A bill face language standard recommendation system based on a term graph

Abstract

The application discloses a kind of based on terminology atlas's ticket surface language standard recommendation system, the system is aimed at the deficiencies of prior art in language specification guiding ability, weak semantic understanding ability, single input mode and isolated verification link, etc., proposed to knowledge graph driven semantic intelligent recommendation mechanism, by constructing the terminology atlas of two ticket business, key terms such as job type, equipment object, safety measures are structured modeling, and combined with semantic analysis, semantic matching, speech recognition, graphical interaction and OCR identification etc. Multi-technology fusion means, realize the intelligent generation of ticket surface language, specification verification and multi-modal input adaptation, effectively improve the accuracy, consistency and efficiency of ticket filling, make the two ticket filling process from experience to knowledge driven, realize the standardization generation and standardization output of ticket surface sentence, significantly improve the intelligent level of system and the quality of operation management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data recommendation, and particularly relates to a ticket surface language standard recommendation system based on a term graph. BACKGROUND

[0002] In the current power system operation and management, the "two-ticket" system (i.e., work ticket and operation ticket) as an important means to ensure on-site operation safety, implement responsibility division and standardize operation process, is widely used in power distribution, maintenance, construction and other business scenarios. However, the existing two-ticket business generally has problems such as non-uniform language expression, missing standard terms, heavy manual input burden and ambiguous safety measure (also known as "safety measure") expression in the process of filling out the ticket surface, which seriously affects the efficiency of ticket review, the accuracy of operation risk assessment and the efficiency of subsequent operation and maintenance data structuring.

[0003] In the existing filling of the ticket surface, personal experience of the operation personnel is mainly relied on for free expression, and there is a lack of structured guidance for standard terms and standard content, resulting in differences in expression of the same type of operation task among different personnel, making it difficult to uniformly archive and accurately identify.

[0004] In addition, the existing system mainly uses static templates or fixed field verification, and fails to fully integrate operation context semantic information for intelligent recommendation and error identification, making it difficult to meet the business needs of multi-source input and multi-scenario fusion.

[0005] At the same time, the actual input forms such as voice input, graphical auxiliary point selection and paper ticket backfilling have insufficient support in the traditional system, and cannot realize full-process closed-loop management of ticket surface information. SUMMARY

[0006] To solve the problems existing in the prior art, the present application provides a ticket surface language standard recommendation system based on a term graph.

[0007] The technical scheme of the present application is as follows:

[0008] A ticket surface language standard recommendation system based on a term graph, comprising a term graph construction module, a text content processing module, a graphical point selection content processing module, a paper image content processing module and a verification module.

[0009] The term graph construction module is configured to construct a term node set containing four types of term nodes of job type, device object, safety measure category, and ticket field according to standard term data of power operation, extract operation action and device collocation relationship and safety term and operation context adaptation relationship based on historical ticket data, construct a semantic relationship edge set, establish a term graph with term nodes in the term node set and semantic relationship edges in the semantic relationship edge set, and form a plurality of standard sentence templates composed of normalized sentences with node fields and formed by connecting term nodes according to semantic relationship edges.

[0010] The text content processing module is configured to perform semantic analysis on user input text and voice converted text, extract key semantic entities including job type, device object, location coordinate, and safety measure category, perform vector matching of the extracted key semantic entities with term nodes in the term graph, filter out target standard sentence templates, generate structured tickets, and push them.

[0011] The graphical point selection content processing module is configured to extract device object and attachment point coordinates from single-line graph node information input by graphical interaction point selection of a user, associate the term graph with standard sentence templates to generate structured tickets, and push them.

[0012] The paper image content processing module is configured to perform text recognition on a paper ticket image input by a user, generate structured tickets, and push them.

[0013] The verification module is configured to perform ticket specification verification on generated structured tickets.

[0014] Further, the specific method of constructing a term node set containing four types of term nodes of job type, device object, safety measure category, and ticket field according to standard term data of power operation, extracting operation action and device collocation relationship and safety term and operation context adaptation relationship based on historical ticket data, constructing a semantic relationship edge set, establishing a term graph with term nodes in the term node set and semantic relationship edges in the semantic relationship edge set, and forming a plurality of standard sentence templates is as follows:

[0015] Standard term data related to two-ticket business of power operation is collected, and standard job type, typical device name, safety measure sentence, operation object attribute, commonly used ticket term, and associated information thereof are extracted to construct an initial term vocabulary.

[0016] The terms in the initial term vocabulary are classified and processed according to four types of structures of job type, device object, safety measure category, and ticket field to divide the term node set.

[0017] Based on historical ticket surface data, extract the semantic association between terms, including the operation action and device collocation relationship and the safety term and operation context adaptation relationship, and construct a set of semantic relationship edges between the terms;

[0018] Organize the set of term nodes and the set of semantic relationship edges into a graph structure data model, and construct a term graph for ticket surface language standard recommendation.

[0019] Further, the text and voice converted text input by the user are semantically analyzed to extract key semantic entities including operation type, device object, location coordinate and safety measure category, the extracted key semantic entities are vectorized matched with the term nodes in the term graph, and the target standard sentence template is screened out, and the specific method of generating and pushing the structured ticket surface is as follows:

[0020] Receive the original ticket surface content input by the user, the original ticket surface content including text and voice converted text;

[0021] Perform text preprocessing operation on the original ticket surface content, the text preprocessing operation including character cleaning, sentence segmentation and stop word filtering, and generate preprocessed sentences;

[0022] Perform morphological analysis on the preprocessed sentences, use a word segmentation algorithm to segment continuous text, and perform part-of-speech tagging on the segmentation results;

[0023] Based on the part-of-speech tagging results, perform named entity recognition operation to identify business key terms including operation action words, device name words, location representation words and safety term words;

[0024] Classify and group the identified business key terms according to operation elements to form key semantic entities including operation type, device object, safety measure category and ticket surface field;

[0025] Vectorize match the key semantic entities with the term nodes in the term graph, screen out the target standard sentence template, and generate and push the structured ticket surface.

