An intelligent diagnosis method and system for power equipment maintenance

Abstract

The application relates to the technical field of intelligent operation and maintenance, and discloses an intelligent diagnosis method and system for power equipment maintenance, which comprises the following steps: a multi-dimensional maintenance model containing a basic state transition graph, a required skill vector and a user skill state vector is constructed; based on historical maintenance event logs, a causal abnormal weight representing an objective risk associated with specific operations is generated; when a maintenance task request of a user is received, the causal abnormal weight and the skill state vector of the user are combined in real time, and a personalized risk weight for the user is calculated; and based on the personalized risk weight, an optimal maintenance path with the lowest comprehensive cost is solved for the user and presented. By fusing objective data mining risks and subjective user capabilities, the application realizes the personalization and precision of maintenance guidance, and utilizes a double closed-loop mechanism to ensure system self-adaptation evolution, so that the safety and efficiency of equipment maintenance are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent operation and maintenance, in particular to an intelligent diagnosis method and system for power equipment maintenance. BACKGROUND

[0002] In the field of power system operation and maintenance, in order to ensure the safety and stability of the power grid and the safety of personnel operation, the maintenance of key power equipment such as transformers, switch devices and relay protection devices by operation and maintenance personnel usually needs to strictly follow the pre-prepared standard operation procedures (SOP). In the prior art, there have appeared auxiliary systems that digitally display these standard operation procedures, which can present step-by-step operation instructions to users on computers or mobile terminals to guide the maintenance work.

[0003] However, in the specific application scenario of the power industry, such methods still have significant limitations. On the one hand, they are usually only static digital presentations of standard operation procedures, and the maintenance paths generated by them are fixed and unchangeable. The operating state of the power system is complex and variable, and these systems are difficult to take into account the implicit operation risks beyond the procedures that are contained in the vast amount of historical operation tickets, defect reports and equipment monitoring data. For example, some operation sequences or parameter configuration combinations that are not explicitly prohibited in the procedures may be associated with serious consequences such as device tripping, insulation breakdown or protection device malfunction in practice, and existing systems cannot identify and avoid such deep risks revealed by actual data.

[0004] On the other hand, existing maintenance guidance systems usually adopt a non-differentiated way to provide the same operation path to all users, and this mode fails to effectively evaluate and utilize the individual skill levels and experience differences of users performing tasks. This is particularly crucial in power equipment maintenance. For example, for a high-voltage insulation test or a complex relay protection device setting verification work, the system cannot distinguish whether the performer is an experienced senior engineer or a newly employed junior technician, so the risk warnings and path planning provided by it lack personalization and adaptability, which may lead to insufficient guidance for junior personnel and potential safety hazards, or excessive intervention for senior personnel and impact on work efficiency, ultimately affecting the overall quality and safety of maintenance. SUMMARY

[0005] The technical problem to be solved by the present application is that the existing power equipment maintenance guidance method may not simultaneously consider the inherent cost of the standard operation procedure, the objective risks mined from historical actual operations, and the individual ability level of the current executing user when generating the maintenance path, thereby being difficult to provide optimal maintenance guidance that takes into account efficiency, safety and personalized adaptability.

[0006] To solve the above technical problems, the application provides an intelligent diagnosis method and system for power equipment maintenance.

[0007] The first aspect of the application provides an intelligent diagnosis method for power equipment maintenance, which comprises:

[0008] A multi-dimensional maintenance model is constructed, which comprises a basic state transition graph representing a standard maintenance process, a required skill vector associated with an operation edge in the basic state transition graph, and a user skill state vector associated with one or more users;

[0009] Based on the event log of one or more historical maintenance events, the causal anomaly weight associated with the operation edge in the basic state transition graph is evolved and generated by process mining;

[0010] In response to receiving a maintenance task request initiated by a user, the personalized risk weight of one or more operation steps in the maintenance task is calculated in real time based on the causal anomaly weight and the user skill state vector of the user;

[0011] Based on the personalized risk weight, an optimal maintenance path is solved and presented for the user;

[0012] After the completion of the maintenance task, the user skill state vector of the user is dynamically updated based on the business result thereof.

[0013] In one specific embodiment, the step of constructing the basic state transition graph comprises: applying natural language processing technology to automatically extract device state entities and operation action entities from one or more standard operating procedure documents; mapping the device state entities to state nodes in the basic state transition graph and mapping the operation action entities to operation edges connecting the state nodes to construct the topological structure of the basic state transition graph.

[0014] In one specific embodiment, the step of evolving and generating the causal anomaly weight based on the event log comprises: comparing the maintenance trajectory in the event log with the standard path of the basic state transition graph for consistency to identify one or more abnormal operation modes; using an association rule mining method to calculate the lift between the abnormal operation mode and the negative business result; screening out strong association modes with a lift greater than a preset threshold, and aggregating and calculating the causal anomaly weight according to the strong association modes.

[0015] In one embodiment, the step of calculating the individualized risk weight in real time comprises: for any operation step in the maintenance task, calculating a skill gap between its required skill vector and the user skill state vector of the current user; and cooperatively calculating the skill gap and the causal abnormality weight of the operation step to generate the individualized risk weight.

[0016] In one embodiment, the cooperative calculation is specifically amplifying the causal abnormality weight by a skill gap modulation function, whose calculation formula is:

[0017] ;

[0018] wherein, is the skill gap modulation function, is the required skill vector, is the user skill state vector, is the total number of dimensions of the required skill vector and the user skill state vector, is a skill dimension index from 1 to , and are the components of the required skill vector and the user skill state vector in the th skill dimension, respectively, is a skill gap sensitivity coefficient.

[0019] In one embodiment, the step of solving and presenting the optimal maintenance path comprises: based on the base cost of the operation edge in the base state transition graph and the individualized risk weight, calculating the comprehensive cost of each operation step in the maintenance task; and applying a graph search algorithm to solve a path with the minimum sum of comprehensive costs from the starting state to the target state on the base state transition graph as the optimal maintenance path.

[0020] In one embodiment, the method further comprises: in the process of solving and presenting the optimal maintenance path, determining whether any operation step meets a preset knowledge base activation condition; the activation condition at least includes that the causal abnormality weight of the operation step is lower than a preset low weight threshold, or the skill gap between its required skill vector and the user skill state vector is greater than a preset skill gap threshold; and if the activation condition is met, presenting a knowledge base associated with the operation step to the user in advance before presenting the execution instruction of the operation step.

