Method and system for constructing a port railway fault knowledge base based on signal microcomputer monitoring

Abstract

The present application relates to the technical field of port railway signal monitoring, and particularly relates to a port railway fault knowledge base construction method and system based on signal microcomputer monitoring. The method collects port railway signal system operation state data in real time through a signal microcomputer monitoring device, cooperatively analyzes and identifies abnormal patterns in time domain and logic domain, and extracts a fault state characteristic vector; matches the characteristic vector with a fault mode library, determines a fault type identifier and a fault propagation path, and obtains an associated fault instance; extracts a cause-effect rule between a fault phenomenon and a fault cause from the associated fault instance in combination with a device topology relationship and a signal interlocking logic, and generates a structured fault knowledge rule; writes the fault knowledge rule into a fault knowledge base, and adaptively updates the characteristic vector accordingly, so as to improve the efficiency and accuracy of port railway fault diagnosis and knowledge base construction.

Description

Technical Field

[0001] This invention relates to the field of port railway signal monitoring technology, and in particular to a method and system for constructing a port railway fault knowledge base based on signal microcomputer monitoring. Background Technology

[0002] Port railway signaling systems handle operations such as shunting, loading / unloading, and interlocking in the port area. Equipment operates frequently, and interlocking relationships are complex, requiring high real-time performance and accuracy in fault diagnosis. Microcomputer-based signal monitoring devices are widely used to collect voltage, current, power, switching parameters, and status information from equipment such as turnouts, signals, track circuits, and power supply panels, providing a data foundation for operation monitoring and fault analysis. Current practices largely rely on fixed threshold alarms, manual review of curves and logs, and the experience-based judgment of maintenance personnel. Fault diagnosis systems primarily use rule-based reasoning, case-based reasoning, or simple statistical models.

[0003] The construction of a routine fault knowledge base typically involves professionals compiling equipment manuals, maintenance procedures, historical fault records, and field experience to create entries corresponding to fault phenomena and causes. These entries are then entered into a database for querying and reasoning. Some systems associate monitoring data with alarm information, triggering diagnostic conclusions through preset logic or threshold conditions, and revising the knowledge content through regular maintenance and manual feedback.

[0004] This approach has significant shortcomings. Monitoring data is often processed from a single dimension, and the correlation between the temporal changes in operational data and the equipment's operational logic and signal interlocking sequence is not fully explored. Anomaly pattern identification is easily affected by fluctuations in operating conditions and occasional interference, making false alarms and missed alarms difficult to avoid. Fault knowledge mainly relies on manual summarization and static input, resulting in long update cycles and limited coverage due to experience. When faced with frequent operations of port railway equipment, changes in work organization, and hidden faults, the knowledge base is difficult to expand and correct quickly, and the interpretability of diagnostic conclusions and the ability to trace the propagation path are also relatively limited. Summary of the Invention

[0005] This invention provides a method and system for constructing a port railway fault knowledge base based on signal microcomputer monitoring, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides a method for constructing a port railway fault knowledge base based on signal microcomputer monitoring, comprising:

[0007] The operation status data of the port railway signaling system is collected in real time by a signal microcomputer monitoring device. The operation status data is analyzed in both the time domain and the logical domain. Abnormal patterns are identified based on the action logic constraints and temporal causal relationships of the signaling equipment, and feature vectors representing fault states are extracted.

[0008] Based on the timing deviation feature and logical consistency deviation feature in the feature vector, the current fault state is matched with the pre-stored fault mode library to determine the fault type identifier and fault propagation path, and obtain the associated fault instance.

[0009] Based on the fault propagation path corresponding to the associated fault instance, and combined with the topological relationship between devices and signal interlocking logic, the causal rules between fault phenomena and fault causes are extracted to generate structured fault knowledge rules.

[0010] The fault knowledge rules are written into the fault knowledge base, and the feature vector representing the fault state is adaptively updated based on the fault knowledge rules.

[0011] The operational status data of the port railway signaling system is collected in real time by a signal microcomputer monitoring device. This operational status data undergoes collaborative analysis in both the time and logical domains. Based on the action logic constraints and temporal causal relationships of the signaling equipment, abnormal patterns are identified, and feature vectors characterizing fault states are extracted, including:

[0012] The operating status data is subjected to time-series analysis in the time domain to extract the time-series relationship between the action time and action state of the signal device, and a time-domain feature set is obtained.

[0013] The operational status data is logically verified in the logical domain dimension. Based on the action logic constraints and interlocking relationship data of the signal equipment, the action status of the signal equipment is verified to see if it meets the preset interlocking logic rules, and a logical domain feature set is obtained.

[0014] The time-domain feature set and the logic-domain feature set are fused to identify consistency deviations between timing relationships and interlocking logic rules, and abnormal patterns are marked.

[0015] Based on the aforementioned abnormal patterns, feature vectors containing timing deviation features and logical consistency deviation features are extracted. The timing deviation features characterize the degree of deviation between the actual action time and the expected action time of the signaling device, and the logical consistency deviation features characterize the degree of conformity between the action state of the signaling device and the interlocking logic rules.

[0016] The time-domain feature set and the logic-domain feature set are fused to identify consistency deviations between timing relationships and interlocking logic rules, and abnormal patterns are marked, including:

[0017] Based on the temporal dependency relationship between the action time and action state of the signal devices in the time-domain feature set, a temporal relationship mapping table is constructed.

[0018] Based on the logical constraint relationship between the action state of the signal devices in the logical domain feature set and the interlocking logical rules, a logical rule constraint table is constructed.

[0019] For the same action event of the same signal device, extract the actual action time and the expected action time recorded in the time sequence mapping table, calculate the time difference between the actual action time and the expected action time, and generate a time sequence deviation feature;

[0020] For the same action event of the same signal device, extract the actual action state recorded in the logical rule constraint table and the expected action state required by the interlocking logical rules, calculate the logical conformity between the actual action state and the expected action state, and generate logical consistency deviation features;

[0021] Based on the fusion analysis of the timing deviation feature and the logical consistency deviation feature, when the timing deviation feature indicates that the signal device has completed the action and the logical consistency deviation feature indicates that the action condition has not been met, the same action event is marked as an abnormal mode.

[0022] Based on the timing deviation feature and logical consistency deviation feature in the feature vector, the current fault state is matched with a pre-stored fault mode library to determine the fault type identifier and fault propagation path, thereby obtaining associated fault instances, including:

[0023] Multiple historical fault modes are obtained from the pre-stored fault mode library. The temporal similarity between the timing deviation feature and the historical timing deviation feature of each historical fault mode, and the logical similarity between the logical consistency deviation feature and the historical logical consistency deviation feature of each historical fault mode are calculated respectively.

[0024] The temporal similarity and the logical similarity are weighted and fused to generate a comprehensive similarity between the current fault state and each historical fault mode;

[0025] Based on the comprehensive similarity, target historical fault modes that meet the preset matching conditions are selected from the multiple historical fault modes. The fault type identifier and fault propagation path corresponding to the target historical fault mode are extracted. The fault propagation path represents the propagation order and propagation association of the fault among multiple signal devices.

[0026] The associated fault instance is determined based on the target historical fault mode, the fault type identifier, and the fault propagation path.

[0027] Based on the fault propagation path corresponding to the associated fault instance, and combined with the topological relationships and signal interlocking logic between devices, causal rules between fault phenomena and fault causes are extracted to generate structured fault knowledge rules, including:

[0028] Obtain the topological relationships and signal interlocking logic between devices in the port railway signaling system;

[0029] Based on the propagation order and propagation correlation recorded in the fault propagation path, and combined with the physical connection relationship and spatial location relationship in the topology, the propagation link of the fault between multiple signal devices is identified;

[0030] Extract fault cause features from the starting device of the propagation link, and extract fault phenomenon features from the terminating device of the propagation link;

[0031] Based on the action constraint relationship and state dependency relationship in the signal interlocking logic, the causal relationship between the fault cause characteristics and the fault phenomenon characteristics is analyzed, and the causal relationship analysis results are obtained.

[0032] Based on the causal correlation analysis results, causal rules between the fault cause characteristics and the fault phenomenon characteristics are established;

[0033] The causal rules are converted into a structured expression containing a condition part and a conclusion part to obtain a structured fault knowledge rule. The condition part records the fault cause characteristics and the corresponding equipment state changes, and the conclusion part records the fault phenomenon characteristics and the corresponding equipment state changes.