[0026] Further, the specific method of vectorizing matching the key semantic entities with the term nodes in the term graph, screening out the target standard sentence template, and generating and pushing the structured ticket surface includes:

[0027] Position correspondence of key semantic entities according to ticket surface fields and node fields of each standard sentence template;

[0028] Construct semantic vector representation for term nodes in the standard sentence template, the semantic vector including node term text information and its semantic context in the connection structure in the graph;

[0029] constructing input semantic vector representation for key semantic entities, generating query vector set;

[0030] adopting similarity calculation strategy, performing semantic matching scoring between semantic vector representation of standard sentence template and query vector set, generating matching score list;

[0031] sorting standard sentence templates in the matching score list, and selecting the one with the highest score as the target standard sentence template;

[0032] embedding the extracted key semantic entities into the reserved position of the target standard sentence template to generate structured ticket surface; recommending the structured ticket surface.

[0033] Further, the specific method of embedding the extracted key semantic entities into the reserved position of the target standard sentence template to generate structured ticket surface comprises:

[0034] receiving the target standard sentence template and the extracted key semantic entities, and replacing the placeholder field in the target standard sentence template with the content of the key semantic entities;

[0035] inserting job type, device object, safety measure and ticket field into the corresponding position according to the field definition order in the target standard sentence template to generate complete ticket surface sentence;

[0036] performing syntax structure arrangement on the complete ticket surface sentence, including conjunction optimization, word order adjustment and unit standardization processing, to generate structured content conforming to language specifications;

[0037] mapping the structured ticket surface content to the ticket information form according to the ticket field to generate structured ticket surface.

[0038] Further, the specific method of extracting device object and hanging point coordinates from the single-line graph node information input by the user through graphical interaction, associating the term graph and the standard sentence template to generate structured ticket surface and push is as follows:

[0039] receiving the single-line graph node information selected by the user in the graphical interaction interface, and extracting the selected device name and hanging point coordinates;

[0040] associating and searching the extracted device name with the device object term node in the term graph to locate the corresponding device object and job type;

[0041] According to the hanging point coordinates, combining the device topology structure and the job range data, filtering the applicable safety measure sentence in the preset safety measure sentence template, and constructing a recommendation list to generate structured ticket surface content;

[0042] synchronously displaying the structured ticket surface content in the form of graphics and text to the ticket filling area;

[0043] Output the final determined structured ticket surface.

[0044] Further, the method for obtaining voice converted text is as follows:

[0045] Receive the ticket filling voice instructions input by the user through voice, collect voice data stream containing job content, operation object, safety measures, and auxiliary query semantic information;

[0046] Perform voice recognition on the voice data stream, execute audio noise reduction, voice segmentation, and semantic feature extraction operations, and generate corresponding text transcription results;

[0047] Format the text transcription results, remove repeated words, mood words, and non-ticket related words, and form voice converted text with clear semantics.

[0048] Further, the specific method for recognizing the paper ticket image input by the user, generating a structured ticket, and pushing is as follows:

[0049] Receive the paper ticket image uploaded by the user, perform image preprocessing operations on the paper ticket image, and extract the text area to be recognized; the image preprocessing operations include image denoising, tilt correction, and region segmentation,

[0050] Perform line-by-line recognition on the text area using a character recognition algorithm to generate corresponding structured text content, including job content, device object, safety measure category, and ticket field;

[0051] Classify and map the structured text content according to the ticket field, and backfill it into the corresponding ticket field of the ticket filling area;

[0052] Perform missing detection and content integrity comparison on the backfilled ticket field, mark the field content that fails to recognize or is incomplete, and generate a structured ticket;

[0053] Push the structured ticket.

[0054] Further, the specific method for performing ticket specification verification on the generated structured ticket is as follows:

[0055] Receive the structured ticket;

[0056] Perform ticket specification verification on the structured ticket, and perform field-level verification according to the term nodes in the term graph and the ticket filling area;

[0057] Perform matching analysis on the job type and device object in the structured ticket to verify the semantic consistency between them;

[0058] The security category in the structured ticket surface is compared, whether there is content that does not conform to the standard term specification or expression is ambiguous, and whether there is a historical high-frequency error language is compared;

[0059] The integrity of the ticket field in the structured ticket surface is detected, and the missing field, the semantic conflict field and the inconsistent field are marked;

[0060] A structured checking report is generated, including problem field, error type, recommended modification suggestion and reference standard template information.

[0061] Compared with the prior art, the present application has the following beneficial effects:

[0062] The present application proposes a ticket language standard recommendation system based on a term graph, which aims at the deficiencies of the prior art in the aspects of poor language specification guidance ability, weak semantic understanding ability, single input mode and isolated verification link, etc. A semantic intelligent recommendation mechanism driven by a knowledge graph is proposed, key terms such as job type, equipment object and safety measures are structured modeled by constructing a term graph for two-ticket business, and combined with semantic analysis, semantic matching, speech recognition, graphical interaction and OCR recognition and other multi-technology fusion means, intelligent generation, specification verification and multi-modal input adaptation of ticket language are realized, which effectively improves the accuracy, consistency and efficiency of ticket filling, makes the two-ticket filling process from experience to knowledge driven, realizes the standardized generation and standardized output of ticket language, and significantly improves the intelligent level of the system and the quality of job management.

