A method for automatically generating an operation ticket and related device

Abstract

The present disclosure provides a method for automatically generating an operation ticket and related equipment, the method comprising: obtaining a target operation diagram and a task instruction, and obtaining structured topology data based on the target operation diagram; performing two-way collaborative retrieval on the task instruction to obtain reference information matched with the task instruction; generating fusion data based on the reference information and the structured topology data; generating an initial operation ticket using a large language model based on the fusion data; checking the initial operation ticket; in response to the check failing, generating feedback information and correcting the initial operation ticket based on the feedback information until the check passes and a target operation ticket is output. The present disclosure realizes autonomous reasoning, real-time checking and dynamic optimization of ticketing logic, effectively improving the intelligent level, safety and adaptability of operation ticket generation.

Description

Technical Field

[0001] This disclosure relates to the field of power system automation technology, and in particular to a method and related equipment for automatically generating operation tickets. Background Technology

[0002] Operation tickets are the basis for operators to switch electrical equipment from one operating mode to another. Correctly filling out operation tickets is a crucial measure and foundation for preventing electrical misoperation accidents. With the deepening and advancement of smart grid construction, the generation mode of operation tickets has gradually evolved from manual paper-based completion to computer-aided generation systems.

[0003] In the manual drafting mode, operation tickets rely on operators to fill them out manually according to the "Power Safety Work Regulations," which suffers from low efficiency, error-proneness, and low standardization, making it difficult to meet the timeliness requirements of substation operation and maintenance. In the rule-driven mode, the expert system based on task decomposition and five-prevention interlocking logic achieves semi-automatic generation, but still requires a lot of manual intervention and rule maintenance. When the wiring method changes, the rule base needs to be reconfigured, the synchronization cycle is long, and it is difficult to adapt to frequent adjustments in operation mode.

[0004] In recent years, intelligent generation technology for operation tickets based on deep learning has emerged, but it still faces core challenges. On the one hand, existing technologies mostly rely on structured data input and cannot directly extract topological relationships from unstructured images such as primary wiring diagrams; on the other hand, such methods lack a deep understanding of the underlying physical logic of the "Electric Power Safety Work Regulations," which can easily lead to seemingly logical but non-compliant operation steps that violate the five-prevention interlocking constraints, seriously threatening power grid safety.

[0005] Therefore, how to achieve autonomous reasoning, real-time verification, and dynamic optimization of invoicing logic has become a key technical challenge that urgently needs to be solved in the field of power dispatch automation.

[0006] It should be noted that the above description of the technical background is only for the purpose of providing a clear and complete explanation of the technical solutions of the present invention and facilitating understanding by those skilled in the art. It should not be assumed that the above technical solutions are known to those skilled in the art simply because they have been described in the background section of this invention. Summary of the Invention

[0007] In view of the above, the purpose of one or more embodiments of this disclosure is to provide an automatic generation method and related equipment for operation tickets, so as to solve or partially solve the problems raised in the background art.

[0008] To achieve the above objectives, this disclosure provides a method for automatically generating operation tickets, the method comprising:

[0009] Obtain the target runtime graph and task instructions, and obtain structured topology data based on the target runtime graph; The task instruction is subjected to dual-path collaborative retrieval to obtain reference information matching the task instruction; and fused data is generated based on the reference information and the structured topology data. Based on the fused data, an initial operation ticket is generated using a large language model; The initial operation ticket is verified; if the verification fails, feedback information is generated, and the initial operation ticket is corrected based on the feedback information until the verification passes, and the target operation ticket is output.

[0010] Based on the same inventive concept, this disclosure also provides an automatic ticket generation system, the system comprising: The acquisition module is configured to acquire the target running graph and task instructions, and obtain structured topology data based on the target running graph; The retrieval module is configured to perform a dual-path collaborative retrieval of the task instruction to obtain reference information matching the task instruction; and generate fused data based on the reference information and the structured topology data. The generation module is configured to generate an initial operation ticket based on the fused data using a large language model; The verification module is configured to verify the initial operation ticket; in response to the verification failing, it generates feedback information and corrects the initial operation ticket based on the feedback information until the verification passes and outputs the target operation ticket.

[0011] Based on the same inventive concept, this disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement an automatic ticket generation method as described in any of the above claims.

[0012] Based on the same inventive concept, this disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute any of the above-described methods for automatically generating operation tickets.

[0013] Based on the same inventive concept, this disclosure also provides a computer program product, including one or more computer programs, which, when executed by one or more processors, implement any of the above-described methods for automatically generating operation tickets.

[0014] As described above, the automatic operation ticket generation method provided in this disclosure obtains structured topology data by parsing the target operation graph, combines dual-path collaborative retrieval to obtain reference information and generate fused data, generates an initial operation ticket using a large language model, and then iteratively corrects it through a verification and feedback mechanism until the target operation ticket is output. This solves the technical problems of being unable to directly extract topological relationships from unstructured images, lacking understanding of the underlying physical logic of safety procedures, and being prone to generating operation steps that violate the five-prevention interlocking constraints. The technical solution of this disclosure realizes autonomous reasoning, real-time verification, and dynamic optimization of the ticketing logic, effectively improving the intelligence level, security, and adaptability of operation ticket generation. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in one or more embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only one or more embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating an automatic operation ticket generation method provided in this embodiment of the disclosure; Figure 2 This is a schematic diagram of the overall operation flow of an automatic operation ticket generation method according to an embodiment of the present disclosure; Figure 3 This is a schematic diagram of the structure of an automatic operation ticket generation system provided in an embodiment of the present disclosure; Figure 4 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0018] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of this disclosure should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar words used in one or more embodiments of this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] Explanation of relevant technical terms: A switching operation ticket is a standard written document that records the operation sequence, verification items, and safety measures when power system equipment status changes.

[0020] Topological semantic parsing uses visual recognition technology to extract the physical connection logic in a primary wiring diagram and transforms it into a machine-understandable structured topological relationship.

