Nuclear power in-service inspection report generation method based on document labeling and tagging technology
Through the method of generating nuclear power in-service inspection report based on document labeling and labeling technology, the problem of time, accuracy and consistency of manual report preparation is solved, and the rapid and accurate report generation and efficient data utilization are achieved, providing intelligent Report generation service.
Patent Information
- Application Number
- CN202410906110.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-08
- Publication Date
- 2024-11-22
AI Technical Summary
The manual compilation of in-service inspection reports in nuclear power plants is difficult to ensure, the update is difficult and the data utilization rate is low.
Using document annotation and labeling technology, a professional vocabulary recognition model is established by analyzing and storing nuclear power field data, and a structured in-service inspection report is generated using NLTK part-of-speech labeler and Transformer model.
Significantly shorten the time for reporting, improve the accuracy, completeness and consistency of data, save costs, improve data utilization, and provide intelligent report generation services.
Smart Images

Figure CN119005180A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of nuclear power in-service inspection report generation, and in particular to a method for generating a nuclear power in-service inspection report based on document annotation and labeling technology. Background Art
[0002] At present, the implementation report of in-service inspection of nuclear power plants is mainly compiled manually in paper form, requiring the person in charge of the work to collect information such as the inspected objects, work outlines, inspection tasks, inspection site conditions and execution inspection records during the in-service inspection work offline, and then write the in-service inspection report through data analysis and processing such as screening and calculation. The report preparation process is time-consuming, cannot guarantee accuracy, is difficult to update, and has low data utilization. The main problems of the existing method are as follows:
[0003] (1) High cost
[0004] Manual preparation of in-service inspection reports is very time-consuming. In the in-service inspection of nuclear power plants, the amount of information is huge, and each step, from data collection and collation to report writing, requires a lot of manpower input. It may take weeks or even months to manually process this information and write a report.
[0005] (2) Accuracy and consistency issues
[0006] Manually compiled reports are easily affected by factors such as personal knowledge level, subjective judgment of experience and fatigue, which may lead to errors and omissions in the reports. In addition, the styles and formats of reports written by different people are different and lack consistency, which makes subsequent reading and understanding difficult.
[0007] (3) Difficulty in updating and low data utilization
[0008] Once a report is compiled, if certain parts need to be updated or modified, the entire document needs to be reviewed again, which is not only time-consuming and laborious, but also likely to introduce new errors. In addition, manual reports often fail to fully utilize all the collected data because the ability to screen and analyze data is limited, resulting in important information being overlooked. Summary of the invention
[0009] The purpose of the present invention is to provide a method for generating a nuclear power in-service inspection report based on document annotation and labeling technology, which solves the problems of high manual preparation cost of nuclear power in-service inspection reports, lack of guarantee of data accuracy and consistency, difficulty in data updating and low utilization rate.
[0010] In order to achieve the above object, the present invention provides the following technical solutions:
[0011] On the one hand, the present invention provides a method for generating a nuclear power in-service inspection report based on document annotation and labeling technology, comprising:
[0012] Step 1: Parse the collected in-service inspection data into text content for structured storage;
[0013] Step 2: Establish a professional vocabulary recognition model and perform text segmentation;
[0014] Step 3: Identify domain-specific nouns based on the text segmentation results;
[0015] Step 4: Adjust and label the words through NLTK part-of-speech tagger according to predefined rules and guidelines;
[0016] Step 5: Convert the annotated text dataset into word vectors and calculate the position vector;
[0017] Step 6: Analyze and process the service inspection record content;
[0018] Step 7: Generate an in-service inspection report.
[0019] In some embodiments, step 1 specifically includes:
[0020] Step 1.1: Collect relevant data in the field of in-service inspection of nuclear power plants;
[0021] Step 1.2: Use Apache Tika to parse the text content in the document and store it in a structured manner.
[0022] In some embodiments, the relevant data in step 1.1 includes documents, reports, papers, journals, and professional books.