[0063] The system of the present application realizes the structured modeling and semantic association expression of key terms such as job type, equipment name and safety measure category (safety measure category) by constructing a term graph for two-ticket business, and the system has the understanding ability and semantic reasoning ability of natural language content. Combined with the semantic analysis and similarity matching mechanism, the system can recommend and automatically fill in the standard language for the free text or voice content input by the user, which significantly improves the consistency and accuracy of the ticket language and avoids common ambiguous expressions and non-standard descriptions.

[0064] The system of the present application integrates speech recognition text, graphical interaction point selection and paper ticket image and other input modes, which meets the operation needs of on-site workers in different environments. Users can generate ticket text by voice command, select equipment and hanging points through graphical method and automatically generate job items, or scan and upload paper ticket to complete structured recognition and backfilling, which truly realizes multi-modal adaptation of ticket content collection, reduces the filling burden of grassroots personnel and improves the input efficiency.

[0065] The system of the present application introduces a face specification checking mechanism, combines a standard term set and a face filling template preset in a term graph, performs semantic consistency, logical integrity and format specification checking on the content of the generated structured face, and generates modification suggestions, so as to ensure that the face content has reached a neat, accurate and standard state before submission, strengthen the first safety checkpoint before face auditing, and improve the safety management level of the overall operation.

[0066] The system of the present application constructs a face language generation and auditing system based on a knowledge graph, realizes a full-process closed loop from semantic analysis to structured content output and from intelligent recommendation to checking feedback, changes the two-ticket management from manual filling and passive checking to intelligent generation and active control, significantly improves the standardization, intelligence and convenience of the two-ticket business in actual operation, overcomes the technical defects of the traditional system in aspects of non-uniform face language, manual filling and weak checking capacity, and has strong engineering promotion value and practical application effect. BRIEF DESCRIPTION OF DRAWINGS

[0067] Figure 1 A processing flow diagram of a face language standard recommendation system based on a term graph of the present application is shown in the figure.

[0068] Figure 2 A block diagram of a face language standard recommendation system based on a term graph of the present application is shown in the figure. DETAILED DESCRIPTION

[0069] The present application will be further illustrated below in combination with the drawings and specific embodiments, and it should be understood that these embodiments are only used to illustrate the present application and not used to limit the scope of the present application, and various modifications of the present application made by those skilled in the art after reading the present application all fall within the scope defined by the appended claims of the present application.

[0070] Embodiment:

[0071] A face language standard recommendation system based on a term graph of the present application, as shown in Figure 1 and Figure 2 , includes a term graph construction module, a text content processing module, a graphical point selection content processing module, a paper image content processing module and a checking module.

[0072] The term graph construction module is used to construct a term node set containing four types of term nodes of job type, device object, safety measure category and ticket field according to the standard term data of power operation, extract operation action and device collocation relationship and safety term and operation situation adaptation relationship based on historical ticket data, construct a semantic relationship edge set, establish a term graph with the term nodes in the term node set and the semantic relationship edges in the semantic relationship edge set, and form a plurality of standard sentence templates, the standard sentence templates are formed by connecting the term nodes according to the semantic relationship edges and are composed of normalized sentences with node fields;

[0073] The text content processing module is used for semantic analysis of the text input by the user and the text converted from voice, extracts key semantic entities including job type, device object, location coordinate and safety measure category, performs vector matching of the extracted key semantic entities and the term nodes in the term graph, screens out a target standard sentence template from the plurality of standard sentence templates, generates a structured ticket and pushes it;

[0074] The graphical point selection content processing module is used for extracting device object and hanging point coordinate from the single-line graph node information input by the user through graphical interaction point selection, generating a structured ticket by associating the term graph and the standard sentence template and pushing it;

[0075] The paper image content processing module (also referred to as "OCR backfilling module") is used for text recognition of the paper ticket image input by the user, generating a structured ticket and pushing it;

[0076] The verification module (also referred to as "normative verification module") is used for ticket normative verification of the generated structured ticket.

[0077] The processing flow of the system adopts a term graph-based ticket language standard recommendation method as follows, as shown in the following figure: Figure 1 The method includes the following steps:

[0078] Step 1) Construct a term node set containing four types of term nodes of job type, device object, safety measure category and ticket field according to the standard term data of power operation, extract operation action and device collocation relationship and safety term and operation situation adaptation relationship based on historical ticket data, construct a semantic relationship edge set, establish a term graph with the term nodes in the term node set and the semantic relationship edges in the semantic relationship edge set, and form a standard sentence template, the standard sentence template is formed by connecting the term nodes according to the semantic relationship edges and is composed of normalized sentences with node fields;

[0079] Step 2) The text input by the user is semantically analyzed, and the text converted from voice is processed in sequence for word segmentation, part-of-speech tagging, and named entity recognition. Key semantic entities including the type of work, the object of equipment, the coordinates of the location, and the category of safety measures are extracted. The extracted key semantic entities are vectorized and matched with the term nodes in the term graph, and the target standard sentence template is selected from a plurality of standard sentence templates, thereby generating a structured ticket and pushing it;

[0080] Step 3) The single-line graph node information input by the user through graphical interaction is extracted for the object of equipment and the coordinates of the hanging point. The structured ticket is generated and pushed by associating the term graph and the standard sentence template;

[0081] Step 4) The paper ticket image input by the user is recognized for text, and the structured ticket is generated and pushed;

[0082] Step 5) The structured ticket is checked for ticket specification.

[0083] In the present embodiment, the specific method of step 1) is as follows:

[0084] Standard term data related to the two-ticket business of power operation is collected, and standard operation types, typical equipment names, safety measure sentences, operation object attributes, commonly used ticket terms, and their upper and lower association information are extracted to construct an initial term vocabulary.

