Electrical work ticket intelligent generation method and device, medium and equipment

Abstract

The invention discloses an electrical work ticket intelligent generation method and device, a medium and equipment in the technical field of power systems, and aims to solve the problems of serious module decoupling, dependence on manual connection and insufficient intelligent generation capability in the prior art. The method comprises the following steps: generating a required digital knowledge base and an electrical work task description generation work ticket based on a pre-constructed work ticket by utilizing a multi-model collaborative generation architecture; the multi-model collaborative generation architecture comprises a rule driving module used for calling a corresponding template based on task element matching; the task execution unit is used for executing a specific subtask according to a pre-deployed lightweight model or rule engine; and the large language model is used for performing semantic understanding and reasoning according to the electrical work task description, calling the rule driving module and the task execution unit according to a semantic understanding and reasoning result, and generating a work ticket in combination with a digital knowledge base required by work ticket generation. The method has both the accuracy of structured knowledge recognition and the flexibility of natural language generation.

Description

TECHNICAL FIELD

[0001] The present application relates to an intelligent generation method, device, medium and equipment of an electrical work ticket, and belongs to the technical field of power systems. BACKGROUND

[0002] The existing technology in the implementation of the power safety measure generation task usually includes the following key steps: work task understanding, device topology modeling, associated device retrieval, safety measure reasoning generation, and historical ticket retrieval generation. The following describes the technical means related to each step: (1) Understanding of the work task (intention recognition): The existing technology in the semantic analysis of the work ticket task text usually relies on the traditional natural language processing process, including Chinese word segmentation, part-of-speech tagging, named entity recognition (NER), and other steps to structurally extract the key elements in the task description. The main entities include: device name, device location, task behavior, and task category. Among them, the entity recognition method is mostly based on the forward maximum matching algorithm combined with the professional term dictionary of the power generation industry to identify the domain-related entities in the text.

[0003] In terms of tool selection, some research attempts to introduce pre-trained language models such as BERT to enhance the semantic modeling ability of the task text and improve the accuracy of entity recognition. At the same time, some work uses the jieba word segmentation tool combined with artificial rules to extract specific fields in a relatively simple scenario. In addition, to reduce the impact of semantic diversity, some methods introduce entity normalization technology, which unifies different forms of expression into a standard term set through the construction of a synonym mapping table or a term standardization dictionary.

[0004] Disadvantages: Traditional methods mainly rely on keyword matching, rule extraction, or basic language models, but have the following disadvantages: ① Lack of robustness: In real production scenarios, task texts often contain multiple devices, complex instructions, or non-standard expressions, making it difficult for traditional methods to accurately identify the semantic boundaries and key actions in multi-task compound sentences; ② Normalization relies on manual maintenance: Term normalization rules rely heavily on manual maintenance and updates, making it difficult to cover a large number of non-standardized expressions and local terms in field data, affecting the generalization ability; ③ Lack of graph alignment support: Existing methods generally decouple the intention recognition and device matching processes, without explicitly considering how to associate the recognized device entities with the standardized node names in the device graph or device atlas, affecting the accuracy of downstream device positioning and operation analysis, and making it difficult to support subsequent device cascading query and structured retrieval tasks (2) Device topology modeling and associated device retrieval strategy: Most existing literature does not explicitly model the topology of associated devices, assuming that the operating device is known or that the safety measures do not involve upstream and downstream device topology relationships. In a small number of works that involve device relationship modeling, explicit modeling is achieved using structures such as linked lists and adjacency tables. Topology information sources include one-time wiring diagrams, historical work ticket data, and SCADA databases, and the connection relationship between devices relies on manual annotation or rule-based reasoning. In addition, some literature implements ticket generation by constructing a knowledge graph with devices, work tickets, risk factors, and safety measures as nodes.

[0005] Associated device retrieval is based on the device topology modeling in the previous step. Traditional methods mostly use a set retrieval order and static rule set, such as searching for devices in the order of "circuit breaker open point → isolation switch → circuit breaker → grounding switch" in a 500kV system, and deriving the operating device sequence. Some research proposes to retrieve associated devices based on a knowledge graph, but does not detail the reasoning strategy. Some methods introduce electrical connectivity analysis algorithms, but the specific analysis process is not disclosed or explicitly stated in existing research. Some literature introduces manual input of the next operation sequence at this step.

[0006] Deficiency: Most existing literature does not explicitly model the topology of associated devices, and the related technology of device topology modeling is incomplete: although it includes the association between devices, work tickets, risk factors, and safety measures, it does not model the topology connection relationship between devices, which cannot support the complex logical reasoning required for work ticket generation, such as electrical isolation.

[0007] Most literature does not explicitly model the topology of associated devices, and the current operating device identification is mostly based on static rules or templates, or the safety measures do not involve upstream and downstream device topology relationships. It is difficult to achieve complete closed-loop analysis of operating devices and support integrated work ticket intelligent generation in real power production scenarios (3) Safety measure reasoning and generation: The guiding knowledge for safety measure generation usually comes from historical work tickets, expert experience summaries, safety regulations texts (such as the five-prevention locking principle), and maintenance operation instruction books. The organization forms include manual induction based on rules and structured knowledge graph modeling. Existing systems rely on manually developed If-Then implicit rules to describe the operation sequence in typical scenarios, and use a combination of rule engines, terminology specifications, and sequence models to generate safety measure content. In specific task scenarios, if the device operation sequence is known, only the term fill-in-the-blank needs to be completed.

