Collaborative fault diagnosis system for multi-lane equipment of highway toll station

Abstract

The invention belongs to the field of equipment diagnosis, and particularly relates to an expressway toll station multi-lane equipment cooperative fault diagnosis system, which comprises an acquisition module used for acquiring multi-lane heterogeneous equipment data in real time through an Internet of Things edge gateway, and generating an equipment state event with a timestamp and a topology identifier after standardization processing; the topology construction module is used for dynamically constructing and maintaining a fault association tree reflecting a fault propagation relationship; the matching module is used for constructing a multi-dimensional query vector according to the fault association tree node features and matching a special diagnosis analysis plug-in in the plug-in diagnosis knowledge base; the response module is used for driving a collaborative diagnosis inference engine through the hit plug-in, performing graph calculation and causal analysis on the cross-lane and cross-equipment fault association tree, and outputting a collaborative diagnosis report and a precise disposal strategy; according to the method, correlation analysis and intelligent root cause positioning of multi-lane equipment faults are realized, and the operation and maintenance response speed and the equipment availability of the toll station are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of equipment diagnostics, and in particular relates to a collaborative fault diagnosis system for multi-lane equipment at highway toll stations. Background Technology

[0002] While current highway tollbooth equipment (such as lane controllers, automatic barrier gates, and toll displays) has achieved basic automation, monitoring and diagnosis still primarily rely on independent operation of individual devices, lacking systematic collaborative analysis and correlation judgment capabilities. This model struggles to effectively identify and locate common faults or potential risks across lanes and equipment. For example, when equipment from the same batch experiences synchronized performance degradation due to hardware batch defects, or when abnormal power supply or network switch failures cause simultaneous malfunctions in multiple lanes, existing systems cannot correlate and integrate these isolated alarm messages for root cause tracing. Maintenance personnel often only receive scattered, superficial alarms, requiring them to rely on experience to troubleshoot one by one, which is not only inefficient but also prone to misjudging the nature of the fault, delaying timely intervention. Therefore, current technology lacks a system capable of real-time collaborative data collection, intelligent correlation analysis, and comprehensive diagnosis of multi-lane equipment status, resulting in the inability to achieve fault early warning, root cause location, and rapid response at the system level, thus hindering the improvement of the overall maintenance level of tollbooths. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a collaborative fault diagnosis system for multi-lane equipment at highway toll stations. This system enables correlation analysis and intelligent root cause localization of multi-lane equipment faults, significantly improving the toll station's operation and maintenance response speed and equipment availability.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A collaborative fault diagnosis system for multi-lane equipment at highway toll stations includes: The data acquisition module collects real-time parallel operation data and real-time alarm streams of heterogeneous devices in multiple lanes through the IoT edge gateway of the toll station, and performs standardization and event processing on the data to generate device status events with timestamps, device topology identifiers and status feature vectors. The topology construction module constructs and updates a fault association tree based on the monitoring cycle and equipment topology relationship of the collaborative fault diagnosis system. The matching module constructs a multi-dimensional collaborative diagnostic query vector based on the state characteristics of the current focused node and its parent and child nodes in the fault association tree, and matches the multi-dimensional collaborative diagnostic query vector with a preset plug-in diagnostic knowledge base to match the dedicated diagnostic analysis plug-in applicable to the current associated fault scenario. The response module, through the hit dedicated diagnostic analysis plugin, drives a preset collaborative diagnostic reasoning engine to perform graph calculation and causal analysis on the fault association tree across lanes and equipment, and generates and outputs a collaborative diagnostic report and handling strategy.

[0005] Specifically, the fault association tree is as follows: If the current monitoring cycle is the first monitoring cycle after the fault diagnosis system is started, then the fault association tree is initialized with the first abnormal or critical equipment status event as the root node; If it is a continuous monitoring process, based on the physical correlation of the device status events, the alarm logic chain, and the temporal proximity, the newly generated device status events are treated as child nodes and attached to the existing nodes in the fault association tree, and the fault association tree is updated in real time.

[0006] Specifically, the process of constructing the fault association tree includes: Based on the first monitoring cycle after the collaborative fault diagnosis system is started, the first device status event generated by the acquisition module is obtained. Based on the preset root node initialization rules, feature analysis is performed on the first device status event. If the status feature vector of the device status event exceeds the preset abnormal threshold, or the alarm flow identifier contained in the device status event belongs to the preset key alarm type, then the device status event is determined as a root node event, and the fault association tree is initialized with the root node event. The root node carries the timestamp, device topology identifier and complete status feature vector corresponding to the root node event. Once the collaborative fault diagnosis system enters the continuous monitoring process, it collects and acquires new equipment status events continuously generated by the acquisition module in real time. Specifically, the process of constructing the fault association tree also includes: Based on a multi-dimensional association rule engine, parent node matching calculation is performed for each newly generated device status event to obtain a set of candidate parent nodes for association; the set of candidate parent nodes for association includes a set of candidate parent nodes for physical association, a set of candidate parent nodes for logical association, and a set of candidate parent nodes for time association; Weighting and intersection analysis are performed on the physical association candidate parent node set, logical association candidate parent node set, and time association candidate parent node set to calculate the matching degree score of each candidate parent node, and the candidate parent node with the highest matching degree score is selected as the optimal parent node. Based on the preset tree structure dynamic maintenance strategy, the current new device status event is attached as a child node to the optimal parent node; If there are multiple candidate parent nodes whose matching scores meet the preset threshold, a virtual convergence node is created according to the fault propagation probability model, and the multiple candidate parent nodes are associated with the current new device status event child nodes through the virtual convergence node. Specifically, the process of constructing the fault association tree also includes: Based on a periodic tree structure optimization mechanism, the fault association tree is structurally reviewed and optimized according to a preset period, specifically including: If the device status corresponding to any child node in the fault association tree recovers to normal within several consecutive monitoring cycles, and its status feature vector remains below a preset stable threshold, then the child node is archived or weakened. The weakening mark includes reducing the node display weight or adding a recovery identifier. If multiple child nodes attached to the same parent node are detected to exhibit highly similar state feature patterns, and the similarity meets a preset similarity threshold, then a node merging operation is triggered to generate a super node representing the common fault pattern. The super node carries the feature summary information of the common fault pattern and the association and tracing information of each original child node.

[0007] Specifically, a multi-dimensional collaborative diagnostic query vector is constructed, including: Determine the current focus node in the fault association tree and obtain the target node set centered on the focus node, the target node set including the current focus node, its parent node and all its child nodes; Based on each node in the target node set, state feature extraction and standardization are performed to obtain a standardized feature set; the standardized feature set includes: node state feature vector, node association feature, and node optimized state feature. The features in the standardized feature set are mapped to pre-divided physical association dimension, logical association dimension, temporal association dimension, node attribute dimension and optimization state dimension according to their attributes, forming feature subsets corresponding to each dimension; Following a preset order from the physical association dimension to the optimized state dimension, feature subsets of each dimension are sequentially concatenated, and the initial multidimensional collaborative diagnostic query vector is optimized using the L2 normalization algorithm to obtain the multidimensional collaborative diagnostic query vector.

[0008] Specifically, the multidimensional collaborative diagnostic query vector is matched with a preset pluggable diagnostic knowledge base, including: Based on the preset plug-in diagnostic knowledge base, a plug-in knowledge unit library after structured preprocessing is obtained; wherein, the fault diagnosis knowledge in the plug-in diagnostic knowledge base is divided into independent plug-in knowledge units according to different fault types and scenarios, each plug-in knowledge unit is bound to a dedicated diagnostic analysis plug-in, and is configured with feature tag sets and matching threshold ranges corresponding to each dimension of the multi-dimensional collaborative diagnostic query vector. Based on the multidimensional collaborative diagnostic query vector, a target query vector with standardized format and dimension alignment is obtained; Based on the target query vector, a subset of initial screening knowledge units is obtained. Specifically, the key character features of the physical and logical association dimensions in the target query vector are extracted as initial screening keywords to traverse the plugin knowledge unit library for screening. If no results are found, the time association dimension features are extracted as auxiliary initial screening keywords to re-traverse and screen.

[0009] Specifically, matching the multidimensional collaborative diagnostic query vector with a preset pluggable diagnostic knowledge base further includes: Based on the initial screening subset of knowledge units and the target query vector, the comprehensive similarity score of each plug-in knowledge unit is obtained. Specifically, the cosine similarity algorithm is used to calculate the single-dimensional similarity between the target query vector and the feature tag set of each plug-in knowledge unit in each dimension, and the weighted comprehensive similarity is calculated by combining the preset weight coefficients of each dimension. If the target query vector contains super node features, the corresponding preset weight is added. Based on the comprehensive similarity score and the matching threshold range of the plug-in knowledge unit, a valid and ranked list of matching knowledge units is obtained. Based on the effective matching knowledge unit ranking list, the final hit dedicated diagnostic analysis plugin is obtained. Specifically, the plugin bound to the knowledge unit ranked first is selected as the hit plugin. If there are ties, the fault propagation probability model is called to calculate the fault propagation contribution and the plugin bound to the one with the highest contribution is selected.

[0010] Specifically, graph computation and causal analysis are performed on fault association trees across lanes and equipment, including: Based on the hit dedicated diagnostic analysis plugin, the initialized collaborative diagnostic reasoning engine and the injected current diagnostic context are obtained; wherein, the collaborative diagnostic reasoning engine completes core initialization by loading the pre-set domain analysis strategy and reasoning rules in the dedicated diagnostic analysis plugin, and the current diagnostic context includes a complete fault association tree topology structure, the multi-dimensional collaborative diagnostic query vector, real-time and historical status data of all nodes in the target node set, and the association information of the device physical topology knowledge base and alarm logic chain rule base; Based on the complete fault association tree topology and the analysis strategy specified by the dedicated diagnostic analysis plugin, an abstractly constructed weighted directed graph model is obtained; wherein, the nodes in the fault association tree are abstracted as nodes in the weighted directed graph model, the parent-child connection relationship in the fault association tree is abstracted as directed edges in the weighted directed graph model, and the weight of each directed edge is comprehensively constructed based on the physical matching degree weight, logical matching degree weight, and time matching degree weight between the nodes corresponding to the directed edge; Based on the weighted directed graph model and the graph computing component of the collaborative diagnostic reasoning engine, the node centrality analysis results and the core suspected fault propagation chain are obtained.