[0034] Based on the action constraint relationship and state dependency relationship in the signal interlocking logic, the causal correlation between the fault cause characteristics and the fault phenomenon characteristics is analyzed, and the causal correlation analysis results are obtained, including:

[0035] Obtain the starting device identifier and starting device status change corresponding to the fault cause characteristics, and the ending device identifier and ending device status change corresponding to the fault phenomenon characteristics;

[0036] Based on the action constraint relationship, starting from the starting device identifier, trace the action triggering condition corresponding to the change in the state of the starting device, identify the set of downstream devices constrained by the action triggering condition, determine whether the terminating device identifier is located in the set of downstream devices, and generate a constraint association metric value that characterizes the action constraint transmission path.

[0037] Based on the state dependency relationship, starting from the state change of the starting device, the transmission process of the state change among multiple signal devices is tracked, the influence intensity of the state change of the starting device on the state change of the terminating device is calculated, and a dependency correlation metric value characterizing the state dependency transmission effect is generated.

[0038] The constraint correlation metric and the dependency correlation metric are fused together. When both the constraint correlation metric and the dependency correlation metric satisfy the preset correlation judgment condition, it is determined that there is a causal correlation between the fault cause feature and the fault phenomenon feature, and a causal correlation analysis result containing the constraint correlation metric and the dependency correlation metric is generated.

[0039] The fault knowledge rules are written into a fault knowledge base, and the feature vector representing the fault state is adaptively updated based on the fault knowledge rules, including:

[0040] The fault knowledge rules are written into the fault knowledge base according to the preset knowledge organization structure. The target feature items corresponding to the fault cause features are extracted from the condition part of the fault knowledge rules, and the target feature items corresponding to the fault phenomenon features are extracted from the conclusion part of the fault knowledge rules.

[0041] Based on the matching analysis between the target feature terms and the feature dimensions in the feature vector representing the fault state, the target feature dimension that needs to be adjusted in the feature vector representing the fault state is identified.

[0042] For the target feature dimension, the feature dimension adjustment amount is calculated based on the equipment state changes and causal relationships recorded in the fault knowledge rules.

[0043] The target feature dimension in the feature vector representing the fault state is updated based on the feature dimension adjustment amount to generate an adaptively updated feature vector.

[0044] A second aspect of this invention provides a port railway fault knowledge base construction system based on signal microcomputer monitoring, comprising:

[0045] The data acquisition and analysis unit is used to acquire the operating status data of the port railway signaling system in real time through the signal microcomputer monitoring device, perform joint analysis of the operating status data in the time domain and logic domain, identify abnormal patterns based on the action logic constraints and temporal causal relationships of the signaling equipment, and extract feature vectors that characterize the fault state.

[0046] The fault matching unit is used to perform similarity matching between the current fault state and the pre-stored fault mode library based on the timing deviation feature and logical consistency deviation feature in the feature vector, determine the fault type identifier and fault propagation path, and obtain the associated fault instance.

[0047] The rule generation unit is used to extract causal rules between fault phenomena and fault causes based on the fault propagation path corresponding to the associated fault instance, combined with the topological relationship between devices and signal interlocking logic, and generate structured fault knowledge rules.

[0048] The knowledge update unit is used to write the fault knowledge rules into the fault knowledge base and adaptively update the feature vector representing the fault state based on the fault knowledge rules.

[0049] A third aspect of the present invention provides an electronic device, comprising:

[0050] processor;

[0051] Memory used to store processor-executable instructions;

[0052] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0053] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0054] This invention utilizes signal microcomputer monitoring data to perform time-domain and logic-domain collaborative analysis of port railway signaling systems. It can identify abnormal patterns from equipment action logic constraints and temporal causal relationships, extract feature vectors that accurately reflect fault states, overcome the problems of missed and false alarms caused by single data sources and single threshold judgments, and improve the timeliness of anomaly detection and the accuracy of fault diagnosis.

[0055] This invention performs similarity matching based on timing deviation characteristics and logical consistency deviation characteristics, which can quickly associate the current fault state with the fault mode library, determine the fault type identifier and fault propagation path, and form associated fault instances. This facilitates the location of the root cause fault in complex interlocking relationships, reduces the scope and time of manual investigation, and improves emergency response efficiency.

[0056] This invention extracts causal rules based on fault propagation paths and combined with equipment topology and signal interlocking logic. It can transform scattered fault phenomena and underlying causes into structured fault knowledge rules, enabling port railway signal fault experience to be expressed in a standardized manner and continuously accumulated, thereby enhancing the usability and interpretability of the knowledge base.

[0057] This invention incorporates fault knowledge rules into a fault knowledge base and adaptively updates feature vectors based on these rules, forming a closed loop of diagnosis, learning, and optimization. This allows feature extraction and the knowledge base to evolve with changes in equipment status and operating environment, reducing reliance on human experience, improving fault early warning and maintenance decision-making, and ensuring safe and efficient port railway transportation. Attached Figure Description

[0058] Figure 1 This is a flowchart illustrating the method for constructing a port railway fault knowledge base based on signal microcomputer monitoring, according to an embodiment of the present invention.

[0059] Figure 2 This is a schematic diagram of the process for obtaining associated fault instances according to an embodiment of the present invention;

[0060] Figure 3 This is a schematic diagram of the topological relationships in an embodiment of the present invention;

[0061] Figure 4 This is a schematic diagram of the fault propagation link according to an embodiment of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0064] Figure 1 This is a flowchart illustrating the method for constructing a port railway fault knowledge base based on signal microcomputer monitoring, according to an embodiment of the present invention. Figure 1 As shown, the method for constructing a port railway fault knowledge base based on signal microcomputer monitoring includes:

[0065] The operation status data of the port railway signaling system is collected in real time by a signal microcomputer monitoring device. The operation status data is analyzed in both the time domain and the logical domain. Abnormal patterns are identified based on the action logic constraints and temporal causal relationships of the signaling equipment, and feature vectors representing fault states are extracted.

[0066] Based on the timing deviation feature and logical consistency deviation feature in the feature vector, the current fault state is matched with the pre-stored fault mode library to determine the fault type identifier and fault propagation path, and obtain the associated fault instance.

[0067] Based on the fault propagation path corresponding to the associated fault instance, and combined with the topological relationship between devices and signal interlocking logic, the causal rules between fault phenomena and fault causes are extracted to generate structured fault knowledge rules.

[0068] The fault knowledge rules are written into the fault knowledge base, and the feature vector representing the fault state is adaptively updated based on the fault knowledge rules.

[0069] The operational status data of the port railway signaling system is collected in real time by a signal microcomputer monitoring device. This operational status data undergoes collaborative analysis in both the time and logical domains. Based on the action logic constraints and temporal causal relationships of the signaling equipment, abnormal patterns are identified, and feature vectors characterizing fault states are extracted, including:

[0070] The operating status data is subjected to time-series analysis in the time domain to extract the time-series relationship between the action time and action state of the signal device, and a time-domain feature set is obtained.

[0071] The operational status data is logically verified in the logical domain dimension. Based on the action logic constraints and interlocking relationship data of the signal equipment, the action status of the signal equipment is verified to see if it meets the preset interlocking logic rules, and a logical domain feature set is obtained.

[0072] The time-domain feature set and the logic-domain feature set are fused to identify consistency deviations between timing relationships and interlocking logic rules, and abnormal patterns are marked.

[0073] Based on the aforementioned abnormal patterns, feature vectors containing timing deviation features and logical consistency deviation features are extracted. The timing deviation features characterize the degree of deviation between the actual action time and the expected action time of the signaling device, and the logical consistency deviation features characterize the degree of conformity between the action state of the signaling device and the interlocking logic rules.

[0074] Port railway signaling systems continuously generate a large amount of status data during operation, including track circuit occupancy status, turnout position status, signal display status, and relay action records. The signal microcomputer monitoring device collects these status quantities in real time at a millisecond-level sampling frequency, forming a raw operational status data stream with precise timestamps. Each data record contains three basic fields: equipment number, status value, and acquisition time, forming the basic data unit for subsequent analysis and processing. Due to the high operational density and frequent shunting operations of port railways, status data exhibits highly dense jump sequences within a short period. Therefore, simultaneous analysis in both the time and logical domains is necessary to accurately capture potential anomaly patterns.