[0021] Retrieval-Augmented Generation (RAG) is an architecture that combines a retrieval system with a generative large model. It provides real-time and accurate background information to the large model by retrieving relevant documents from a knowledge base, thereby reducing model "illusion".

[0022] Five-prevention logic (electrical interlocking) refers to the five basic functions of power systems to prevent misoperation, including hard constraints such as preventing accidental opening and closing of circuit breakers and opening and closing of disconnecting switches under load.

[0023] As described in the background section, the relevant technologies have the following problems: First, the grammatical templates are rigid and inflexible, lacking adaptability to irregular and non-standard power grid structures. Current template-based and symbol-replacement-based solutions become completely ineffective when faced with scenarios such as non-standard bay designs, unconventional bypass power supply switching, or sudden temporary wiring. This necessitates significant manpower for customized table maintenance. For example, adding a new non-typical bay in a 220kV substation requires 8-12 man-hours and 3-5 working days for template reconstruction and testing, resulting in high maintenance costs and difficulty adapting to frequent adjustments in power grid operation modes.

[0024] Second, there is a severe disconnect between graphical perception and logical decision-making, and an information gap exists between physical drawings and digital models. Current operation ticket generation solutions rely on manual selection of equipment or manual creation of typical bay models, failing to achieve a direct end-to-end understanding of the primary wiring diagram. The system's backend static database frequently lags behind the actual physical state, with long synchronization delays. When the wiring diagram changes, the operation ticket status input source deviates significantly from physical reality, causing a sharp drop in the accuracy of the generated operation tickets and posing a significant safety hazard.

[0025] Third, the state machine solution path is rigidly fixed, lacking the ability for complex temporal reasoning and self-healing mechanisms. Currently, the system is based on fixed rule tree traversal or univariate linear operating condition derivation. When handling complex long-term time-series operations such as full-station bus switching and multi-loop closure, it is prone to computational deadlock due to unmet boundary conditions, resulting in a low success rate. More critically, after triggering security rule interception, the system can only passively report errors, unable to transform violation information into self-healing correction clues, relying excessively on manual intervention, and thus generating inefficient data.

[0026] Fourth, there is an over-reliance on historical samples, and the response mechanism for updating procedures is severely lagging. Current machine learning-based solutions heavily depend on the coverage of training samples. Samples for low-frequency, high-risk scenarios (such as handling severe faults and isolating special anomalies) are naturally scarce, leading to decision-making failures. Furthermore, updating safety procedures requires restarting the entire data collection, training, and verification process, resulting in long update cycles and compliance risks associated with the generated operation tickets. Therefore, there is an urgent need to develop a technical solution for automatically generating operation tickets to achieve intelligent, automated, and safe substation operation ticket generation, significantly improving the safety and efficiency of power grid operation and providing core technical support for the intelligent operation and maintenance of new power systems.

[0027] Based on some implementations of this disclosure, a scheme for the automatic generation of operation tickets based on image topological semantic parsing and RAG enhanced reasoning is provided. In this scheme, a closed-loop architecture is constructed, which includes conversational human-computer interaction, multimodal visual topology extraction, dual-path collaborative knowledge retrieval, and generation of a large model temporal thought chain. This enables end-to-end, interactive, and highly reliable intelligent generation of power switching operation tickets in a visualized large model dialogue interface.

[0028] refer to Figure 1 The present disclosure discloses an automatic generation method for operation tickets, which includes the following steps: Step S101: Obtain the target operation diagram and task instructions, and obtain structured topology data based on the target operation diagram. In some embodiments, this step uses a web-based or client-side conversational human-computer interaction window as the front-end interaction interface. The original main wiring operation diagram submitted by the user is obtained through a file drag-and-drop upload area, and macroscopic operation instructions input by the user are received through a natural language conversation input box. The topology parsing results, retrieval results, and operation ticket text are dynamically displayed in the conversation stream using a graphics rendering engine.

[0029] In some embodiments, the interactive interface adopts a React (a JavaScript library for building user interfaces) + WebSocket (a full-duplex communication protocol) architecture, with an interactive response latency of less than 300ms.

[0030] In some embodiments, obtaining the target runtime graph and task instructions, and obtaining structured topology data based on the target runtime graph, includes: Step S1011: Obtain the target running graph, and preprocess the target running graph using a normalization algorithm to obtain the target running graph with uniform resolution.

[0031] In some embodiments, a target running image is obtained through the graphics upload interface of a human-computer interaction control terminal. The target running image supports high-resolution PNG, BMP bitmap, or SVG vector image formats. A normalization algorithm is used to adjust its pixel coordinate system to a uniform resolution.

[0032] Specifically, operators manually submit real-time main wiring diagrams of the target substation or distribution network through the graphic upload interface of the human-machine interaction control terminal; after receiving the image file, the system calls a normalization algorithm to adjust the global pixel coordinate system of the canvas to a uniform 2560×1440 spatial resolution.

[0033] Step S1012: Based on the pre-trained multimodal large model, identify the connection relationships of each electrical device in the pre-processed target running diagram.

[0034] In some embodiments, the preprocessed target running graph is input into a pre-trained multimodal large model; wherein the multimodal large model is a network model based on vision-language joint training, used to perform primitive recognition and topology parsing tasks of electrical drawings. For example, a pre-trained multimodal large model such as Qwen-35b can be used.

[0035] In some embodiments, to enable the multimodal large model to output deterministic electrical feature analysis results, this disclosure provides location-sensitive structured visual analysis prompts. These prompts include at least the following constraints: Role constraints instruct the large model to perform identification tasks as an expert in electrical drawing standards and primary wiring diagram analysis. Identify task constraints, instruct the large model to identify electrical symbols in the operation diagram, and extract the equipment type, spatial location, and on / off status of each symbol and switch element; Topological constraints instruct the large model to trace the wire routing path and identify device pairs with direct electrical connections. Output format constraints instruct large models to return parsed results in a structured data format (such as JSON) and prohibit the inclusion of unstructured descriptive text.