[0021] In one embodiment, the step of dynamically updating the user skill state vector comprises: determining an execution performance score for each operation step performed in the maintenance task according to the final business result of the maintenance task; and adjusting the corresponding dimension of the user skill state vector of the user according to the execution performance score and the required skill vector of the operation step.

[0022] In one embodiment, the method further comprises: encapsulating the execution process and result of the completed maintenance task as a new maintenance track and adding it to the event log; and using the expansion of the event log as a trigger signal to re-execute the step of evolving and generating the causal abnormal weight based on the event log, so as to realize closed-loop evolution of the causal abnormal weight.

[0023] The second aspect of the present application provides an intelligent diagnosis system for power equipment maintenance, which comprises:

[0024] a model construction module, configured to construct a multi-dimensional maintenance model, the multi-dimensional maintenance model comprising at least: a basic state transition graph representing a standard maintenance process, a required skill vector associated with an operation edge in the basic state transition graph, and a user skill state vector associated with one or more users;

[0025] a weight evolution module, configured to evolve and generate a causal abnormal weight associated with an operation edge in the basic state transition graph based on an event log of one or more historical maintenance events through process mining;

[0026] a dynamic navigation module, configured to, in response to receiving a maintenance task request initiated by a user, calculate a personalized risk weight of one or more operation steps in the maintenance task based on the causal abnormal weight and the user skill state vector of the user in real time, and solve and present an optimal maintenance path for the user based on the personalized risk weight;

[0027] a state updating module, configured to, after the completion of the maintenance task, dynamically update the user skill state vector of the user based on the business result thereof.

[0028] In summary, the present application has at least one of the following beneficial technical effects:

[0029] 1. The present application can identify potential operation risks that are not explicitly defined in the standard procedure. By performing process mining on the historical maintenance event log, the present application identifies abnormal operation patterns that have strong association with negative business results and quantifies them as causal abnormal weights associated with specific operation edges. This mechanism enables the system's risk assessment to be based on actual operation data, revealing implicit risk associations and improving the objectivity and accuracy of risk identification.

[0030] 2.The application can provide a personalized maintenance path that adapts to the current user's specific ability level, calculate the skill gap between the required skill vector of the operation and the skill state vector of the user in real time, and use the gap to collaboratively calculate the objective causal abnormal weight, and the application generates a personalized risk weight for the user. Therefore, when the system solves the optimal maintenance path for users with different abilities, it can assign a higher comprehensive cost to the operation steps that the user is not capable of, thereby guiding the user to avoid potential risks and achieving personalized and adaptive maintenance guidance, thereby improving the safety of task execution.

[0031] 3.The application constructs a self-adaptive system that can evolve collaboratively and in a closed loop, dynamically updates the user's skill state vector after completing the task, and simultaneously feeds back the task execution result to the event log to trigger the re-evolution of the causal abnormal weight, thereby realizing the bidirectional iteration of the user model and the system knowledge model. This mechanism ensures that the system can continuously learn from new practices, so that the accuracy of personalized risk assessment improves over time, thereby enhancing the long-term effectiveness and reliability of the entire diagnostic method and system. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 FIG. 1 is a functional module diagram of an embodiment of the intelligent diagnostic system for power equipment maintenance according to the present application.

[0033] Figure 2 FIG. 2 is a flowchart of an embodiment of the intelligent diagnostic method for power equipment maintenance according to the present application.

[0034] Wherein, 10, data processing layer; 20, model construction and evolution layer; 30, dynamic navigation and interaction layer; 40, data storage layer. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical scheme and advantages of the present application clearer, the embodiments of the present application will be described in detail below. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0036] Referring to the accompanying drawings Figure 1 , Figure 1 FIG. 1 is a functional module diagram of an embodiment of the intelligent diagnostic system for power equipment maintenance according to the present application. The intelligent diagnostic system provided by the present application, as an entity for executing the method of the present application, can include: a data processing layer 10, a model construction and evolution layer 20, a dynamic navigation and interaction layer 30, and a data storage layer 40.

[0037] A data processing layer 10 is configured to obtain raw power operation and maintenance data from external data sources and pre-process the raw data. In an embodiment, the raw power operation and maintenance data can include structured power equipment standard maintenance procedures, historical operation tickets, equipment defect reports, and unstructured text or voice form field work logs. The data processing layer 10 performs parsing, format conversion and data cleaning operations on the raw data to generate standardized data for subsequent processing by other modules.

[0038] A model construction and evolution layer 20 is connected to the data processing layer 10 and the dynamic navigation and interaction layer 30. The layer receives the standardized data output by the data processing layer 10 to construct and continuously update the core parameters required for the diagnostic model. The model construction and evolution layer 20 can include a model initialization module 21 and a causal evolution module 22.

[0039] The model initialization module 21 is configured to generate a basic state transition graph based on structured documents such as power equipment standard maintenance procedures. The module is also configured to construct and initialize a multi-dimensional skill vector system that describes operational requirements and power operation and maintenance personnel capabilities.

[0040] The causal evolution module 22 is configured to receive event sequences transformed from historical operation tickets, field work logs and the like, to perform consistency comparison between the event sequences and the basic state transition graph, to identify abnormal causal patterns by calculating statistical correlation degrees, and to generate or update causal abnormal weights stored in the data storage layer 40 based on the identified abnormal causal patterns.

[0041] The dynamic navigation and interaction layer 30 is connected to the model construction and evolution layer 20 and the data storage layer 40. The layer generates and presents maintenance operation guidance to users based on maintenance task requests initiated by power operation and maintenance personnel, and calls model parameters. The dynamic navigation and interaction layer 30 can include a navigation generation module 31 and a user interaction module 32.

[0042] The navigation generation module 31 is configured to obtain the skill state vector of the current operation and maintenance personnel and the causal abnormal weight corresponding to the task when receiving the task request. The module calculates the individualized risk weight and the comprehensive cost of each operation step in the maintenance path in real time through a preset calculation formula, and solves the optimal maintenance path based on the comprehensive cost.