[0008] Some literature attempts to use advanced technologies such as introducing graph neural networks, LSTM sequence models, or reinforcement learning strategies to optimize the model for the generation process of safety measures.

[0009] Insufficient: The reasoning generation part, most of the literature still relies on rules, poor generalization; while the advanced technology based on machine learning, deep learning still needs to rely on a large number of labeled data, and the reasoning process is not easy to explain, and it is difficult to generalize to non-standardized scenarios. The safety standard term is matched through the slot, and the content is manually combed more, and the generalization is insufficient, and the modification is difficult.

[0010] (4) Similar historical ticket retrieval generation: In the existing system, the station name, voltage level, work site and work content fields are often used to match similar samples from historical tickets to generate the current work ticket through character overlap rate, TF-IDF similarity, cosine similarity and other methods. In order to improve the accuracy of text matching, some studies have proposed using bag-of-words model or neural network model to extract feature vectors of work content and location to alleviate the interference of natural language expression diversity on matching results.

[0011] Insufficient: This kind of technical feature is characterized by lack of real modeling and analysis capability of device topology, and poor effect when encountering scenes outside the coverage of historical tickets; it can only reuse existing ticket content and cannot generate new reasonable safety measures, and the quality depends on the quality of historical tickets, and cannot adapt to the update of equipment and rules.

[0012] Overall reasoning generation framework, the existing system module is seriously decoupled and relies on manual connection: the interfaces between functions are not unified, the process is fragmented, and the whole process from device identification, device reasoning to safety measure generation cannot be automatically completed according to the task text; the intelligent generation capability is insufficient: the traditional rule method shows obvious inadaptability when facing new tasks, new devices and new sentence patterns. SUMMARY

[0013] The technical problem to be solved by the present application is to overcome the defects of the prior art and provide an electrical work ticket intelligent generation method, system, medium and equipment.

[0014] To solve the above technical problems, the present application is realized by the following technical scheme.

[0015] In a first aspect, the present application discloses an electrical work ticket intelligent generation method, comprising: obtaining an electrical work task description and inputting it into a multi-model collaborative generation architecture, the multi-model collaborative generation architecture being used to generate a work ticket based on a pre-constructed work ticket generation digital knowledge base and the electrical work task description; The multi-model collaborative generation architecture comprises a rule-driven module, a task execution unit and a large language model. The rule-driven module is used to call corresponding templates based on task element matching, and the corresponding templates are general rules and templates after abstract processing of high-frequency, complex or / and patternable tasks in business. The task execution unit is configured to execute a specific subtask according to a pre-deployed lightweight model or a rule engine. The large language model is configured to perform semantic understanding and reasoning according to the description of the electrical work task, and to call the rule-driven module and the task execution unit according to the results of the semantic understanding and reasoning, and to generate a work ticket in combination with a required digital knowledge base.

[0016] Further, the construction of the required digital knowledge base for the work ticket generation includes: The collected historical electrical work ticket text data is subjected to field analysis to extract key field information and construct a structured historical ticket database. Based on the station electrical primary wiring diagram, implicit logical knowledge including the connection relationship between drawings and the business operation process is fused to construct a unified topology graph reflecting the physical structure of the equipment and the operation semantics. Based on the unified topology graph entity reflecting the physical structure of the equipment and the business logic constructed by the electrical wiring drawing, guided by the equipment operation characteristics and the operation semantics, and in combination with the general safety measure sentences in the actual work ticket, a safety measure standard rule library with an extensible template structure and programmable execution logic is constructed in a natural language expression form.

[0017] Further, the construction of the unified electrical equipment topology graph reflecting the physical structure of the equipment and the operation semantics based on the station electrical primary wiring diagram and fusing the implicit logical knowledge including the connection relationship between drawings and the business operation process includes: For each station drawing to be processed, the drawing legend area is identified and analyzed to extract the drawing name and its type annotation information, and the drawing is classified according to the drawing name and its type annotation information. The drawing types of the drawing classification include six types of bus principle wiring diagram, bus configuration wiring diagram, UPS principle wiring diagram, UPS configuration wiring diagram, DC principle wiring diagram, and DC feeder / screen configuration wiring diagram. According to the drawing type, the device objects in the drawing are preliminarily extracted, and the basic attributes of the devices in the drawing are assigned. In combination with the text and graphical annotation information in the drawing, the associated metadata between the devices and the drawing is constructed. On the basis of the preliminary extraction of the device information, the device attributes are completed and standardized in combination with the device number, name, voltage level, and installation location information to obtain the device entities after attribute enhancement. Based on the connection, control, and subordinate relationship logic implied by the drawing, in combination with the relationship types defined in the graph schema, the semantic relationship between the device entities after attribute enhancement is reasoned and annotated to obtain the device entities after attribute enhancement and relationship construction. The device entity with the completed attribute reinforcement and relationship construction is uniformly mapped as a node object in a graph database, an edge relationship between devices is created according to a relationship type and an attribute field defined in a graph schema, and a unified electrical device topology graph reflecting a physical structure and an operation semantic of the device is constructed.