[0036] The multimodal large model parses the input operation diagram based on the prompt words and outputs the type, location, status of each electrical device and the connection relationship between the devices.

[0037] For example, the following is a sample of the logical architecture design for prompt words: [System Role Declaration] You are a spatial topology analysis expert who is proficient in the national electrical drawing standard (GB / T 4728) and the characteristics of substation primary wiring diagrams.

[0038] [Input Background] Real-time electrical main wiring diagram of a substation manually uploaded by the user.

[0039]

Refined Recognition Task

[0040]

Spatial Connection Constraints

[0041]

Strict Output Format Requirements

[0042] Step S1013: A quadratic verification algorithm based on geometric neighborhood metric is used to verify the connection relationship; In some embodiments, in order to eliminate the inherent spatial perception error of multimodal large models in complex and dense visual scenes, a secondary verification based on geometric neighborhood metric is further performed after the multimodal large model outputs the connection relationship.

[0043] Specifically, the continuous pixel trajectories representing electrical connection wires in the target running graph are extracted to obtain the set of coordinates of the end breakpoints of the wire trajectories. Obtain the bounding box coordinates of each device component. For any two device components, calculate the shortest Euclidean distance from the set of wire end breakpoints to the bounding box of the first device component, and the shortest Euclidean distance to the bounding box of the second device component. If both distances are less than their respective preset geometric tolerance thresholds, it is determined that there is an electrical connection between the two device components; otherwise, it is determined that there is no electrical connection.

[0044] The topology quantization calculation formula for determining whether a deterministic electrical conduction relationship exists between any two device components is as follows:

[0045] in, For equipment components With equipment components The result of determining the connection relationship between them is 1, which indicates that a connection exists, and 0 indicates that a connection does not exist. The first of the trajectories Coordinates of the terminal breakpoints; Let L be the coordinates of the l-th end breakpoint of the traverse trajectory; For equipment components The bounding rectangle of the outer rectangle; For equipment components The bounding rectangle of the outer rectangle; This represents a second-order Euclidean distance spatial metric. and The preset geometric tolerance pixel thresholds for the lead ports of devices at different voltage levels are as follows:

[0046] when At step 1, an electrical connection is determined to exist between the two device components. This geometric verification algorithm is used to review and correct the connection relationships output by the multimodal large model, thereby improving the accuracy of topology analysis.

[0047] Step S1014: Serialize and encapsulate the verified connection relationships to obtain structured topology data.

[0048] In some embodiments, the connection relationships confirmed through geometric verification are subjected to structured encapsulation.

[0049] Specifically, isolated nodes with a degree of zero are removed from the connectivity relationships; these are device components that have no connection to any other device. Device node pairs with confirmed connectivity through geometric verification are then matrix-assembled to form standardized topology data. The assembled topology data is then encapsulated into a structured format file using a serialization middleware.

[0050] In some embodiments, the structured topology data is concurrently transmitted to two paths: one path feeds back to a visual dialog interface for graphical status presentation, and the other path sends it to a subsequent processing module for operation ticket generation. For example, the interface can be displayed in real time via the WebSocket protocol, and data can be sent to the context assembler via an internal message queue.

[0051] In some embodiments, the structured topology data includes the following core fields: substation metadata (including substation name, voltage level and generation timestamp), equipment node list (including unique identifier, type, status, bay, coordinates and voltage level of each device), topology edge list (including unique identifier of connection relationship, starting device, target device and connection type), and bus system information (including bus identifier, voltage level, wiring method and set of connected devices).

[0052] Through the above encapsulation process, structured topology data for subsequent operation ticket generation is obtained.

[0053] For example, the following is an example of how core schema fields are organized: { "Substation_Metadata": { Name: "XXX Substation", "Voltage_Level": "220kV", "Generation_Timestamp": "2026-05-18T11:42:00+08:00" }, "Equipment_Nodes": [ { "ID": "2211", "Type": "Breaker", "Status": "Closed", "Bay": "Anamental spacing", "Coordinates": [0.35, 0.42, 0.38, 0.45], "Voltage_Level": "220kV" }, { "ID": "22111", "Type": "Disconnector", "Status": "Closed", "Bay": "Line I Bay", "Coordinates": [0.32, 0.43, 0.34, 0.44], "Voltage_Level": "220kV" } , "Topology_Edges": { "Edge_ID": "E01", "From": "2211", "To": "22111", "Connection_Type": "Direct" } , "Bus_Systems": { "Bus_ID": "BUS_220_I", "Voltage_Level": "220kV", "Configuration": "Double Bus", "Connected_Equipment": ["2211", "22111", "2212", "22121"] } } ​In some embodiments, by introducing a multimodal large-scale model (VLM) for end-to-end semantic parsing of electrical drawings and combining it with geometric verification algorithms to automatically extract equipment types, spatial coordinates, and connection relationships, the reliance on preset templates and symbol mapping rules is completely eliminated. This mechanism can automatically identify any complex wiring configuration (such as T-buses, angled arrangements, temporary sidings, etc.) and generate compliant structured topology data without manual intervention, significantly improving the system's adaptability to irregular non-standard network structures. Simultaneously, by allowing users to directly upload on-site main wiring diagrams, the system automatically completes the conversion from visual perception to structured topology data and overlays voltage-level adaptive geometric verification, ensuring that the constructed digital topology is strictly consistent with the on-site physical state. This avoids "model-on-site" deviations caused by lag in backend database modeling or omissions in manual selection, achieving near real-time linkage between the graphic display layer and the logical decision-making layer. The accuracy of topology extraction is significantly improved, and the risk of data distortion at the source of the operation ticket is effectively controlled.

[0054] In some embodiments, this step can also use a serial pipeline of "target detection network (such as YOLO series or Faster R-CNN) + optical character recognition + image morphology wire tracing algorithm" instead of a single multimodal large model. First, the target detection network locates the spatial bounding box of components such as switches and knife switches. Then, the optical character recognition is used to extract the physical code of the device. Next, the wire skeleton refinement algorithm is used to extract the physical link of the wire. Finally, the general text parsing module assembles these discrete geometric and text features into standard structured topology data.