[0075] In the time domain, the collected operational status data is reorganized according to equipment number and time axis to construct the action timing sequence of each signaling device. Taking a turnout as an example, its normal transition process should sequentially go through three states: "start command issued," "transition in progress," and "indicator relay activated." The duration of each state should fall within the time window given by the design specifications. For a signal, there is usually a fixed response delay interval from the activation of the drive relay to the lighting of the signal filament. For each device, all state transition events occurring within the observation window are extracted, the start and end times of each transition are recorded, and the time interval between adjacent actions is calculated to form a time domain feature set. The core elements of the time domain feature set include: the device action time sequence, the action duration sequence, and the timing deviation value between the action and the reference template. If the completion time of an action is significantly earlier or later than the expected reference time, the deviation is marked as a candidate abnormal feature and included in the subsequent fusion analysis.

[0076] In the logical domain dimension, the core constraints of the signal interlocking system are embodied in a set of strict logical rules: signals are only allowed to open when conditions such as route establishment, turnout locking, and section vacancy are simultaneously met. These rules originate from the interlocking table and are pre-stored in the verification rule base in a condition-conclusion format. For each collected action status record, its current state is verified against the corresponding interlocking preconditions based on the role of the equipment in the interlocking table. If a signal is open before the route is locked, or a turnout changes position within a train-occupied section, the record is deemed to violate the preset interlocking logical rules, generating a logical domain anomaly flag. The results of the logical verification are output in a structured form, with each record accompanied by a compliance flag and a specific number indicating the rule violation, collectively forming a logical domain feature set. This feature set not only records whether the status value itself is compliant but also the interlocking context in which the violation occurred, providing correlation information for subsequent causal analysis.

[0077] When fusing temporal and logical feature sets, the device ID and timestamp are used as the association key to align and merge the temporal and logical features of the same device within the same time period. The fusion process focuses on two types of consistency deviations: first, the co-occurrence relationship between timing deviations and logical violations, i.e., whether a device experiences both timing anomalies and logical rule violations simultaneously; if both occur, it indicates that the device is in an active fault state; second, the situation where timing is normal but logical violations occur, which usually points to interlocking configuration errors or abnormal state transmission from upstream devices. The fusion algorithm calculates the number of timing deviation events and logical violation events for each device within the observation window and calculates the overlap ratio of the two types of events as a quantitative basis for judging the severity of the anomaly. For devices with an overlap ratio exceeding a set threshold, they are prioritized as high-confidence anomaly nodes and proceed to the subsequent feature vector extraction process.

[0078] After completing the fusion analysis, feature vectors are extracted for each device marked as having an abnormal mode. These feature vectors consist of two core types of features: timing deviation features and logical consistency deviation features. Timing deviation features quantitatively characterize the degree of deviation between the actual action time of the device and its corresponding expected action time. Specifically, they include four components: maximum positive deviation, maximum negative deviation, mean deviation, and standard deviation of deviation, reflecting the statistical deterioration of the device's response speed. Suppose a device experiences a total of [number missing] occurrences within the observation window. The next action, the first The actual completion time of this action is The corresponding expected completion time is Then the timing deviation of this action is The mean deviation is The standard deviation of the deviation is The logical consistency deviation feature quantifies the degree of conformity between the device's operational state and the interlocking logic rules. It uses the ratio of the number of violations to the total number of actions as the primary feature component, while also adding the rule category code involved in the violation to identify the fault type in subsequent matching stages. For devices exhibiting both timing deviations and logical violations, the weight allocation of the two types of features in their feature vector is dynamically adjusted according to the severity of the violation, ensuring that high-risk interlocking violations receive priority responses during the matching process.

[0079] In actual engineering deployments, the aforementioned temporal analysis and logical domain verification processes are executed in parallel in a pipeline manner. The fusion of the temporal and logical domain feature sets is triggered once after each sampling period to ensure the real-time performance and integrity of the feature vectors. For the common scenario of establishing multiple routes in short periods during port railway shunting operations, a sliding time window mechanism is also introduced. The window length is dynamically adjusted based on the average cycle of route establishment, avoiding feature loss due to an excessively short window or abnormal overload due to an excessively long window.

[0080] The time-domain feature set and the logic-domain feature set are fused to identify consistency deviations between timing relationships and interlocking logic rules, and abnormal patterns are marked, including:

[0081] Based on the temporal dependency relationship between the action time and action state of the signal devices in the time-domain feature set, a temporal relationship mapping table is constructed.

[0082] Based on the logical constraint relationship between the action state of the signal devices in the logical domain feature set and the interlocking logical rules, a logical rule constraint table is constructed.

[0083] For the same action event of the same signal device, extract the actual action time and the expected action time recorded in the time sequence mapping table, calculate the time difference between the actual action time and the expected action time, and generate a time sequence deviation feature;

[0084] For the same action event of the same signal device, extract the actual action state recorded in the logical rule constraint table and the expected action state required by the interlocking logical rules, calculate the logical conformity between the actual action state and the expected action state, and generate logical consistency deviation features;

[0085] Based on the fusion analysis of the timing deviation feature and the logical consistency deviation feature, when the timing deviation feature indicates that the signal device has completed the action and the logical consistency deviation feature indicates that the action condition has not been met, the same action event is marked as an abnormal mode.

[0086] In the actual operating environment of port railway signaling systems, the action events of signaling equipment carry both temporal and state attributes. However, existing single-dimensional analysis methods often only capture one type of deviation, and their ability to identify interlocking logic failures such as "equipment has acted but the conditions are not met" is clearly insufficient. To address this issue, we fuse the temporal and logical domain feature sets by constructing two structured mapping tables. Then, for the same action event, we extract temporal deviation features and logical consistency deviation features respectively. Finally, we use a two-dimensional fusion judgment to mark abnormal patterns.

[0087] The construction of the time-series mapping table uses the equipment action events recorded in the time-domain feature set as the original input. When the signal microcomputer monitoring device collects operating status data, it adds a high-precision timestamp to each equipment action, recording its actual occurrence time. Simultaneously, based on the shunting route arrangement logic, block section occupancy order, and the causal chain of turnout switching in the port railway signaling system, the expected time when each signaling device should complete its action under normal operating procedures can be pre-calculated. The actual time and expected time of the same action of the same equipment are stored as a key-value pair in the mapping table, with the equipment number, action type (such as turnout switching, signal opening, track circuit occupancy, etc.), and the associated route number added as index fields, forming a time-series mapping table that can be quickly retrieved. The granularity of this mapping table should be refined to the level of a single action event, rather than the equipment level, to ensure the accuracy of subsequent deviation calculations.

[0088] The construction of the logical rule constraint table is based on the logical domain feature set. The action logic of the port railway signal interlocking system is described by a set of strict Boolean constraint rules, such as "before the signal is opened, all switches in the relevant route must be in the specified position and locked" and "the relevant routes in this section are prohibited from being arranged when the track circuit is occupied". These interlocking logic rules are formalized, and the set of preconditions on which each type of action event depends, as well as the expected action state corresponding to the set of conditions, are extracted. The actual collected action states of each device in the logical domain feature set (such as switch positioning / reversal state, track circuit idle / occupied state, signal closed / open state, etc.) and the expected action states required by the interlocking rules are written into the logical rule constraint table. Similarly, the index structure is consistent with the timing relationship mapping table, using the device number, action type, and route number as indexes, to facilitate subsequent cross-table joint queries.

[0089] After the two mapping tables are established, for the same action event of the same signal device, the corresponding records are extracted from the two tables respectively, and the timing deviation characteristics are calculated. Let the actual completion time of a certain action event be... The corresponding expected completion time is Then the time difference of this action Defined as:

[0090] ;

[0091] when When, it indicates that the action is lagging behind expectations; when When, it indicates that the action is ahead of schedule; when Exceeding the preset threshold (This threshold is set according to equipment type and route complexity classification) when a significant timing deviation is considered to exist. Timing deviation characteristics include not only... The scalar value also includes its directionality (leading or lagging), providing directional semantics for subsequent fusion analysis.