[0011] Specifically, graph computation and causal analysis of fault association trees across lanes and equipment also include: Based on the causal determination rule base pre-set in the plugin and the multidimensional features of the nodes in the fault association tree, the causal analysis results across lanes and across devices are obtained, forming a fault causal network with confidence weights. Based on the core propagation path and the fault causal network, the root cause node of the fault and the defined scope of the fault's impact are obtained. Based on the node centrality analysis results, the core suspected fault propagation chain, the fault causal network, the root cause node of the fault, and the scope of the fault's impact, a structured collaborative diagnostic report and a sequence of targeted handling strategies are obtained.

[0012] Compared with the prior art, the beneficial effects of the present invention are: This invention addresses the shortcomings of existing technologies by constructing a multi-dimensional collaborative diagnostic query vector based on the state features of fault association tree nodes. This provides accurate and comprehensive feature input for the diagnostic process, effectively solving the problems of complex equipment fault representation and poor correlation in multi-lane highway environments. In particular, by introducing a plug-in diagnostic knowledge base and a multi-dimensional matching mechanism, it achieves modular organization and efficient retrieval of massive fault diagnosis knowledge, enabling rapid matching of the most suitable dedicated analysis plug-in for the current complex fault scenario, significantly improving the accuracy and relevance of diagnostic strategy invocation. Furthermore, through a collaborative diagnostic reasoning engine driving graph computation and multi-dimensional causal analysis, the abstract fault association tree is transformed into a quantifiable and computable weighted directed graph. Combined with physical, logical, and temporal dimensions, it performs in-depth reasoning and verification of fault propagation paths and root causes, thereby achieving precise location and scope of impact definition from scattered alarms to associated root causes. Finally, the system can automatically generate a structured collaborative diagnostic report containing clear root causes, propagation paths, confidence assessments, and handling strategies, greatly improving the diagnostic efficiency, accuracy, and timeliness of handling faults in multi-equipment systems at toll stations, ensuring the continuous and stable operation of highway toll collection services. Attached Figure Description

[0013] Figure 1 This is a module diagram of a multi-lane equipment collaborative fault diagnosis system for highway toll stations according to the present invention; Figure 2 This is a logical structure diagram of a multi-lane equipment collaborative fault diagnosis system for highway toll stations according to the present invention. Detailed Implementation

[0014] Example 1 Please see Figure 1 , Figure 2The present invention provides an embodiment of a collaborative fault diagnosis system for multi-lane equipment at highway toll stations, comprising the following steps: The data acquisition module collects real-time parallel operation data and real-time alarm streams of heterogeneous devices in multiple lanes through the IoT edge gateway of the toll station, and performs standardization and event processing on the data to generate device status events with timestamps, device topology identifiers and status feature vectors. The topology construction module, based on the current collaborative fault diagnosis system's monitoring cycle and the equipment topology relationship, constructs and maintains a fault association tree, specifically: If the current monitoring cycle is the first monitoring cycle after the fault diagnosis system is started, then the fault association tree is initialized with the first abnormal or critical equipment status event as the root node; If it is a continuous monitoring process, based on the physical correlation of the device status event, the alarm logic chain and the time proximity, the newly generated device status event is treated as a child node and attached to the existing node in the fault association tree, and the fault association tree is updated in real time. The matching module constructs a multi-dimensional collaborative diagnostic query vector based on the state characteristics of the current focused node and its parent and child nodes in the fault association tree, and matches the multi-dimensional collaborative diagnostic query vector with a preset plug-in diagnostic knowledge base to match the dedicated diagnostic analysis plug-in applicable to the current associated fault scenario. The response module, through the hit dedicated diagnostic analysis plugin, drives a preset collaborative diagnostic reasoning engine to perform graph calculation and causal analysis on the fault association tree across lanes and equipment, and generates and outputs a collaborative diagnostic report and handling strategy.

[0015] It should be further explained that the fault association tree in this embodiment is specifically as follows: If the current monitoring cycle is the first monitoring cycle after the fault diagnosis system is started, then the fault association tree is initialized with the first abnormal or critical equipment status event as the root node; If it is a continuous monitoring process, based on the physical correlation of the device status events, the alarm logic chain, and the temporal proximity, the newly generated device status events are treated as child nodes and attached to the existing nodes in the fault association tree, and the fault association tree is updated in real time.

[0016] It should be further explained that the process of constructing the fault association tree in this embodiment includes: S1. Based on the first monitoring cycle after the collaborative fault diagnosis system is started, obtain the first device status event generated by the acquisition module; It should be further explained that the device status event in this embodiment includes at least a status feature vector, an alarm flow identifier, a timestamp, and a device topology identifier, which are used to be analyzed in the first monitoring cycle of the collaborative fault diagnosis system to determine whether the root node initialization conditions are met, thereby determining the root node event and initializing the fault association tree; For example, in the first monitoring cycle after the multi-lane equipment collaborative fault diagnosis system of the highway toll station is started, the acquisition module collects the operating data and real-time alarm streams of heterogeneous equipment in each lane in real time and in parallel through the IoT edge gateway of the toll station. Assuming that the core equipment is the controller of lane 3, the acquisition module obtains its communication connection parameters, hardware operating temperature, power supply voltage and other operating data, and captures the communication abnormality alarm signal issued by the equipment to form a raw data set. The acquisition module performs standardization processing on the raw data, converts different types of parameters into a unified data format to eliminate the difference in dimensions, and then binds the standardized data with the alarm signal through event processing to generate a device status event. The device status event contains four core elements: the device topology identifier is clearly a unique identifier of the controller of lane 3, which is used to locate the lane and equipment type of the faulty equipment; the alarm stream identifier corresponds to the communication interruption type, accurately representing the equipment alarm category; the timestamp records the specific time when the alarm is first detected after the system starts, ensuring that the event sequence is traceable; the status feature vector includes key parameters such as communication connection strength, data transmission rate, hardware temperature, power supply stability and so on, which comprehensively reflects the operating status of the equipment.

[0017] S2. Based on the preset root node initialization rules, feature analysis is performed on the first device status event. If the status feature vector of the device status event exceeds the preset abnormal threshold, or the alarm flow identifier contained in the device status event belongs to the preset key alarm type, then the device status event is determined as the root node event, and the fault association tree is initialized with the root node event. The root node carries the timestamp, device topology identifier, and complete status feature vector corresponding to the root node event. It should be further noted that the complete status feature vector in this embodiment includes key device operating parameters, hardware component status data, communication connection indicators, and performance characterization data, which are used for feature analysis during root node initialization to determine whether the device status is abnormal, providing a core basis for the determination of the root node event, and serving as a basis for node feature comparison, fault propagation path analysis, and periodic optimization of the tree structure in subsequent fault association tree construction. S3. After the collaborative fault diagnosis system enters the continuous monitoring process, it collects and acquires new equipment status events continuously generated by the acquisition module in real time. S4. Based on the multi-dimensional association rule engine, perform parent node matching calculation for each newly generated device status event to obtain a set of candidate parent nodes. In this embodiment, the set of candidate parent nodes includes a physical association candidate parent node set, a logical association candidate parent node set, and a time association candidate parent node set. The specific process of obtaining the parent node matching calculation using the multi-dimensional association rule engine in this embodiment includes: A1 invokes a preset device physical topology knowledge base, using the device identifier corresponding to the current new device status event as the search keyword, and performs a related information query operation to obtain all device information that has a physical connection relationship or a common source relationship with the device identifier. The physical connection relationship refers to the direct or indirect connection between devices through hardware interfaces, transmission lines, etc., and the common source relationship includes, but is not limited to, belonging to the same power distribution cabinet or the same network switch. It should be further explained that the device physical topology knowledge base in this embodiment is constructed by collecting the identifier information, physical connection relationships (such as direct or indirect connections formed by hardware interfaces and transmission lines), and common source relationships (such as belonging to the same power distribution cabinet or the same network switch) of various devices in multiple lanes of highway toll stations. This knowledge base is used to query related device information using the device identifier of the current new device status event as the search keyword when constructing the physical association candidate parent node set. This supports active device filtering and matching of corresponding event nodes in the fault association tree, ultimately helping to form a physical association candidate parent node set, providing data support for the physical association dimension of the optimal parent node matching and dynamic construction of the fault association tree.

[0018] For example, in the continuous monitoring process of the multi-lane equipment collaborative fault diagnosis system at a highway toll station in this embodiment, the acquisition module generates a new equipment status event. Assuming the event corresponds to the automatic barrier gate in lane 4, its equipment identifier is a unique identifier. The system's pre-built physical topology knowledge base for equipment has been completed in advance. During the construction process, core information on various types of equipment in the toll station's multi-lane network is comprehensively collected, covering the unique identifier information of all equipment, the physical connection relationships between devices, and their common source relationships. Specifically, the physical connection relationship is manifested as a direct or indirect connection between devices via dedicated hardware interface transmission lines. For example, the lane controller and the automatic barrier gate are directly connected via a dedicated communication interface, while the lane controller and the toll station's main server are indirectly connected via fiber optic transmission lines. The common source relationship is manifested as devices belonging to the same power distribution cabinet or the same network switch. For example, the automatic barrier gate in lane 4 and the toll display in lane 3 both belong to power distribution cabinet 2, and the controllers in lane 4 and 5 both belong to network switch 3. This knowledge base uses a structured storage method to establish a mapping relationship between equipment identifiers and associated information, providing efficient data support for subsequent retrieval operations.

[0019] The system invokes a pre-defined device physical topology knowledge base to initiate a related information query process, using the device identifier of the automatic barrier gate machine in lane 4 corresponding to the current new device status event as the unique search keyword. Through the knowledge base's built-in search algorithm, it precisely matches all device identifiers stored in the database. After locating the information entry corresponding to the automatic barrier gate machine in lane 4, it further extracts the physical connection relationships and common-source relationship data bound to that entry. The search reveals that the device information with a physical connection relationship to this device identifier is the lane 4 controller. The two are directly physically connected via an RS485 hardware interface and a matching transmission line, allowing bidirectional data transmission through this link. The device information with a common-source relationship to this device identifier is the toll display in lane 3 and the toll display in lane 4. All three belong to the same power distribution cabinet (distribution cabinet 2) and are provided with stable power by the same cabinet, thus forming a common-source relationship. After the search is completed, the system summarizes all the above-mentioned related device information to form an initial list of related devices.