[0055] Step S102: Perform dual-path collaborative retrieval on the task instruction to obtain reference information matching the task instruction; and generate fused data based on the reference information and the structured topology data.

[0056] In some embodiments, this step employs a Retrieval-Augmented Generation (RAG) technical architecture. A power operation and maintenance knowledge base is pre-built, and unstructured text is sliced ​​and stored quantitatively. Simultaneously, a heterogeneous knowledge collaborative retrieval server is constructed, comprising a high-dimensional dense vector database (Milvus) and a switching rule hash index engine (Redis). Responding to user-inputted operation task commands, a dual-path collaborative retrieval is performed from the knowledge base via the retrieval server; the dual-path retrieval results are merged and combined with the structured topology data as fused data for subsequent operation ticket generation.

[0057] In some embodiments, the dual-path collaborative retrieval of the task instruction to obtain reference information matching the task instruction; and the generation of fused data based on the reference information and the structured topology data, including: Step S1021: Based on the task instructions, perform semantic vector retrieval and attribute filtering retrieval to obtain candidate reference information.

[0058] In some embodiments, a power operation and maintenance knowledge base is pre-built. Unstructured texts such as the "Power Safety Work Regulations," historical typical operation ticket stubs, and substation end operation rules are overlapped and sliced ​​to form dry text fragments. The chunk size can be set to 400 tokens, and the overlap can be set to 80 tokens. NLTK library is used for protective word segmentation of power-related terminology. A deep text embedding model (e.g., sentence-transformers / all-mpnet-base-v2) is called to convert each text fragment into dense vectors, and the generated vectors are stored in a high-dimensional vector database (e.g., Milvus), thus completing the construction of the power operation and maintenance knowledge base. Each record in the power operation and maintenance knowledge base is attached with metadata tags; the metadata tags include attribute fields such as voltage level, equipment type, and operation type.

[0059] In some embodiments, this disclosure employs a hybrid retrieval and enhancement generation technology architecture specifically designed for high-risk switching tasks. This architecture involves a dual-link retrieval process: the first link calculates generalized cosine similarity between the current dialogue command text and a high-dimensional vector database (containing all safety procedures and typical case slices); the second link uses named entity recognition technology to extract strong physical attributes such as voltage levels and equipment spacing, performing hard Boolean hash mask pruning. After dual-link cross-compilation and rearrangement, the top K (Top-K) core knowledge blocks with the highest scores are dynamically concatenated with the structured data exchange format (JSON) topology structure generated in the first stage, serving as enhanced factual background fed to the large language model.

[0060] Specifically, the first approach is semantic vector retrieval, which calls a deep text embedding model to transform the task instruction T into a query feature vector. The query feature vector is compared with the knowledge slice vector in the pre-built power operation and maintenance knowledge base. The cosine similarity is used to retrieve semantically relevant candidate reference information. The calculation formula is as follows:

[0061] in, This is the query feature vector obtained by converting the user's task instructions; The feature vector corresponding to any text fragment in the pre-built power operation and maintenance knowledge base; The cosine similarity score ranges from -1 to 1, with larger values ​​indicating greater semantic relevance.

[0062] In some embodiments, the retrieval uses GPU-accelerated approximate nearest neighbor search (IVF_FLAT index), with a response time controlled within 800ms.

[0063] The second approach is strong attribute Boolean filtering retrieval (i.e., attribute filtering retrieval). It employs a Named Entity Recognition (NER) model to accurately extract explicit physical constraints from the task instruction T. Attribute matching is performed on the metadata of the pre-built power operation and maintenance knowledge base, directly eliminating irrelevant clauses and filtering out candidate reference information that meets the constraints. The explicit physical constraints include voltage level, equipment type, and operational target state (e.g., voltage level "220kV", equipment "line", operating condition transition "operation to maintenance"). The named entity recognition model is a BERT-BiLSTM-CRF architecture with an F1 score of 92.4%. The mathematical expression for Boolean filtering is:

[0064] in, For the pre-built power operation and maintenance knowledge base Individual metadata attributes (such as voltage level, device type); For the first one extracted from the task instruction One explicit physical constraint; For the filtered results, True means to keep and False means to remove.

[0065] In some embodiments, a dual-path parallel search is initiated in response to a user-inputted operation task command. Specifically, when a user inputs a natural language macro-level switching task command (e.g., "Please generate an operation ticket to change the 220kV Line A from operating status to maintenance status") in the input field of a visual dialog box, the search server initiates a parallel dual-path search. The dual-path search results are then merged to obtain candidate reference information.

[0066] Step S1022: Reorder the candidate reference information based on the cross-coding reordering model to determine the reference information that matches the task instruction.

[0067] In some embodiments, the candidate reference information merged from the dual-path retrieval is input into a cross-encoder reranker (CER) model. The CER model employs a reranking network architecture based on a deep association attention mechanism to calculate the deep business association score between each candidate reference and the current operation task instruction. The reranking score is calculated using the following formula:

[0068] in, For the rearrangement model function, it represents the mapping relationship in which the cross-coding rearrangement model jointly encodes the input task instructions and candidate reference information and calculates the association score; For the first One candidate reference information fragment; The operation task instructions entered by the user; The score is rearranged to indicate the degree of business relevance between a candidate reference and the task instruction; the higher the score, the more relevant it is.

[0069] The candidate reference information is sorted in descending order based on the correlation score, and the top K groups (K is usually 3-5) of core knowledge fragments with the highest scores are selected as reference information for matching the task instructions. The reference information includes relevant mandatory clauses of safety procedures and matching historical typical operating steps.

[0070] In some embodiments, the dual-path retrieval method combining semantic retrieval and attribute filtering, along with the deep association scoring of candidate results by the cross-coding rearrangement model, can significantly improve the knowledge relevance between retrieval results and operation task instructions. Compared with single semantic vector retrieval or single attribute filtering retrieval, it has higher matching accuracy and retrieval efficiency.