[0092] The calculation of logical consistency deviation characteristics is relatively complex because action states are discrete enumerations rather than continuous quantities. For the same action event, the actual action state is read from the logical rule constraint table. The expected action state required by the interlocking logic rules Define logical compliance As follows: When and When a perfect match is achieved, ;when and When there is a mismatch, For complex interlocking rules involving multiple preconditions, assume the rule requires that the following conditions be met. There are several conditions, among which the actual conditions are met. If there are 1, then the logical conformity is:

[0093] ;

[0094] when At that time, it is considered that a logical consistency deviation exists, that is, there is a non-compliance between the actual action state and the interlocking logic requirements. The characteristics of logical consistency deviation are as follows: The value and the specific number of the condition not met are used together to express the error. The latter is used to locate the source of the bias when extracting causal rules in the future.

[0095] The fusion analysis phase combines the two types of features mentioned above, focusing on a specific combination of anomalies: timing deviation features indicate that the device has completed its action (i.e., (Recorded, action state has changed), but logical consistency deviation characteristics are shown. (That is, not all preconditions were met when the action was triggered). This combination reveals a dangerous interlocking logic failure mode, in which the equipment completes the action even though the interlocking conditions are not met. This should not happen during normal operation and points to deeper faults such as relay sticking, control circuit abnormalities, or software logic errors.

[0096] When the above combined conditions are simultaneously met, the action event is marked as an abnormal mode, and the following information is written into the marking record: device number and action type, actual action time. Time difference Logical compliance This includes a list of unmet interlocking preconditions. This tag record will serve as input for subsequent similarity matching and causal rule extraction, providing a concrete evidentiary basis for fault type identification and fault propagation path inference.

[0097] It is worth noting that not all and All such situations will be marked as abnormal modes. Simple timing deviations (interlocking conditions are fully met but there is a slight delay in the timing of the action) are considered performance degradation rather than logical failures and need to be handled according to different procedures. Similarly, However, if the equipment does not perform any action (the action is correctly prevented), it indicates that the interlocking logic is operating normally and should not be mistakenly marked as abnormal. The core of fusion judgment lies in the simultaneity of "action performed" and "condition not met." This constraint effectively ensures the accuracy of abnormal marking, reduces the false alarm rate, and ensures that no real interlocking logic failure events are missed.

[0098] Figure 2 This is a schematic diagram illustrating the process of obtaining associated fault instances according to an embodiment of the present invention. Figure 2 As shown, based on the timing deviation feature and logical consistency deviation feature in the feature vector, the current fault state is matched with a pre-stored fault mode library to determine the fault type identifier and fault propagation path, thus obtaining associated fault instances, including:

[0099] Multiple historical fault modes are obtained from the pre-stored fault mode library. The temporal similarity between the timing deviation feature and the historical timing deviation feature of each historical fault mode, and the logical similarity between the logical consistency deviation feature and the historical logical consistency deviation feature of each historical fault mode are calculated respectively.

[0100] The temporal similarity and the logical similarity are weighted and fused to generate a comprehensive similarity between the current fault state and each historical fault mode;

[0101] Based on the comprehensive similarity, target historical fault modes that meet the preset matching conditions are selected from the multiple historical fault modes. The fault type identifier and fault propagation path corresponding to the target historical fault mode are extracted. The fault propagation path represents the propagation order and propagation association of the fault among multiple signal devices.

[0102] The associated fault instance is determined based on the target historical fault mode, the fault type identifier, and the fault propagation path.

[0103] When retrieving multiple historical fault modes from a pre-stored fault mode library, the extracted feature vector needs to be decomposed into two types of sub-features: timing deviation features and logical consistency deviation features. Timing deviation features reflect the degree of deviation of equipment actions in the time dimension, while logical consistency deviation features characterize the degree of inconsistency between equipment actions and expected behavior at the interlocking logic level. Each record in the historical fault mode library also contains historical descriptive values ​​of these two types of features, stored in vector form. For the current fault state, the similarity is calculated with the corresponding sub-features of each historical fault mode in the library, which forms the basis for subsequent matching judgments.

[0104] The temporal similarity is calculated using the vector distance metric. Let the temporal deviation feature vector of the current fault state be... , No. The historical time-series deviation feature vector of each historical failure mode is: Then the temporal similarity Defined as:

[0105] ;

[0106] in This represents the Euclidean distance. The formula maps the distance to... The smaller the distance between intervals, the closer the similarity is to 1, indicating that the current time series deviation is highly consistent with the historical pattern.

[0107] The similarity calculation method for logical consistency deviation features differs. Logical consistency deviation features typically consist of two parts: a discrete conformity label and a continuous deviation amount. Let the current logical consistency deviation feature vector be... , No. The historical logical consistency deviation feature vector of each historical failure mode is: Logical similarity Cosine similarity is used as the metric.

[0108] ;

[0109] Cosine similarity is not sensitive to vector magnitude but focuses more on directional consistency. It can better reflect the structural similarity between two sets of logical deviation modes and has a certain robustness to cases where there are differences in the number of interlocking constraints in the fault modes.

[0110] After obtaining the temporal similarity respectively Logical similarity Then, the two are weighted and fused to generate a comprehensive similarity score. :

[0111] ;

[0112] in The fusion weights for temporal similarity Let be the fusion weight for logical similarity, and satisfy . , , The weight values ​​are determined based on the specific fault type distribution and historical experience of the port railway signaling system. For faults primarily characterized by timing anomalies (such as turnout operation timeouts), the weight can be appropriately increased. For faults primarily manifested as interlocking logic errors (such as route establishment failure), the threshold should be appropriately increased. This weighted fusion mechanism enables the overall similarity to balance the diagnostic contributions of the two types of features in different fault scenarios.

[0113] After the overall similarity calculation is completed, historical fault modes are filtered according to preset matching conditions. The preset matching conditions usually include two levels: first, setting a lower threshold for overall similarity. ,Require The candidate set is determined by two criteria: first, historical failure patterns are included; second, within the candidate set, patterns are sorted in descending order of comprehensive similarity, and the top few (e.g., the top 3 to 5) are selected as target historical failure patterns. If the candidate set is empty, the threshold is appropriately lowered or the search range is expanded before re-matching to avoid matching failures due to an overly strict threshold.

[0114] From the selected target historical fault modes, corresponding fault type identifiers and fault propagation paths are extracted. Fault type identifiers are pre-coded enumerated labels, such as "turnout malfunction," "track section occupancy anomaly," and "signal lighting failure," used to clarify the functional category of the current fault. The fault propagation path is represented by a directed graph structure, with nodes corresponding to signaling equipment (turnouts, track circuits, signals, computer interlocking units, etc.), and directed edges describing the direction of fault propagation and their relationships between devices. The propagation path not only records the originating device of the fault but also depicts the order in which the fault spreads downstream along the electrical connections and interlocking logic dependencies between devices. For example, the propagation path of a turnout malfunction fault is represented as: turnout control unit malfunction → turnout indication circuit open circuit → interlocking logic determines that the route condition is not met → relevant signal is prohibited from opening. This path clearly presents the process of the fault propagating step by step from the equipment layer to the logic layer.

[0115] In port railway scenarios, signaling equipment has dense interlocking dependencies, and the same physical device may participate in multiple interlocking routes simultaneously. Therefore, an initial fault point often generates multiple branches in the fault propagation path. The fault propagation path extracted from the target's historical fault modes retains this multi-branch structure and can fully describe the extent of fault spread in the topology.

[0116] Based on the target historical fault patterns, fault type identifiers, and fault propagation paths, associated fault instances are ultimately determined. An associated fault instance is a structured data object containing: historical fault pattern records matching the current fault state (including comprehensive similarity scores), a determined fault type identifier, a fault propagation path represented as a directed graph, and maintenance and handling reference information inherited from historical records. When multiple target historical fault patterns exist, the associated fault instance can contain multiple candidate records, sorted by comprehensive similarity for use in subsequent causal rule extraction stages. The matching result with the highest comprehensive similarity is used as the primary associated instance, and the others are used as candidate associated instances, ensuring sufficient reference data in the subsequent fault knowledge rule generation stage.

[0117] In actual deployments, the fault mode library expands continuously with the accumulation of new fault instances, so the similarity matching process needs to consider retrieval efficiency. For scenarios with a large library, historical fault modes can be pre-indexed by grouping them according to fault type identifiers. Before calculating the comprehensive similarity, a coarse-grained screening can be used to narrow down the candidate range, and then a precise similarity calculation can be performed within the candidate range, thereby reducing computational overhead while ensuring matching accuracy.