[0020] A2 verifies the activity status of all device information obtained from the query, filters out devices that are in operation monitoring status and have not been marked as offline or disabled, and forms a list of active devices. It should be further explained that, in this embodiment, combined with the collaborative diagnosis scenario of multi-lane equipment at highway toll stations, the active status verification adopts a multi-dimensional data cross-validation technique. Specifically, the acquisition module obtains real-time operating data of the query devices through the IoT edge gateway, extracts key indicators such as device heartbeat signals, communication connection status, and core power supply parameters, and simultaneously retrieves the system's device status management ledger to check whether each device has an offline marker or a fault shutdown registration. The real-time operating indicators are compared with the preset active judgment threshold to confirm that the device heartbeat signal is continuous, the communication link is smooth, the power supply is stable, and the status ledger has no offline or fault shutdown records. Devices with abnormal real-time indicators or invalid status markers are eliminated, and finally, devices that meet the dual conditions of "real-time operation monitoring + no offline / fault shutdown markers" are summarized to form an active device list, providing effective device range support for subsequent fault association tree node matching.

[0021] A3. In the initialized fault association tree, match the event nodes corresponding to each device in the active device list one by one, that is, find the event nodes in the fault association tree whose device topology identifier is consistent with the device identifier in the active device list. A4, summarize all successfully matched event nodes to form a set of candidate parent nodes for physical association; A5, invokes the preset alarm logic chain rule base, uses the alarm type of the current new device status event as the search keyword, performs an alarm type search operation, and obtains all alarm types that have a preset causal or sequential logical relationship with that alarm type; wherein, in this embodiment, the causal logical relationship refers to the preceding alarm type being the fault cause of the subsequent alarm type, and the sequential logical relationship refers to the time sequence of fault propagation between the preceding alarm type and the subsequent alarm type. An example of the logical relationship in this embodiment is that the "lane controller communication interruption" alarm type and the "automatic barrier machine unresponsive" alarm type have a causal logical relationship; it should be further noted that this embodiment... The alarm logic chain rule base of the example is constructed by collecting various alarm types of multi-lane equipment at highway toll stations, sorting out the preset causal logical relationships (preceding alarms are the causes of subsequent alarm failures) and sequential logical relationships (the order of fault propagation time) between different alarm types, and using the alarm type of the current new equipment status event as the search keyword to query related alarm types when constructing the logical association candidate parent node set. It supports the effective filtering of related alarm types and the matching of corresponding event nodes in the fault association tree, and finally helps to form a logical association candidate parent node set, providing data support for the alarm logic association dimension for optimal parent node matching and dynamic construction of the fault association tree.

[0022] A6. Perform validity verification on all retrieved alarm types, filter out alarm types that are enabled and match the current fault diagnosis scenario, and form a list of valid associated alarm types. A7. In the initialized fault association tree, match the event nodes corresponding to each alarm type in the list of valid associated alarm types one by one, that is, find the event nodes in the fault association tree whose alarm type corresponding to the alarm flow identifier is consistent with the alarm type in the list of valid associated alarm types. A8 summarizes all successfully matched event nodes to form a logically related candidate parent node set; A9, enable time window and event flow analysis strategy, configure preset time neighborhood parameters. The preset time neighborhood is a fixed duration of tracing back from the time when the current new device status event is generated. In this embodiment, the initial fixed duration of tracing back is set to 30 seconds in advance, and its duration can be dynamically adjusted according to the actual fault diagnosis accuracy requirements. A10. Based on the preset time neighborhood parameters, compare the timestamps of all event nodes that have been added to the fault association tree, and filter out event nodes whose timestamps are within the preset time neighborhood. A11 performs status verification on the selected event nodes, excluding nodes that have been archived, weakened, or expired, and retaining valid event nodes; A12 summarizes all valid event nodes that have passed the verification, forming a set of time-related candidate parent nodes.

[0023] S5. Perform weight allocation and intersection analysis on the physical association candidate parent node set, logical association candidate parent node set and time association candidate parent node set, calculate the matching degree score of each candidate parent node, and select the candidate parent node with the highest matching degree score as the optimal parent node. It should be further explained that the process of calculating the matching score of each candidate parent node in this embodiment includes: The system pre-determines weighting coefficients for three dimensions: physical association, logical association, and temporal association. These weighting coefficients are determined based on historical fault diagnosis data, equipment fault propagation patterns, and engineering application experience, and the sum of the weighting coefficients for the three dimensions is 1. For each node in the physical association candidate parent node set, the initial score of the physical dimension is calculated by multiplying the physical association weight coefficient by the physical matching degree of the physical association candidate parent node. In this embodiment, the physical matching degree is quantitatively determined by calling a preset device physical topology knowledge base. The specific steps are as follows: First, determine whether the association type between the device corresponding to the candidate parent node and the device corresponding to the current new device status event is a physical connection or a co-source relationship. If it is a physical connection, query the physical connection path between the two and divide the connection level into direct connection, first-level indirect connection, and multi-level indirect connection. The fewer the levels, the higher the tightness and the higher the quantitative score is assigned. If it is a co-source relationship, query the co-source carrier information of the two and divide the co-source relationship level into the same core carrier or the same secondary association carrier. The closer the level, the higher the association and the higher the quantitative score is assigned. Finally, take the quantitative score under the corresponding association type as the physical matching degree.

[0024] For each node in the set of logically associated candidate parent nodes, an initial score for the logical dimension is calculated by multiplying the logical association weight coefficient by the logical matching degree of the candidate parent node. In this embodiment, the logical matching degree is quantitatively determined by calling a preset alarm logic chain rule library. The specific steps are as follows: First, determine whether the association type between the alarm type corresponding to the candidate parent node and the alarm type of the current new device status event is a causal logical relationship or a sequential logical relationship. If it is a causal logical relationship, the probability of the preceding alarm type inducing the subsequent alarm type is statistically calculated based on the historical fault diagnosis data. The higher the probability, the stronger the causal relationship. If it is a sequential logical relationship, the fault propagation path length and the time interval between the two in historical events are combined. The shorter the path and the smaller the time interval, the stronger the sequential relationship. Then, according to the preset strength level and quantitative score mapping rules, a specific quantitative score is assigned to the strength of the corresponding association type. This score is the logical matching degree.

[0025] For each node in the time-related candidate parent node set, an initial score for the time dimension is calculated by multiplying the time-related weight coefficient by the time matching degree of the candidate parent node. In this embodiment, the time matching degree is determined by timestamp comparison and quantitative scoring. The specific steps are as follows: First, extract the timestamp of the candidate parent node and the timestamp of the current new device status event, and calculate the absolute value of the time difference between the two; then check whether the time difference is within a preset time neighborhood. If it exceeds the neighborhood, the time matching degree is assigned a value of 0; if it is within the neighborhood, based on the preset time difference and matching quantification mapping rules, a specific quantitative score is assigned according to the principle that the smaller the time difference, the higher the matching degree. This score is the time matching degree. The intersection of the candidate parent node sets for physical association, logical association, and temporal association is performed to identify single nodes that exist only in a single set, double-intersection nodes that exist in two sets, and triple-intersection nodes that exist in three sets. The set affiliation of each node is then labeled. The overall matching score of a single node is the initial score of its respective set in a single dimension. The overall matching score of a double-intersection node is the sum of the initial scores of its two respective sets in a single dimension. The overall matching score of a triple-intersection node is the sum of the initial scores of its three respective sets in a single dimension. Sort all candidate parent nodes by their overall matching scores in descending order, and select the node with the highest score as the optimal parent node. S6. Based on the preset tree structure dynamic maintenance strategy, the current new device status event is attached as a child node to the optimal parent node. It should be further explained that the tree structure dynamic maintenance strategy of this embodiment is constructed by sorting out the node hierarchy rules of the fault association tree, the child node attachment standard, and the creation logic of virtual aggregation nodes in multi-source association scenarios. It is combined with the device fault propagation characteristics and tree structure integrity requirements for structured design. It is used to accurately attach the current new device status event as a child node to the optimal parent node. When there are multiple parent nodes that meet the conditions, a virtual aggregation node is created to represent the complex fault scenario of multi-source association, ensuring that the fault association tree can be dynamically updated in real time and accurately present the fault propagation path and association relationship.

[0026] S7. If there are multiple candidate parent nodes whose matching scores meet the preset threshold, a virtual convergence node is created according to the fault propagation probability model. The multiple candidate parent nodes are associated with the current new device status event child nodes through the virtual convergence node to represent the complex fault scenario of multi-source association and realize the real-time update of the fault association tree. It should be further explained that the fault propagation probability model in this embodiment is used to quantify the contribution of multiple candidate parent nodes that meet preset thresholds to the fault propagation of the current new device state event, supporting the creation of virtual convergence nodes and the establishment of association relationships between multiple parent nodes and child nodes. The specific construction process includes: S71. Collect historical fault data of multi-lane equipment at highway toll stations, including fault occurrence time, fault equipment identification, alarm type, fault propagation path, and status change data of related equipment; S72. Based on the collected historical fault data, extract fault propagation correlation features, including the physical topology correlation strength of the equipment, the alarm logic chain relationship type, the fault propagation time interval, and the historical co-occurrence frequency. S73. Using a statistical learning algorithm, with the extracted fault propagation correlation features as input and the actual propagation probability of multiple parent nodes to subnet faults in historical faults as labels, a fault propagation probability calculation model is trained to determine the influence weight of each correlation feature on the propagation probability. S74. Validate the trained model using newly collected fault diagnosis data, and adjust the model parameters and feature weights based on the actual fault propagation situation to improve the accuracy of propagation probability calculation.

[0027] S8. Based on the periodic tree structure optimization mechanism, the fault association tree is structurally reviewed and optimized according to a preset period, specifically including: If the device status corresponding to any child node in the fault association tree recovers to normal within several consecutive monitoring cycles, and its status feature vector remains below a preset stable threshold, then the child node is archived or weakened. The weakening mark includes reducing the node display weight or adding a recovery identifier. If multiple child nodes attached to the same parent node are detected to exhibit highly similar state feature patterns, and the similarity meets a preset similarity threshold, then a node merging operation is triggered to generate a super node representing the common fault pattern. The super node carries the feature summary information of the common fault pattern and the association and tracing information of each original child node. The topology of the fault association tree is updated in real time to ensure that the optimized tree structure accurately reflects the current fault association status of the equipment. At the same time, optimization operation logs are recorded, including optimization execution time, optimized node identifier, optimization type (archiving / weakening / merging), optimization trigger condition parameters, and description of tree structure changes after optimization, which are used for fault diagnosis tracing and optimization mechanism parameter iteration.