[0071] Step S1023: Generate fused data based on the reference information and the structured topology data.

[0072] In some embodiments, the reference information is concatenated with the structured topology data to form fused data containing device topology status, safety procedure constraints, and operational task objectives, which serves as input features for subsequent operation ticket generation.

[0073] In some embodiments, this disclosure enhances the generation architecture by constructing a "topology-semantic" dual-path collaborative retrieval system for the power vertical domain. Real-time topology data is vectorized and fused with unstructured text such as power safety regulations. Employing a dual-path collaborative mechanism of semantic retrieval and attribute filtering, safe and compliant operation steps can be generated solely based on the current equipment status and the latest regulations, even without historical sample support. The knowledge base supports automated incremental updates; new regulations can be vectorized and take effect quickly after import. In low-frequency, high-risk scenarios, the error rate of operation tickets is significantly reduced, and the ticket preparation cycle for newly commissioned substations is greatly shortened, ensuring that the system always complies with the latest safety standards and greatly improving knowledge maintenance efficiency and compliance assurance levels.

[0074] In some embodiments, this step can also employ a hybrid retrieval of power knowledge graphs to replace pure vector retrieval and enhance the generation architecture. Specifically, a knowledge graph or graph database is used to replace or supplement a single dense vector database, transforming the provisions of the "Power Safety Regulations" and historical operation ticket experience into a semantic topological graph containing "equipment—operating conditions—safety constraints—operation sequence". During retrieval, by parsing the entities in the task instructions, subgraph matching and path query are directly performed in the graph, replacing text vector similarity calculation with structured association retrieval to achieve more deterministic rule recall.

[0075] Step S103: Based on the fused data, generate an initial operation ticket using a large language model.

[0076] In some embodiments, the fused data is input into a pre-trained large language model, which possesses power semantic understanding capabilities and is used to perform step-by-step time-series deduction based on structured topology data, reference information, and user task instructions in the fused data. The large language model generates an initial operation ticket according to preset inference criteria and outputs the generated initial operation ticket to a visual dialogue interface in real-time in a streaming manner, realizing a progressive presentation of the operation steps. The context window length of the large language model is configured to be sufficient to accommodate the complete processing of substation topology data.

[0077] In some embodiments, generating an initial operation ticket based on the fused data using a large language model includes: Step S1031: Based on the fused data, construct prompt words that include identity constraints, reasoning rules, and output format.

[0078] In some embodiments, structured topology data, reference information, and user task instructions in the fused data are dynamically concatenated to construct a composite prompt word. The prompt word employs a hierarchical structure and includes at least the following components: The identity constraint layer instructs the large language model to perform the operation ticket generation task as a power operation and maintenance expert, enabling it to make inferences and judgments from a professional perspective.

[0079] The factual basis layer uses the structured topology data as the real-time physical state of the current power grid and the reference information as safety procedure constraints and typical operation references, together forming the factual basis for the generation of operation tickets.

[0080] The task description layer takes the user-input operation task instructions as the target to be executed and clarifies the operation start state and target state.

[0081] The reasoning criteria layer instructs the large language model to perform step-by-step logical deduction according to a preset thought chain before generating operation steps. The reasoning criteria include: verifying the impact of target equipment state transitions on related electrical circuits, analyzing the disconnection sequence of load power supplies, arranging timing actions that comply with the five-prevention interlocking rules, and embedding safety verification points before key operation steps.

[0082] The output format layer instructs the large language model to output operation tickets in a structured format. The structured format includes at least the operation sequence number, the device operation content, and the status verification items after the operation, and prohibits the output of unstructured descriptive text.

[0083] For example, here are some examples of prompt words: [Expert Identity Injection] You are a senior operation and maintenance shift leader who has worked in a power supply company for many years and is responsible for power grid operation monitoring and intrinsic safety protection.

[0084] [Multiple sources of factual evidence] 1. Real-time digital network topology status on site (JSON file obtained from parsing the runtime graph): {{Current_JSON_Topology}} 2. Based on the current dispatch order task, the rigid institutional constraints and typical reference steps matched from the safety procedure database are: {{RAG_Regulations}} [User-issued task] {{User_Task}}

Clock of Reasoning (CoT) Step-by-Step Reasoning Principle

[0085] 2. Perform the [Load Power Supply Stripping Constraint Analysis]: Determine whether, during operation, it is necessary to prioritize disconnecting the load-side circuit before isolating the power supply-side circuit breaker.

[0086] 3. Implement the

Safety Arrangement of Timing Actions

[0087] 4. Perform [Safety Verification Point Implantation]: Automatically insert status check items before critical operation steps, such as "Check that the voltage indication of device XX is zero".

[0088] [Hard requirements for output format] After the logical derivation is completed, the final operation ticket details must be output strictly according to the following format, and any unnecessary or deviating terms are strictly prohibited: [Operation Sequence Number] | [Specific Controlled Equipment Number and Action Standard Terminology] | [Physical Status Checks and Verifications After Execution] In some embodiments, through the above-described composite prompt word architecture design, this disclosure explicitly defines the senior expert identity in the meta prompt words and forces the large language model to follow the three-stage thought chain evolution rule of "power flow topology verification - load power stripping constraint analysis - timing action safety arrangement". At the same time, it stipulates that the model must strictly follow the standard triplet format containing operation sequence number, specific controlled equipment number and action standard terminology, and physical state check and verification items after execution for streaming output, thereby achieving effective constraint on the long time series generation evolution trajectory.

[0089] Step S1032: Input the prompt words into the large language model to generate an initial operation ticket containing the operation sequence number, operation content, and security checkpoint.

[0090] In some embodiments, the prompt word is input into a large language model. The large language model performs step-by-step temporal deduction under the constraints of the prompt word to generate an initial operation ticket. Each step in the initial operation ticket... It includes three core fields: operation sequence number, operation content, and security checkpoint, in the following format:

[0091] The steps for the initial operation ticket are as follows: , For the first operation ticket One operation step; This is the operation sequence number; For operation content; This is a security checkpoint.