[0043] The user interaction module 32 is configured to provide a human-computer interaction interface, receive text or voice input of power operation and maintenance personnel as a task request, and present the optimal maintenance path and the corresponding risk warning information output by the navigation generation module 31 to the operation and maintenance personnel in the form of step-by-step instructions.

[0044] Data storage layer 40, for persistently storing various types of data required for the operation of the system and generated. The data includes: raw power operation and maintenance data, basic state transition graph, user skill state vector library, causal anomaly weight library, and various logs generated during system operation.

[0045] Referring to the accompanying drawings Figure 2 , Figure 2 is a flowchart of an intelligent diagnosis method for power equipment maintenance according to an embodiment of the present application. The intelligent diagnosis method provided by the present application can include the following steps:

[0046] S100, initialization and construction of a multi-dimensional maintenance model.

[0047] This step automatically constructs a basic state transition graph representing the standard maintenance process according to the structured standard operation procedure document; at the same time, a multi-dimensional skill vector system is established, and for each operation action in the basic state transition graph and each user in the system, the required skill vector and the user skill state vector are initialized respectively, thereby forming the basic model required for subsequent calculation.

[0048] S200, evolution of causal anomaly weight based on process mining.

[0049] This step continuously extracts actual operation event sequences from unstructured operation and maintenance logs, compares the event sequences with the standard path of the basic state transition graph, identifies abnormal operation patterns with strong association with negative business results through a preset statistical association measurement method, and generates or dynamically updates the causal anomaly weight associated with the specific operation action according to the identified abnormal operation patterns.

[0050] S300, dynamic generation and navigation of real-time personalized maintenance path.

[0051] This step, when receiving a maintenance task request initiated by a user, real-time acquires the causal anomaly weight evolved by step S200 and the skill state vector of the current user, generates a personalized risk weight for the user and the task through a preset collaborative calculation model, and calculates the optimal maintenance path with the lowest comprehensive cost, and finally presents the navigation to the user in the form of step-by-step instructions.

[0052] S400, execution feedback and closed-loop iteration of the model.

[0053] The actual execution process and result of the user completing the task under the guidance of step S300 are recorded and fed back as new event sequence data. The new event sequence data is used in step S200 to drive the continuous evolution of the causal abnormal weight, and is used to update the skill state vector of the user according to the task execution performance, thereby forming a collaborative and closed-loop iteration of the system knowledge model and the user model.

[0054] The steps in the embodiment method of the present application will be described in detail below.

[0055] In an embodiment of the present application, the initialization and construction of the multi-dimensional maintenance model in step S100 can include the following steps:

[0056] S110, automatic generation of a basic state transition graph.

[0057] S120, construction and initialization of a skill vector system.

[0058] In an embodiment of the present application, step S110 aims to automatically convert the standard operating procedure (SOP) document written in human language and unstructured into a machine-readable and structured graphical representation, i.e., a basic state transition graph. This graph constitutes the basis for all subsequent analysis, calculation and navigation. Step S110 can include:

[0059] S111, document acquisition and text normalization. The system acquires the standard operating procedure document in electronic format from a specified data source. After acquisition, the system performs text normalization processing to uniformly convert documents of different formats into a pure text stream and remove elements unrelated to the operation process to generate pure operation procedure text.

[0060] S112, text structured analysis and operation unit identification. The system analyzes the normalized pure operation procedure text, and according to the structured markers such as chapter numbers, bullets or numbers in the text, divides the continuous text stream into independent and minimum granularity operation units. Each operation unit corresponds to a specific operation step in the procedure.

[0061] S113, extraction of state nodes and operation edges. The system applies natural language processing techniques to deeply analyze the text content of each operation unit, in order to extract the key elements required for constructing the graph. Specifically, through a pre-trained Named Entity Recognition (NER) model, the system identifies and extracts predefined entity categories from the text description of the operation unit. The entity categories include: device state entities, which describe the physical or logical state of a device or its components at a certain point in time, such as "power is off", "housing screws are tightened", or "lubrication oil level is normal"; and operation action entities, which describe a specific physical operation, such as "tighten", "disassemble", "clean", or "measure".

[0062] S114, construction of state transition relationships and generation of the graph. Based on the extracted entities and their contextual relationships in the original document, the system constructs the topology of the graph. Each identified device state entity is mapped to a state node in the basic state transition graph . Each identified operation action entity is mapped to a directed operation edge in the graph .

[0063] According to the sequential relationship of the operation units in the original document, the system determines the timing logic between the operation edge and the state nodes connected to it. The operation edge described by an operation unit has a starting node that is the state node generated by the previous operation unit, and an ending node that is the state node reached after the execution of the current operation unit. By traversing all operation units, the system generates nodes and edges in sequence and connects them to form a directed graph representing the entire standard maintenance process, i.e., the basic state transition graph . The basic state transition graph can be formally represented as:

[0064] ;

[0065] where, is the set of all state nodes in the graph. is the set of all operation edges in the graph, each edge connecting a pair of state nodes . is a function that assigns a basic cost to each edge .

[0066] S115, assignment of basic costs. After the topology of the basic state transition graph is generated, the system assigns a basic cost to each operation edge The base cost is a benchmark index representing the standard effort required to perform the operation. In different embodiments of the present invention, the base cost can be determined in any of the following ways or their combinations:

[0067] The first embodiment: direct extraction based on explicit information in the text.

[0068] This approach is suitable for cases where the standard operating procedure document contains explicit quantitative indicators. The system retrieves the pre-set cost-related keywords, such as "standard man-hours," "rated time," "minutes," "hours," etc., in the text description of each operation unit through text matching or regular expression techniques. When the keywords and their associated values are matched, the system extracts the values and normalizes them by unit, which are then used as the base cost of the operation edge. .

[0069] The second embodiment: mapping query based on operation type.

[0070] This approach is suitable for cases where the procedure document does not contain explicit quantitative indicators. The system maintains a pre-defined "operation type-cost" mapping table. This mapping table maps the operation action entities identified in step S113 to a pre-set cost value. When assigning the cost of the operation edge, , the system queries the mapping table according to the operation action entity type corresponding to the edge and assigns the queried cost value to . The cost values in the mapping table can be pre-set and entered by technical personnel with relevant domain knowledge based on historical experience.