[0018] Further, before the device entity with the completed attribute reinforcement and relationship construction is uniformly mapped as a node object in a graph database, the method further comprises: Entity alignment between multiple graph papers is performed based on a device number, a similar name, a location, and an association relationship, so that multiple expressions of a same real device in different graph papers in the topology graph are uniformly abstracted as a single node.

[0019] Further, in the process of uniformly mapping the device entity with the completed attribute reinforcement and relationship construction as a node object in a graph database, a unique identification attribute is used for de-duplication verification and attribute merging to ensure data consistency. In the process of creating an edge relationship between devices, the same relationship is merged and optimized to create a non-repetitive edge relationship between devices.

[0020] Further, the safety measure standard rule library is divided into three categories according to operation objects and safety measure targets: device self-safety measures, control power safety measures, and grounding and hanging plate safety measures. The template of the safety measure adopts a combination of condition matching and field filling. The condition matching is used to identify constraint conditions including the type of the device, the voltage level, and the control relationship, and the field filling is used to generate complete and standardized natural language safety measure sentences in combination with the information extracted from the graph.

[0021] Further, the method of generating the required digital knowledge base based on the pre-constructed work ticket and generating the work ticket based on the electrical work task description comprises: A large language model is called to analyze the electrical work task description, search for historical tickets with a similarity exceeding a threshold, and determine safety measure content of the historical tickets that can be reused by the current task, A large language model is called to perform semantic analysis on the electrical work task description, extract devices and operation types involved in the task, and use a hybrid retrieval enhancement mechanism to perform entity alignment using a vector library constructed based on the unified topology graph reflecting the physical structure of the device and the business logic, and select the most relevant graph device entity from the vector library; according to the most relevant graph device entity and the operation type thereof, and in combination with the device topology graph, a targeted safety measure related device query is performed to obtain a set of operation device objects. The large language model calls the attribute information of each device in the operation device object set, matches the corresponding standard expression in the safety measure standard rule library, and generates natural language by combining the task context and the safety measure content of the historical ticket that can be reused in the current task to obtain a work ticket.

[0022] In a second aspect, the present application further discloses an electrical work ticket intelligent generation device, comprising: An acquisition module is configured to acquire an electrical work task description. A generation module is configured to input the electrical work task description into a multi-model collaborative generation architecture, and generate a work ticket based on a pre-constructed work ticket generation required digital knowledge base and the electrical work task description by using the multi-model collaborative generation architecture. The multi-model collaborative generation architecture comprises a rule-driven module, a task execution unit and a large language model. The rule-driven module is configured to call a corresponding template based on task elements matching, wherein the corresponding template is a general rule and template after abstract processing of high-frequency, complex or / and patternable tasks in business. The task execution unit is configured to execute specific sub-tasks according to a pre-deployed lightweight model or a rule engine. The large language model is configured to perform semantic understanding and reasoning according to the electrical work task description, call the rule-driven module and the task execution unit according to the results of semantic understanding and reasoning, and generate a work ticket in combination with a work ticket generation required digital knowledge base.

[0023] In a third aspect, the present application further discloses a computer readable storage medium storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the method of the first aspect.

[0024] In a fourth aspect, the present application further discloses a computer device, comprising, One or more processors, a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method of the first aspect.

[0025] The present application has the following beneficial effects: The present application constructs an intelligent and highly reliable safety measure generation scheme. The scheme has the precision of structured knowledge recognition and the flexibility of natural language generation, effectively improves the generation efficiency and accuracy of work ticket safety content, provides strong support for the improvement of intelligent level of electric power safety production, and has significant engineering application value and popularization prospect. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a schematic diagram of the overall framework of the present application; Figure 2 is a schematic diagram of the overall flow of the present application; Figure 3 is a schematic diagram of the Schema design of the atlas structure; Figure 4 is a specific reasoning generation flowchart taking busbar equipment as an example. DETAILED DESCRIPTION

[0027] The present application will be further described below in conjunction with the accompanying drawings. The following examples are only used to more clearly illustrate the technical solutions of the present application, and cannot be used to limit the protection scope of the present application.

[0028] Example 1, this embodiment introduces an intelligent generation method of electrical work ticket, comprising: An electrical work task description is obtained and input into a multi-model collaborative generation architecture, the multi-model collaborative generation architecture is used to generate a required digital knowledge base based on a pre-constructed work ticket and generate a work ticket based on the electrical work task description; The multi-model collaborative generation architecture comprises a rule-driven module, a task execution unit and a large language model; The rule-driven module is used to call a corresponding template based on task element matching, and the corresponding template is a general rule and template after abstract processing of high-frequency, complex or / and patternable tasks in business; The task execution unit is used to execute specific subtasks according to a pre-deployed lightweight model or rule engine; The large language model is used to perform semantic understanding and reasoning according to the electrical work task description, and call the rule-driven module and the task execution unit according to the results of semantic understanding and reasoning, and generate a work ticket in combination with a required digital knowledge base for generating a work ticket.