[0085] The terms in the initial term vocabulary are classified and processed. The term node set is divided according to four types of structures, i.e., operation type, equipment object, safety measure category, and ticket field, and each term node is given a unique code identifier.

[0086] Based on historical ticket data, the semantic associations between terms are extracted, including the operation action and equipment collocation relationship and the safety term and operation context adaptation relationship, and the semantic relationship edge set between terms is constructed. The historical ticket data generally includes ticket images, ticket records, safety regulation documents, and field language records.

[0087] The term node set and the semantic relationship edge set are organized as a graph structure data model to construct a term graph for standard recommendation of ticket language.

[0088] The term graph data model is imported into a graph database, the term node attribute field and the semantic relationship edge attribute field are set, and the graph index is established and stored for supporting subsequent semantic calculation and graph calling operations.

[0089] In the present embodiment, the specific method of step 2) is as follows:

[0090] The original ticket surface content received by the user input includes text and voice converted text. The text is generally a ticket surface description sentence input through a text input interface, and the voice converted text is a text transcribed through voice recognition.

[0091] The original ticket surface content is subjected to text preprocessing operations, including character cleaning, sentence segmentation, and stop word filtering, to generate preprocessed sentences for subsequent analysis.

[0092] The preprocessed sentences are subjected to lexical analysis, using a segmentation algorithm to segment continuous text and performing part-of-speech tagging on each segmentation result.

[0093] Based on the part-of-speech tagging results, a named entity recognition operation is performed to identify business key terms including job action words, device name words, location representation words, and safety terminology words, and to identify the category to which the vocabulary belongs.

[0094] The identified business key terms are classified and grouped according to job elements to form key semantic entities including job type, device object, safety category, and ticket surface field.

[0095] The key semantic entities are vectorized and matched with the term nodes in the term graph, and the target standard sentence template is selected, thereby generating a structured ticket surface and pushing it.

[0096] In this embodiment, the specific steps of vectorizing and matching the key semantic entities with the term nodes in the term graph, selecting the target standard sentence template, and thereby generating a structured ticket surface and pushing it include:

[0097] The key semantic entities are positionally corresponding to the ticket surface fields and the node fields of each standard sentence template.

[0098] The term nodes in the standard sentence template are constructed into semantic vector representations, including node term text information and semantic context of the connection structure in the graph.

[0099] The key semantic entities corresponding to the original ticket surface content are constructed into input semantic vector representations to generate a query vector set.

[0100] Using a similarity calculation strategy, the semantic vector representations of the standard sentence templates and the query vector set are subjected to semantic matching scoring to generate a matching score list.

[0101] The standard sentence templates in the matching score list are sorted, and the one with the highest score is selected as the target standard sentence template.

[0102] The extracted key semantic entities are embedded into the reserved positions of the target standard sentence template according to the preset field rules to generate a structured ticket surface candidate result.

[0103] The structured ticket surface is recommended for user interface display or subsequent content confirmation processing.

[0104] In this embodiment, the extracted key semantic entities are embedded into the reserved positions of the target standard sentence template, and the specific steps of generating the structured ticket surface content include:

[0105] The target standard sentence template and the extracted key semantic entities are received, and the placeholder field in the target standard sentence template is replaced with the content of the key semantic entities;

[0106] According to the field definition order in the target standard sentence template, the job type, device object, safety measure and ticket field are inserted into the corresponding positions according to the matching fields to generate a complete ticket surface sentence;

[0107] The complete ticket surface sentence is grammatically structured, including conjunction optimization, word order adjustment and unit standardization processing, to generate a structured content conforming to language specifications;

[0108] The structured ticket surface content is grouped and mapped to the ticket information form according to the ticket field, including work content and safety measures, to generate a structured ticket surface;

[0109] Further, the ticket field of the ticket information form can be subjected to integrity detection, and missing or unsuccessfully matched field items are marked and prompt information is recorded.

[0110] The structured ticket surface is pushed for user confirmation, supplement or correction editing.

[0111] In this embodiment, the specific method of step 3) is as follows:

[0112] The single-line graph node information selected by the user in the graphical interactive interface is received, and the selected device name and hanging point coordinates are extracted;

[0113] The extracted device name is associated with the device object term node in the term graph to locate the corresponding device object and job type;

[0114] According to the hanging point coordinates, the device topology structure and the job range data are combined to filter the applicable safety measure sentences in the preset safety measure sentence template, and a recommendation list is constructed to generate a structured ticket surface content;

[0115] The structured ticket surface content is displayed in the form of graphics and text in the ticket filling area;

[0116] The final determined structured ticket surface is output.

[0117] In this embodiment, the method for obtaining voice converted text is as follows:

[0118] The voice inputted by the user is received, and the voice data stream containing the operation content, operation object, safety measure, and auxiliary query semantic information is collected;

[0119] The voice data stream is subjected to voice recognition, audio noise reduction, voice segmentation, and semantic feature extraction operations to generate a corresponding text transcription result;

[0120] The text transcription result is subjected to formatting processing to remove repeated words, emotional words, and non-ticket surface related words, thereby forming a semantic clear voice conversion text.