[0071] The third embodiment: weighted calculation based on multi-dimensional cost factors.

[0072] To represent the inherent effort of the operation, this approach defines the base cost as a comprehensive index composed of multiple cost factors. The base cost of each operation edge is calculated by the following formula:

[0073] ;

[0074] Where:

[0075] represents the time cost factor, representing the standard time required to perform the operation . Its value can be determined by the first or second embodiment.

[0076] represents the resource cost factor, representing the standard resource required to perform the operation ​​The required level of special resources or consumables. For example, an operation that requires the use of special tools has a higher value; only regular tools are required, and its value is lower. The value is higher; only regular tools are required, and its value is lower.

[0077] represents a safety risk factor, characterizing the inherent, objective safety risk level of performing the operation itself. For example, an operation involving high-voltage electricity, high-altitude work, or heavy lifting has a higher value; a regular inspection operation has a lower value. The value is higher; only regular tools are required, and its value is lower. The value is higher; only regular tools are required, and its value is lower.

[0078] and The specific numerical values can be obtained from the "operation type-multidimensional factor" knowledge base pre-built by domain experts.

[0079] , , are preset weight coefficients corresponding to time, resources, and safety risk, respectively, used to adjust the importance of different cost factors in the comprehensive cost calculation. These weight coefficients are positive real numbers, and their sum can be normalized to 1.

[0080] In system implementation, priority strategies can be set; for example, the first method is preferred; if the first method cannot obtain the value, the third method is used as a fallback; if the knowledge base required by the third method is incomplete, the second method is used as a basis. Through the combination of one or more of the above methods, it is ensured that each operation edge can be assigned a basic cost value.

[0081] In an embodiment of the present application, step S120 aims to establish a quantifiable framework describing the requirements of the operation action on the personnel's ability and the personnel's own ability level. Step S120 can include:

[0082] S121, definition of skill space. The system first defines a unified, multidimensional skill space. This space is composed of a set of mutually orthogonal basic skill dimensions related to the equipment maintenance field. The selection and definition of skill dimensions can be pre-set by technical personnel with corresponding field knowledge. For example, a dimensional skill space can include the following dimensions: {mechanical disassembly ability, electrical system diagnosis ability, hydraulic system knowledge, precision instrument operation ability, software and communication debugging ability, safety procedure familiarity...}. Each dimension represents an independent ability or knowledge field.

[0083] S122, construction and assignment of required skill vectors. The system assigns a required skill vector to each operation edge of the basic state transition graph generated in step S110. ​Construct and assign required skill vectors The vector is in the defined... Operation in the dimensional skill space Quantify the required abilities:

[0084] ;

[0085] in, A non-negative real number indicates the operation to be performed. In the The system assigns skill requirements across multiple skill dimensions. Specifically, the system maintains a predefined "operation type - skill requirement" knowledge base. This knowledge base maps the type of operation (e.g., "disassembling a motor," "calibrating a sensor," "reading fault codes") to a specific required skill vector. This is done by assigning a value to each operation dimension. During assignment, the system searches the knowledge base based on the entity type of the operation action corresponding to the edge, and uses the retrieved vector as... The vector data in the knowledge base is pre-labeled and entered by domain experts.

[0086] S123. Construction and initialization of user skill state vectors. The system initializes the user skill state vector for each registered operations and maintenance user. Create and maintain user skill state vectors This vector is also defined in the above... In the dimensional skill space, it is used to quantify users. Current overall ability level:

[0087] ;

[0088] in, It is a non-negative real number, representing the user. In the The current skill level across each skill dimension. This vector can be initialized based on one or more of the following information sources:

[0089] One approach is based on user profile information. The system obtains structured information such as the user's job title, rank, and qualifications. The system maintains a "profile information - initial skills" mapping table, mapping different job titles or certificate combinations to preset initial skill vectors.

[0090] Another approach is through standardized competency assessments. When users first use the system, they can complete standardized online quizzes or simulated tasks. Based on their performance in the tests related to each competency dimension, the system calculates their initial score for each skill dimension, thus forming their initial user skill state vector.

[0091] The user skill state vector is not static, and it is dynamically updated and iterated in subsequent step S400 according to the actual task execution performance of the user.

[0092] In an embodiment of the present application, the process mining based causal anomaly weight evolution of step S200 can include the following steps:

[0093] S210, extraction and alignment of maintenance event logs.

[0094] S220, quantitative discovery of abnormal causal patterns.

[0095] S230, generation and evolution mechanism of causal anomaly weight.

[0096] In an embodiment of the present application, step S210 aims to convert the original operation and maintenance logs of diverse sources and different formats into machine-readable event logs. Step S210 can include:

[0097] S211, heterogeneous data acquisition and normalization processing. The system acquires the original operation and maintenance logs from one or more preset data sources. The data sources include but are not limited to operation and maintenance management system databases, text reports entered by operation and maintenance personnel through client applications, or field recorded voice notes.

[0098] For voice format logs, the system calls an automatic speech recognition (ASR) engine to convert voice data streams into unstructured text data.

[0099] For all text data, the system performs normalization processing, including converting to a unified character encoding, removing irrelevant metadata or tags, and performing text cleaning, such as correcting common domain term spelling errors.

[0100] S212, identification and extraction of event elements. The system processes the normalized text data to extract key elements that constitute the event sequence for each independent maintenance case (e.g., a work order). Key elements include case ID, activity (operation), timestamp, and business result.

[0101] Extraction of case ID: The system directly extracts the work order number, equipment unique identifier, or task ID from the data meta information as the case ID.

[0102] Extraction and alignment of activity (operation): The system identifies the specific operation action from the text sentence describing the operation process, and aligns it with the standard operation edge in the basic state transition graph constructed in step S110. ​Alignment. The process is as follows: the system adopts a deep learning-based semantic similarity calculation model, which inputs an operation description text and outputs a standard operation edge with the closest semantic to the text. The unique identifier of the standard operation edge . In this way, non-standard expressions (e.g., "I replaced that old valve") are mapped to standard operation activities (e.g., the operation edge "replace valve").

[0103] Timestamp extraction: the system parses the time point description (e.g., "3:15 pm") from the text through time expression recognition technology and converts it into a standard timestamp format. If there is no explicit time in the text, the system generation time of the log record is used as the timestamp.