[0029] The construction of the required digital knowledge base for generating the work ticket comprises: The collected historical electrical work ticket text data is field parsed to extract key field information and construct a structured historical ticket database; Based on the station electrical primary wiring diagram, implicit logical knowledge including the connection relationship between drawings and equipment, business operation process, etc. is fused to construct a unified topology atlas reflecting the physical structure of the equipment and the operation semantics; Based on the unified topology atlas entity reflecting the physical structure of the equipment and the business logic constructed by the electrical wiring drawing, guided by the equipment operation characteristics and operation semantics, combined with the general safety measure sentences in the actual work ticket, the safety measure standard rule library with extensible template structure and programmable execution logic is constructed by classifying and abstracting and using natural language as the expression form.

[0030] The unified electrical equipment topology graph reflecting the physical structure and operation semantics of the equipment is constructed based on the electrical primary wiring diagram of the power station, and the implicit logical knowledge such as the connection relationship between the devices in the drawings and the business operation process is integrated, including: For each drawing to be processed, the legend area of the drawing is recognized and analyzed, the drawing name and type annotation information are extracted, the drawing is classified according to the drawing name and type annotation information, and the drawing types of the drawing classification include six types of bus principle wiring diagram, bus configuration wiring diagram, UPS principle wiring diagram, UPS configuration wiring diagram, DC principle wiring diagram, and DC feeder / screen configuration wiring diagram; According to the drawing type, the device objects in the drawing are preliminarily extracted, and the basic attributes of the devices in the drawing are given, and the associated metadata between the devices and the drawings is constructed in combination with the text and graphical annotation information in the drawing; On the basis of preliminarily extracting the device information, the device attributes are completed and standardized in combination with the device number, name, voltage level, and installation location information, and the device entity after attribute enhancement is obtained; Based on the connection, control, and subordinate relationship logic implied by the drawing, in combination with the relationship types defined in the graph schema, the semantic relationship between the device entities after attribute enhancement is reasoned and annotated, and the device entity after attribute enhancement and relationship construction is obtained; The device entity after attribute enhancement and relationship construction is uniformly mapped to a node object in the graph database, the edge relationship between the devices is created according to the relationship types and attribute fields defined in the graph schema, and the unified electrical equipment topology graph reflecting the physical structure and operation semantics of the equipment is constructed.

[0031] Before the device entity after attribute enhancement and relationship construction is uniformly mapped to a node object in the graph database, it further includes: Based on the device number, similar name, location, and associated relationship, the entity alignment between multiple drawings is performed, so that the multiple expressions of the same real device in different drawings in the topology graph are uniformly abstracted as a single node.

[0032] In the process of uniformly mapping the device entity after attribute enhancement and relationship construction to a node object in the graph database, a unique identification attribute is used to ensure data consistency, and the de-duplication verification and attribute merging are performed; In the process of creating the edge relationship between the devices, the same relationship is merged and optimized, and the non-repeated edge relationship between the devices is created.

[0033] The safety measure standard rule library is divided into three categories according to the operation object and the safety measure target: device safety measure, control power safety measure, and grounding and hanging plate safety measure. The template of the safety measure adopts a combination of condition matching and field filling; the condition matching is used to identify constraint conditions including the type of equipment, voltage level, control relationship, and the field filling is used to generate complete and standard natural language safety measure sentences in combination with the information extracted from the atlas.

[0034] The generation of the required digital knowledge base based on the pre-constructed work ticket and the generation of the work ticket based on the electrical work task description include: The large language model is called to understand and analyze the electrical work task description, search for historical tickets with a similarity exceeding a threshold, and determine the safety measure content of the historical tickets that can be reused by the current task, The large language model is called to perform semantic analysis on the electrical work task description, extract the equipment and operation types involved in the task, and use the vector library constructed based on the unified topology atlas of the reaction equipment physical structure and business logic (the equipment atlas vector library is established by RAG (Retrieval-Augmented Generation) technology) to perform entity alignment in combination with a hybrid retrieval enhancement mechanism, select the most relevant atlas equipment entity from the task, and perform targeted safety measure related equipment query based on the most relevant atlas equipment entity of the task and the operation type thereof in combination with the equipment topology atlas to obtain an operation equipment object set. The large language model is called to match the corresponding standard expressions in the safety measure standard rule library based on the attribute information (such as the equipment type, connection level, and control relationship information) of each equipment in the operation equipment object set, and generate natural language in combination with the task context and the safety measure content of the historical tickets that can be reused by the current task to obtain the work ticket.