[0121] In this embodiment, the specific method of step 4) is as follows:

[0122] The paper ticket surface image uploaded by the user is received, and the paper ticket surface image is subjected to image preprocessing operations to extract the text area to be recognized; the image preprocessing operations include image denoising, tilt correction, and region segmentation,

[0123] The text area is subjected to line-by-line recognition using a character recognition algorithm to generate a corresponding structured text content, which includes the operation content, device object, safety measure category, and ticket surface field;

[0124] The structured text content is automatically classified and mapped according to the ticket surface field and backfilled into the ticket surface field corresponding to the ticket surface filling area;

[0125] The backfilled ticket surface field is subjected to missing detection and content integrity comparison, and the field content that fails to be recognized or is incomplete is marked for secondary recording or confirmation by the user, thereby generating a structured ticket surface;

[0126] The structured ticket surface is pushed to support user manual correction and content confirmation operations.

[0127] In this embodiment, the specific method of step 5) is as follows:

[0128] The structured ticket surface is received;

[0129] The structured ticket surface is subjected to ticket surface specification verification;

[0130] The operation type and device object in the structured ticket surface are subjected to matching analysis to verify the semantic consistency between them;

[0131] The safety measure category statement in the structured ticket surface is subjected to specification comparison to identify whether there is content that does not conform to the standard terminology specification or is ambiguous in expression, and whether there is a historical high-frequency error language;

[0132] The ticket surface field in the structured ticket surface is subjected to integrity detection to mark missing fields, semantic conflict fields, and format inconsistent fields;

[0133] Generate a structured check report, including problem fields, error types, recommended modification suggestions, and reference standard template information.

[0134] Output the check report together with the original ticket content to the user interface, supporting the user to make content corrections, confirmations, or secondary recommendations according to the prompts.

[0135] To verify the feasibility of the application in practice, the application is applied to the daily work of a certain city power distribution operation center. The team operation personnel need to write and submit a large number of work tickets and operation tickets every day to cover various operation scenarios such as high and low voltage line inspection, planned power outage, repair and recovery, equipment replacement, and grounding installation. In such "two-ticket" operations, the quality of ticket filling not only relates to the accuracy and safety of the on-site operation, but also directly affects the subsequent review process and operation traces. In the past, due to the manual input of ticket content by operation personnel, some team members have limited cultural level or insufficient understanding of safety regulations, resulting in problems such as non-standard expression, missing important information, and vague safety measure description, which leads to delayed safety audit and frequent ticket return and re-filling, seriously affecting operation efficiency and management quality.

[0136] To improve the standardization and intelligent level of two-ticket management, the power distribution operation center introduces the application method and system, which is deployed under its integrated dispatching system platform, covering 8 operation areas and 37 basic operation teams in the city, accessing team mobile terminals and background approval systems, and integrating with the "Power Grid Map" platform through data interfaces.

[0137] In actual application process, team personnel do not need to manually input long work content when writing work tickets, but complete main information collection through voice input or graphical point selection. For example, in the "East City #4 Transformer Replacement" operation, the operation responsible person only needs to input "replace East City No. 4 transformer low-voltage side switch" through voice input, and the system immediately calls voice recognition and converts the text into standard operation description. After semantic analysis, it identifies "operation action = replacement", "device object = East City No. 4 transformer", and "operation part = low-voltage side switch", and then automatically matches the standard sentence template: replace the low-voltage side switch of East City No. 4 transformer, involving device number A-0408, and need to be powered off and confirmed before execution. The whole process takes less than 5 seconds, which is greatly shortened compared with the average time of 6-8 minutes for manual writing.

[0138] In the graphic auxiliary scene, for example, in the operation of "South Street #23 pole grounding wire disassembly", the operator directly selects #23 pole on the single-line diagram and marks the grounding connection point, the system automatically identifies the associated device "Nanyuan Street 10kV branch line", calls the relevant safety measure sentence template, and judges in combination with the intelligent grounding wire state information returned from the scene (hanging state = removed, live detection = 0.03mA, hanging height = 3.1m, positioning accuracy = ±0.6m). After the system confirms that the disassembly conditions are met, it automatically fills in the safety measure sentence, and after confirming that there is no residual current on the Nanyuan Street #23 pole grounding wire and the safety distance is met, the on-site operator performs the disassembly operation. The whole process takes about 12 seconds, while the traditional manual filling and auditing process takes at least 10 minutes, and there is a risk of missing the hanging state judgment step.

[0139] The application of the OCR backfill module in historical ticket processing is also remarkable. As of the end of December 2024, the system has completed the photographing and recognition of 1683 paper tickets and structured storage, with an average recognition accuracy of 94.7%, of which the accuracy rate of clear handwriting tickets is higher than 98.2%. In particular, during the special safety inspection in November 2024, through the combination of OCR identification and standardization verification module, 23 historical work tickets issued in 2023 were found to have inconsistent semantics or incorrect expressions. The system provided correction suggestions and completed batch rectification.

[0140] In terms of standardization verification, the system combines the embedded rules in the terminology atlas and the historical high-frequency error sentence sample library to perform semantic verification and structural review on the real-time generated ticket content. As of December 2024, the system has accumulated 4127 non-standard language prompts, discovered 158 serious semantic errors, and prevented invalid ticket circulation 246 times, effectively ensuring the closed-loop compliance and safety of the operation process.

[0141] By applying the ticket language standard recommendation system provided by the present application in the actual team ticket generation and processing process, a series of problems such as low efficiency of manual filling, non-standard language, and lagging audit are significantly solved, which not only improves the efficiency of ticket generation, but also realizes the standardization of structure, the standardization of language, and the automation of content verification, laying a solid foundation for the digital and intelligent management of power system operation and maintenance.

[0142] This example also applies the present application to the standardized operation management project of a certain power operation and maintenance unit. To solve the problems of non-uniform ticket language, ambiguous expression, semantic conflict, and missing standard terminology in the "two-ticket" filling process for a long time, the unit first deployed and tested the present application method and system in the front-line operation team. The introduction of this system aims to automatically recommend standard language, standardize ticket language, and reduce the operation risk caused by non-standard filling by using semantic understanding and knowledge graph technology to assist team members in filling operation tickets and work tickets.