[0104] Business result extraction: the system analyzes the sentence describing the completion of the task through a text classification model to determine the business result of the maintenance task. The text classification model can be trained to identify "success", "failure", "rework", "safety alarm", etc. preset business result categories.

[0105] S213, structured generation of events and trajectories. The system combines the multiple event elements extracted for the same case ID in the previous step into structured event records. Each event record can be represented as a tuple:

[0106] ;

[0107] Where, is the unique identifier of the case; is the unique identifier of the operation edge in the basic state transition graph ; is the timestamp of the event occurrence; is the final business result of the case.

[0108] The system arranges all event records belonging to the same in ascending order of to form a maintenance trajectory :

[0109] ;

[0110] The collection of all generated trajectories forms a structured event log for subsequent steps of analysis.

[0111] In one embodiment of the invention, step S220 is intended to extract the structured event log generated in step S210. In this process, non-standard operating patterns that are significantly correlated with negative business outcomes are identified. Step S220 may include:

[0112] S221. Comparison of trajectory with standard path. System traverses event logs. Each maintenance track For each trajectory The system combines its operation sequence with the basic state transition diagram constructed in step S110. Align and compare all valid standard paths.

[0113] This comparison process aims to identify trajectories. Deviations exist within the system. These deviations include, but are not limited to, the following types:

[0114] Missing step: One or more operation edges required in the standard path Not on track It appeared in the middle.

[0115] Steps out of order: Trajectory The order in which two or more consecutive operation edges appear does not match the order in the standard path.

[0116] Steps repeated: trajectory A certain operation edge It was executed multiple times, while the standard path only requires it to be executed once.

[0117] Non-standard step insertion: trajectory The basic state transition diagram appears in the diagram. Operations that do not exist in the system.

[0118] S222. Definition and Extraction of Abnormal Operating Modes. The system defines a specific combination of one or more deviant behaviors as an abnormal operating mode. For example, patterns It can be defined as "in performing an operation" An operation was missed previously. ",model It can be defined as "operations performed consecutively" "Twice." The system traverses all maintenance trajectories, extracts all abnormal operation patterns that have occurred, and statistically analyzes each pattern. Throughout the event log Frequency of occurrence.

[0119] S223, Quantification of the correlation between patterns and results. This is for identifying anomalous operation patterns. whether there is a non-accidental causal association with negative business outcome (e.g., "rework" or "failure"), the system quantitatively evaluates the association using the Lift metric in association rule mining. The formula of Lift is:

[0120] ;

[0121] wherein:

[0122] represents the support of pattern , which is calculated by dividing the number of traces containing pattern in the event log by the total number of traces.

[0123] represents the support of negative business outcome , which is calculated by dividing the number of traces with business outcome in the event log by the total number of traces.

[0124] represents the support of pattern and negative business outcome occurring simultaneously, which is calculated by dividing the number of traces containing pattern and with business outcome in the event log by the total number of traces.

[0125] represents the confidence, i.e., the proportion of traces containing pattern whose business outcome is also .

[0126] S224, screening of strongly associated patterns. The system sets a predefined threshold of Lift (e.g., ). The system calculates the Lift value of all extracted abnormal operation patterns with respect to negative business outcome .

[0127] If , the system determines that there is a strong positive correlation between pattern and negative business outcome , i.e., the occurrence of pattern significantly increases the probability of negative outcome. The system screens out all patterns satisfying this condition to form the set of abnormal causal patterns for use in subsequent steps. In this way, the system discovers from the data operation combinations that are practically high-risk.

[0128] In one embodiment of the application, step S230 aims to transform the abnormal causal patterns discovered in step S220 into numerical risk indicators attached to the basic state transition graph Step S230 can include:

[0129] S231, initialization of causal abnormality weight. The system creates in the data storage layer 40 a causal abnormality weight variable corresponding to each operation edge in the basic state transition graph and sets its initial value to 0.

[0130] S232, calculation and aggregation of weight. The system traverses each operation edge in the basic state transition graph . For each edge , the system retrieves from the set of abnormal causal patterns generated in step S224 all abnormal causal patterns containing the operation edge , forming a pattern subset associated with the edge.

[0131] The system calculates the causal abnormality weight of the operation edge according to the pattern subset by a preset aggregation formula. The aggregation formula aims to accumulate the risks implied by multiple relevant patterns. The specific calculation formula is:

[0132] ;

[0133] wherein:

[0134] is the causal abnormality weight of the operation edge to be calculated, which is a non-negative real number.

[0135] is a specific abnormal causal pattern in the pattern subset .

[0136] is the support of the pattern , whose value is calculated in step S223 and represents the universality of the occurrence of the pattern.

[0137] is the negative business result relative to the pattern ​The lift degree, whose value is calculated by step S223, represents the strength of the association between the pattern and the negative result.

[0138] This item normalizes the lift degree, so that the risk contribution of a pattern with no association (lift degree = 1) is 0.

[0139] Through the formula, an operation edge The associated risk is the sum of the risk contributions of all high-risk operation patterns that it participates in. The risk contribution of a pattern is jointly determined by its prevalence (support) and its strength of association with the negative result (lift).

[0140] S233, dynamic evolution mechanism of weight. The causal anomaly weight is dynamically changing, and the system realizes the evolution of the weight through a continuous updating mechanism. The evolution mechanism can be realized in the following way:

[0141] One way is periodic full update. The system sets a fixed update period (for example, every 24 hours). At the end of each period, the system triggers a complete weight update process. This process calls steps S210 to S230 to re-perform consistency comparison, pattern discovery and association degree calculation based on all historical event logs accumulated in the data storage layer 40, and finally replaces the original all causal anomaly weight values with new calculation results.

[0142] Another way is incremental update. To improve computing efficiency, the system can use an incremental update strategy. The system only processes the new event logs generated since the last update. For new logs, the system analyzes the abnormal operation patterns contained therein and updates the support of the existing patterns and other statistics accordingly, or discovers new abnormal causal patterns. Subsequently, the system only recalculates and adjusts the causal anomaly weight of the operation edge affected by the new logs, rather than globally recalculates.