[0028] It should be further explained that the construction of the multi-dimensional collaborative diagnostic query vector in this embodiment includes: The process of determining the current focus node in the fault association tree in this embodiment includes: Step 1: Collect real-time alarm signals of all devices corresponding to all nodes in the fault association tree, and extract nodes whose alarm intensity exceeds the preset alarm threshold as initial candidate nodes; Step 2: Synchronously retrieve historical data of each initial candidate node, including comprehensive matching score, node type and optimization status, and remove nodes whose comprehensive matching score is lower than the preset score threshold, are in the archived state or have been weakened, to obtain the first round of candidate node set; Step 3: Perform association integrity verification on each node in the first round of candidate node set, query the node level record of the fault association tree, confirm that each node has a parent node (except the root node) and at least one child node, or although the number of child nodes is zero, the feature information of physical association, logical association and time association dimensions is complete and traceable, remove the nodes whose association information missing rate exceeds the preset missing threshold, and form the second round of candidate node set. Step 4: If the second round of candidate node set contains only one node, directly use it as the current focus node; if it contains multiple nodes, call the fault propagation probability model, input the physical matching degree, logical matching degree, temporal matching degree of each candidate node and the common fault mode features of the super node (if it is a super node), and calculate the fault propagation core degree of each candidate node, where fault propagation core degree = physical matching degree × physical weight coefficient + logical matching degree × logical weight coefficient + temporal matching degree × temporal weight coefficient + super node feature matching degree × super node weight coefficient, where the super node feature matching degree is 0 when it is not a super node; Step 5: Calculate the fault impact range associated with the device topology identifier of each candidate node, that is, query the number and level of devices directly or indirectly connected to the node based on the device physical topology knowledge base, and count the total number of devices potentially affected by the fault. Step 6: Sort the nodes by fault propagation core in descending order, and sort the nodes with the same core in descending order by fault impact range. Select the node at the top of the sorted list as the current focus node.

[0029] Based on each node in the target node set, state feature extraction and standardization are performed to obtain a standardized feature set. The standardized feature set includes: node state feature vector, node association features, and node optimized state features. The node state feature vector includes key equipment operation parameters, hardware component status data, communication connection indicators, and performance characterization data. The node association features include the physical matching degree, logical matching degree, and temporal matching degree between the focused node and its parent node, the feature similarity between the focused node and its child nodes, and the summary information of common fault mode features corresponding to the super node. The node optimized state features include whether the node is archived, whether it is weakened (including display weight and recovery identifier), whether it is a merged super node, and the correlation degree of the original child node traceability information. The features in the standardized feature set are mapped to pre-divided physical association dimension, logical association dimension, temporal association dimension, node attribute dimension and optimization state dimension according to their attributes, forming feature subsets corresponding to each dimension; The physical association dimension is used to map the device topology identifier association, physical matching degree, device physical connection tightness, and common source relationship hierarchy features between the current focus node and its parent node and all child nodes in the fault association tree. The logical association dimension is used to map the alarm type, logical matching degree, and alarm logical chain relationship type corresponding to the alarm flow identifier between the current focus node and its parent node and all child nodes in the fault association tree. The alarm logical chain relationship type includes causal logical relationship and sequential logical relationship. The time association dimension is used to map the timestamp, time difference, time matching degree, and time within a preset time neighborhood between the current focus node and its parent node and all child nodes in the fault association tree. Node activity features; the node attribute dimension is used to map the node type of the current focused node, its parent node, and all child nodes in the fault association tree. The node type includes root node, ordinary child node, virtual convergence node, and super node. In addition, it also includes mapping the single-dimensional initial score and comprehensive matching score of each node, as well as the core feature parameters of the common fault mode corresponding to the super node; the optimization state dimension is used to map the archived state and weakened marking state of the current focused node, its parent node, and all child nodes in the fault association tree. The weakened marking state includes display weight and recovery flag, and also maps the similarity threshold compliance features when multiple child nodes attached to the same parent node trigger the merging operation.

[0030] Following a preset order from the physical association dimension to the optimized state dimension, feature subsets of each dimension are sequentially concatenated, and the initial multidimensional collaborative diagnostic query vector is optimized using the L2 normalization algorithm to obtain the multidimensional collaborative diagnostic query vector.

[0031] It should be further explained that this embodiment matches the multi-dimensional collaborative diagnostic query vector with a preset pluggable diagnostic knowledge base, including: B1. Based on the preset plug-in diagnostic knowledge base, obtain the plug-in knowledge unit library after structured preprocessing; wherein, the fault diagnosis knowledge in the plug-in diagnostic knowledge base is divided into independent plug-in knowledge units according to different fault types and scenarios, each plug-in knowledge unit is bound to a dedicated diagnostic analysis plug-in, and is configured with feature tag sets and matching threshold ranges corresponding to each dimension of the multi-dimensional collaborative diagnostic query vector. B2. Based on the multidimensional collaborative diagnostic query vector, obtain a target query vector with standardized format and aligned dimensions; wherein, the multidimensional collaborative diagnostic query vector is sorted into elements according to a preset dimensional order, missing feature values ​​are filled in, and normalization processing is performed on all elements to make its dimensional order consistent with the feature tag set order of the plugin knowledge unit library. B3. Based on the target query vector, obtain a preliminary subset of knowledge units. Specifically, perform dimensional analysis on the target query vector, extract the device topology identifier (e.g., "Switch A-Port1") in the physical association dimension and the alarm type code (e.g., "ERR_Comm_Timeout") in the logical association dimension as key character features, and construct a structured query keyword combination. Subsequently, call the inverted index mechanism of the plug-in knowledge unit library, and use an efficient retrieval algorithm based on a hash mapping table. Using the above structured query keyword combination as input, match all preset feature tag sets of plug-in knowledge units in parallel, and filter out the set of units that contain both physical and logical dimension keywords. If the retrieval result is an empty set, trigger an auxiliary filtering process: extract the features in the time association dimension of the target query vector, including the time window identifier (e.g., "T_Last_30s") and the node activity level, as a new keyword combination, and re-traverse the plug-in knowledge unit library for secondary matching. Finally, take the union of the sets obtained from the two retrieval operations to generate a preliminary subset of knowledge units for use in the subsequent precise matching stage.

[0032] B4. Based on the initially screened subset of knowledge units and the target query vector, obtain the comprehensive similarity score of each plug-in knowledge unit. Specifically, use the cosine similarity algorithm to calculate the single-dimensional similarity between the target query vector and the feature tag set of each plug-in knowledge unit in each dimension, and combine it with preset weight coefficients for each dimension to calculate the weighted comprehensive similarity. If the target query vector contains super node features, then the corresponding preset weight is added. Further, in this embodiment, the cosine similarity algorithm is used to calculate the single-dimensional similarity between the target query vector and the feature tag set of each plug-in knowledge unit in each dimension, and the weighted comprehensive similarity is calculated with preset weight coefficients for each dimension, including: B41. Based on the target query vector and the feature tag set of the pluggable knowledge unit to be calculated, obtain feature subsets of each preset dimension; the preset dimensions include physical association dimension, logical association dimension, time association dimension, node attribute dimension and optimization state dimension; the specific feature elements in the feature subset correspond one-to-one with the features mapped to each dimension when constructing the multi-dimensional collaborative diagnostic query vector; B42. Based on feature subsets of each dimension, the cosine similarity algorithm is used to calculate the single-dimensional similarity scores between the target query vector and the pluggable knowledge unit feature tag set in the physical association dimension, logical association dimension, temporal association dimension, node attribute dimension, and optimization state dimension; wherein the cosine similarity calculation of each dimension is performed based on the vector composed of all feature elements under that dimension. B43. Based on the preset weight coefficients for each dimension, obtain the basic score of the weighted comprehensive similarity; specifically, multiply each single-dimensional similarity score by the preset weight coefficients for physical association dimension, logical association dimension, temporal association dimension, node attribute dimension, and optimization state dimension, and sum all the product results to obtain the basic weighted comprehensive similarity score; the sum of the weight coefficients for each dimension is 1; B44. Based on the node attribute dimension features of the target query vector, determine whether it contains the core feature parameters of the common fault mode corresponding to the super node. B45. If included, obtain the super node feature matching degree and its preset weight coefficient, and calculate the super node feature contribution value; the super node feature matching degree is obtained by calculating the cosine similarity between the common fault mode features of super nodes in the target query vector and the feature tag set of the plug-in knowledge unit corresponding to the feature of the same type; the super node feature contribution value is the product of the super node feature matching degree and the preset super node weight coefficient. B46. Based on the basic weighted comprehensive similarity score and the super node feature contribution value, calculate the final weighted comprehensive similarity; specifically: add the basic weighted comprehensive similarity score and the super node feature contribution value to obtain the final weighted comprehensive similarity of the plug-in knowledge unit; if the target query vector does not contain super node features, then the super node feature contribution value is 0, and the final weighted comprehensive similarity is equal to the basic weighted comprehensive similarity score.

[0033] B5. Based on the comprehensive similarity score and the matching threshold range of the plug-in knowledge unit, obtain a valid matching knowledge unit ranking list; wherein, firstly, plug-in knowledge units whose comprehensive similarity scores do not fall within their own matching threshold range are removed, then the feature label set of the remaining knowledge units is checked to see if it conflicts with the actual state of the current fault association tree target node set, and finally the verified ones are arranged in descending order of comprehensive similarity score. B6. Based on the effective matching knowledge unit ranking list, obtain the final hit dedicated diagnostic analysis plugin. Specifically, select the plugin bound to the knowledge unit ranked first as the hit plugin. If there are ties, call the fault propagation probability model to calculate the fault propagation contribution and select the plugin bound to the one with the highest contribution. B7. Based on the matching information and results obtained from B1 to B6, obtain the updated plug-in diagnostic knowledge base; in this process, record the matching process and diagnostic application effects, feed back the new feature associations of successful matches to the knowledge base to update the existing plug-in knowledge units, and create new plug-in knowledge units and bind plug-ins for new fault scenarios that fail to match.