[0118] Based on the fault propagation path corresponding to the associated fault instance, and combined with the topological relationships and signal interlocking logic between devices, causal rules between fault phenomena and fault causes are extracted to generate structured fault knowledge rules, including:

[0119] Obtain the topological relationships and signal interlocking logic between devices in the port railway signaling system;

[0120] Based on the propagation order and propagation correlation recorded in the fault propagation path, and combined with the physical connection relationship and spatial location relationship in the topology, the propagation link of the fault between multiple signal devices is identified;

[0121] Extract fault cause features from the starting device of the propagation link, and extract fault phenomenon features from the terminating device of the propagation link;

[0122] Based on the action constraint relationship and state dependency relationship in the signal interlocking logic, the causal relationship between the fault cause characteristics and the fault phenomenon characteristics is analyzed, and the causal relationship analysis results are obtained.

[0123] Based on the causal correlation analysis results, causal rules between the fault cause characteristics and the fault phenomenon characteristics are established;

[0124] The causal rules are converted into a structured expression containing a condition part and a conclusion part to obtain a structured fault knowledge rule. The condition part records the fault cause characteristics and the corresponding equipment state changes, and the conclusion part records the fault phenomenon characteristics and the corresponding equipment state changes.

[0125] Obtaining the topological relationships and signal interlocking logic among devices in a port railway signaling system is a fundamental preparatory step for extracting causal rules. Topological relationships describe the physical connection methods and spatial distribution of signaling equipment, such as the upstream and downstream connections between switches, track sections, and signals, as well as the positional order of equipment within a route. Signal interlocking logic is stored in the form of interlocking tables or a logic rule base, recording the state constraints that each device must satisfy under each route condition. For example, a switch must be in a specified position before an open signal, and an empty track section is a prerequisite for route establishment. Only after these two types of information are structured and stored can subsequent link identification and causal analysis have a complete topological background and logical basis.

[0126] Figure 3 This is a schematic diagram illustrating the topological relationships in an embodiment of the present invention. For example... Figure 3 As shown in the diagram, from left to right, the following are arranged: X1 entry signal, 1G track section, P1 turnout (including DJ1 contact check relay), 2G track section, X2 exit signal, 3G track section, and X3 shunting signal. Below P1 turnout, a branch connects to the 4G track section, which in turn connects to the X4 shunting signal. P1 serves as a throat junction, connecting to 2G in the straight track direction and 4G in the side track direction. The contacts of the contact check relay DJ1 inside P1 are connected in series in the turnout indication circuit. Each node in the diagram is labeled with its equipment identifier and type. The lines represent the upstream and downstream physical connections; for example, X1 is downstream of 1G, 1G is downstream of P1, P1's straight track is downstream of 2G, 2G is downstream of X2, X2 is downstream of 3G, and 3G is downstream of X3. This diagram overlays the physical connections with the interlocking logic background, providing a physical layer reachability verification basis for subsequent fault propagation link mapping.

[0127] After obtaining the topology and interlocking logic, based on the fault propagation paths recorded in the associated fault instances, and combining the physical connections and spatial relationships in the topology, the propagation links of the fault among multiple signaling devices are identified. The fault propagation paths exist as an ordered sequence of nodes, with each node corresponding to a signaling device. Directed edges between nodes represent the propagation association of the fault spreading from one device to the next. Based on this, each directed edge in the path is mapped to its corresponding physical connection in the topology graph, verifying whether the propagation direction is consistent with the electrical connection or control dependency direction between devices. If a propagation association does not have a direct connection in the topology graph, the existence of an indirect connection path through intermediate devices is further checked to confirm the reachability of the propagation link at the physical layer. After the above mapping and verification, a complete fault propagation link is obtained, identifying the starting and ending devices of the link.

[0128] Figure 4 This is a schematic diagram of a fault propagation chain according to an embodiment of the present invention. Figure 4 As shown, the fault propagation process from the starting device to the ending device is illustrated using a horizontal chain structure. From left to right, the diagram shows DJ1 (close-fitting inspection relay), P1 (turnout / representation circuit), IL (interlocking logic unit), and X2 (departure signal). Adjacent nodes are connected by directed edges with arrows, labeled E1, E2, and E3 respectively. DJ1 is located at the beginning of the link, P1 is an intermediate device, IL is the interlocking logic unit that receives the turnout's indication status and outputs control commands, and X2 is located at the end of the link. DJ1 contacts are connected in series in the P1 indication circuit, and the propagation direction is consistent with the electrical connection direction. E2 is an indirect connection; there is no direct connection between P1 and IL, requiring an intermediate path of "turnout indication relay contact → indication cable → interlocking input board → IL," with a path length of 3, and the topology is reachable. E3 is a control-dependent direct connection; the IL output interface directly drives X2 through the signal's lighting circuit. The graph has 4 link nodes and 3 directed edges, including 1 direct physical connection edge, 1 indirect connection edge, and 1 control dependency edge. All of them have passed the physical layer reachability verification. The starting device is DJ1 and the ending device is X2.

[0129] Fault causal characteristics are extracted from the initiating device in the fault propagation chain, and fault phenomenon characteristics are extracted from the terminating device. The initiating device is the device where the fault initially occurs, and its fault causal characteristics include the device's state change records at the time of the fault, the timing deviation of its actions, and the degree of deviation from the normal operating baseline. Taking a turnout as an example, if the turnout control circuit experiences a break in the circuit, the fault causal characteristics of the initiating device include the timing deviation of the switch machine not responding after the turnout start command is issued, and the abnormal current characteristics indicating the break in the circuit. The terminating device is the device affected at the end of the fault propagation chain, and its fault phenomenon characteristics are manifested as states that should not occur under normal interlocking logic, such as the inability of relevant route signals to open due to a turnout fault, or the closure of train routes due to misoccupation of track sections. These two types of characteristics are expressed in the form of structured attribute sets, each attribute containing device identifier, state name, deviation direction, and deviation magnitude, forming the input elements for subsequent causal analysis.

[0130] Based on the action constraint and state dependency relationships in signal interlocking logic, this study analyzes the causal relationship between fault cause characteristics and fault phenomenon characteristics. Action constraints stipulate that a specific device must perform or is prohibited from performing a certain action when a certain set of preconditions is met. State dependencies describe how the state of one device affects the permissible state range of another device. By substituting fault cause characteristics into the interlocking logic rules, the study progressively deduces how state changes caused by fault causes propagate along the interlocking chain and compares them item by item with fault phenomenon characteristics. If the deduction result matches the actual record of the fault phenomenon characteristics, a causal relationship is established. If multiple propagation branches appear during the deduction, the deduction path that best matches the measured data is selected by combining the actual state records of each node in the propagation link, resulting in a uniquely determined causal relationship analysis result. In this process, for composite preconditions in the interlocking logic, each precondition is checked to see if it is satisfied or not satisfied at the time of the fault, thus distinguishing between necessary and sufficient causes and providing a traceable logical basis for the causal relationship analysis results.

[0131] Based on the causal correlation analysis results, causal rules are established between fault cause characteristics and fault phenomenon characteristics. Each causal rule is expressed in the implication form of "if the set of cause characteristics is true, then the set of phenomenon characteristics will appear." The antecedent of the rule corresponds to the set of fault cause characteristics, and the consequent of the rule corresponds to the set of fault phenomenon characteristics. The confidence level of the rule is determined by the proportion of historical fault instances that satisfy both the antecedent and consequent of the rule, reflecting the reliability of the rule. For multiple causal rules with the same consequent but different antecedents, they are organized into a multi-cause explanation set for the same fault phenomenon, so that the conditions for the validity of each candidate cause can be checked one by one during actual diagnosis, avoiding missed diagnosis caused by a single cause assumption.