[0143] In the pilot process, 10 different operation and maintenance scenarios were selected, covering typical business types such as high-voltage line power-on and power-off, transformer routine maintenance, switch cabinet unlocking and hanging, GIS equipment testing, etc. Operation and maintenance personnel enter ticket information through portable terminals or dispatch support systems. In this process, the ticket recommendation system provides three main input methods: manual input (including free text and graphical selection), voice input (through a head-mounted voice terminal or a dispatch voice recognition interface), and paper ticket photo upload (suitable for electronic archiving and correction suggestion generation in the later stage).

[0144] In practical applications, the system first establishes a graph system covering nearly 3000 standard terms, including 760 typical safety precaution category terms, 430 operation type terms, 1200 device name terms, and more than 600 semantic edges. The term graph is stored and managed in the form of a graph database, and is linked in real time with the field feedback mechanism for continuous expansion and supplementation. Semantic analysis supports word segmentation, part-of-speech tagging, and named entity recognition, which can accurately extract key semantic elements from user input sentences, such as "power-on and power-off", "bus A", "10kV switch cabinet", etc. For example, in the 10kV I section bus power-off maintenance scenario, the user selects the 10kV I section bus node on the single-line diagram and inputs "sentence to perform power-off operation on the bus". The system first identifies the key semantic entities "bus", "power-off", and "operation", matches the associated standard sentence template "perform operation on device name" in the term graph, and then automatically generates the standard ticket sentence "perform power-off operation on 10kV I section bus" and fills it into the "operation type" field. At the same time, it automatically pushes safety precaution suggestions related to bus power-off, such as "hang a grounding line after checking the voltage", "hang a power-off signboard on both sides of the device", etc.

[0145] If the operation and maintenance personnel use voice input "we are ready to perform power-on on 10kV II section", the system will perform real-time voice recognition and convert the result to text "ready to perform power-on on 10kV II section". Then it enters the semantic analysis process, matches the standard sentence template through the term graph, and finally generates the standard ticket sentence "perform power-on operation on 10kV II section bus" and fills it into the ticket. In this process, the system will prompt that the sentence involves a high-risk operation and recommend adding the note "confirm again that the isolation measures are in place before operation".

[0146] If the user uploads a paper bill picture, the system identifies the bill surface through the OCR module, and the extracted field content will be automatically classified into the corresponding electronic field, and the semantic comparison will be performed in the standard statement library. For the content with non-standard terms or ambiguous expression, such as cable head electric test, the system identifies the non-standard part and suggests to modify it to "electric test on cable head", and gives the semantic consistency analysis score and the non-compliance item prompt of safety regulations.

[0147] To verify the effectiveness of the system, the unit selects all 232 work tickets in a month for comparative experiment. Among them, there are 38 statements with ambiguous language, semantic conflict or term missing in the manual ticket surface without the aid of the system, and the error rate is 16.38%; while in the ticket surface generated with the aid of the system, only 4 statements are marked as expression optimization by manual review, and the error rate is reduced to 1.72%. In addition, in terms of ticket filling efficiency, the average time spent by manual filling is about 19 minutes, and after using the recommended system, the average time is shortened to 12 minutes, and the efficiency is improved by 36.8%. In the voice input scene, the average time spent from voice transcription to structured recommendation output is about 8.3 seconds, and the system response delay is less than 1.5 seconds, which greatly improves the convenience of filling. The system realizes dynamic learning and feedback on frequently occurring errors through linkage with the historical ticket library. In the continuous three weeks of data monitoring, the system adds 28 high-frequency error correction rules and 13 new scene term nodes through the expansion of the term graph, realizes continuous iteration and optimization, and forms a closed loop of ticket surface semantic recommendation.

[0148] In summary, the term graph-based ticket language standard recommendation system effectively solves the core problems of non-uniformity of terms, non-standardization of expression and single input form in the filling of "two tickets" in power operation and maintenance, realizes the change of ticket content from human understanding to system intelligent analysis and recommendation, not only improves the standardization of ticket language, but also greatly reduces the burden of manual review and potential operation risk, and has wide application value.

[0149] The example also applies the application to the daily operation of a certain power system. The operation ticket and the work ticket (collectively referred to as "two tickets") are important credentials to ensure safe operation and standardize operation procedures. However, in actual application, due to differences in writing habits of on-site workers, inconsistent terminology and lack of intelligent assistance in filling procedures, there are a lot of non-standard descriptions on the ticket surface. For example, device names are not uniform, safety measures are not accurate, and semantic ambiguity is difficult to identify, which often leads to misunderstanding of operation or failure of safety review, affecting operation efficiency and quality.

[0150] A power supply company pilot deployed the method and system in a spring maintenance concentrated operation stage, and carried out a joint test for a month under the cooperation of multiple departments such as actual dispatching, operation and maintenance, safety supervision, etc. The system is deployed on the unit's ticket editing platform, on-site mobile operation terminal and dispatching center voice station, and runs through the whole process of "two tickets" input, review and backfilling.

[0151] In a 110kV substation protection device replacement task, the on-site personnel use the tablet terminal to fill in the operation ticket. The system automatically calls the graphical interaction module, parses the 110kV I section bus protection device node selected in the single-line diagram into a device entity, and binds it with the terminology atlas. After the user inputs "remove the protection device", the system detects that this statement lacks specific expression and target device information, and prompts "it is recommended to supplement the device number and location". Subsequently, the system automatically matches "remove the device name at the device number", and recommends the ticket statement as "remove the 110kV I section bus K01 protection device", which the user only needs to confirm to fill in one key.