[0035] Embodiment 2 is based on the same inventive concept as Embodiment 1, and this embodiment introduces an electrical work ticket intelligent generation method, as shown in Figure 1 The method constructs an equipment topology atlas that integrates the power plant electrical primary wiring diagram and business logic, realizes equipment recognition and alignment through vectorized semantic matching, identifies the target equipment related to the task through multi-hop path reasoning in combination with the equipment topology atlas, exports the operation equipment sequence that conforms to the business logic, and finally generates standard safety measure content by using a large model in combination with the safety measure terminology rule library, thereby realizing end-to-end intelligent generation of the closed loop from task analysis to safety measure generation.

[0036] To achieve the above purpose, the technical framework shown in Figure 2 is used to realize the present application, which specifically includes: Step 1: Based on the original data related to work ticket preparation, including plant station electrical system wiring diagram, historical work ticket, static data knowledge base such as safety regulations and operation regulations, and unstructured business logic, a digital knowledge base required for work ticket generation is constructed.

[0037] Step 2: Based on the above-mentioned constructed knowledge base, with the powerful semantic understanding and reasoning ability of the large language model and suitable large model application technology, the required safety measures content in the first electrical work ticket is intelligently generated, forming an end-to-end closed-loop processing flow from task input to safety measure output.

[0038] Step 3: In addition, to ensure the accuracy and business controllability of the generated safety measures, the present application further designs a multi-model collaborative generation architecture, which comprehensively utilizes the capabilities of rule logic, large language model and small model module to collaboratively complete the intelligent generation process: Rule-driven module (fixed logic): Abstract high-frequency, complex, and patternable tasks in business into general rules and templates. The system can match and call corresponding templates based on task elements to quickly output standardized safety measures, ensuring stable business logic and generation consistency.

[0039] Large language model: With the help of the large language model's ability in natural language understanding and reasoning, it can complete core functions such as task semantic recognition, keyword extraction, synonym normalization, device type identification, and safety measure type prediction. At the same time, the large model also plays the role of system scheduling and organization, responsible for coordinating the call of each related function module, improving the naturalness of language, the accuracy of context, and the generalization ability of task processing in the safety measure content generation process.

[0040] Task execution unit (small model): Deploy lightweight models or rule engines to execute specific sub-tasks, such as sub-tasks in graph construction and graph query execution, to ensure data consistency and operation accuracy in the overall process.

[0041] The present application deeply integrates the semantic understanding and reasoning ability of the large language model with fixed logic rules and precise execution modules for key tasks. Based on accurate device positioning, it fully utilizes the advantages of the large model in task understanding, element extraction and language generation, and efficiently outputs standardized safety measures. This method significantly improves the intelligent level of the work ticket preparation process, enhances the generalization ability of the system in various task scenarios, and effectively guarantees the standardization of safety measures and the controllability of production safety requirements.

[0042] Specifically, the present application realizes the construction of the knowledge base by the following steps: Step 11: Structured processing of historical work ticket data. Field parsing is performed on the collected historical electrical first work ticket text data to extract key field information (including work tasks, work content, work location, electrical work conditions, safety measures, etc.), and a structured historical ticket database is constructed to provide data support for subsequent historical ticket retrieval generation and safety measure generation.

[0043] Step 12: Constructing the device topology map. Based on the electrical one-line diagram of the station, the implicit logical knowledge such as the connection relationship between devices and the business operation process is integrated to construct a unified topology map reflecting the physical structure and operation semantics of the device. In the topology construction process, the dependency of the key device path and the associated relationship in the subsequent safety measure reasoning is fully considered to ensure that the map has good structure readability and reasoning convenience. As shown in FIG. 12, the specific construction steps are as follows: Figure 3 Step 121: Preprocessing and type identification of drawings. For the station drawing data to be processed, the drawing legend area is identified and analyzed one by one to extract the drawing name and type annotation information. According to the drawing annotation content, the drawing type is judged, including six types of bus principle wiring diagram, bus configuration wiring diagram, UPS principle wiring diagram, UPS configuration wiring diagram, DC principle wiring diagram, and DC feeder / screen configuration wiring diagram. According to this classification management, entity extraction and attribute reasoning tasks under different drawing structures are supported.

[0044] Step 122: Entity extraction and preliminary attribute assembly. According to the drawing type, different entity recognition strategies are applied to preliminarily extract the device objects in the drawing and assign them basic attributes such as device ID, name, coordinate position, drawing source, etc. At the same time, combined with the text, graphical annotation, and other information in the drawing, the association metadata between the device and the drawing is constructed to provide support for subsequent attribute reasoning and entity alignment.

[0045] Step 123: Attribute reasoning and entity reinforcement. Based on the preliminary extraction of entities, combined with device number, name, voltage level, installation location, and other information, corresponding reasoning strategies are designed for different drawing types to realize the completion and standardization of device attributes.

[0046] Step 124: Relationship construction. Based on the connection, control, and subordinate relationship logic implied in the drawing, combined with the relationship types defined in the schema of the map, the reasoning and annotation of the semantic relationship between devices are completed. The focus of relationship construction in different drawing types is different, such as the emphasis on the electrical connection relationship between bus and device in the bus principle diagram, and the emphasis on the installation and control relationship between screen and device in the feeder screen / DC screen configuration wiring diagram. Through entity position, topology link, name / number matching, and other information, high-quality relationship construction is realized.