[0132] Causal rules are transformed into a structured expression containing conditional and conclusion parts, resulting in structured fault knowledge rules. The conditional part records the characteristics of the fault cause and its corresponding equipment state changes, specifically including equipment identification, fault trigger time, state attribute name, comparison of measured value with normal baseline value, and the duration or trigger frequency threshold required to trigger the condition. The conclusion part records the characteristics of the fault phenomenon and its corresponding equipment state changes, including the identification of the affected equipment, abnormal state attribute name, expected deviation direction, and the manifestation of the phenomenon at the interlocking logic level, such as route establishment failure, signal malfunction, or section occupancy anomaly. The structured expression uses unified field definitions and data type constraints, enabling each fault knowledge rule to be directly processed by the parsing, retrieval, and inference engine, thus supporting subsequent knowledge base writing and adaptive update processes.

[0133] Taking a turnout contact check circuit failure as an example, the complete execution process of the above procedure is illustrated. The topology records the connection relationships between the turnout and the upstream and downstream track sections and route signals. The interlocking logic stipulates that the turnout contact condition is a necessary prerequisite for establishing the relevant route. The fault propagation path shows that the fault starts from the demagnetization of the turnout contact check relay and ends when the corresponding route signal cannot be opened. The fault cause characteristics extracted from the starting device include the open circuit state of the contact check circuit and the time of relay demagnetization. The fault phenomenon characteristics extracted from the terminating device include the signal remaining in a red state and the route application response timeout. After substituting the contact check failure into the interlocking logic, the deduction results confirm that the route establishment prerequisite is not met, which is completely consistent with the phenomenon characteristics of the signal not being able to open, thus verifying the causal relationship. The finally generated structured fault knowledge rule records the state change of the turnout contact check relay demagnetization in the condition part and the state change of the corresponding route signal remaining in a red state and the route being blocked in the conclusion part. This rule is then written into the fault knowledge base to guide the rapid location and handling of subsequent similar faults.

[0134] Based on the action constraint relationship and state dependency relationship in the signal interlocking logic, the causal correlation between the fault cause characteristics and the fault phenomenon characteristics is analyzed, and the causal correlation analysis results are obtained, including:

[0135] Obtain the starting device identifier and starting device status change corresponding to the fault cause characteristics, and the ending device identifier and ending device status change corresponding to the fault phenomenon characteristics;

[0136] Based on the action constraint relationship, starting from the starting device identifier, trace the action triggering condition corresponding to the change in the state of the starting device, identify the set of downstream devices constrained by the action triggering condition, determine whether the terminating device identifier is located in the set of downstream devices, and generate a constraint association metric value that characterizes the action constraint transmission path.

[0137] Based on the state dependency relationship, starting from the state change of the starting device, the transmission process of the state change among multiple signal devices is tracked, the influence intensity of the state change of the starting device on the state change of the terminating device is calculated, and a dependency correlation metric value characterizing the state dependency transmission effect is generated.

[0138] The constraint correlation metric and the dependency correlation metric are fused together. When both the constraint correlation metric and the dependency correlation metric satisfy the preset correlation judgment condition, it is determined that there is a causal correlation between the fault cause feature and the fault phenomenon feature, and a causal correlation analysis result containing the constraint correlation metric and the dependency correlation metric is generated.

[0139] When obtaining the starting device identifier corresponding to the fault cause characteristics, it is necessary to extract the source device number and its state transition record that triggered the fault propagation from the associated fault instances. The state change of the starting device describes the state transition process of the device before and after the fault occurred, such as a turnout switching from a positioning state to an abnormal state, or a track circuit device changing from an occupied state to a lost-indication state. Simultaneously, the terminating device identifier and its state change corresponding to the fault phenomenon characteristics are obtained. The terminating device is the device that observed the fault phenomenon, and its state change is the external manifestation of the fault at the end of the propagation link. The accurate extraction of these two sets of identifiers and state changes is the fundamental input for subsequent causal correlation analysis, ensuring that the analysis scope is limited to the actual pairs of devices where fault propagation occurred, rather than an indiscriminate exhaustive search of all device combinations.

[0140] When performing constraint association measurement based on action constraint relationships, the process starts from the starting device identifier and traces downwards level by level along the action triggering conditions recorded in the interlocking logic. Action triggering conditions describe the prerequisites for a certain device's action to occur; for example, route establishment requires the relevant turnout to be in place, the section to be clear, and the signal to be open. The tracing process unfolds layer by layer along the directed edges of the interlocking relationship graph, identifying the entire set of downstream devices constrained by the starting device's state change. During this process, it is determined whether the terminating device identifier is within this downstream device set. If the terminating device exists in the downstream set, it indicates that there is a traceable action constraint transmission path from the starting device to the terminating device; if not, it indicates that the two are not directly related in the action constraint dimension. Constraint Association Measurement Value The strength of this transitive relationship is quantified by considering both the hierarchical depth of the transitive path and the constraint weights of each node along the path. The shorter the path and the higher the weight of the intermediate constraint nodes, the stronger the transitive relationship. The larger the value, the stronger the likelihood that the action constraint is transmitted from the starting device to the ending device.

[0141] When performing dependency correlation measurement based on state dependencies, the process starts from the state change of the initial device and traces the transmission process of that state change among multiple signaling devices. State dependencies reflect how a state change of one device triggers state responses from adjacent or more distant devices through interlocking logic. For example, occupancy of a track circuit can cause related signals to automatically close, thus affecting the arrangement of subsequent routes. The tracing process unfolds step-by-step along the directed graph of state dependencies, recording the intermediate device nodes and their state response amplitudes. When calculating the impact strength of the initial device's state change on the terminating device's state change, the attenuation effect of the transmission path must be considered. The impact strength decreases according to the state transmission coefficient of each intermediate node. Let the state transmission path from the initial device to the terminating device contain... The intermediate node, the first The state transit coefficients of the intermediate nodes are: Then it depends on the associated metric. satisfy:

[0142] ;

[0143] in The value is 1 when a node has no attenuation effect on state changes. If there are multiple state transmission paths between the starting and ending devices, the maximum value of the calculated results for each path is taken as the final value. This is to reflect the strongest dependency transmission effect. The closer it is to 1, the more direct and stronger the impact of the change in the state of the starting equipment on the change in the state of the ending equipment.

[0144] Associate constraints with metrics Dependency-related metrics A fusion calculation is performed to obtain the comprehensive causal correlation strength. The fusion method uses weighted summation. To constrain the fusion weights of related metrics, The fusion weights depend on the associated metrics and satisfy the following conditions: Then, considering the strength of the causal relationship for:

[0145] ;

[0146] and The value can be configured based on actual operation and maintenance experience of the port railway signaling system. For fault propagation scenarios dominated by interlocking logic, the value should be appropriately increased. For fault propagation scenarios dominated by state-cascade response, appropriately increase... The preset association determination criteria include two independent thresholds: the constraint association determination threshold. Dependency association threshold Only when and Only when both the cause characteristics and the phenomenon characteristics of the failure are established can a causal relationship be determined. This dual-determination mechanism avoids the situation where an overemphasis in one dimension masks a deficiency in the other, ensuring that the confirmed causal relationship has sufficient supporting evidence in both the action constraint and state dependency dimensions.

[0147] After passing the dual threshold test, the causal relationship analysis results are generated. These results are recorded in a structured format, including the starting device identifier, the ending device identifier, and the constraint relationship metric. Dependency-related metrics Comprehensive causal relationship strength This includes the specific node sequences of the action constraint transmission path and the state dependency transmission path. The node sequence of the constraint transmission path details each interlocking constraint node and its constraint type traversed from the starting device to the terminating device, such as route locking constraints, turnout position constraints, and section occupancy constraints. The node sequence of the state dependency transmission path records the state response type of each intermediate device and its corresponding transmission coefficient. These two sets of path information together constitute the evidence chain of causal rules, providing an interpretable reasoning basis for the subsequent generation of fault knowledge rules.

[0148] In actual port railway operations, freight train shunting involves the coordinated operation of multiple sets of switches and multiple sections, with complex signal interlocking relationships. When a switch experiences a timeout anomaly, action constraint tracing can quickly pinpoint which route establishment conditions are affected, while state dependency tracing can identify which track circuits and signals generate state responses as a result. Through the aforementioned causal correlation analysis process, it is possible to effectively distinguish between the direct cause equipment and the affected apparent equipment, avoiding misjudging fault phenomena as root causes, thus ensuring the accuracy of causal rules in the knowledge base. After the analysis results are written into the fault knowledge base, the corresponding causal rules can be directly invoked when similar faults occur subsequently, enabling rapid tracing and location of fault propagation paths.