[0034] It should be further explained that this embodiment performs graph calculation and causal analysis on fault association trees that cross lanes and cross equipment, including: D1. Based on the matched dedicated diagnostic analysis plugin, obtain the initialized collaborative diagnostic reasoning engine and the injected current diagnostic context. The collaborative diagnostic reasoning engine completes its core initialization by loading the pre-set domain analysis strategies and reasoning rules within the dedicated diagnostic analysis plugin. The current diagnostic context includes a complete fault association tree topology, the multi-dimensional collaborative diagnostic query vector, real-time and historical status data of all nodes in the target node set, and the association information between the equipment physical topology knowledge base and the alarm logic chain rule base. A specific implementation example scenario can be understood as follows: In a multi-lane equipment collaborative fault diagnosis scenario at a highway toll station, the system monitors the operating status of each lane's equipment in real time through the acquisition module. When three associated alarm events are detected—communication interruption for lane 3 controller, no response from lane 4 automatic barrier machine, and black screen for lane 3 toll display—the matching module matches the constructed multi-dimensional collaborative diagnostic query vector with the preset plug-in diagnostic knowledge base, ultimately matching the dedicated diagnostic analysis plugin adapted for multi-lane equipment linkage fault diagnosis. This dedicated diagnostic analysis plugin pre-configures exclusive content for the physical connection logic and fault propagation characteristics between multi-lane equipment in toll stations. The core consists of two parts: domain analysis strategy and inference rules. The domain analysis strategy clarifies the analysis framework for cross-lane and cross-equipment faults, covering node association identification, path weight calculation, causal verification standards, etc. The inference rules sort out the causal and temporal relationships and matching judgment conditions between common alarm types in toll stations, providing core basis for subsequent engine initialization.

[0035] The collaborative diagnostic inference engine initiates its core initialization process based on the matched dedicated diagnostic analysis plugin. This initialization process loads pre-defined content and configures engine modules through the plugin's standardized interface. Specifically: First, the engine calls the plugin's built-in domain analysis strategy loading interface to read the analysis logic for multi-lane equipment linkage faults at toll stations. This includes abstract rules for fault association tree nodes to weighted directed graph nodes, weight allocation across dimensions, standard graph calculation execution processes, etc., and configures these rules into the engine's graph calculation component and strategy scheduling module. Next, the engine loads the pre-defined inference rules within the plugin. These rules include causal judgment conditions for communication interruption alarms and non-response alarms, correlation verification rules for device faults and controller status, and timing matching requirements for multi-node alarms, constructing a temporary rule set which is then injected into the engine's rule inference component. Simultaneously, the engine completes the initialization configuration of its core algorithm components, enabling it to possess dedicated diagnostic capabilities for the current multi-lane linkage fault, ensuring that subsequent analysis processes are tailored to the scenario's characteristics.

[0036] After engine initialization, the integration and injection of the current diagnostic context are performed synchronously. This context is formed by the collaborative integration of data from multiple system modules, comprehensively covering the topological data, feature data, knowledge base, and related data required for diagnosis. Firstly, there is a complete fault association tree topology structure, generated by the topology construction module. This structure uses the communication interruption of lane 3 controller as the root node, the unresponsive automatic barrier machine in lane 4, and the black screen of the toll display in lane 3 as child nodes. It clearly records the device topology identifier, alarm flow identifier, timestamp, and parent-child connection relationship of each node, fully representing the initial fault association state. Secondly, there is a multi-dimensional collaborative diagnostic query vector, constructed by the matching module based on the current focused node, lane 3 controller, and its parent and child nodes. After state feature extraction, standardization, and multi-dimensional mapping, it covers five dimensions of features: physical association, logical association, time association, node attributes, and optimized state, accurately extracting the core features of the target node set. Thirdly, there is the real-time and historical status data of the target node set. Real-time data includes communication connection indicators, hardware component operating parameters, and alarm intensity of the corresponding devices for each node. Historical data includes the trajectory of the comprehensive matching score status change of each node in the previous monitoring period, fault recovery records, etc., providing data support for trend analysis and anomaly verification. Fourthly, there is the association information between the device physical topology knowledge base and the alarm logic chain rule base. The system calls the device physical topology knowledge base to obtain the direct physical connection method between lane controller No. 3 and automatic barrier machine No. 4, and the common power supply relationship between lane controller No. 3 and toll display No. 3. It calls the alarm logic chain rule base to obtain the causal logic strength of communication interruption alarm and no response alarm, and the sequential logic association basis of communication interruption alarm and black screen alarm. The above knowledge base information is integrated and injected into the engine as supplementary context content.

[0037] The initialization of the collaborative diagnostic inference engine and the injection of the current diagnostic context establish a diagnostic foundation adapted to the current multi-lane linkage fault. The pre-built domain analysis strategies and inference rules in the plugin provide dedicated methodological support for the diagnosis, ensuring that the analysis direction aligns with the fault propagation patterns of the toll station equipment. The injected context data covers all dimensions of diagnostic elements, including topological feature time-series knowledge bases, providing complete data input for subsequent steps. This process effectively solves the problems of insufficient adaptability and inadequate data support in general diagnostic engines, enabling the engine to accurately focus on the current fault scenario. This lays a solid foundation for subsequent steps such as constructing a weighted directed graph model based on the fault association tree to perform node centrality analysis and conduct cross-device causal verification, ensuring the accuracy and efficiency of the subsequent diagnostic process.

[0038] It should be further explained that the collaborative diagnostic inference engine in this embodiment is not built from a single algorithm, but rather by integrating and collaboratively scheduling multiple mature algorithm components and strategy frameworks. Its core construction method includes: building a graph computation component based on graph theory algorithms such as Dijkstra's shortest path algorithm, PageRank, or BetweennessCentrality to analyze the weighted directed graph model abstracted from the fault association tree; integrating a rule-based inference engine such as Drools as a rule inference component to execute the causal determination rule library loaded by a dedicated diagnostic analysis plugin; and introducing statistical methods such as Granger causality tests and temporal correlation analysis as causal analysis components to quantify temporal causal relationships. The engine's "collaborative" capability originates from its internal workflow scheduler, which sequentially calls the above components according to predefined diagnostic logic and integrates related information such as the device physical topology knowledge base and alarm logic chain rule base, thereby achieving graph computation, causal inference, and root cause localization for complex faults across devices.

[0039] D2. Based on the complete fault association tree topology and the analysis strategy specified by the dedicated diagnostic analysis plugin, an abstractly constructed weighted directed graph model is obtained. Nodes in the fault association tree are abstracted as nodes in the weighted directed graph model, and parent-child relationships in the fault association tree are abstracted as directed edges in the weighted directed graph model. The weight of each directed edge is comprehensively constructed based on the physical matching weight, logical matching weight, and temporal matching weight between the nodes corresponding to the directed edge. It should be further noted that in this embodiment, in the scenario of collaborative fault diagnosis of multi-lane equipment at highway toll stations, the topology construction module has completed the construction of the fault association tree. This fault association tree contains three core nodes: the communication interruption node for lane 3 controller, the unresponsive node for lane 4 automatic barrier machine, and the black screen node for lane 3 toll display. The communication interruption node for lane 3 controller is the parent node, and the unresponsive node for lane 4 automatic barrier machine and the black screen node for lane 3 toll display are its child nodes, forming a clear parent-child relationship. At this point, the system has been matched with a dedicated diagnostic analysis plugin for multi-lane equipment linkage fault diagnosis. The analysis strategy specified by this plugin includes graph model abstraction rules and weight calculation logic. The graph model abstraction rules clarify the mapping standard between fault association tree nodes and weighted directed graph nodes. The weight calculation logic stipulates that physical matching degree weight, logical matching degree weight, and time matching degree weight must be integrated to construct directed edge weights. Moreover, the weight coefficients of each dimension are determined based on historical fault diagnosis data and equipment fault propagation patterns. The sum of all weight coefficients is 1, providing a clear basis for the abstract construction of the weighted directed graph model.

[0040] Based on the graph model abstraction rules specified by the dedicated diagnostic analysis plugin, the process of abstracting fault association tree nodes into weighted directed graph nodes is initiated. The abstraction process is based on the core attributes of the fault association tree nodes, ensuring that the abstracted graph nodes fully retain the fault association characteristics of the original nodes. Specifically, the communication interruption node of lane 3 controller in the fault association tree is abstracted into the corresponding graph node in the weighted directed graph; the unresponsive automatic barrier gate node of lane 4 is abstracted into the corresponding graph node in the weighted directed graph; and the black screen node of the toll display of lane 3 is abstracted into the corresponding graph node in the weighted directed graph. Each abstracted graph node carries the core information of the original fault association tree node, including the device topology identifier, alarm flow identifier, timestamp, and state feature vector. For example, the graph node corresponding to the communication interruption of the controller in lane 3 carries its device identifier, alarm type "communication interruption", and the timestamp of the corresponding event. The graph node corresponding to the non-responsive automatic barrier machine in lane 4 carries its device identifier, alarm type "non-responsive", and the timestamp of the corresponding event. The graph node corresponding to the black screen of the toll display in lane 3 carries its device identifier, alarm type "black screen", and the timestamp of the corresponding event. This ensures that the attributes of the graph nodes correspond one-to-one with the attributes of the original fault association tree nodes, providing data support for subsequent directed edge construction and weight calculation.

[0041] Based on the parent-child relationship in the fault association tree and the analysis strategy specified by the plugin, the abstract construction of directed edges is completed. The core of the directed edge abstraction is mapping the direction of fault propagation, that is, the fault propagates from the parent node to the child node. Therefore, the direction of the directed edge is set from the abstracted parent graph node to the child graph node. For the parent-child relationship between the communication interruption node of lane 3 controller and the unresponsive automatic barrier machine node of lane 4 in the fault association tree, it is abstracted into a directed edge in the weighted directed graph from the graph node corresponding to the communication interruption of lane 3 controller to the graph node corresponding to the unresponsive automatic barrier machine of lane 4. For the parent-child relationship between the communication interruption node of lane 3 controller and the black screen node of the toll display of lane 3, it is abstracted into a directed edge in the weighted directed graph from the graph node corresponding to the communication interruption of lane 3 controller to the graph node corresponding to the black screen of the toll display of lane 3. Each directed edge is associated with its corresponding original parent-child node connection relationship identifier, which is used to trace the actual relationship type between nodes during subsequent weight calculation, ensuring that the directed edge can accurately represent the direction of fault propagation and the source of the relationship.