[0143] Through the above mechanism, when new operation and maintenance data are continuously collected and processed, the causal anomaly weight of each operation edge in the system also evolves dynamically, so that the risk index can reflect the latest knowledge learned from real-world operations.

[0144] In an embodiment of the present application, the dynamic generation and navigation of the real-time personalized maintenance path of step S300 can include the following steps:

[0145] S310, real-time collaborative calculation of personalized risk weight.

[0146] S320, intelligent activation and application of knowledge base.

[0147] S330, solution and presentation of the optimal maintenance path.

[0148] In one embodiment of the present invention, step S310 aims to combine objective risks mined from historical data with the user's current personal ability level in real time when the user initiates a maintenance task, generating a personalized risk indicator for that user. This indicator is the core of achieving individualized risk warning and path planning. Step S310 may include:

[0149] S311. Receiving and Acquiring Task Requests. When a user selects and initiates a maintenance task through the client application, the navigation generation module 31 receives the task request. This request contains the current user's unique identifier. and the unique identifier of the target task .

[0150] Navigation generation module 31 based on The user's current skill state vector is retrieved from the user skill state vector library in data storage layer 40. At the same time, according to The system determines the basic state transition diagram corresponding to the task. All operation edges The set of causal anomaly weights is obtained by querying and retrieving the causal anomaly weights corresponding to these operation edges, which are generated by step S200. .

[0151] S312. Quantitative calculation of skill gaps. For each operation edge in the maintenance task. The system needs to quantify the current user The gap between the system's current capability level and the capability requirements for performing the operation. The system retrieves the operation edge from data storage layer 40. Required skill vector .

[0152] The system modulates the skill gap using a preset function. To calculate this gap, the function uses the user's skill state vector. And the skill vector required for operation As input, the output is a skill gap modulation factor. The specific skill gap modulation function... The calculation formula is:

[0153] ;

[0154] in:

[0155] It is an operation edge The required skill vector.

[0156] is the user 's skill state vector.

[0157] and are the components of these two vectors in the i-th skill dimension.

[0158] The function is used to calculate the capability gap in a single skill dimension, which ensures that the gap is only counted when the user skill is lower than the requirement , while skill redundancy has no effect.

[0159] The summation and square root operation calculates the Euclidean distance of the capability gap in all dimensions, which comprehensively quantifies the user's overall skill gap.

[0160] is a preset hyperparameter greater than zero, called the skill gap sensitivity coefficient. This coefficient is used to adjust the amplification effect of the skill gap on the final risk. The larger the value, the more significant the risk weight will be amplified even by a small skill deficiency.

[0161] S313, collaborative generation of personalized risk weight. The system collaboratively calculates the objective causal anomaly weight and the quantified subjective skill gap to generate a personalized risk weight for the user and operation , which is calculated by the following formula:

[0162] ;

[0163] where is the skill gap modulation factor calculated by step S312.

[0164] The mechanism of this calculation formula is:

[0165] If the user's skills fully meet or exceed the operation requirements (i.e., all ), the value of the skill gap modulation factor is 1, and the personalized risk weight is equal to the objective causal anomaly weight .

[0166] If the user has deficiencies in one or more skill dimensions (i.e., there are ), the value of the skill gap modulation factor will be greater than 1, thereby amplifying the objective risk. The larger the user's skill gap, the more significant the amplification effect. ​​

[0167] Through this collaborative computation, the system generates, for each operational step in the task, a dynamic, individualized risk indicator that reflects both the inherent risk of the operation and the current executor's individual ability.

[0168] In one embodiment of the invention, step S320 aims at handling two specific situations: one is when the operation involved in the task is newly introduced or the historical data is sparse, leading to insufficient objective risk assessment; the other is when there is a significant gap between the user's ability and the operation's requirement, proactively providing pre-emptive knowledge supplementation. This step can include:

[0169] S321, Preposition and association of knowledge scaffolds. The system maintains a knowledge scaffold library in data storage layer 40. A knowledge scaffold is a structured teaching material associated with a specific operation edge or a specific skill dimension in the underlying state transition graph . The teaching material can include: detailed theoretical principle explanation, exploded view of equipment components, video of correct usage method of special tools, or common fault case analysis related to the operation. Each knowledge scaffold is mapped and associated with the identifier of one or more operation edges through its unique identifier.

[0170] S322, Activation condition determination. Before generating the navigation instructions for each operation edge in the repair task, the navigation generation module 31 executes the activation condition determination logic. This logic is used to determine whether the knowledge scaffold associated with the operation needs to be activated for the current user . The knowledge scaffold is activated when at least one of the following two conditions is met:

[0171] Condition one: New business or low data density operation determination.

[0172] This condition is determined by the causal abnormality weight of the operation edge . The system sets a low weight threshold . If the calculation results in:

[0173] ;

[0174] then condition one is met. The basis for this determination is that a very low causal abnormality weight value (e.g., close to 0) indicates that there are very few or no abnormal patterns related to the operation that lead to negative outcomes in the historical event log. This directly corresponds to the operation being a new or rarely executed business step, whose potential risks have not been fully revealed by historical data.

[0175] Condition two: User skill significantly insufficient determination.

[0176] This condition is determined by the skill gap modulation factor calculated in step S312. The system makes a judgment. A skill gap threshold is set. (This threshold is a preset value greater than 1, for example, 1.5). If the calculation yields:

[0177] ;

[0178] Then condition two is satisfied. The basis for this judgment is that the skill gap modulation factor directly quantifies the degree to which a user's ability is insufficient relative to the operational requirements. A modulation factor value exceeding a threshold indicates that the user has a significant skill gap in that operation, and performing that operation carries a high risk of failure due to insufficient personal ability.

[0179] S323. Integration and presentation of the knowledge framework. If the determination result of step S322 indicates that the knowledge framework needs to be activated, the navigation generation module 31 will adjust the original navigation instruction sequence. In this operation... Before the formal execution instruction is issued, the system will communicate with the operation side. Related knowledge framework content is pushed to the user interaction module 32 first.

[0180] On the user interface, the user interaction module 32 first presents the activated knowledge framework content to the user. Only after the user completes the learning of this knowledge material and sends a confirmation input (e.g., clicking the "I have learned" button) does the user interaction module 32 proceed to present the execution operation. The system provides specific step-by-step instructions. In this way, the necessary knowledge training is seamlessly and proactively integrated into the actual task execution process.