[0047] ​Step 125: entity disambiguation. For the same device entity that may appear repeatedly in multiple drawings (but there are differences in naming expressions), this step is based on device number, similar name, location, association relationship and other prior knowledge to align entities between multiple drawings, complete the disambiguation and fusion of the device. Through entity merging, it is ensured that the multiple expressions of the same real device under different drawings in the topology graph are abstracted as a single node, which significantly enhances the integrity and logical consistency of the graph, and is one of the key improvements of the present application compared to the original drawing structure.

[0048] Step 126: graph generation and storage. The device entity with completed attribute enhancement and relationship construction is uniformly mapped to a node object in a graph database (such as Neo4j), and the edge relationship between devices is created according to the relationship type and attribute field defined in the graph Schema. Unique identification attributes (such as device ID) are used for de-duplication verification and attribute merging during node storage to ensure data consistency. The same relationship is merged and optimized during relationship creation to avoid repeated edge creation, and finally a unified electrical device topology graph reflecting the structure and business semantics of the station is constructed, which serves as an important supporting foundation for subsequent safety measure reasoning and path analysis.

[0049] Step 13: Constructing a safety measure standard rule base. To realize the structured expression and automatic generation of safety measures in work tickets, a safety measure standard rule base for electrical equipment operation scenarios is constructed. This rule base is based on graph entities, guided by device operation characteristics and operation semantics, and abstracted by classifying common safety measures in actual tickets. It finally forms a safety measure generation knowledge system with expandable template structure and programmable execution logic.

[0050] The template library is divided into three categories according to the operation object and safety measure target: device itself safety measure, control power safety measure, and grounding and hanging card safety measure. The safety measure template adopts the form of "condition matching + field filling" and has good interpretability and expandability. The matching field in the template is used to identify the type, voltage level, control relationship and other constraint conditions of the device, and the filling field generates complete and standardized natural language safety measure sentences combined with the information extracted from the graph.

[0051] Specifically, the present application uses a knowledge base enhanced retrieval collaborative large language model reasoning technology framework to realize the overall reasoning generation, Step 21: historical ticket applicability decision. After receiving the work task description, the system first calls the large language model to understand and analyze the task, combines RAG, keyword retrieval and other retrieval techniques to find historical tickets with high similarity, and judges whether the current task can reuse the safety measure content based on the LLM to reduce the generation risk.

[0052] Step 22: Work task analysis and entity linking. In addition to historical ticket retrieval, the system also enters an automated generation process. After receiving the work task description, the system first calls a large language model for semantic analysis to extract key information such as device description (including device name, device location, etc.), operation task type, etc. On this basis, combined with mixed retrieval enhancement mechanisms such as RAG and keyword retrieval, the vector library constructed by the device graph is used to realize automatic entity alignment, and the most relevant device is selected from it. The final output is the graph device entity and its operation type, providing basic input for subsequent safety measure generation.

[0053] Step 23: Safety-related device query. After the system completes semantic analysis and entity alignment of the task device, it uses the input device entity and its operation type as the basis, combined with the device topology graph, to perform targeted safety-related device queries. As shown in Figure 4 , this step takes "busbar" as an example to illustrate the specific processing logic.

[0054] First, the system makes suitability judgment and pre-classification on the input device based on the pre-defined typical device types and their ticketing logic. If it is identified as a "busbar" type device, the system will enter the busbar safety logic processing module. This module includes the following two sub-steps: Step 231: Main device safety-related device query: starting from the busbar device, the system extracts the principle connection diagram formed by the connection relationship and expands the query. In the principle connection graph, a depth-first traversal algorithm is used to recursively find all electrical devices connected to the operating busbar entity and calculate their connection levels. When encountering other busbars, oil-immersed transformers and other devices during traversal, recursion is terminated, and special devices such as transformers are skipped.

[0055] Step 232: Control power device query: based on the "control" relationship in the graph, the system retrieves the control power nodes for all related devices mentioned above, supplements control power safety information, and forms a complete set of operation device objects.

[0056] Step 24: Standard safety template matching and large model reasoning generation. For each device in the input operation device object set, the system calls a large language model to match the corresponding standard expression in the safety template library according to device type, connection level, control relationship, etc., and generates natural language based on the task context.

[0057] Taking the busbar maintenance task as an example, first, for the device ontology in the set of all operating device objects, the corresponding power switch safety precaution term is matched; in the second step, the control power safety precaution is generated. Taking the busbar as the maintenance ontology, for other devices in the set of operating device objects, if there is a device directly connected to the busbar ontology, the direct current screen control power safety precaution is generated according to the rules; the rest of the devices use the ontology control power safety precaution template. Finally, the general safety precaution content such as grounding and posting is automatically supplemented, and a set of safety precaution sentences with complete structure and standard language is finally output. In order to improve the accuracy and professionalism of the generated, the system will also combine the similar safety precaution content retrieved from the historical work ticket as auxiliary reference, enhance the context relevance and consistency of the reasoning generation.