[0042] Following the weight calculation logic specified by the plugin, the weight of each directed edge is constructed in three steps. The weight calculation result ranges from 0 to 1, with higher values ​​indicating stronger correlation in fault propagation between nodes. The first step obtains the physical matching degree weight by calling a preset device physical topology knowledge base to query the physical connection relationship between the directed edge and the original node device. For the directed edge representing the connection between lane 3 controller and lane 4 automatic barrier machine, it is found that they are directly connected through the corresponding bus, and the corresponding physical connection tightness quantification value is obtained as the physical matching degree weight. For the directed edge representing the connection between lane 3 controller and lane 3 toll display, it is found that they belong to the same power distribution cabinet and are indirectly connected through the corresponding network, and the corresponding physical connection tightness quantification value is obtained as the physical matching degree weight. The second step obtains the logical matching degree weight by calling a preset alarm logic chain rule base to query the logical relationship of the directed edge corresponding to the original node alarm type. For the directed edge representing the association between the communication interruption of the lane 3 controller and the lack of response of the automatic barrier gate in lane 4, the alarm types corresponding to the two in the knowledge base have a clear causal logical relationship, and the corresponding logical causal strength quantification value is obtained as the logical matching degree weight. For the directed edge representing the association between the communication interruption of the lane 3 controller and the black screen of the toll display in lane 3, the alarm types corresponding to the two in the knowledge base have a sequential logical relationship, and the corresponding logical association strength quantification value is obtained as the logical matching degree weight. The third step is to obtain the time matching degree weight. Based on the timestamp of the original node corresponding to the directed edge, the time difference is calculated. Combined with the preset time difference and matching quantification mapping rules, the time matching degree weight is determined according to the principle that the smaller the time difference, the higher the matching degree. Finally, the final weight of each directed edge is calculated using a linear weighted summation formula: Weight = Physical matching degree weight × Preset physical association dimension weight + Logical matching degree weight × Preset logical association dimension weight + Time matching weight × Preset time association dimension weight, thus completing the calculation of the final weight of each directed edge.

[0043] The abstracted graph nodes corresponding to each device event, the directed edges representing the direction of fault propagation, and the calculated edge weights are integrated to form a complete weighted directed graph model. This model includes the graph node corresponding to the communication interruption of the lane 3 controller, the unresponsive automatic barrier gate of lane 4, and the black screen of the toll display in lane 3, as well as two directed edges pointing from the node corresponding to the communication interruption of the lane 3 controller to the other two graph nodes. Each directed edge carries a corresponding comprehensive calculated weight. This model fully preserves the topological structure of the original fault association tree and the fault association characteristics between nodes. Simultaneously, it quantifies the association strength of fault propagation between nodes through weights, and can be directly used as input to the graph calculation component of the subsequent collaborative diagnostic inference engine. This supports subsequent diagnostic steps such as node centrality analysis and identification of core suspected fault propagation chains, ensuring that subsequent analysis can be conducted based on accurate graph model data. D3. Based on the weighted directed graph model and the graph computation component of the collaborative diagnostic reasoning engine, obtain the node centrality analysis results and the core suspected fault propagation chain; wherein, by performing node centrality analysis, calculate and obtain the degree centrality, proximity centrality, and between centrality indices of each node in the weighted directed graph model; sort the nodes according to the centrality indices to identify and obtain the key hub nodes and edge nodes in the fault propagation network; at the same time, perform path search based on the weight of the directed edges to obtain the top N paths with the highest total path weight from the root node to each leaf node as the core suspected fault propagation chain.

[0044] For example, in the practical application scenario of multi-lane equipment collaborative fault diagnosis at highway toll stations, the topology construction module has completed the construction and maintenance of the fault association tree. Assuming that the fault association tree in this embodiment takes the communication interruption event of lane 3 controller as the root node, the child nodes include the unresponsive event of the automatic barrier gate of that lane, the black screen event of the toll display, and the communication abnormality event of lane 4 controller. All of the above nodes are derived from the equipment status events generated by the acquisition module. Each node carries the corresponding equipment topology identifier timestamp and status feature vector, and the parent node matching and attachment have been completed through the multi-dimensional association rule engine to form a clear fault association topology.

[0045] Based on the existing fault association tree topology and the analysis strategy specified by the targeted diagnostic analysis plugin, the abstract construction of a weighted directed graph model is initiated. First, each device status event node in the fault association tree is abstracted one-to-one into a node in the weighted directed graph model, ensuring that the node attributes are consistent with the original fault association tree nodes. Then, the parent-child connection relationships representing the fault propagation direction in the fault association tree are abstracted into directed edges in the weighted directed graph, with the direction of the directed edges consistent with the fault propagation direction, i.e., from the parent node to the child node. Finally, the weight of each directed edge is calculated using a linear weighted summation algorithm, with physical matching degree weight, logical matching degree weight, and time matching degree weight as input. The weight coefficients are consistent with the coefficients used in the parent node matching calculation above, and their sum is 1. This algorithm yields the final weight of each directed edge, completing the construction of the weighted directed graph model.

[0046] The collaborative diagnostic inference engine utilizes the built-in graph computation component and employs classic graph theory algorithms to perform node centrality analysis on the constructed weighted directed graph model. Degree centrality is directly obtained by counting the number of directed edges connecting a node, reflecting the tightness of its connection within the fault propagation network. Proximity centrality is calculated using Dijkstra's shortest path algorithm, taking the inverse of the average shortest path length, reflecting the ease with which a node reaches other nodes in the network. Since centrality is calculated using Brandes' algorithm, iterates through all node pairs and counts the frequency of the target node on these paths, reflecting the node's control over the fault propagation path. Based on the weighted sum of these three indicators, all nodes are ranked. Lane controller node 3 is identified as a critical hub node, possessing the highest weighted sum of its three centrality indicators, playing a core supporting role in fault propagation across equipment and lanes. Lane controller node 4 is an edge node, with the lowest weighted sum of its three centrality indicators, only affected by association and not becoming a critical node for fault propagation.

[0047] After completing the node centrality analysis, the graph computing component performs path search using a variant of Dijkstra's shortest path algorithm, based on the weights of directed edges in the weighted directed graph model. This variant algorithm aims to maximize the sum of path weights. Starting from the root node of the fault association tree (the communication interruption node for lane 3 controllers), it traverses to all leaf nodes (the unresponsive automatic barrier gate node, the black screen node for the toll display, and the communication anomaly node for lane 4 controllers). The algorithm calculates the sum of weights for each path and selects the two paths with the highest sums: Lane 3 controller communication interruption → Unresponsive automatic barrier gate → Black screen for lane 4 controllers. These two paths are identified as the core suspected fault propagation chain. The results of this node centrality analysis, along with the core suspected fault propagation chain, provide crucial data support for the subsequent collaborative diagnostic inference engine to perform causal analysis, locate the root cause of the fault, and define the scope of impact, ensuring the accuracy and relevance of the diagnostic results.

[0048] D4. Based on the pre-set causal judgment rule base of the plugin and the multi-dimensional features of the nodes in the fault association tree, obtain the causal analysis results across lanes and across devices to form a fault causal network with confidence weights. Specifically: based on the device physical topology knowledge base, analyze whether there is a definite physical connection fault transmission relationship between nodes on the core propagation path; based on the alarm logic chain rule base, verify whether the logical sequence of alarm events on the path conforms to the preset causal paradigm; based on the timestamp sequence, use Granger causality test or time-series correlation analysis statistical methods to analyze the leading and lagging relationship of node state changes in time; and comprehensively analyze the results of the physical, logical, and temporal dimensions to score the confidence of the causal relationship between nodes, thus forming the fault causal network.

[0049] For example, in the scenario of collaborative fault diagnosis of multi-lane equipment at a highway toll station, the core suspected fault propagation chain has been obtained through the graph calculation and analysis described above, namely: communication interruption of lane 3 controller → no response of automatic barrier gate → black screen of toll display. At this time, the collaborative diagnosis inference engine loads the hit dedicated diagnosis analysis plugin. This plugin has a pre-built causal judgment rule base, which contains causal association rules for common faults of toll station equipment, such as abnormal controller communication easily causing unresponsive downstream execution equipment, and faults of toll display equipment often being related to power supply or communication links. At the same time, multi-dimensional features of each node on the core propagation path in the fault association tree are extracted, including the equipment topology identifier, physical matching degree, logical matching degree, timestamp, and state feature vector corresponding to the node, providing complete data input for subsequent multi-dimensional causal analysis, clarifying that the analysis object is the relationship between nodes on the core propagation path, and the goal is to form a fault causal network with confidence weights.

[0050] A physical connection fault propagation analysis was conducted based on a device physical topology knowledge base. The knowledge base was used to query the physical connection information of each node on the core propagation path, confirming that the No. 3 lane controller and the automatic barrier gate have a direct hardware interface connection via RS485 bus communication; the No. 3 lane controller and the fare display are powered by the same distribution cabinet and have an indirect communication link. The analysis focused on verifying the integrity of the device physical connections and the possibility of fault propagation, determining whether a communication interruption in the No. 3 lane controller could be propagated to the automatic barrier gate through a direct physical connection, causing it to become unresponsive, or whether it could be propagated to the fare display through shared power supply or an indirect communication link, causing it to go black. The analysis confirmed a definite physical connection fault propagation relationship between the No. 3 lane controller and the automatic barrier gate; and a physical connection fault propagation relationship between the No. 3 lane controller and the fare display based on shared power supply and indirect communication. The physical dimension analysis results were recorded as either "existence" or "non-existence" of a definite propagation relationship.

[0051] The alarm logic chain rule base is used to verify whether the logical sequence of alarm events conforms to a preset causal paradigm. This knowledge base contains pre-defined templates for causal and sequential logical relationships between alarm types. For example, the preset causal paradigm includes communication interruption alarms taking precedence over device unresponsive alarms and communication anomaly alarms taking precedence over display anomaly alarms. The alarm types corresponding to the alarm flow identifiers of each node on the core propagation path are extracted, and the order of alarm events is analyzed as follows: the communication interruption alarm for lane 3 controller is triggered first, followed by the automatic barrier machine unresponsive alarm and the toll display black screen alarm. This order is compared with the preset causal paradigm in the knowledge base, verifying that the logical sequence of the communication interruption alarm for lane 3 controller and the automatic barrier machine unresponsive alarm conforms to the preset causal paradigm. The logical sequence of the communication interruption alarm for lane 3 controller and the toll display black screen alarm also conforms to the preset causal paradigm. The logical dimension analysis results are recorded as conforming to or not conforming to the preset causal paradigm.

[0052] Based on the timestamp sequences of each node, the Granger causality test was used to analyze the lead-lag relationship of node state changes. The Granger causality test is a statistical method used to determine whether a change in one time-series variable can predict a change in another. If a change in variable A can predict a change in variable B, then A is considered a Granger cause of B. Timestamp data for each node along the core propagation path were extracted. The timestamp for the communication interruption of lane 3 controller was T1, the timestamp for the automatic barrier gate's lack of response was T2, and the timestamp for the toll display's blackout was T3. T2 is 2 seconds later than T1, and T3 is 3 seconds later than T1. The Granger causality test was performed using the timestamp sequences as input. The test results showed that the communication interruption of lane 3 controller was a Granger cause of both the automatic barrier gate's lack of response and the toll display's blackout. This clarified the lead-lag relationship in time for node state changes: the change in lane 3 controller state preceded both the automatic barrier gate and the toll display. The time dimension analysis results indicated the existence or non-existence of a lead-lag causal relationship.