[0181] In one embodiment of the present invention, step S330 aims to plan an execution path with the lowest overall cost for the user based on all cost and risk information calculated in the preceding steps, and to provide guidance. Step S330 may include:

[0182] S331, Calculation of the comprehensive cost of operation steps. The navigation generation module 31 generates the cost for each operation step involved in the maintenance task. Calculate the comprehensive cost This overall cost will be the standard incurred in performing this operation compared to the current user's cost. The individualized risks of performing this operation are uniformly quantified. This calculation is performed using the following formula:

[0183] ;

[0184] in:

[0185] The operation edge to be calculated The total cost for the user .

[0186] is the base cost determined by step S115.

[0187] is the personalized risk weight for the user calculated by step S313.

[0188] is a preset risk cost conversion coefficient greater than zero. This coefficient is used to convert the unitless personalized risk weight into a cost value with the same dimension as the base cost, whose physical meaning can be understood as quantifying the risk into equivalent time or resource loss.

[0189] S332, solving the optimal maintenance path. The navigation generation module 31 processes the base state transition graph corresponding to the current maintenance task as a weighted directed graph. In this graph, each state node represents a device state, and the weight of each operation edge is set to its comprehensive cost calculated in step S331.

[0190] The system determines the start state and the target state of this maintenance task as the start node and the target node in the graph, respectively. Then, the system solves the minimum cost path from to on this weighted directed graph. This solving process can apply standard graph search algorithms, such as Dijkstra's algorithm or A* algorithm.

[0191] The output result of the algorithm is an ordered sequence of operation edges , whose sum of comprehensive costs is the minimum among all possible valid paths. This sequence is exactly the optimal maintenance path for the current user and the current task.

[0192] S333, serialized presentation of navigation instructions. The navigation generation module 31 sends the solved optimal maintenance path to the user interaction module 32. The user interaction module 32 serializes and presents the operation instructions in the path step by step.

[0193] For the first operation edge in the path sequence, the user interaction module 32 displays the operation instruction text corresponding to on its user interface. At the same time, the system obtains the personalized risk weight and a preset risk alert threshold are compared. If , the system will attach a high-risk warning mark, such as a highlighted color or a specific warning icon, next to the instruction.

[0194] After the user completes the step operation and submits a confirmation input to the system, the user interaction module 32 continues to present the next operation edge instruction and its corresponding risk warning information in the sequence, and so on. This process is repeated until all operations in the sequence are completed, thereby guiding the user to perform the maintenance task according to the optimal path.

[0195] In an embodiment of the present application, the execution feedback of step S400 is iterated with the model in a closed loop, ensuring that the system can learn and optimize from each actual maintenance task. The following steps can be included:

[0196] S410, dynamic updating of the user skill state.

[0197] S420, feedback of the event log and triggering of the model iteration.

[0198] Step S410 aims to objectively and quantitatively adjust the user's personal ability model according to the user's actual task execution performance, so as to realize the continuous evolution of the user model. Step S410 can include:

[0199] S411, collection of task execution data. After the user completes the entire maintenance task according to the navigation of step S330 and the system records the final business result thereof, the system collects the complete execution data of the task. The execution data includes: a unique identifier of the user who executed the task, the operation edge sequence actually executed by the user, and the final business result of the task (e.g., "successfully completed" or "failed to rework").

[0200] S412, determination of execution performance score. The system assigns an execution performance score to each operation edge included in the task according to the final business result of the task. The score is a scalar for quantifying the contribution of a single-step operation. In a specific embodiment, if the final business result is "successfully completed", all operation edges in the task path are assigned a preset positive score . Conversely, if the final business result is "failed to rework", all operation edges are assigned a preset negative score .

[0201] S413, update calculation of skill state vector. The system updates the user skill state vector of the user according to the execution performance score. The update process traverses each operation edge actually performed in the current task. For each operation edge

[0202] ;

[0203] wherein:

[0204] is the updated ability level of the user in the skill dimension .

[0205] is the pre-update ability level of the user in the skill dimension .

[0206] is a preset learning rate, which is a small positive real number, used to control the step size of a single update to ensure the smoothness of vector update.

[0207] is determined by step S412, which is the performance score of the operation edge .

[0208] is the component value of the required skill vector corresponding to the operation edge in dimension .

[0209] The principle of this update mechanism is to feed back the performance of a single task to the ability dimension directly related to the operation performed. When the user successfully completes an operation with a higher requirement for a certain skill, the ability value of the user in that skill dimension will be positively enhanced. Conversely, if the task fails, the ability value of the related skill dimension will be weakened accordingly. This update process is repeated after each task, and the user's skill state vector can be continuously adjusted to more truly reflect the evolution of the user's ability with the accumulation of practical experience. The updated user skill state vector will be written back to the user skill state vector library in the data storage layer 40 for subsequent task calls.

[0210] In an embodiment of the present application, step S420 aims to feed back the execution result of a single task to the global knowledge base of the system, drive the system to continuously learn and self-improve the cognition of objective risks, and form a self-adaptive closed loop. Step S420 can include:

[0211] S421. Structured encapsulation of task execution trajectory. Once a maintenance task is completed and its final business result is determined, the system encapsulates the complete execution process of the task into a structured maintenance trajectory. The specific implementation of this encapsulation process is as follows: The system collects the unique case ID of this task, the sequence of operation edges actually executed by the user, the execution timestamp corresponding to each operation edge, and the final business result applied to the entire task. The system converts this information into one or more event records. An ordered set whose data structure is completely consistent with the structure defined in step S213.

[0212] S422: Feedback and expansion of the event log. The system will encapsulate the newly generated maintenance trajectory in step S421. As a new data sample, it is fed back to the data storage layer 40 and added to the global structured event log. In this process, each actual maintenance task completed under system guidance becomes a data source of historical system experience, making the event log... The scale and diversity of these systems continue to grow as they are used.

