of locked objects. This refers to a specific target being blocked. This function represents the violation severity of the blocking priority rule, outputting a non-negative value; a larger value indicates a more severe violation.

[0081] Step S9: Iteratively optimize the velocity field predictor based on the total loss function to obtain the station structure-preserving velocity field; after obtaining the station structure-preserving velocity field, use the source station map after structure alignment as the initial input, and perform finite-step rectification integral update according to the preset time step to obtain the endpoint station representation at the termination time; perform discretization projection and interlocking rule correction on the endpoint station representation to make the adjacency relationship, node state and edge feature satisfy the equipment type value range and interlocking constraints to obtain the vehicle base station map.

[0082] Example 3 differs from Example 2 in that: in step S5, the fusion Gromov-Wasserstein optimal transmission mechanism is invoked, and a fusion cost function is constructed based on the intra-graph structural cost matrix and the cross-graph feature cost matrix to solve for the structurally preserving coupling relationship between the noisy prior station map and the observation station map; step S5 differs from Example 2, specifically in this example: the node alignment mechanism based on graph matching is invoked, and a comprehensive matching cost matrix is ​​constructed based on the intra-graph structural cost matrix and the cross-graph feature cost matrix to perform the optimal matching function on the node correspondence between the noisy prior station map and the observation station map. The optimal matching solution involves: under the condition of uniform weight constraint on the nodes of the two graphs, the Hungarian algorithm is used to perform global minimum cost matching calculation on the comprehensive matching cost matrix to obtain the optimal matching relationship between the nodes of the two graphs; in the process of constructing the comprehensive matching cost matrix, the intra-graph structural cost and cross-graph feature cost are weighted and fused by setting a trade-off coefficient, so that the node matching result takes into account both the similarity of node attributes and the consistency of station topology; based on the obtained node matching relationship, a node correspondence relationship is generated, and node rearrangement and alignment processing is performed on the noisy prior station map and the observation station map accordingly to obtain structurally aligned paired station map pairs.

[0083] Example 4 differs from Example 2 in that: in step S8, an endpoint consistency supervision term and an interlocking feasibility regularization constraint term are introduced and weightedly fused with the local velocity matching loss function to form the total loss function for training the velocity field predictor; step S8 differs from Example 2 in that step S8 specifically involves: introducing a node attribute reconstruction loss term and an L2 parameter regularization term, and weightedly fused with the local velocity matching loss function to form the total loss function; wherein, the node attribute reconstruction loss term is used to constrain the difference between the predicted node attributes and the observed node attributes, and the L2 parameter regularization term is used to suppress excessively large model parameters and improve training stability, and the velocity field predictor is trained by minimizing the total loss function.

[0084] Example 5, based on Example 2, describes the edge relationship set of the vehicle base station map, which consists of topological connection edges, control dependency edges, interlocking constraint edges, and fault propagation edges. Specifically: topological connection edges represent the spatial connectivity between turnouts and sections, and between sections and signals; control dependency edges represent the influence of equipment control commands on equipment representation quantities and interlocking zone states; interlocking constraint edges represent the constraint relationships between turnout positions, section occupancy, and signal opening conditions; and fault propagation edges represent the influence of power supply anomalies, communication anomalies, and control cabinet anomalies on adjacent equipment and equipment within the same interlocking zone.

[0085] Example 6, based on Example 5, includes a field signal acquisition and protocol adaptation unit comprising a DI / DO acquisition board, an AI acquisition board, a serial communication module, and an Ethernet communication module. It supports the acquisition of turnout indication, locking indication, switching indication, signal filament / lamp position indication, track section occupancy, axle counting status, control power status, and access control / protection status. It also supports protocol adaptation for field equipment control interfaces, including relay acquisition, dry contact, RS485 / RS422, and Ethernet industrial protocols.

[0086] The present invention and its embodiments have been described above. This description is not restrictive. The accompanying drawings are only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this description and designs a similar structure and embodiment without departing from the spirit of the present invention, such design should fall within the protection scope of the present invention.

Claims

1. A signal takeover control system for emergency conditions in vehicle depots, characterized in that, The system includes: The emergency judgment module constructs multi-source status data; performs verification, discrimination and comparison on the multi-source status data to form an emergency condition judgment result; makes a judgment based on the emergency condition judgment result, generates a takeover request signal, and switches the system status from monitoring state to takeover ready state. Upon receiving the takeover request signal, the isolation switching module performs electrical isolation and logic interlocking operations, switching the system state from the takeover preparation state to the takeover operation state. After entering the takeover operation state, the station map construction module performs full refresh acquisition and structured mapping; it introduces a structure-preserving rectifier flow matching graph generation method to perform graph generation processing, and sequentially completes structure migration and endpoint generation through the graphette prior generation mechanism, the fusion of the Gromov-Wasserstein structure optimal transmission mechanism and continuous-time rectifier flow matching reconstruction constraints, to obtain the vehicle base station map; The graph reasoning boundary module completes the construction of the vehicle base station map, performs multi-round message passing calculations on the vehicle base station map, and performs joint reasoning at the graph level and subgraph level to generate station map reasoning results. The control permit generation module generates a manual control permit matrix and step constraints based on the station map reasoning results, and forms a control script; The command execution closed-loop module receives the control script and then executes the control operation.