[0053] The confidence score of the causal relationship between nodes is calculated based on the analysis results of the physical, logical, and temporal dimensions. The weight coefficients for each dimension are set to be consistent with those used in the parent node matching calculation mentioned earlier. The sum of the weights for the physical, logical, and temporal dimensions is 1. A linear weighted summation algorithm is used to calculate the confidence score, ranging from 0 to 100. A higher score indicates a more reliable causal relationship between nodes. For example, the physical dimension analysis results for lane controller 3 and the automatic barrier gate show a definite transmission relationship, the logical dimension conforms to the preset causal paradigm, and the temporal dimension shows a lead-lag relationship. After weighted calculation, the confidence score is 85 points, and the overall confidence score for lane controller 3 and the toll display is 80 points. Each node is treated as a node in the fault causal network, and the causal relationship between nodes is treated as a directed edge. The weight of the directed edge corresponds to the confidence score, ultimately forming a fault causal network with confidence weights, clearly presenting the causal relationships and reliability on the core propagation path.

[0054] D5. Based on the core propagation path and the fault causal network, obtain the most likely root cause node and the defined scope of fault impact; specifically, call the root cause inference algorithm encapsulated in the plugin, comprehensively consider the centrality of the node in the weighted directed graph model, the source attribute in the fault causal network, the severity of its abnormal state characteristics, and whether it is a virtual convergence node or a super node, to locate and obtain the root cause node; simultaneously, based on the topology of the fault association tree and the physical connection relationship of the equipment, starting from the root cause node and the key hub node, simulate the fault propagation, obtain the set of affected equipment and the lane range, and assess the potential scale of the fault impact.

[0055] For example, in the scenario of collaborative fault diagnosis of multi-lane equipment at highway toll stations, this embodiment has obtained the core propagation path and fault causal network based on the previous analysis. The core propagation path is manifested in the communication interruption of lane 3 controller causing the automatic barrier gate to become unresponsive and the toll display to go black. In the fault causal network, lane 3 controller has a clear causal relationship with the latter two and the confidence level is high. All nodes are ordinary nodes without virtual convergence nodes or super nodes. At the same time, the node centrality analysis of the weighted directed graph model shows that lane 3 controller has the highest level in the weighted sum of degree centrality, proximity centrality, and between centrality, and is a key hub node in the fault propagation network. This root cause localization uses the core propagation path and fault causal network as the core input and calls the root cause inference algorithm encapsulated in the dedicated diagnostic analysis plugin. This algorithm adopts a multi-feature fusion weighted decision mode. The core logic is to convert the multi-dimensional features of the nodes into quantitative indicators and then calculate the root cause score by weighting. The node with the highest score is the most likely root cause node of the fault. The algorithm first quantifies four core features: the node's central position in the weighted directed graph model, its source attribute in the fault causal network, the severity of its own state anomalies, and its node type. The source attribute focuses on whether the node is at the top of the causal chain and has no other fault causes. The node type determines whether it is a virtual convergence node or a supernode to avoid multiple source overlap. Then, combining the weight allocation logic used in the parent node matching calculation, corresponding weight coefficients are assigned to the four features. Based on the quantification results and weight coefficients, the root cause score for each node is calculated. After sorting, the score of lane 3 controller is significantly higher than that of the automatic barrier gate and the toll display, thus identifying it as the most likely root cause node. In the fault impact range definition stage, starting from the identified root cause node and the key hub node in the weighted directed graph, namely lane 3 controller, a breadth-first search algorithm based on the equipment physical topology is used to simulate fault propagation. The simulation first invoked the equipment physical topology knowledge base to query the direct and indirect physical connection information of the lane 3 controller. Then, taking the communication interruption fault of this node as the cause of the spread, it traversed the related devices layer by layer based on the physical connection relationship, prioritizing the impact probability of directly connected communication devices, and decreasing the impact probability of indirectly connected devices according to the physical connection level. Through simulation calculations, the set of affected devices and the lane range were determined, clarifying that all equipment in lane 3 and the core control equipment in lane 4 were affected by cross-lane issues. Finally, it was concluded that the root cause node should be addressed first to prevent further spread of the fault.

[0056] D6. Based on the node centrality analysis results, the core suspected fault propagation chain, the fault causal network, the root cause node, and the fault impact range, obtain a structured collaborative diagnostic report and a sequence of targeted handling strategies. Specifically, according to the report template defined by the plugin, automatically generate the structured collaborative diagnostic report, which includes the identified root cause node and its device identifier, a complete visual description of the fault propagation path, a sequence of key related events, a confidence level description of the causal relationship, and a list of affected devices and lanes. Then, call the handling strategy library associated with the plugin to generate the sequence of targeted handling strategies, which includes priority maintenance recommendations for the root cause device, instructions for isolating or checking the status of related devices, and temporary operational adjustment schemes to prevent fault spread. Output the diagnostic report and handling strategies to the response module to complete this collaborative diagnostic process.

[0057] For example, in the scenario of collaborative fault diagnosis of multi-lane equipment at highway toll stations, this embodiment has completed node centrality analysis, identification of core suspected fault propagation chains, construction of fault causal networks, location of root cause nodes, and definition of the scope of fault impact. The core input information is as follows: The node centrality analysis results show that lane 3 controller is the key hub node, with the highest weighted sum of the three centrality indicators; the core suspected fault propagation chain is: lane 3 controller communication interruption → automatic barrier gate unresponsive, lane 3 controller communication interruption → toll display black screen; in the fault causal network, the causal confidence score between lane 3 controller and automatic barrier gate is 85 points, and the causal confidence score between lane 3 controller and toll display is 80 points; the root cause node is lane 3 controller, with equipment identification LC-3-001; the scope of fault impact covers all equipment in lane 3 and lane 4 controller. Based on the above core inputs, this embodiment utilizes a dedicated diagnostic analysis plugin to generate a structured collaborative diagnostic report and construct a targeted handling strategy sequence, ultimately outputting to the closed-loop diagnostic process of the response module.

[0058] The standardized report template built into the dedicated diagnostic analysis plugin is invoked. This template is a pre-set structured format adapted to multi-lane fault diagnosis scenarios at toll stations, containing five fixed modules: root cause information, propagation path, event sequence, confidence level description, and impact list. It supports automatic filling and formatting based on input data. The report generation steps are as follows: First, fill in the root cause node information, clearly identifying the fault root cause node as the No. 3 lane controller, device identifier LC-3-001, and simultaneously marking the core anomaly points in its status feature vector where the communication connection index exceeds the threshold by 50% and the communication module is abnormal; Second, generate a visual description of the fault propagation path, presenting the core suspected propagation chain in text topology form, namely, No. 3 lane controller (root cause node) → direct physical connection / logical causality → automatic barrier gate (no response), No. 3 lane controller (root cause node) → common power supply / indirect communication → toll display (black screen), clearly marking the causal relationship direction corresponding to the path; Third, sort out the key related event sequence. The events are ordered chronologically by timestamp: Lane 3 controller communication interruption (T1), automatic barrier gate no response (T1+2 seconds), and fee display black screen (T1+3 seconds). The alarm stream identifier and status anomaly type of each event are marked simultaneously. Fourth, the confidence level of the causal relationship is supplemented, clarifying the confidence score and calculation basis of each propagation path, which is obtained by weighting the results of the three-dimensional analysis of physical connection, logical paradigm, and time series. Fifth, a list of affected equipment and lanes is provided. The equipment list includes Lane 3 controller, automatic barrier gate fee display, and Lane 4 controller. The lane range is the entire Lane 3 and the core control area of ​​Lane 4. The current operating status of each device is marked.

[0059] The system calls upon the handling strategy library associated with the dedicated diagnostic analysis plugin. This strategy library stores preset handling rules categorized by toll station equipment type, fault type, and scope of impact, and supports dynamic matching of strategies based on the root cause of the fault and the scope of impact. The strategy sequence generation steps are as follows: First, formulate a priority maintenance recommendation for the root cause device. For the communication interruption fault of the No. 3 lane controller, match the communication module fault maintenance rules, and clarify the maintenance process as follows: power off and restart the controller → check the RS485 communication interface connection status → check the power supply voltage of the communication module → replace the faulty communication module. Mark the priority level as the highest level and require maintenance to be started within 10 minutes. Second, generate isolation or status check instructions for related devices. Perform a status check on the toll display of the automatic barrier machine in No. 3 lane. The instructions are to check the physical connection link between the device and the controller, read the hardware status data of the device, and verify the communication signal strength. Perform temporary isolation on the No. 4 lane controller, that is, block its communication link with the No. 3 lane controller to prevent the fault from spreading. Mark the execution time limit as within 5 minutes after the maintenance starts. Third, formulate a temporary operation adjustment plan to prevent the fault from spreading. Match the lane redundancy scheduling rules and instruct the toll station management system to guide vehicles in No. 3 lane to the backup lanes 1 and 2. Activate the backup controller of No. 4 lane to temporarily take over the core control function. Mark the adjustment plan as valid until the root cause device maintenance is completed and normal operation is restored. The above strategies are arranged logically as “root cause repair → related inspection → spread prevention”, forming a sequence of targeted treatment strategies.

[0060] The generated structured collaborative diagnostic report and targeted handling strategy sequences are standardized and converted into JSON data format recognizable by the response module. The report content is encoded according to the field codes specified in the plugin template, and the strategy sequences are labeled with attributes such as execution priority, execution subject, and time limit requirements. The standardized data is then output to the response module, which distributes it according to a preset mechanism: on the one hand, the structured collaborative diagnostic report is pushed to the toll station monitoring center terminal for a visual interface, allowing maintenance personnel to view detailed diagnostic information; on the other hand, the targeted handling strategy sequences are broken down into device-executable instructions and distributed to the corresponding maintenance terminals and device controllers via IoT edge gateways, driving maintenance personnel to perform maintenance, device isolation, and temporary adjustment operations. Simultaneously, the entire diagnostic process information is recorded, including input data, report content, strategy content, output objects, and execution status, and fed back to the plug-in diagnostic knowledge base. This provides data support for subsequent diagnostic template optimization and strategy iteration, completing the closed loop of this collaborative diagnostic process.