[0152] In another 10kV line power transmission operation scene, the on-duty dispatcher issues the operation task through the voice system, and the voice content is "restore 101 line power supply, pay attention to the on-site personnel evacuation situation". After the system recognizes and transcribes the text through the voice recognition module, it extracts the semantic elements such as "restore" and "101 line power supply", and queries the 101 line node through the semantic atlas to obtain its corresponding line attribute and switch number, and recommends the statement "close the 101 line switch and restore power transmission". At the same time, the system prompts "it is recommended to supplement the safety confirmation measures", and recommends the safety measure statement "confirm that the line operation personnel have been evacuated and the electricity has been verified".

[0153] In the operation archiving link, the safety supervision personnel upload the paper work ticket to the OCR recognition module through the background system, and the system automatically completes image recognition and ticket field mapping. In the recognition result, there is a handwritten statement "hang the ground wire at the switch" in the "safety measure content" column. The system recognizes it and prompts "the semantics are incomplete, it is recommended to use standard expression to hang the ground wire on both sides of the switch and hang the ground identification card", and provides correction suggestions for the reference of the auditors, improving the quality of the later archiving.

[0154] In the actual use process, the system accumulatively processes 689 ticket tasks, generates 3124 structured ticket statements, and the semantic matching accuracy rate reaches 97.2%. Among them, the proportion of tickets generated through graphical interaction assistance reaches 41.3%, the proportion of voice generated tickets reaches 22.7%, and the proportion of OCR recognition and backfilling tasks reaches about 10.6%. In terms of error recognition and modification, the system prompts 173 ticket expressions that are not standardized or have ambiguous items, among which 71 are safety measure item statements, 46 are incomplete operation content descriptions, and the rest are inconsistent device terminology expressions.

[0155] Through comparative analysis, it is found that the average generation time of the ticket surface under the system assistance is 11.8 minutes, which is shortened by about 6 minutes compared with the traditional manual filling; the proportion of language repetition of the ticket surface is reduced by about 45%, the compliance rate of the safety measure sentence is increased from the original 82.4% to 97.6%; in the auditing process, the cumulative trigger times of the "high-risk language" prompt items actively identified by the system are 92, of which 15 items are confirmed to be misleading in the manual review.

[0156] The operation and maintenance team in the pilot summary feedback that the system is particularly helpful for new employees, through the standard template prompt and field automatic filling, the new person can also quickly complete the complex ticket filling task, significantly reducing the rejection rate caused by unfamiliar terms or sentence splicing errors. At the same time, the safety supervision personnel generally reflect that the system auxiliary positioning is efficient and reasonable, which greatly relieves the work pressure of the previous manual reading and auditing.

[0157] Through the deployment and testing of the system in the actual field environment, it is fully verified that it has significant effect in improving the standardization degree of the ticket surface, reducing human filling errors and optimizing the operation process. Relying on the semantic understanding ability of the term graph, it truly realizes the change from writing tickets by experience to filling tickets with system assistance, and lays a solid foundation for future power operation automation and intelligentization.

[0158] The above describes only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A term map-based face language standard recommendation system, characterized in that, The term graph construction module, the text content processing module, the graphical point selection content processing module, the paper image content processing module and the verification module are included. The term graph construction module is used for constructing a term node set containing four types of term nodes of operation type, device object, safety measure category and ticket field according to standard term data of power operation, and extracting operation action and device matching relationship and safety term and operation situation adaptation relationship based on historical ticket data to construct a semantic relationship edge set, and establishing a term graph with term nodes in the term node set and semantic relationship edges in the semantic relationship edge set, and forming several standard sentence templates which are formed by connecting term nodes according to semantic relationship edges and consist of normalized sentences with node fields. The text content processing module is used for performing semantic analysis on user input text and voice converted text, extracting key semantic entities including operation type, device object, location coordinate and safety measure category, performing vector matching of the extracted key semantic entities with term nodes in the term graph, screening out target standard sentence templates, generating structured tickets and pushing them. The graphical point selection content processing module is used for extracting device object and hanging point coordinate from single-line graph node information input by graphical interaction point selection of the user, generating structured tickets by associating the term graph and the standard sentence templates and pushing them. The paper image content processing module is used for performing text recognition on paper ticket image input by the user, generating structured tickets and pushing them. The verification module is used for performing ticket specification verification on the generated structured tickets. 2.The term graph-based faceplate language standard recommendation system according to claim 1, wherein, The specific method for constructing a term node set containing four types of term nodes of operation type, device object, safety measure category and ticket field according to standard term data of power operation, and extracting operation action and device matching relationship and safety term and operation situation adaptation relationship based on historical ticket data to construct a semantic relationship edge set, and establishing a term graph with term nodes in the term node set and semantic relationship edges in the semantic relationship edge set, and forming several standard sentence templates is as follows: Standard term data related to two-ticket business of power operation is collected, and standard operation type, typical device name, safety measure sentence, operation object attribute, commonly used ticket term and its upper and lower association information are extracted to construct an initial term vocabulary; The terms in the initial term vocabulary are classified and processed according to four types of structures of operation type, device object, safety measure category and ticket field to divide the term node set; Based on historical ticket data, semantic associations between terms are extracted, including operation action and device matching relationship and safety term and operation situation adaptation relationship, to construct a semantic relationship edge set between the terms; The term node set and the semantic relationship edge set are organized as a graph structure data model to construct a term graph for ticket language standard recommendation. 3.The term graph-based faceplate language standard recommendation system according to claim 2, characterized in that, The semantic analysis of the user input text, voice conversion text is performed, key semantic entities including job type, device object, location coordinate and safety category are extracted, the extracted key semantic entities are vectorized matched with the term nodes in the term graph, and the target standard sentence template is screened out, and the specific method for generating and pushing the structured ticket surface is as follows: Receiving the original ticket content input by the user, the original ticket content includes text and voice conversion text; Text preprocessing operation is performed on the original ticket content, the text preprocessing operation includes character cleaning, sentence segmentation and stop word filtering, and preprocessed sentences are generated; The preprocessed sentences are subjected to morphological analysis, continuous text is segmented by using a segmentation algorithm, and the segmentation results are subjected to part-of-speech tagging; Based on the part-of-speech tagging results, a named entity recognition operation is performed to identify business key terms including job action words, device name words, location representation words and safety term words; The identified business key terms are classified and grouped according to job elements to form key semantic entities including job type, device object, safety category and ticket field; The key semantic entities are vectorized matched with the term nodes in the term graph, and the target standard sentence template is screened out, thereby generating and pushing the structured ticket surface. 4.The term graph-based faceplate language standard recommendation system according to claim 3, characterized in that, The specific method for vectorizing matching the key semantic entities with the term nodes in the term graph, screening out the target standard sentence template, and generating and pushing the structured ticket surface includes: Position correspondence of key semantic entities according to ticket fields and node fields of each standard sentence template; Constructing semantic vector representation of term nodes in the standard sentence template, the semantic vector includes node term text information and its semantic context of connection structure in the graph; Constructing input semantic vector representation of key semantic entities to generate a query vector set; Using a similarity calculation strategy, the semantic vector representation of the standard sentence template and the query vector set are subjected to semantic matching scoring to generate a matching score list; The standard sentence templates in the matching score list are sorted, and the one with the highest score is selected as the target standard sentence template; Embedding the extracted key semantic entities into the reserved positions of the target standard sentence template to generate a structured ticket surface; and recommending the structured ticket surface.