[0023] In some embodiments, the text extraction process in step 1.2 is:
[0024] Step 1.2.1: Apache Tika receives a file as input, which can be a local file path or an input stream. It supports a variety of file formats, including common Office documents (such as .docx, .xlsx, .pptx), PDF, HTML, XML, etc.
[0025] Step 1.2.2: After receiving the file, determine the file type based on the file characteristics and metadata (such as file header information);
[0026] Step 1.2.3: After determining the file type, start extracting the file content. For text files, directly extract the file content. For non-text files such as pictures or videos, perform OCR processing to obtain the text information.
[0027] Step 1.2.4: Extract the file’s metadata, including file name, creation date, author, keywords, and other information;
[0028] Step 1.2.5: Output the extracted text content and metadata.
[0029] In some embodiments, step 2 specifically includes:
[0030] Step 2.1: Establish a professional vocabulary recognition model based on regular expressions, perform text segmentation through Jieba, use the PaddleNLP part-of-speech analysis tool to analyze the word parts of speech, and extract proper nouns based on the collocation of adjectives and nouns;
[0031] Step 2.2: Extract the information in the database in the form of structural nouns or phrases through data rules;
[0032] Step 2.3: Normalize the words with irregular expressions through the professional term normalization rules, and map different expressions with the same meaning;
[0033] Step 2.4: Professional vocabulary cleaning rules remove meaningless special characters in the database.
[0034] In some embodiments, the information contained in the database includes equipment name, inspection personnel, component name, inspection method, inspection result, and professional terminology.
[0035] In some embodiments, step 5 specifically includes:
[0036] Step 5.1: Train a text classification model based on Transformer and obtain word input through word vectors and position vectors;
[0037] Step 5.2: Convert the annotated text dataset into word vectors through word2vec;
[0038] Step 5.3: Get the position vector by the following calculation formula:
[0039]
[0040] Where pos represents the position of the word in the sentence, d represents the dimension of PE, 2i represents the even dimension, and 2i+1 represents the odd dimension.
[0041] In some embodiments, in step 6, the text content of the in-service inspection work record is recognized through a pre-trained model to obtain structured equipment names, inspection personnel, component names, inspection methods, and inspection results.
[0042] On the other hand, the present invention provides a computer device including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the method for generating a nuclear power in-service inspection report based on document annotation and labeling technology are implemented.
[0043] On the other hand, the present invention provides a computer-readable storage medium having computer-readable instructions stored thereon, which, when executed, implement the steps of the method for generating a nuclear power in-service inspection report based on document annotation and labeling technology.
[0044] Compared with the prior art, the method for generating a nuclear power in-service inspection report based on document annotation and labeling technology provided by the present invention has the following beneficial effects:
[0045] The present invention is based on the digital data of in-service inspections. After the nuclear power plant inspectors complete the inspection of the inspected objects, they use the report automatic generation tool developed in this project to generate an in-service inspection report, saving the workload of report preparation and providing a basis for subsequent data analysis.
[0046] Furthermore, the present invention uses natural language processing technology and machine learning algorithms to automatically generate in-service inspection reports that meet standards and business requirements based on document annotated information and structured data stored after labeling. Instead of manual compilation, the report compilation time is shortened by several weeks to several months, and a single overhaul is equivalent to reducing the input of 3 to 5 people, effectively ensuring the accuracy, completeness, consistency and traceability of the data, saving costs while improving data utilization, and providing more intelligent report generation services for nuclear power in-service inspectors. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for the technical description.
[0048] Figure 1 A flow chart of a method for generating a nuclear power in-service inspection report based on document annotation and labeling technology provided by the present invention;
[0049] Figure 2 A text extraction flow chart provided by the present invention;
[0050] Figure 3 The present invention provides a part-of-speech tagging flow chart. DETAILED DESCRIPTION
[0051] The following is further explained in detail through specific implementation methods.