[0058] The work ticket intelligent generation method of the application can: greatly improve the accuracy and professionalism of the generated: by constructing the power equipment graph and the vectorized knowledge base, the high-precision recognition of the device entity and the automatic recognition of the upstream and downstream relationship are realized. Combined with the standard safety precaution template library and the large model reasoning ability, the standardization of the template is retained, and the language flexibility and context understanding ability of the large model are also integrated, which ensures that the generated safety precaution content meets the industry standard requirements in terms of semantic expression, logical structure and safety compliance, avoiding the one-sidedness and mistakes caused by the traditional method of relying on rule or template matching; significantly enhance the generalization and adaptability of the generation system: compared with the traditional rule system which can only cover limited scenarios, the application uses the semantic understanding and generation ability of the large language model, and combines the graph knowledge to dynamically complete the generated elements, which can automatically adapt to the task description in new devices and new scenes, realize the automatic generation of safety precautions in complex and fuzzy task semantics, and improve the system intelligence level and application range; effectively reduce the manual intervention and dependence, and improve the work ticket preparation efficiency: compared with the traditional manual preparation or semi-automatic auxiliary generation method, the application can realize one-time automatic generation of safety precaution content under natural language task description, greatly reduce the cost of manual arrangement, completion and audit, and improve the overall work ticket processing efficiency; has good system expansibility and industry generalizability: the scheme has modular and reusable design in graph construction, task understanding, device query, natural language safety precaution template use, etc., is easy to migrate to different power plants and different type work ticket safety precaution generation tasks, and has strong engineering landing value and promotion prospect.

[0059] In summary, the application constructs an intelligent and highly reliable safety precaution generation scheme by introducing device topology graph, RAG vectorization retrieval and large language model reasoning and other key technologies. The scheme has both the accuracy of structured knowledge recognition and the flexibility of natural language generation, not only effectively improves the generation efficiency and accuracy of the work ticket safety precaution content, but also provides a strong support for the improvement of the intelligent level of power safety production, has a significant engineering application value and promotion prospect.

[0060] Embodiment 3, based on the same inventive concept as other embodiments, introduces an electrical work ticket intelligent generation device, comprising: An acquisition module is configured to acquire an electrical work task description. A generation module is configured to input the electrical work task description into a multi-model collaborative generation architecture, and use the multi-model collaborative generation architecture to generate a work ticket based on a pre-constructed work ticket generation required digital knowledge base and the electrical work task description. The multi-model collaborative generation architecture comprises a rule-driven module, a task execution unit, and a large language model. The rule-driven module is configured to call a corresponding template based on task element matching, wherein the corresponding template is a general rule and template after abstract processing of high-frequency, complex, or / and patternable tasks in business. The task execution unit is configured to execute specific subtasks according to a pre-deployed lightweight model or rule engine. The large language model is configured to perform semantic understanding and reasoning according to the electrical work task description, call the rule-driven module and the task execution unit according to the results of semantic understanding and reasoning, and generate a work ticket in combination with a work ticket generation required digital knowledge base.

[0061] Embodiment 4, based on the same inventive concept as other embodiments, introduces a computer-readable storage medium storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the method of embodiment 1.

[0062] Embodiment 5, based on the same inventive concept as other embodiments, introduces a computer device, comprising one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method of embodiment 2.

[0063] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0064] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0065] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0066] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0067] The above description is only preferred embodiment of the present application, it should be pointed out that, for those skilled in the art, without departing from the technical principles of the present application, can also make a number of improvements and variations, these improvements and variations should also be considered as the protection scope of the present application.