2. The signal takeover control system for emergency conditions in a vehicle depot according to claim 1, characterized in that: The station map construction module is as follows: after entering the takeover operation state, a full refresh acquisition is performed to obtain refreshed multi-source status data; based on the refreshed multi-source status data, consistency verification and anomaly cleaning are performed, and structured mapping is carried out. A structure-preserving rectified flow matching graph generation method is introduced, and a noisy prior station map is constructed through the graphette prior generation mechanism. By integrating the Gromov-Wasserstein structure optimal transmission mechanism, structural migration and endpoint generation are completed under the constraints of structural coupling and node alignment and continuous-time rectified flow matching reconstruction, thus obtaining the vehicle base station map.

3. A signal takeover control system for emergency conditions in a vehicle depot according to claim 1, characterized in that: The station map reasoning results include risk assessment results and control boundaries; wherein, the risk assessment results include risk scores and risk source explanation information, and the control boundaries include a set of controllable equipment, a set of prohibited equipment, a suggested operation sequence, and a set of key interlocking constraints.

4. A signal takeover control system for emergency conditions in a vehicle depot according to claim 1, characterized in that: The process of obtaining the vehicle depot site map specifically includes: Step E1: Construct an observation station map based on refreshed multi-source state data, and generate a corresponding noise prior station map; Step E2: Perform structural alignment between the noise prior station map and the observation station map, and construct the rectified flow matching trajectory; Step E3: Based on the rectified flow matching trajectory, generate the vehicle base station map through rectified integration.

5. A signal takeover control system for emergency conditions in a vehicle depot according to claim 4, characterized in that: Step E1, specifically: Each piece of status data that is refreshed from multiple sources is appended with a collection timestamp, a data source identifier, and a quality flag to form a station status observation set. Consistency verification and anomaly cleaning are performed on the station status observation set, a structured mapping index is established, and a set of node feature vectors and a set of edge relationships are generated to obtain the observation station map; The graphette prior generation mechanism is invoked to perform sampling, resulting in a noise prior station map.

6. A signal takeover control system for emergency conditions in a vehicle depot according to claim 5, characterized in that: Step E2, specifically: The noise prior station map and the observation station map are encoded separately, and the node-level structure embedding representations of the two are extracted. Based on the node-level structure embedding representations, the in-graph structure cost matrix of the noise prior station map and the observation station map are calculated respectively. Calculate the cross-map feature cost matrix based on the node features of the noise prior station map and the node observation features of the observation station map; By invoking the optimal transport mechanism of the fusion Gromov-Wasserstein structure, a fusion cost function is constructed based on the intra-graph structural cost matrix and the cross-graph feature cost matrix. The structural preservation coupling relationship between the noisy prior station map and the observation station map is solved to obtain structurally aligned paired station map pairs. The structurally aligned and paired station site map includes the source station site map and the target station site map after structural alignment; A continuous-time rectified flow matching trajectory is constructed on the structurally aligned and paired station map pairs to generate a linear interpolated station representation.

7. A signal takeover control system for emergency conditions in a vehicle depot according to claim 6, characterized in that: Step E3, specifically: Construct a velocity field predictor, input the linear interpolated station field representation and the corresponding time parameters into the velocity field predictor, and output the adjacent velocity, node feature velocity and edge feature velocity; Construct a local velocity matching loss function; by performing a minimization optimization on the local velocity matching loss function, constrain the velocity field predictor to output the instantaneous transmission direction that conforms to the rectified flow matching trajectory at any intermediate time. An endpoint consistency supervision term and an interlocking feasibility regularization constraint term are introduced, and combined with the local velocity matching loss function for weighted fusion to form the total loss function; Based on the total loss function, the velocity field predictor is iteratively optimized to obtain the station structure-preserving velocity field. The source station map after structural alignment is used as the initial input, and a finite-step rectification integral update is performed according to the preset time step to obtain the endpoint station representation at the termination time. Discretize the terminal station representation and correct it using interlocking rules to obtain the vehicle base station map.

8. A signal takeover control system for emergency conditions in a vehicle depot according to claim 5, characterized in that: The set of edge relationships consists of topological connection edges, control dependency edges, interlocking constraint edges, and fault propagation edges.

9. A signal takeover control system for emergency conditions in a vehicle depot according to claim 5, characterized in that: The graphette prior in the graphette prior generation mechanism consists of a graph function, a sparse control sequence, and a graph editing function. The graph function is used to generate an initial topological skeleton that matches the scale of the vehicle base station. The sparse control sequence is used to constrain the connectivity density of the initial topological skeleton to match the typical density of the vehicle base tracks and interlocking areas. The graph editing function is used to perform structural editing on the initial topological skeleton to inject repeating structural motifs into the station.

10. A signal takeover control system for emergency conditions in a vehicle depot according to claim 7, characterized in that: The interlocking feasibility regularity constraints include constraints prohibiting switching under occupancy conditions, constraints prohibiting open signals when the turnout is not locked, constraints restricting concurrent operations in the same interlocking zone, and constraints prioritizing blocking.

Images