[0092] The generated initial operation ticket is streamed to the visual dialogue interface in real time. The streamed output is presented character by character at a preset rate to simulate the rhythm of human writing and improve the user experience. The streamed output rate can be controlled at 125ms / character.

[0093] Step S104: Verify the initial operation ticket; if the verification fails, generate feedback information and correct the initial operation ticket based on the feedback information until the verification passes, and output the target operation ticket.

[0094] In some embodiments, this step utilizes a first-order predicate five-defense logic state machine built based on the graph database (Neo4j) to perform physical-level security control on each operation step generated in real time by the large model. By monitoring the output stream of the inference engine through shared memory, when a violation is detected, the intercepted violation motive is immediately formatted as a feedback prompt word and incrementally sent back to the inference engine via a callback function, guiding it to perform self-healing correction within the same session context. The verification latency does not exceed 200ms to ensure a smooth user experience.

[0095] In some embodiments, the step of verifying the initial operation ticket; in response to the verification failing, generating feedback information, and correcting the initial operation ticket based on the feedback information until the verification passes, and outputting the target operation ticket, includes: Step S1041: Construct a virtual state machine based on the structured topology data, convert each operation step in the initial operation ticket into a state modification instruction, and perform simulation and rehearsal in the virtual state machine to verify whether the five-prevention rules are violated.

[0096] In some embodiments, a virtual state machine is constructed based on the structured topology data simultaneously with the generation of the initial operation ticket. The virtual state machine embeds power safety rules for performing physical-level security checks on the operation steps.

[0097] Each operation step in the initial operation ticket The sequence is converted into state modification instructions for the corresponding devices in the structured topology data, and then arranged in the virtual state machine. The state transition function is defined as follows:

[0098] in, express The system state at any given moment; The system state after execution; This is a state transition function used to describe the state update rules after performing operation steps.

[0099] With extremely high safety standards, "open the disconnect switch on a certain line side" Taking this action as an example, its anti-misoperation interlocking verification function is defined as follows:

[0100] in, Indicates the location at the disconnect switch A collection of all circuit breakers (switches) that are in the same electrical compartment and directly connected in series; Indicates circuit breaker The current status; Open indicates the circuit is open, Closed indicates the circuit is closed.

[0101] The logic of this verification function is as follows: if the set If all circuit breakers are in the open state, the verification passes and the operation of opening the disconnecting switch is allowed; if any circuit breaker is in the closed state, the verification fails and the operation is prohibited to prevent accidents caused by opening the disconnecting switch under load.

[0102] If the verification passes, proceed to the next operation step; if the verification fails, intercept the current operation step and output violation information, triggering the subsequent correction process.

[0103] In some embodiments, while the operation step sequence is output in a streaming manner, the verification network element simultaneously performs five-prevention rule-based error prevention judgment in the background. Once an operation step triggers the five-prevention blocking rule, the verification network element immediately intercepts the output and extracts the violation step number, conflicting device code, and violated error prevention blocking mechanism as violation information. This information is then converted into self-healing correction information and input into the large language model, driving it to perform autonomous rewriting and repair of the violation step and subsequent steps within the same session context.

[0104] Step S1042: In response to any of the operation steps failing the verification, the sequence number of the violation step and the reason for the violation are extracted, and feedback information is generated.

[0105] In some embodiments, when the virtual state machine fails to verify a certain operation step, violation information is extracted. The violation information includes: the sequence number of the violation step, the type of the five-prevention rule violated, and the identifier and status of the currently conflicting device.

[0106] Based on the violation information, feedback information containing correction instructions is generated, instructing the large language model to partially rewrite the violation step and subsequent steps. The correction instructions include at least: the specific location of the violation step, a description of the reason for the violation, and a correction suggestion.

[0107] For example, highlight the text output in the conversation interface in red to block it, and automatically append a "correction feedback prompt" pop-up to the conversation flow: [System hard check interception and error correction notification] Warning: The initial operation ticket sequence you compiled is in step m (pulling the isolation switch). This location violates the hard constraints of the substation's five-prevention interlocking mechanism! A circuit breaker currently connected in series with this location has been detected. It is still in the "Closed" state.

[0108] [Self-healing correction command] The system has paused output. Please review the current topology. Before executing step m, you must add "Disconnect the circuit breaker". For the operation ", please rewrite the conflicting step and all subsequent linked steps locally.

[0109] Step S1043: Based on the feedback information, use the large language model to locally modify the initial operation ticket until all steps pass the verification, and output the target operation ticket.

[0110] In some embodiments, the feedback information is input into the large language model. Based on the correction instructions in the feedback information, the large language model utilizes its context window's incremental update capability to partially rewrite the initial operation ticket within the same session context. Specifically, only the violating steps and subsequent affected steps are regenerated; the validated correct steps are retained.

[0111] Repeat the verification, feedback, and local correction steps until all operations pass verification. The iteration termination condition is:

[0112] Where n is the total number of operation steps; For the first One operation step; To execute the first The system state before the step; This is a verification function; returning True indicates that the step has passed the five-prevention rule verification.

[0113] When all operation steps meet the above conditions, that is, when the pass rate of the full-step pre-simulation of the entire operation ticket in the state machine reaches 100%, the iteration terminates, the target operation ticket is output, and finally a digital operation ticket with electronic check boxes and conforming to the industry standard format will be rendered, which can be exported as PDF or printed by the operators for execution.

[0114] In some embodiments, this step constructs a closed-loop control process of "generation-interception-feedback-correction" by integrating the heuristic reasoning capabilities of a large language model with the hard-locking verification mechanism of a five-defense rule state machine. While generating operation steps, virtual topology state evolution is performed synchronously, and illegal operations are intercepted in real time through a verification function. Once a conflict is detected, the system automatically generates error correction feedback information, driving the large language model to locally correct the illegal steps and subsequent steps within the same session context. This mechanism significantly improves the success rate of generating complex operation tickets (such as full-site bus reversal and multi-loop closure) and greatly reduces the frequency of manual intervention, thereby significantly enhancing the system's robustness and autonomy in high-risk, long-term scenarios.