5. The term graph-based faceplate terminology standard recommendation system according to claim 4, characterized in that, The specific method for embedding the extracted key semantic entities into the reserved positions of the target standard sentence template to generate a structured ticket surface includes: Receiving the target standard sentence template and the extracted key semantic entities, and replacing the content of the placeholder field in the target standard sentence template with the key semantic entities; According to the field definition order in the target standard sentence template, inserting job type, device object, safety measure and ticket field into the corresponding positions to generate a complete ticket sentence; Performing syntax structure arrangement on the complete ticket sentence, including connection word optimization, syntax order adjustment and unit standardization processing, to generate structured content conforming to language specifications; Grouping the structured ticket content according to the ticket field and mapping it to the ticket information form to generate a structured ticket. 6.The term graph-based faceplate language standard recommendation system according to claim 5, wherein, The specific method for extracting the coordinates of device objects and attachment points from the single-line graph node information input by the user through graphical interaction, as well as generating a structured ticket by associating the terminology map with the standard statement template and pushing it, is as follows: Receive the single-line graph node information selected by the user in the graphical interactive interface, and extract the selected device name and the coordinates of the attachment point; The extracted equipment names are associated with the equipment object term nodes in the terminology graph to locate the corresponding equipment objects and operation types. Based on the coordinates of the attachment point, combined with the equipment topology and work scope data, applicable safety measures statements are selected from the preset safety measure statement templates, and a recommendation list is constructed to generate structured ticket content. The structured ticket content will be displayed simultaneously in the ticket filling area in the form of graphics and text; Output the final structured ticket. 7.The term graph-based faceplate language standard recommendation system according to claim 6, wherein, The method for obtaining the speech-to-text conversion is as follows: It receives voice commands from users to fill out tickets via voice input, and collects voice data streams containing semantic information such as job content, operation objects, safety measures, and auxiliary query information; The speech data stream is subjected to speech recognition, audio noise reduction, speech segmentation and semantic feature extraction operations are performed to generate the corresponding text transcription results; The text transcription results are formatted to remove duplicate words, modal particles, and words not related to the ticket price, resulting in semantically clear speech-to-text. 8.The term graph-based faceplate language standard recommendation system according to claim 7, characterized in that, The specific method for performing text recognition on the user-input paper ticket image, generating a structured ticket, and then pushing it out is as follows: The system receives images of paper tickets uploaded by users, performs image preprocessing on the images, and extracts the text regions to be recognized. The image preprocessing includes image denoising, tilt correction, and region segmentation. The text area is identified line by line using a character recognition algorithm to generate corresponding structured text content, which includes work content, equipment objects, safety measure categories, and ticket fields. The structured text content is categorized and mapped according to the ticket fields and then populated back into the corresponding ticket fields in the ticket filling area. Perform missing detection and content integrity comparison on the backfilled ticket fields, mark the fields that failed to be identified or were incomplete, and generate a structured ticket; The structured ticket is then pushed out. 9.The term graph-based faceplate language standard recommendation system according to claim 8, wherein, The specific method for verifying the standardization of the generated structured ticket is as follows: Receive structured tickets; The structured ticket is subjected to ticket standardization verification, and field-level verification is performed according to the terminology nodes in the terminology graph and the ticket filling area. A matching analysis is performed on the operation type and equipment object in the structured ticket to verify the semantic consistency between the two. The security measure categories in the structured ticket are compared against the norms to identify whether there is any content that does not conform to the standard terminology or is vague, and to compare whether there are any historically frequently used erroneous terms. Perform integrity checks on the fields in the structured ticket, and mark missing fields, semantically conflicting fields, and fields with inconsistent formats; Generate a structured validation report, which includes the problem fields, error types, recommended modifications, and reference standard template information.

Images