[0052] like Figure 1As shown, the present invention provides a method for generating a nuclear power in-service inspection report based on document annotation and labeling technology, comprising the following steps:
[0053] Step 1: Text collection. Specifically including:
[0054] Step 1.1: Collect relevant data in the field of in-service inspection of nuclear power plants, including documents, reports, papers, journals, professional books and other professional terms in the nuclear power field.
[0055] Step 1.2: Use Apache Tika to parse the text content in the document and store it in a structured manner. The text extraction process is as follows: Figure 2 shown.
[0056] Step 1.2.1: Apache Tika receives a file as input, which can be a local file path or an input stream. It supports a variety of file formats, including common Office documents (such as .docx, .xlsx, .pptx), PDF, HTML, XML, etc.
[0057] Step 1.2.2: After receiving the file, determine the file type based on the file characteristics and metadata (such as file header information);
[0058] Step 1.2.3: After determining the file type, start extracting the file content. For text files, directly extract the file content. For non-text files such as pictures or videos, perform OCR processing to obtain the text information.
[0059] Step 1.2.4: Extract the file’s metadata, including file name, creation date, author, keywords, and other information;
[0060] Step 1.2.5: Output the extracted text content and metadata.
[0061] Step 2: Text segmentation. Specifically includes:
[0062] Step 2.1: Based on the existing regular expressions, a professional vocabulary recognition model is established for semantic analysis and understanding. The regularization rules define the combination structure of adjectives and names such as "** equipment". First, the text is segmented through Jieba, and then the word part of speech analysis tool Paddlenlp is used to analyze the word part. Finally, the proper nouns are extracted according to the collocation of adjectives and nouns.
[0063] Step 2.2: Extract the information such as equipment name, inspection personnel, component name, inspection method, inspection result and professional terminology from the existing service inspection business system database in the form of structural nouns or phrases through data rules.
[0064] Step 2.3: Normalize the non-standard expressions in equipment names, inspection personnel, component names, inspection methods, inspection results, and professional terms through the professional term normalization rules, map different expressions that express the same meaning, and achieve the standardization of professional vocabulary.
[0065] Step 2.4: Professional vocabulary cleaning rules remove meaningless special characters from information such as equipment names, inspectors, component names, inspection methods, inspection results, and professional terms in the database.
[0066] Step 3: Manual identification.
[0067] Domain experts identify domain-specific nouns based on text segmentation results, customize dictionaries based on actual needs, add domain-specific nouns and terms, and merge and split current segmentation results to improve segmentation accuracy.
[0068] Step 4: Part-of-speech tagging.
[0069] Domain experts read the structured text segmentation information (including the original content and segmentation results), and adjust and label the words through the NLTK part-of-speech tagger according to predefined rules and guidelines. The specific process is as follows: Figure 3 Step 4 includes:
[0070] Step 4.1: Import the corpus or text information that needs to be POS tagged;
[0071] Step 4.2: Divide the text to be annotated into blocks and words and clean and filter them;
[0072] Step 4.3: Use the part-of-speech tagger provided by NLTK to tag the segmented words and determine the grammatical role of each word in the context.
[0073] In the preprocessing stage, the sources of materials include corpora, local libraries, and online content. The preprocessing methods used are block segmentation, word segmentation, cleaning and filtering. Block segmentation and word segmentation can divide the text into words or phrases, and process different types of text data through word segmenters, such as English, Chinese, or other languages. Cleaning and filtering can remove noise and unnecessary parts in the text to improve the quality and efficiency of analysis, such as removing special characters and punctuation marks, and removing stop words.