Claims

1. An intelligent generation method of an electrical work ticket, characterized by, The method comprises the following steps: An electrical work task description is acquired and input into a multi-model collaborative generation architecture, which is used to generate a required digital knowledge base and a work ticket based on a pre-constructed work ticket and the electrical work task description; The multi-model collaborative generation architecture comprises a rule-driven module, a task execution unit, and a large language model; The rule-driven module is used to call a corresponding template based on task element matching, wherein the corresponding template is a general rule and template obtained by abstracting high-frequency, complex, or / and patternable tasks in business; The task execution unit is used to execute specific sub-tasks according to a pre-deployed lightweight model or rule engine; The large language model is used to perform semantic understanding and reasoning according to the electrical work task description, call the rule-driven module and the task execution unit according to the results of semantic understanding and reasoning, and generate a work ticket based on the required digital knowledge base generated by the work ticket. 2.The intelligent generation method of the electrical work ticket according to claim 1, characterized in that, The construction of the required digital knowledge base for the work ticket generation comprises the following steps: Field analysis is performed on collected historical electrical work ticket text data to extract key field information and construct a structured historical ticket database; Based on the station electrical primary wiring diagram, implicit logical knowledge including the connection relationship between drawings and the business operation process is fused to construct a unified topology graph reflecting the physical structure of the equipment and the operation semantics; Based on the unified topology graph entity reflecting the physical structure of the equipment and the business logic constructed by the electrical wiring diagram, guided by the operation characteristics and operation semantics of the equipment, and combined with the general safety measure sentences in the actual work ticket, a safety measure standard rule library with an extensible template structure and programmable execution logic is constructed by classifying and abstracting the sentences and expressing them in natural language. 3.The intelligent generation method of the electrical work ticket according to claim 2, characterized in that, The construction of the unified topology graph reflecting the physical structure of the equipment and the operation semantics based on the station electrical primary wiring diagram comprises the following steps: For each station drawing to be processed, the drawing legend area is recognized and analyzed to extract the drawing name and type annotation information, and the drawings are classified according to the drawing name and type annotation information, wherein the drawing types of the drawing classification include six types of bus principle wiring diagram, bus configuration wiring diagram, UPS principle wiring diagram, UPS configuration wiring diagram, DC principle wiring diagram, and DC feeder / screen configuration wiring diagram; According to the drawing type, the device objects in the drawing are preliminarily extracted, and the basic attributes of the devices in the drawing are assigned, and the associated metadata between the devices and the drawings are constructed in combination with the text and graphical annotation information in the drawings; On the basis of preliminarily extracting the device information, the device attributes are completed and standardized in combination with the device number, name, voltage level, and installation location information to obtain device entities with enhanced attributes; Based on the connection, control, and subordinate relationship logic implied by the drawing, in combination with the relationship types defined in the graph schema, the semantic relationships between the device entities with enhanced attributes are reasoned and labeled to obtain device entities with completed attribute enhancement and relationship construction. The device entity with the completed attribute reinforcement and relationship construction is uniformly mapped as a node object in a graph database, edge relationships between devices are created according to relationship types and attribute fields defined in a graph schema, and a unified electrical device topology graph reflecting the physical structure and operation semantics of the devices is constructed. 4.The method of claim 3, wherein, Before the device entity with the completed attribute reinforcement and relationship construction is uniformly mapped as a node object in a graph database, the method further includes: Based on the device number, similar name, location and association relationship, the entity alignment between multiple graphs is performed, so that the multiple expressions of the same real device in the topology graph under different graphs are uniformly abstracted as a single node.

5. The intelligent generation method of the electrical work ticket according to claim 3, characterized in that, In the process of uniformly mapping the device entity with the completed attribute reinforcement and relationship construction as a node object in a graph database, a unique identification attribute is used for de-duplication verification and attribute merging to ensure data consistency; In the process of creating the edge relationships between the devices, the same relationships are merged and optimized to create non-repeated edge relationships between the devices. 6.The intelligent generation method of an electrical work ticket according to claim 2, characterized in that, The safety measure standard rule library is divided into three categories according to operation objects and safety measure targets: device self-safety measure, control power safety measure, and grounding and hanging plate safety measure; The template of the safety measure adopts a combination of condition matching and field filling; the condition matching is used to identify constraint conditions including the type of the device, the voltage level and the control relationship, and the field filling is used to generate complete and standard natural language safety measure sentences in combination with the information extracted from the graph. 7.The intelligent generation method of an electrical work ticket according to claim 2, characterized in that, The generation of the required digital knowledge base based on the pre-constructed work ticket and the generation of the work ticket based on the electrical work task description include: A large language model is called to understand and analyze the electrical work task description, search for historical tickets with a similarity exceeding a threshold, and determine the safety measure content of the historical tickets that can be reused by the current task, A large language model is called to perform semantic analysis on the electrical work task description, extract the devices and operation types involved in the task, and use a hybrid retrieval enhancement mechanism to perform entity alignment using a vector library constructed based on the unified topology graph reflecting the physical structure and business logic of the devices, and select the most relevant graph device entity of the task; according to the most relevant graph device entity of the task and the operation type thereof, and in combination with the device topology graph, a targeted safety measure related device query is performed to obtain a set of operation device objects; A large language model is called to match the corresponding standard expressions in the safety measure standard rule library based on the attribute information of each device in the set of operation device objects, and to generate a natural language based on the task context and the safety measure content of the historical tickets that can be reused by the current task, thereby obtaining a work ticket.

8. An electric work ticket intelligent generation device characterized by comprising: The method includes: An acquisition module is configured to acquire an electrical work task description; A generation module is configured to input the electrical work task description into a multi-model collaborative generation architecture, and use the multi-model collaborative generation architecture to generate a required digital knowledge base based on a pre-constructed work ticket and generate a work ticket based on the electrical work task description; The multi-model collaborative generation architecture includes a rule-driven module, a task execution unit and a large language model; The rule driving module is configured to call a corresponding template based on matching of task elements, the corresponding template being a general rule and template after abstract processing of high-frequency, complex, or / and patternable tasks in a business; The task execution unit is configured to execute specific subtasks according to a pre-deployed lightweight model or rule engine; The large language model is configured to perform semantic understanding and reasoning according to an electrical work task description, call the rule driving module and the task execution unit according to a result of the semantic understanding and reasoning, and generate a work ticket by combining a required digital knowledge base.

9. A computer-readable storage medium storing one or more programs, the one or more programs comprising instructions that when executed by a computer cause the computer to perform a method of any of claims 1-8. The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods of claims 1-8.

10. A computer device, comprising: including, one or more processors, memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods of claims 1-8.

Images