[0061] This embodiment constructs a fault association tree to topologically organize multi-source heterogeneous fault events, and builds a multi-dimensional collaborative diagnostic query vector based on the state features of the current focused node and its parent and child nodes. This provides accurate and structured feature input for the diagnostic process, effectively solving the key problems of complex fault representation and poor correlation in multi-lane equipment. By performing multi-stage matching between the query vector and the plug-in diagnostic knowledge base, the most suitable dedicated diagnostic analysis plug-in for the current complex fault scenario can be quickly matched, realizing accurate invocation of diagnostic strategies and efficient reuse of knowledge, significantly improving the targeting and efficiency of diagnosis. Furthermore, through collaborative diagnosis... The inference engine drives graph computation and cross-dimensional causal analysis, transforming abstract fault correlations into quantifiable weighted directed graph models. It integrates physical, logical, and temporal correlations for deep reasoning and root cause localization, thereby achieving accurate determination from scattered alarms to core root causes and the scope of impact. Finally, the system automatically generates structured diagnostic reports containing clear root causes, propagation paths, confidence assessments, and sequential handling strategies, forming a complete closed loop from intelligent analysis to precise handling. This greatly improves the diagnostic efficiency, decision-making accuracy, and timeliness of operation and maintenance for multi-equipment system faults at highway toll stations, effectively ensuring the continuous and stable operation of toll collection services.

[0062] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the claims. All of these variations are within the protection scope of the present invention.

Claims

1. A collaborative fault diagnosis system for multi-lane equipment at highway toll stations, characterized in that, include: The data acquisition module collects real-time parallel operation data and real-time alarm streams of heterogeneous devices in multiple lanes through the IoT edge gateway of the toll station, and performs standardization and event processing on the data to generate device status events with timestamps, device topology identifiers and status feature vectors. The topology construction module constructs and updates a fault association tree based on the monitoring cycle and equipment topology relationship of the collaborative fault diagnosis system. The matching module constructs a multi-dimensional collaborative diagnostic query vector based on the state characteristics of the current focused node and its parent and child nodes in the fault association tree, and matches the multi-dimensional collaborative diagnostic query vector with a preset plug-in diagnostic knowledge base to match the dedicated diagnostic analysis plug-in applicable to the current associated fault scenario. The response module, through the hit dedicated diagnostic analysis plugin, drives a preset collaborative diagnostic reasoning engine to perform graph calculation and causal analysis on the fault association tree across lanes and equipment, and generates and outputs a collaborative diagnostic report and handling strategy.

2. The multi-lane equipment collaborative fault diagnosis system for highway toll stations as described in claim 1, characterized in that, The fault association tree is specifically as follows: If the current monitoring cycle is the first monitoring cycle after the fault diagnosis system is started, then the fault association tree is initialized with the first abnormal or critical equipment status event as the root node; If it is a continuous monitoring process, based on the physical correlation of the device status events, the alarm logic chain, and the temporal proximity, the newly generated device status events are treated as child nodes and attached to the existing nodes in the fault association tree, and the fault association tree is updated in real time.

3. The multi-lane equipment collaborative fault diagnosis system for highway toll stations as described in claim 2, characterized in that, The process of constructing the fault association tree includes: Based on the first monitoring cycle after the collaborative fault diagnosis system is started, the first device status event generated by the acquisition module is obtained. Based on the preset root node initialization rules, feature analysis is performed on the first device status event. If the status feature vector of the device status event exceeds the preset abnormal threshold, or the alarm flow identifier contained in the device status event belongs to the preset key alarm type, then the device status event is determined as a root node event, and the fault association tree is initialized with the root node event. The root node carries the timestamp, device topology identifier and complete status feature vector corresponding to the root node event. Once the collaborative fault diagnosis system enters the continuous monitoring process, it collects and acquires new equipment status events continuously generated by the acquisition module in real time.

4. The multi-lane equipment collaborative fault diagnosis system for highway toll stations as described in claim 3, characterized in that, The process of constructing the fault association tree also includes: Based on a multi-dimensional association rule engine, parent node matching calculation is performed for each newly generated device status event to obtain a set of candidate parent nodes for association; the set of candidate parent nodes for association includes a set of candidate parent nodes for physical association, a set of candidate parent nodes for logical association, and a set of candidate parent nodes for time association; Weighting and intersection analysis are performed on the physical association candidate parent node set, logical association candidate parent node set, and time association candidate parent node set to calculate the matching degree score of each candidate parent node, and the candidate parent node with the highest matching degree score is selected as the optimal parent node. Based on the preset tree structure dynamic maintenance strategy, the current new device status event is attached as a child node to the optimal parent node; If there are multiple candidate parent nodes whose matching scores meet the preset threshold, a virtual aggregation node is created based on the fault propagation probability model, and the multiple candidate parent nodes are associated with the current new device status event child nodes through the virtual aggregation node.

5. The multi-lane equipment collaborative fault diagnosis system for highway toll stations as described in claim 4, characterized in that, The process of constructing the fault association tree also includes: Based on a periodic tree structure optimization mechanism, the fault association tree is structurally reviewed and optimized according to a preset period, specifically including: If the device status corresponding to any child node in the fault association tree recovers to normal within several consecutive monitoring cycles, and its status feature vector remains below a preset stable threshold, then the child node is archived or weakened. The weakening mark includes reducing the node display weight or adding a recovery identifier. If multiple child nodes attached to the same parent node are detected to exhibit highly similar state feature patterns, and the similarity meets a preset similarity threshold, then a node merging operation is triggered to generate a super node representing a common fault pattern. The super node carries the feature summary information of the common fault pattern and the association and tracing information of each original child node.

6. The multi-lane equipment collaborative fault diagnosis system for highway toll stations as described in claim 5, characterized in that, Construct a multidimensional collaborative diagnostic query vector, including: Determine the current focus node in the fault association tree and obtain the target node set centered on the focus node, the target node set including the current focus node, its parent node and all its child nodes; Based on each node in the target node set, state feature extraction and standardization are performed to obtain a standardized feature set; the standardized feature set includes: node state feature vector, node association feature, and node optimized state feature. The features in the standardized feature set are mapped to pre-divided physical association dimension, logical association dimension, temporal association dimension, node attribute dimension and optimization state dimension according to their attributes, forming feature subsets corresponding to each dimension; Following a preset order from the physical association dimension to the optimized state dimension, feature subsets of each dimension are sequentially concatenated, and the initial multidimensional collaborative diagnostic query vector is optimized using the L2 normalization algorithm to obtain the multidimensional collaborative diagnostic query vector.

7. The multi-lane equipment collaborative fault diagnosis system for highway toll stations as described in claim 6, characterized in that, Matching the multidimensional collaborative diagnostic query vector with a preset pluggable diagnostic knowledge base includes: Based on the preset plug-in diagnostic knowledge base, a plug-in knowledge unit library after structured preprocessing is obtained; wherein, the fault diagnosis knowledge in the plug-in diagnostic knowledge base is divided into independent plug-in knowledge units according to different fault types and scenarios, each plug-in knowledge unit is bound to a dedicated diagnostic analysis plug-in, and is configured with feature tag sets and matching threshold ranges corresponding to each dimension of the multi-dimensional collaborative diagnostic query vector. Based on the multidimensional collaborative diagnostic query vector, a target query vector with standardized format and dimension alignment is obtained; Based on the target query vector, a subset of initial screening knowledge units is obtained. Specifically, the key character features of the physical and logical association dimensions in the target query vector are extracted as initial screening keywords to traverse the plugin knowledge unit library for screening. If no results are found, the time association dimension features are extracted as auxiliary initial screening keywords to re-traverse and screen.

8. The multi-lane equipment collaborative fault diagnosis system for highway toll stations as described in claim 7, characterized in that, Matching the multidimensional collaborative diagnostic query vector with a preset pluggable diagnostic knowledge base also includes: Based on the initial screening subset of knowledge units and the target query vector, the comprehensive similarity score of each plug-in knowledge unit is obtained. Specifically, the cosine similarity algorithm is used to calculate the single-dimensional similarity between the target query vector and the feature tag set of each plug-in knowledge unit in each dimension, and the weighted comprehensive similarity is calculated by combining the preset weight coefficients of each dimension. If the target query vector contains super node features, the corresponding preset weight is added. Based on the comprehensive similarity score and the matching threshold range of the plug-in knowledge unit, a valid and ranked list of matching knowledge units is obtained. Based on the effective matching knowledge unit ranking list, the final hit dedicated diagnostic analysis plugin is obtained. Specifically, the plugin bound to the knowledge unit ranked first is selected as the hit plugin. If there are ties, the fault propagation probability model is called to calculate the fault propagation contribution and the plugin bound to the one with the highest contribution is selected.

9. A multi-lane equipment collaborative fault diagnosis system for highway toll stations as described in claim 8, characterized in that, Graph computation and causal analysis are performed on fault association trees that cross lanes and equipment, including: Based on the hit dedicated diagnostic analysis plugin, the initialized collaborative diagnostic reasoning engine and the injected current diagnostic context are obtained; wherein, the collaborative diagnostic reasoning engine completes core initialization by loading the pre-set domain analysis strategy and reasoning rules in the dedicated diagnostic analysis plugin, and the current diagnostic context includes a complete fault association tree topology structure, the multi-dimensional collaborative diagnostic query vector, real-time and historical status data of all nodes in the target node set, and the association information of the device physical topology knowledge base and alarm logic chain rule base; Based on the complete fault association tree topology and the analysis strategy specified by the dedicated diagnostic analysis plugin, an abstractly constructed weighted directed graph model is obtained; wherein, the nodes in the fault association tree are abstracted as nodes in the weighted directed graph model, the parent-child connection relationship in the fault association tree is abstracted as directed edges in the weighted directed graph model, and the weight of each directed edge is comprehensively constructed based on the physical matching degree weight, logical matching degree weight, and time matching degree weight between the nodes corresponding to the directed edge; Based on the weighted directed graph model and the graph computing component of the collaborative diagnostic reasoning engine, the node centrality analysis results and the core suspected fault propagation chain are obtained.

10. A multi-lane equipment collaborative fault diagnosis system for highway toll stations as described in claim 9, characterized in that, Graph computation and causal analysis of fault association trees across lanes and equipment also include: Based on the causal determination rule base pre-set in the plugin and the multidimensional features of the nodes in the fault association tree, the causal analysis results across lanes and across devices are obtained, forming a fault causal network with confidence weights. Based on the core propagation path and the fault causal network, the root cause node of the fault and the defined scope of the fault's impact are obtained. Based on the node centrality analysis results, the core suspected fault propagation chain, the fault causal network, the root cause node of the fault, and the scope of the fault's impact, a structured collaborative diagnostic report and a sequence of targeted handling strategies are obtained.

Images