[0115] In some embodiments, this step can also use a dual-model adversarial verification mechanism of "Generation Agent + Supervision Agent" to replace the hard-coded first-order predicate five-defense state machine. The supervision model is dedicated to loading the full set of security procedures as system constraint prompts. When the generation model streams the operation steps, the supervision model, as an independent reviewer, engages in multiple rounds of semantic review and debate with it in the conversation stream until the two models reach a conflict-free consensus, thereby achieving security control based on semantic adversarial mechanisms.

[0116] In some embodiments, this step can also use a formal verification engine (such as the Z3 theorem solver or the SMV model checker) instead of the traditional Boolean five-defense matrix. The operation sequence produced by the large model is transformed into a first-order temporal logic formula, and formal deduction of it in the full state space is performed using satisfiability model theory, thereby discovering hidden dynamic locking conflicts and generating differential error information.

[0117] This disclosure addresses the problem of insufficient adaptive capability under irregular and non-standard grid structures by introducing a multimodal large model for end-to-end semantic parsing of electrical drawings and automatically extracting equipment types, spatial coordinates, and connection relationships using a geometric verification algorithm. It also resolves the safety risks caused by information gaps between physical drawings and digital models by constructing a precise digital topology model through pixel-level analysis of primary wiring diagrams. Furthermore, it addresses the issues of insufficient reasoning ability and lack of self-repair mechanisms in complex time-series operation tasks by integrating thought chain reasoning and closed-loop verification mechanisms, combined with real-time feedback correction of deterministic five-prevention rules, and establishing a self-healing closed loop. Finally, it solves the problems of excessive reliance on historical samples and delayed response to procedure updates by constructing a "topology-semantics" collaborative retrieval and enhanced generation architecture that deeply semantically fuses structured topology data with unstructured procedure text and dynamically updates the knowledge base.

[0118] This disclosure constructs an intelligent switching operation ticket compilation system, consisting of a visual large-scale model dialogue interaction interface, digital reconstruction of the operational graph space topology, enhanced multi-dimensional knowledge retrieval, and self-healing correction. It deeply integrates the semantic reasoning capabilities of a large language model with the rigid physical constraints of a deterministic electrical rule state machine, systematically solving the core bottlenecks of existing technologies in terms of adaptability, security, intelligence, and timeliness. This achieves a paradigm shift from "passive response" to "proactive intelligence," providing solid technical support for the inherent safety and efficient operation and maintenance of new power systems. It is understood that this method can be executed by any device, equipment, platform, or equipment cluster with computing and processing capabilities.

[0119] It should be noted that the methods of one or more embodiments of this disclosure can be executed by a single device, such as a computer or server. The methods of this embodiment can also be applied in a distributed scenario, where multiple devices cooperate to complete the process. In such a distributed scenario, one of these devices may execute only one or more steps of the methods of one or more embodiments of this disclosure, and the multiple devices will interact with each other to complete the method described.

[0120] It should be noted that the above description pertains to specific embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0121] Figure 2 This is a schematic diagram illustrating the overall operation flow of an automatic operation ticket generation method according to an embodiment of this disclosure. Figure 2 As shown, the method provided in this embodiment generally includes the following steps: First, we enter the first stage, namely "manual interactive uploading and spatial topology analysis of the runtime graph based on the multimodal large model". Users manually upload the target runtime graph, and the multimodal large model performs spatial topology analysis on the runtime graph to generate structured topology data.

[0122] Secondly, the process moves to the second stage, namely "RAG knowledge base construction and matching based on heterogeneous dual-path retrieval and semantic reordering". According to the natural language task instructions input by the user, heterogeneous dual-path retrieval and semantic reordering are performed to obtain relevant reference information from the knowledge base and then merge it with the structured topological data generated in the first stage.

[0123] Next, the process moves to the third stage, namely "synthesis of heuristic thought chain operation tickets driven by visual dialogue tasks." This involves assembling compound prompts to drive the large language model to perform step-by-step temporal deductions along the thought chain, generating initial operation tickets and outputting them in a streaming manner.

[0124] Subsequently, the process moves to the fourth stage, namely "deterministic state machine hard-locking interception and dialogic self-healing correction closed loop." The verification network element simultaneously performs digital twin state transition simulation and five-prevention rule verification in the background, using state machine calculations to determine whether the operation ticket passes the five-prevention lockout verification. If the state machine does not return an error (i.e., the verification passes), a standard digital operation ticket is rendered and output, and the process ends. If the state machine returns an error (i.e., the verification fails), an interception is immediately executed, error correction feedback information is generated, and the process returns to the third stage. This drives the large language model to perform local self-healing rewriting of the violation steps and subsequent steps, forming a closed loop of iteration until all steps pass the verification and the operation ticket is output.

[0125] Based on the same inventive concept, corresponding to any of the above-described embodiments, this disclosure also provides an automatic operation ticket generation system. For example... Figure 3 As shown, the above system includes: The acquisition module 301 is configured to acquire the target running graph and task instructions, and obtain structured topology data based on the target running graph; The retrieval module 302 is configured to perform a dual-path collaborative retrieval of the task instruction to obtain reference information matching the task instruction; and generate fused data based on the reference information and the structured topology data. The generation module 303 is configured to generate an initial operation ticket based on the fused data using a large language model; The verification module 304 is configured to verify the initial operation ticket; in response to the verification failing, generate feedback information and correct the initial operation ticket based on the feedback information until the verification passes, and output the target operation ticket.

[0126] For ease of description, the above system is described by dividing it into various modules based on their functions. Of course, when implementing one or more embodiments of this disclosure, the functions of each module can be implemented in one or more software and / or hardware.