[0074] In the analysis phase, the analysis dimensions include words, sentences, and chapters. The analysis methods for words include bigram extraction and same-context extraction, the analysis methods for sentences include syntactic analysis, and the analysis methods for chapters include statistical analysis and chapter analysis. Bigram extraction: Use NLTK's bigrams function to extract bigrams and identify common phrases and collocations to improve the expressiveness of text features. Same-context word extraction: By analyzing the same-context words, identify the relevance and co-occurrence relationship of words, understand the semantic network of words, and use the co-occurrence matrix or point mutual information (PMI) to quantify the common relationship of words. Syntactic analysis: Use the pos_tag provided by NLTK for part-of-speech tagging, and RegexpParser for syntactic parsing to identify the subject, predicate, and object components, help understand the rules and structure of sentences, and then understand the semantics of sentences. Statistical analysis: Use NLTK to count the word frequency, sentence length, and paragraph structure in the text to obtain the overall characteristics of the text and help understand the style, complexity, and information density of the text. Discourse analysis: Use NLTK to segment the text into paragraphs and extract key sentences, analyze the macro structure and coherence of the text, including the transition between paragraphs, the development of the theme, etc., which helps to understand the overall logic and information flow of the text.
[0075] In the output stage, NLTK is used to generate various forms of text interpretation and visual display, which can better understand the analysis results. The text interpretation content includes part-of-speech tagging, syntactic analysis tree, word frequency statistics, etc. The visual display includes syntactic tree, word frequency distribution map, etc.
[0076] Step 5: Vector calculation. Specifically including:
[0077] Step 5.1: Train a text classification model based on Transformer and obtain word input through word vectors and position vectors.
[0078] Step 5.2: Convert the annotated text dataset into word vectors through word2vec.
[0079] Step 5.3: Get the position vector through the calculation formula. The calculation formula is as follows:
[0080]
[0081] Among them, pos represents the position of the word in the sentence, d represents the dimension of PE, 2i represents the even dimension, and 2i+1 represents the odd dimension. By adding the word vector and the position vector of the word, we can get the word representation vector x, which is used as the input of Transformer.
[0082] Step 6: Analyze and process the service inspection record content.
[0083] The pre-trained model is used to identify the text content of the in-service inspection work record to obtain structured information such as equipment name, inspection personnel, component name, inspection method, inspection result, etc. For example, the following content in the work record:
[0084] "1# main water supply pump auxiliary oil pump power switch EC38-GZD-105-3-FLB adjust the switch handle to indicate the position".
[0085] The following results are obtained through model analysis:
[0086] Words Part of Speech 1# Main feed water pump auxiliary oil pump Device Name EC38-GZD-105-3-FLB Part Number Power switch Part Name Adjust the switch handle indicator position Job Description
[0087] For information that is inaccurately recognized by large models, manual labeling is performed and training is strengthened to continuously improve machine learning capabilities.
[0088] Step 7: Generate an in-service inspection report.
[0089] According to the standard template of nuclear power in-service inspection business, query the structured database, automatically fill in the report content, and generate the in-service inspection report online.
[0090] The present invention also includes three parts: constructing a professional vocabulary library in the field of nuclear power in-service inspection (step 3), a labeled document structured storage library (step 4), and a report generation module (step 7).
[0091] Build a professional vocabulary database in the field: collect relevant data in the field of nuclear power plant in-service inspection, use the jieba word segmentation tool to perform preliminary word segmentation on the document, and obtain basic vocabulary information. Automatically combine manual verification to identify the special terms and terminology in the field of nuclear power in-service inspection, and summarize and build a professional vocabulary database in the field of nuclear power plant in-service inspection.
[0092] Labeled document structured repository: Domain experts label the nouns and terms in the professional vocabulary of nuclear power plant in-service inspection to form specific labels and obtain high-quality standard data. Based on the data characteristics, different data structures are used to structure the data, and the data is stored in different media through storage strategies such as partitioning and indexing, finally forming a structured repository.
[0093] Report generation module: Based on the structured repository, using natural language processing technology and machine learning algorithms, and then according to the preset report template, the data is cleaned, converted and analyzed, and the processed data is filled into the template to automatically generate a complete report. The generated report can be displayed through the user interface and supports export to PDF, WORD, EXCEL and other formats.
[0094] In addition, the present invention also provides a computer device, including a memory and a processor, the memory stores computer-readable instructions, and the processor executes the steps of the above-mentioned method for generating a nuclear power in-service inspection report based on document annotation and labeling technology.