[0149] The fault knowledge rules are written into a fault knowledge base, and the feature vector representing the fault state is adaptively updated based on the fault knowledge rules, including:

[0150] The fault knowledge rules are written into the fault knowledge base according to the preset knowledge organization structure. The target feature items corresponding to the fault cause features are extracted from the condition part of the fault knowledge rules, and the target feature items corresponding to the fault phenomenon features are extracted from the conclusion part of the fault knowledge rules.

[0151] Based on the matching analysis between the target feature terms and the feature dimensions in the feature vector representing the fault state, the target feature dimension that needs to be adjusted in the feature vector representing the fault state is identified.

[0152] For the target feature dimension, the feature dimension adjustment amount is calculated based on the equipment state changes and causal relationships recorded in the fault knowledge rules.

[0153] The target feature dimension in the feature vector representing the fault state is updated based on the feature dimension adjustment amount to generate an adaptively updated feature vector.

[0154] When writing fault knowledge rules into the fault knowledge base, the rules need to be categorized and stored according to a pre-defined knowledge organization structure. This structure typically uses fault type as the primary classification dimension and equipment category and signal subsystem as secondary classification dimensions, forming a hierarchical rule index system. Each fault knowledge rule is broken down into a condition part and a conclusion part when written: the condition part records the preconditions and causal chain that trigger the fault phenomenon, including abnormal actions of specific equipment, signal status deviations, and violations of interlocking logic; the conclusion part records the manifestation of the fault phenomenon and corresponding handling suggestions. After storing the rules in the knowledge base in a structured format, subsequent matching and retrieval rely on this index structure, ensuring the scalability of the rules and facilitating incremental updates when new fault cases are added to the database without disrupting the existing rule system.

[0155] The process of extracting target feature terms from fault knowledge rules is a prerequisite for adaptive feature vector updating. In the condition section, feature terms directly related to fault causes are typically descriptions of state changes in a certain type of equipment, such as an abnormal drop in turnout starting current within a specified time window, track circuit shunt resistance exceeding a preset range, or signal relay action occurring later than the triggering time specified by the interlocking logic. These descriptions correspond to specific feature dimensions in the feature vector, with each dimension characterizing the quantified state of a particular device within a specific signal cycle. In the conclusion section, fault phenomenon feature terms correspond to abnormal behaviors directly observable at the signal acquisition layer, such as route establishment failure, abnormally persistent track occupancy, or signals remaining in a prohibited state for an extended period. By extracting the fault cause feature terms from the condition section and the fault phenomenon feature terms from the conclusion section, it is possible to determine which dimensions in the feature vector have semantic association with the current fault knowledge rules.

[0156] The identification of target feature dimensions relies on matching analysis between feature terms and feature vector dimensions. The feature vector representing the fault state is composed of timing deviation feature components and logical consistency deviation feature components. Each component contains multiple specific dimensions, and each dimension corresponds to the quantization deviation description of a certain device on a certain type of signal event. The target feature terms extracted from the rules are compared one by one with the dimension labels of the feature vector. If the device identifier, event type and feature term description of a certain dimension match, then that dimension is marked as the target feature dimension that needs to be adjusted. The core of the matching analysis is to determine whether the feature terms correspond to the vector dimensions at the semantic level, rather than just performing a precise comparison at the string level. For devices located at intermediate nodes in the fault propagation path, their corresponding dimensions should be associated and marked according to the topological relationship, even if they are not explicitly mentioned in the rule conditions, to ensure that the feature vector can completely cover the range of devices affected by the fault propagation.

[0157] After identifying the target feature dimensions, the adjustment amount for each target feature dimension needs to be calculated based on the equipment state changes and causal relationships recorded in the fault knowledge rules. Let the magnitude of a certain equipment state change recorded in the fault knowledge rules be... The original feature value of the device in the current feature vector corresponding to the target feature dimension is Then the adjustment amount of this dimension Based on the strength of causal association in the rules The weighted average is then applied. Specifically, if the fault knowledge rules record the causal relationship strength between the fault cause device and the fault phenomenon device as follows: Then the adjustment amount of the target feature dimension satisfies ,in This indicates the degree of attenuation in causal propagation; a value closer to 1 indicates a stronger causal relationship and a closer adjustment magnitude to the original state change. For devices that pass through multiple intermediate nodes along the fault propagation path, the adjustment amount for the corresponding dimension needs to be attenuated sequentially at each node, with the final adjustment amount being the result of cumulative attenuation along the path. Furthermore, when the fault knowledge rule involves both the timing deviation dimension and the logical consistency deviation dimension, the adjustment amounts for the two dimensions are calculated independently, without cross-type interference during the calculation process.

[0158] Based on the above adjustments, the original feature values ​​of the target feature dimension in the feature vector are updated to generate an adaptively updated feature vector. The update method uses incremental stacking rather than full replacement: the updated target feature dimension values... From the original value With adjustment amount The result of superposition is that The advantage of this incremental approach lies in preserving the cumulative shaping of feature vectors by historical fault information, avoiding the complete overwriting of existing information by updates introduced by a single rule. For non-target feature dimensions, their feature values ​​remain unchanged, and the update operation has precise locality, avoiding interference with dimensions semantically unrelated to the current fault knowledge rules. After updating all target feature dimensions, the adaptively updated feature vectors are written back to the feature representation layer for subsequent similarity matching and fault type identification.

[0159] When multiple fault knowledge rules are written to the knowledge base within the same time window, the update operations for feature vectors of each rule are executed sequentially according to the order in which the rules were written into the knowledge base. Each execution is based on the feature vector after the previous update, rather than always using the initial feature vector as the basis. This mechanism allows the feature vectors to exhibit dynamic evolution characteristics as knowledge accumulates. In port and railway operation scenarios with diverse fault modes, the expressive power of the feature vectors continuously increases with the enrichment of the knowledge base. At the same time, to prevent feature dimension values ​​from deviating from normal physical meaning due to continuous accumulation, boundary constraint checks are performed on the values ​​of each dimension after each update. If the updated value exceeds the preset effective range of that dimension, it is truncated to the boundary value, thereby ensuring the overall rationality of the feature vectors.

[0160] During the continuous writing of fault knowledge rules, the knowledge base periodically triggers rule consistency checks to examine whether newly written rules contradict or are redundant with existing rules in terms of condition coverage and conclusion direction. For rule pairs with logical contradictions, the rule with stronger causal relationship is retained, while the rule with weaker causal relationship is placed in a pending review state. For redundant rules that are highly similar in both condition and conclusion parts, they are merged into a single generalized rule through weighted fusion to control the size of the knowledge base and reduce computational overhead during subsequent retrieval. This management mechanism works in conjunction with the adaptive update process of feature vectors to support the continuous evolution and self-improvement of the fault knowledge base during the long-term operation of the port railway signaling system.

[0161] A second aspect of this invention provides a port railway fault knowledge base construction system based on signal microcomputer monitoring, comprising:

[0162] The data acquisition and analysis unit is used to acquire the operating status data of the port railway signaling system in real time through the signal microcomputer monitoring device, perform joint analysis of the operating status data in the time domain and logic domain, identify abnormal patterns based on the action logic constraints and temporal causal relationships of the signaling equipment, and extract feature vectors that characterize the fault state.

[0163] The fault matching unit is used to perform similarity matching between the current fault state and the pre-stored fault mode library based on the timing deviation feature and logical consistency deviation feature in the feature vector, determine the fault type identifier and fault propagation path, and obtain the associated fault instance.

[0164] The rule generation unit is used to extract causal rules between fault phenomena and fault causes based on the fault propagation path corresponding to the associated fault instance, combined with the topological relationship between devices and signal interlocking logic, and generate structured fault knowledge rules.

[0165] The knowledge update unit is used to write the fault knowledge rules into the fault knowledge base and adaptively update the feature vector representing the fault state based on the fault knowledge rules.