[0127] The system described above is used to implement the corresponding methods in the foregoing embodiments and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0128] Figure 4This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0129] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure.

[0130] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this disclosure are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0131] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0132] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0133] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0134] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this disclosure, and not necessarily all the components shown in the figures.

[0135] The electronic devices described above are used to implement the corresponding methods in the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0136] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0137] Based on the same inventive concept, corresponding to the automatic operation ticket generation method of any of the above embodiments, this disclosure also provides a computer program product, which includes one or more computer programs. In some embodiments, the one or more computer programs are executable by one or more processors to cause the one or more processors to execute the automatic operation ticket generation method. Corresponding to the execution entity for each step in each embodiment of the automatic operation ticket generation method, the processor executing the corresponding step may belong to the corresponding execution entity. The computer program product of the above embodiments is used to cause the processor to execute the automatic operation ticket generation method as described in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0138] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure (including the claims) is limited to these examples; within the framework of this disclosure, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this disclosure as described above, which are not provided in detail for the sake of brevity.

[0139] Additionally, to simplify the description and discussion, and to avoid obscuring one or more embodiments of this disclosure, the provided drawings may or may not show well-known power / ground connections to integrated circuit (IC) chips and other components. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring one or more embodiments of this disclosure, and this also takes into account the fact that the details of implementation of these block diagram apparatuses are highly dependent on the platform on which one or more embodiments of this disclosure will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuitry) are set forth to describe exemplary embodiments of this disclosure, it will be apparent to those skilled in the art that one or more embodiments of this disclosure may be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0140] Although this disclosure has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0141] This disclosure includes one or more embodiments intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for automatically generating operation tickets, characterized in that, The method includes: Obtain the target runtime graph and task instructions, and obtain structured topology data based on the target runtime graph; The task instruction is subjected to dual-path collaborative retrieval to obtain reference information matching the task instruction; and fused data is generated based on the reference information and the structured topology data. Based on the fused data, an initial operation ticket is generated using a large language model; The initial operation ticket is verified; if the verification fails, feedback information is generated, and the initial operation ticket is corrected based on the feedback information until the verification passes, and the target operation ticket is output.

2. The method according to claim 1, characterized in that, The step of obtaining the target runtime graph and task instructions, and obtaining structured topology data based on the target runtime graph, includes: The target running graph is obtained, and the target running graph is preprocessed using a normalization algorithm to obtain the target running graph with uniform resolution; Based on a pre-trained multimodal large model, the connection relationships of various electrical devices in the pre-processed target running diagram are identified; A quadratic verification algorithm based on geometric neighborhood metric is used to verify the connection relationship; The verified connection relationships are serialized and encapsulated to obtain structured topology data.

3. The method according to claim 1, characterized in that, The task instruction is subjected to dual-path collaborative retrieval to obtain reference information that matches the task instruction; Based on the reference information and the structured topology data, fused data is generated, including: Based on the task instructions, semantic vector retrieval and attribute filtering retrieval are performed to obtain candidate reference information; The candidate reference information is reordered based on the cross-coding reordering model to determine the reference information that matches the task instruction; Based on the reference information and the structured topology data, fused data is generated.

4. The method according to claim 3, characterized in that, The semantic vector retrieval includes: The task instructions are transformed into query feature vectors by calling a deep text embedding model. The query feature vector is compared with the knowledge slice vector in the pre-built power operation and maintenance knowledge base. The cosine similarity is used to retrieve semantically relevant candidate reference information. The calculation formula is as follows: in, This is the query feature vector obtained by converting the user's task instructions; The feature vector corresponding to any text fragment in the pre-built power operation and maintenance knowledge base; The cosine similarity score has a value range of [-1, 1].

5. The method according to claim 3, characterized in that, The attribute filtering retrieval includes: A named entity recognition model is used to accurately extract explicit physical constraints from the task instructions. Attribute matching is performed on the metadata of a pre-built power operation and maintenance knowledge base to directly eliminate irrelevant clauses and filter out candidate reference information that meets the constraint conditions. The mathematical expression of Boolean filtering is as follows: in, For the pre-built power operation and maintenance knowledge base Each metadata attribute; For the first one extracted from the task instruction One explicit physical constraint; For the filtered results, True means to keep and False means to remove.

6. The method according to claim 1, characterized in that, The step of generating an initial operation ticket based on the fused data using a large language model includes: Based on the fused data, prompt words containing identity constraints, reasoning rules, and output formats are constructed; The prompt words are input into the large language model to generate an initial operation ticket containing the operation sequence number, operation content, and security checkpoints.

7. The method according to claim 1, characterized in that, The process of verifying the initial operation ticket, and generating feedback information in response to a failed verification, and correcting the initial operation ticket based on the feedback information until the verification passes, and then outputting the target operation ticket, includes: A virtual state machine is constructed based on the structured topology data. Each operation step in the initial operation ticket is converted into a state modification instruction. The simulation is performed in the virtual state machine to verify whether the five-prevention rules are violated. If any of the operation steps fails the verification, the sequence number of the violation step and the reason for the violation are extracted, and feedback information is generated. Based on the feedback information, the initial operation ticket is locally modified using the large language model until all steps pass the verification, and the target operation ticket is output.

8. An automatic ticket generation system, characterized in that, The system includes: The acquisition module is configured to acquire the target running graph and task instructions, and obtain structured topology data based on the target running graph; The retrieval module is configured to perform a dual-path collaborative retrieval of the task instruction to obtain reference information matching the task instruction; and generate fused data based on the reference information and the structured topology data. The generation module is configured to generate an initial operation ticket based on the fused data using a large language model; The verification module is configured to verify the initial operation ticket; in response to the verification failing, it generates feedback information and corrects the initial operation ticket based on the feedback information until the verification passes and outputs the target operation ticket.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executed by the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions for causing the computer to perform the method according to any one of claims 1 to 7.

11. A computer program product, characterized in that, It includes one or more computer programs that, when executed by one or more processors, implement the method as described in any one of claims 1 to 7.

Images