[0213] S423, Triggering and Execution of Model Iteration Event Log The expansion will serve as a trigger signal to initiate the adjustment of the system's knowledge model, namely the causal anomaly weighting. The update iteration. This iterative process is achieved by re-executing the complete process of step S200 (i.e., the evolution of causal anomaly weights based on process mining). Specifically:

[0214] Upon receiving the trigger signal, the system will update the event log with the latest maintenance data. As input data, step S220 (quantitative discovery of anomalous causal patterns) is re-executed. During this process, the system recalculates all anomalous operational patterns based on a richer dataset. The support, confidence, and elevation of these factors can change the risk correlation of existing anomalous patterns, or reveal new strong correlation patterns that were previously missed due to insufficient data.

[0215] Subsequently, based on the updated set of abnormal causal patterns and their statistical attributes, the system continues to execute step S230 (the generation and evolution mechanism of causal anomaly weights). The system will recalculate the basic state transition diagram based on the latest correlation calculation results. All operation edges Causal anomaly weights And overwrite the old value with the newly calculated value.

[0216] The trigger and execution mechanism can adopt periodic full update or event-driven incremental update strategy according to system configuration. Through the closed-loop feedback mechanism, the knowledge model of the system is no longer static, but can dynamically evolve from each successful or failed practice, so that its assessment of operational risk more and more accurately reflects the objective reality.

[0217] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications, changes, variations, substitutions, and equivalents will occur to those of ordinary skill in the art without departing from the spirit and scope of the application as defined by the following claims and their equivalents.

Claims

1. A smart diagnosis method for power equipment maintenance, characterized in that, The method comprises the following steps: constructing a multi-dimensional maintenance model, the multi-dimensional maintenance model comprising: a basic state transition graph representing a standard maintenance process, a required skill vector associated with an operation edge in the basic state transition graph, and a user skill state vector associated with one or more users; evolutionarily generating a causal abnormality weight associated with an operation edge in the basic state transition graph based on an event log of one or more historical maintenance events through process mining; in response to receiving a maintenance task request initiated by a certain user, calculating a personalized risk weight of one or more operation steps in the maintenance task in real time based on the causal abnormality weight and the user skill state vector of the user; based on the personalized risk weight, solving and presenting an optimal maintenance path for the user; after the completion of the maintenance task, dynamically updating the user skill state vector of the user based on the business result thereof; the construction step of the basic state transition graph comprises: mapping device state entities into state nodes in the basic state transition graph and mapping operation action entities into operation edges connecting the state nodes to construct a topological structure of the basic state transition graph; the calculation step of the personalized risk weight comprises: for any operation step in a maintenance task, calculating a skill gap between the required skill vector thereof and the user skill state vector of the current user; co-calculating the skill gap and the causal abnormality weight of the operation step to generate the personalized risk weight; the method further comprises: in the process of solving and presenting the optimal maintenance path, determining whether any operation step meets a preset knowledge base activation condition; the activation condition at least comprises that the causal abnormality weight of the operation step is lower than a preset low weight threshold, which directly corresponds to the operation step being a brand-new business link; if the activation condition is met, presenting a knowledge base associated with the operation step to the user in advance before the presentation of an execution instruction of the operation step; the generation step of the causal abnormality weight comprises: consistently comparing maintenance trajectories in the event log with standard paths of the basic state transition graph to identify one or more abnormal operation modes; adopting an association rule mining method to calculate the lift between the abnormal operation modes and negative business results; screening strong association modes with a lift greater than a preset threshold and aggregating and calculating the strong association modes to generate the causal abnormality weight; 2. The intelligent diagnosis method for power equipment maintenance according to claim 1, characterized in that, the construction step of the basic state transition graph further comprises: applying natural language processing technology to automatically extract device state entities and operation action entities from one or more standard operating procedure documents. 3.The power equipment maintenance oriented intelligent diagnosis method according to claim 1, characterized in that, the co-calculation specifically is: amplifying the causal abnormality weight through a skill gap modulation function, a calculation formula of the skill gap modulation function being: ； wherein, is the skill gap modulation function, is the required skill vector, is the user skill state vector, is the total number of dimensions of the required skill vector and the user skill state vector, is the skill dimension index from 1 to is the skill dimension index from 1 to and are the components of the required skill vector and the user skill state vector, respectively, on the th skill dimension, is the skill gap sensitivity coefficient. 4.The power equipment maintenance oriented intelligent diagnosis method according to claim 1, characterized in that, the step of solving and presenting an optimal maintenance path for the user comprises: calculating a comprehensive cost of each operation step in the maintenance task based on a basic cost of an operation edge in the basic state transition graph and the personalized risk weight; On the basis state transition graph, a graph search algorithm is applied to solve a path with a minimum sum of comprehensive costs from a start state to a target state as the optimal maintenance path.

5. The intelligent diagnosis method for power equipment maintenance according to claim 1, characterized in that, The step of dynamically updating the user skill state vector of the user comprises: According to the final business result of the maintenance task, an execution performance score is determined for each operation step performed in the task; According to the execution performance score and the required skill vector of the operation step, a corresponding dimension of the user skill state vector of the user is adjusted in strength or weakness.

6. The power equipment maintenance oriented intelligent diagnosis method according to claim 1, characterized in that, The method further comprises: The execution process and result of the completed maintenance task are encapsulated as a new maintenance track and added to the event log; The expansion of the event log is taken as a trigger signal to re-execute the step of generating the causal abnormal weight, so as to realize closed-loop evolution of the causal abnormal weight.

7. A smart diagnosis system for power equipment maintenance, configured to perform the method according to any one of claims 1-6, characterized in that, Comprise: A model construction module is configured to construct a multi-dimensional maintenance model, which comprises a basis state transition graph representing a standard maintenance process, a required skill vector associated with an operation edge in the basis state transition graph, and a user skill state vector associated with one or more users; A weight evolution module is configured to generate, based on an event log of one or more historical maintenance events, a causal abnormal weight associated with an operation edge in the basis state transition graph through process mining evolution; A dynamic navigation module is configured to, in response to receiving a maintenance task request initiated by a user, calculate, based on the causal abnormal weight and a user skill state vector of the user, a personalized risk weight of one or more operation steps in the maintenance task in real time, and solve and present an optimal maintenance path for the user based on the personalized risk weight; A state updating module is configured to, after the maintenance task is completed, dynamically update the user skill state vector of the user based on the business result thereof.

Images