[0095] In addition, the present invention also provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed, the steps of the above-mentioned method for generating a nuclear power in-service inspection report based on document annotation and labeling technology are implemented.
[0096] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. 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 technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.
[0097] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A method for generating a nuclear power in-service inspection report based on document annotation and labeling technology, characterized in that: include: Step 1: Parse the collected in-service inspection data into text content for structured storage; Step 2: Establish a professional vocabulary recognition model and perform text segmentation; Step 3: Identify domain-specific nouns based on the text segmentation results; Step 4: Adjust and label the words through NLTK part-of-speech tagger according to predefined rules and guidelines; Step 5: Convert the annotated text dataset into word vectors and calculate the position vector; Step 6: Analyze and process the service inspection record content; Step 7: Generate an in-service inspection report.
2. The method for generating a nuclear power in-service inspection report based on document annotation and labeling technology according to claim 1 is characterized in that: Step 1 specifically includes: Step 1.1: Collect relevant data in the field of in-service inspection of nuclear power plants; Step 1.2: Use Apache Tika to parse the text content in the document and store it in a structured manner.
3. The method for generating a nuclear power in-service inspection report based on document annotation and labeling technology according to claim 1 is characterized in that: The relevant data in step 1.1 include documents, reports, papers, journals, and professional books.
4. The method for generating a nuclear power in-service inspection report based on document annotation and labeling technology according to claim 1 is characterized in that: The text extraction process in step 1.2 is: Step 1.2.1: Apache Tika receives a file as input; Step 1.2.2: After receiving the file, determine the file type based on the file characteristics and metadata; Step 1.2.3: After determining the file type, start extracting the file content. For text files, directly extract the file content. For non-text files, perform OCR processing to obtain the text information. Step 1.2.4: Extract the metadata of the file; Step 1.2.5: Output the extracted text content and metadata.
5. The method for generating a nuclear power in-service inspection report based on document annotation and labeling technology according to claim 1 is characterized in that: Step 2 specifically includes: Step 2.1: Establish a professional vocabulary recognition model based on regular expressions, perform text segmentation through Jieba, use the PaddleNLP part-of-speech analysis tool to analyze the word parts of speech, and extract proper nouns based on the collocation of adjectives and nouns; Step 2.2: Extract the information in the database in the form of structural nouns or phrases through data rules; Step 2.3: Normalize the words with irregular expressions through the professional term normalization rules, and map different expressions with the same meaning; Step 2.4: Professional vocabulary cleaning rules remove meaningless special characters in the database.
6. The method for generating a nuclear power in-service inspection report based on document annotation and labeling technology according to claim 5 is characterized in that: The information contained in the database includes equipment name, inspector, component name, inspection method, inspection result and professional terminology.
7. The method for generating a nuclear power in-service inspection report based on document annotation and labeling technology according to claim 1 is characterized in that: Step 5 specifically includes: Step 5.1: Train a text classification model based on Transformer and obtain word input through word vectors and position vectors; Step 5.2: Convert the annotated text dataset into word vectors through word2vec; Step 5.3: Get the position vector by the following calculation formula: Where pos represents the position of the word in the sentence, d represents the dimension of PE, 2i represents the even dimension, and 2i+1 represents the odd dimension.
8. The method for generating a nuclear power in-service inspection report based on document annotation and labeling technology according to claim 1 is characterized in that: In step 6, the text content of the in-service inspection work record is recognized through the pre-trained model to obtain the structured equipment name, inspection personnel, component name, inspection method, and inspection result.
9. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, characterized in that: When the processor executes the computer-readable instructions, the steps of the method for generating a nuclear power in-service inspection report based on document annotation and labeling technology as described in any one of claims 1-8 are implemented.
10. A computer-readable storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed, the steps of the method for generating a nuclear power in-service inspection report based on document annotation and labeling technology as described in any one of claims 1 to 8 are implemented.
Citation Information
Cited By
Evaluation report automatic generation system and method
CN119180621A
Monitoring management method for nuclear power equipment
CN120299765A