[0166] A third aspect of the present invention provides an electronic device, comprising:

[0167] processor;

[0168] Memory used to store processor-executable instructions;

[0169] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0170] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0171] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a port railway fault knowledge base based on signal microcomputer monitoring, characterized in that, include: The operation status data of the port railway signaling system is collected in real time by a signal microcomputer monitoring device. The operation status data is analyzed in both the time domain and the logical domain. Abnormal patterns are identified based on the action logic constraints and temporal causal relationships of the signaling equipment, and feature vectors representing fault states are extracted. Based on the timing deviation feature and logical consistency deviation feature in the feature vector, the current fault state is matched with the pre-stored fault mode library to determine the fault type identifier and fault propagation path, and obtain the associated fault instance. Based on the fault propagation path corresponding to the associated fault instance, and combined with the topological relationship between devices and signal interlocking logic, the causal rules between fault phenomena and fault causes are extracted to generate structured fault knowledge rules. The fault knowledge rules are written into the fault knowledge base, and the feature vector representing the fault state is adaptively updated based on the fault knowledge rules.

2. The method according to claim 1, characterized in that, The operational status data of the port railway signaling system is collected in real time by a signal microcomputer monitoring device. This operational status data undergoes collaborative analysis in both the time and logical domains. Based on the action logic constraints and temporal causal relationships of the signaling equipment, abnormal patterns are identified, and feature vectors characterizing fault states are extracted, including: The operating status data is subjected to time-series analysis in the time domain to extract the time-series relationship between the action time and action state of the signal device, and a time-domain feature set is obtained. The operational status data is logically verified in the logical domain dimension. Based on the action logic constraints and interlocking relationship data of the signal equipment, the action status of the signal equipment is verified to see if it meets the preset interlocking logic rules, and a logical domain feature set is obtained. The time-domain feature set and the logic-domain feature set are fused to identify consistency deviations between timing relationships and interlocking logic rules, and abnormal patterns are marked. Based on the aforementioned abnormal patterns, feature vectors containing timing deviation features and logical consistency deviation features are extracted. The timing deviation features characterize the degree of deviation between the actual action time and the expected action time of the signaling device, and the logical consistency deviation features characterize the degree of conformity between the action state of the signaling device and the interlocking logic rules.

3. The method according to claim 2, characterized in that, The time-domain feature set and the logic-domain feature set are fused to identify consistency deviations between timing relationships and interlocking logic rules, and abnormal patterns are marked, including: Based on the temporal dependency relationship between the action time and action state of the signal devices in the time-domain feature set, a temporal relationship mapping table is constructed. Based on the logical constraint relationship between the action state of the signal devices in the logical domain feature set and the interlocking logical rules, a logical rule constraint table is constructed. For the same action event of the same signal device, extract the actual action time and the expected action time recorded in the time sequence mapping table, calculate the time difference between the actual action time and the expected action time, and generate a time sequence deviation feature; For the same action event of the same signal device, extract the actual action state recorded in the logical rule constraint table and the expected action state required by the interlocking logical rules, calculate the logical conformity between the actual action state and the expected action state, and generate logical consistency deviation features; Based on the fusion analysis of the timing deviation feature and the logical consistency deviation feature, when the timing deviation feature indicates that the signal device has completed the action and the logical consistency deviation feature indicates that the action condition has not been met, the same action event is marked as an abnormal mode.

4. The method according to claim 1, characterized in that, Based on the timing deviation feature and logical consistency deviation feature in the feature vector, the current fault state is matched with a pre-stored fault mode library to determine the fault type identifier and fault propagation path, thereby obtaining associated fault instances, including: Multiple historical fault modes are obtained from the pre-stored fault mode library. The temporal similarity between the timing deviation feature and the historical timing deviation feature of each historical fault mode, and the logical similarity between the logical consistency deviation feature and the historical logical consistency deviation feature of each historical fault mode are calculated respectively. The temporal similarity and the logical similarity are weighted and fused to generate a comprehensive similarity between the current fault state and each historical fault mode; Based on the comprehensive similarity, target historical fault modes that meet the preset matching conditions are selected from the multiple historical fault modes. The fault type identifier and fault propagation path corresponding to the target historical fault mode are extracted. The fault propagation path represents the propagation order and propagation association of the fault among multiple signal devices. The associated fault instance is determined based on the target historical fault mode, the fault type identifier, and the fault propagation path.

5. The method according to claim 1, characterized in that, Based on the fault propagation path corresponding to the associated fault instance, and combined with the topological relationships and signal interlocking logic between devices, causal rules between fault phenomena and fault causes are extracted to generate structured fault knowledge rules, including: Obtain the topological relationships and signal interlocking logic between devices in the port railway signaling system; Based on the propagation order and propagation correlation recorded in the fault propagation path, and combined with the physical connection relationship and spatial location relationship in the topology, the propagation link of the fault between multiple signal devices is identified; Extract fault cause features from the starting device of the propagation link and extract fault phenomenon features from the terminating device of the propagation link; Based on the action constraint relationship and state dependency relationship in the signal interlocking logic, the causal relationship between the fault cause characteristics and the fault phenomenon characteristics is analyzed, and the causal relationship analysis results are obtained. Based on the causal correlation analysis results, causal rules are established between the fault cause characteristics and the fault phenomenon characteristics; The causal rules are converted into a structured expression containing a condition part and a conclusion part to obtain a structured fault knowledge rule. The condition part records the fault cause characteristics and the corresponding equipment state changes, and the conclusion part records the fault phenomenon characteristics and the corresponding equipment state changes.

6. The method according to claim 5, characterized in that, Based on the action constraint relationship and state dependency relationship in the signal interlocking logic, the causal correlation between the fault cause characteristics and the fault phenomenon characteristics is analyzed, and the causal correlation analysis results are obtained, including: Obtain the starting device identifier and starting device status change corresponding to the fault cause characteristics, and the ending device identifier and ending device status change corresponding to the fault phenomenon characteristics; Based on the action constraint relationship, starting from the starting device identifier, trace the action triggering condition corresponding to the change in the state of the starting device, identify the set of downstream devices constrained by the action triggering condition, determine whether the terminating device identifier is located in the set of downstream devices, and generate a constraint association metric value that characterizes the action constraint transmission path. Based on the state dependency relationship, starting from the state change of the starting device, the transmission process of the state change among multiple signal devices is tracked, the influence intensity of the state change of the starting device on the state change of the terminating device is calculated, and a dependency correlation metric value characterizing the state dependency transmission effect is generated. The constraint correlation metric and the dependency correlation metric are fused together. When both the constraint correlation metric and the dependency correlation metric satisfy the preset correlation judgment condition, it is determined that there is a causal correlation between the fault cause feature and the fault phenomenon feature, and a causal correlation analysis result containing the constraint correlation metric and the dependency correlation metric is generated.

7. The method according to claim 1, characterized in that, The fault knowledge rules are written into a fault knowledge base, and the feature vector representing the fault state is adaptively updated based on the fault knowledge rules, including: The fault knowledge rules are written into the fault knowledge base according to the preset knowledge organization structure. The target feature items corresponding to the fault cause features are extracted from the condition part of the fault knowledge rules, and the target feature items corresponding to the fault phenomenon features are extracted from the conclusion part of the fault knowledge rules. Based on the matching analysis between the target feature terms and the feature dimensions in the feature vector representing the fault state, the target feature dimension that needs to be adjusted in the feature vector representing the fault state is identified. For the target feature dimension, the feature dimension adjustment amount is calculated based on the equipment state changes and causal relationships recorded in the fault knowledge rules. The target feature dimension in the feature vector representing the fault state is updated based on the feature dimension adjustment amount to generate an adaptively updated feature vector.

8. A port railway fault knowledge base construction system based on signal microcomputer monitoring, used to implement the method as described in any one of claims 1-7, characterized in that, include: The data acquisition and analysis unit is used to acquire the operating status data of the port railway signaling system in real time through the signal microcomputer monitoring device, perform joint analysis of the operating status data in the time domain and logic domain, identify abnormal patterns based on the action logic constraints and temporal causal relationships of the signaling equipment, and extract feature vectors that characterize the fault state. The fault matching unit is used to perform similarity matching between the current fault state and the pre-stored fault mode library based on the timing deviation feature and logical consistency deviation feature in the feature vector, determine the fault type identifier and fault propagation path, and obtain the associated fault instance. The rule generation unit is used to extract causal rules between fault phenomena and fault causes based on the fault propagation path corresponding to the associated fault instance, combined with the topological relationship between devices and signal interlocking logic, and generate structured fault knowledge rules. The knowledge update unit is used to write the fault knowledge rules into the fault knowledge base and adaptively update the feature vector representing the fault state based on the fault knowledge rules.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images