Intelligent auxiliary generation method for power grid project review opinions based on information extraction

Abstract

The invention provides a power grid project review opinion intelligent auxiliary generation method based on information extraction, and relates to the field of power grid projects. The invention discloses an intelligent auxiliary generation method for power grid project review opinions based on information extraction. The method comprises the following steps: S1, collecting a material file carding index extraction rule; s2, on the basis of a universal information extraction model UIE, index extraction in the unstructured file is achieved, and an engineering index library is formed; and S3, constructing a power grid feasibility research and power grid initial review opinion template library, and customizing a configuration directory chapter. According to the method, key indexes are obtained from files such as a design specification and a budget book, intelligent auxiliary generation of the review opinions is achieved, multi-specialty and online collaborative editing of the review opinions is achieved through an online editing control, the text format is standardized, each item in the process is traceable, the text quality and efficiency of the review opinions are improved, and the method is suitable for popularization and application. And the opinion issuing time is greatly shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid engineering, and in particular to an intelligent auxiliary generation method for power grid engineering review opinions based on information extraction. Background Art

[0002] The power grid project review is an important part of the power grid investment decision-making, covering technical feasibility, economic benefits, environmental impact and social benefits, and finally issuing a review opinion. The power grid project review opinion is highly professional and technical, involves many professions, and is highly dependent on the level and experience of experts. There are problems such as large workload and low efficiency. Therefore, the NLP natural language processing technology is used to automatically extract the key information of the design results, realize the intelligent auxiliary generation of power grid project review opinions and online collaborative compilation of opinions, and the process is "traceable and traceable", which improves the quality and efficiency of power grid project review opinions.

[0003] At present, the review opinions of power grid projects are still compiled offline, which requires review experts to read, analyze and process a large number of different types of documents, data tables, and drawings, and finally form professional opinions in written form. The final review opinions are formed after professional verification, assistant summary and manager review.

[0004] The traditional way of compiling opinions has obvious shortcomings. First, the opinions involve many indicators and are not highly structured. Business indicators are scattered in different design results documents. Reviewers need to find relevant documents and compile professional opinions, which results in low work efficiency and inconsistent indicators. Second, the quality of opinion compilation varies and is highly dependent on the level and experience of experts. Traditional opinions are still compiled offline, relying on expert experience. They must pay attention to both the quality of opinions and the format of documents. At the same time, manual merging is inefficient and no traces can be left in the process, which increases the application burden of each editor and reviewer and causes a large workload. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the deficiencies in the prior art, the present invention provides an intelligent auxiliary generation method for power grid project review opinions based on information extraction, which solves the problems of low work efficiency, heavy workload and inconsistent indicators among reviewers.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for intelligently assisting the generation of power grid project review opinions based on information extraction, comprising the following steps:

[0009] S1. Collect material files, sort out index extraction rules;

[0010] S2. Based on the universal information extraction model UIE, the indicators in the unstructured files are extracted to form an engineering indicator library;

[0011] S3. Build a template library for power grid feasibility study and power grid preliminary design review opinions, and customize the configuration catalog chapter;

[0012] S4. According to different individual projects, form the secondary and tertiary directories of the main technical solution chapters;

[0013] S5. Combine the automatically extracted indicators to automatically generate review comments in a format that meets the requirements;

[0014] S6. Use the online editing controls to complete the editing of review comments.

[0015] Preferably, the material files collected in step 1 are sorted out to extract indexes, and the specific contents are as follows:

[0016] Collect historical power grid feasibility studies and preliminary design review opinions, sort out the technical and economic indicators involved in the review opinions, and form preliminary design opinion templates, 330 kV and above feasibility study opinion templates, and 220 kV and below feasibility study opinion templates; combine design results documents, analyze the data sources of engineering indicators such as substations, overhead cables, and cables, and sort out indicator extraction rules from design specifications, budget estimates, and equipment and material inventory documents.

[0017] Preferably, in step 2, based on the universal information extraction model UIE, the index extraction in the unstructured file is realized to form an engineering index library, and the specific contents are as follows:

[0018] PaddleNLP natural language processing technology is selected to realize the extraction of indicators from unstructured files based on the universal information extraction model UIE. In the data preprocessing stage, Aspose is first used to convert the editable PDF version of the design manual and the budget estimate into txt format, and then input it into the UIE model. Sample training and data labeling are carried out, and the UIE model is tuned to realize the automatic extraction of indicators and form an engineering indicator library.

[0019] Preferably, in step 3, a template library for power grid feasibility study and power grid preliminary design review opinions is constructed, and a custom configuration directory chapter is configured, and the specific contents are as follows:

[0020] In accordance with the document publication serial number requirements, establish a multi-level title serial number dictionary, and set the directory chapter compilation authority according to [one, to twenty,], [(one) to (twenty)], [1. to 20.], [(1) to (20)]. Automatically generate first-level and second-level directory chapters based on the review type and voltage level.

[0021] Preferably, in step 4, secondary and tertiary directories of the main technical solution chapters are formed according to different individual projects, and the specific contents are as follows:

[0022] The first-level title "Main Technical Solution" chapter catalog consists of two parts: non-fixed and fixed. For the non-fixed part, create the name and type of individual projects, including 9 categories: new substation construction, main transformer expansion, interval modification, new line construction, line modification, interval modification, protection modification, cable line, and safety control system. Set the corresponding third-level titles and editing permissions in the configuration file; the fixed part includes the system and secondary part, and the communication part, which are configured in the opinion template library.

[0023] Preferably, in step 5, the automatically extracted indicators are combined to automatically generate review opinions in a format that conforms to the format, and the specific contents are as follows:

[0024] Set the indicator bookmark in the opinion template to realize automatic filling and display of the indicator in the opinion, and according to the document format requirements, the first line is indented by 2 characters, the line spacing is 28.5, the first-level title is in Fangzheng bold GBK, the font size is No. 3, the second-level title is in Fangzheng Kaiti GBK, the font size is No. 3 and it should be bold, the main text is in Fangzheng Fangsong_GBK, the font size is No. 3, and the review opinions that conform to the format are automatically generated.

[0025] Preferably, in step 6, an online editing control is used to complete the review and editing of the review opinions, and the specific contents are as follows:

[0026] By adopting online editing controls, coordinated editing by multiple disciplines is achieved, and different writing areas are controlled by editing permissions of catalog chapters. Review experts improve the content and adopt revision models to record the improved content, leaving traces of the process. After the professional review experts complete the preparation of professional opinions, the review opinions are edited after verification by professional reviewers, summary by assistants, and review by project managers.

[0027] Preferably, the index extraction in step 2 is specifically performed by:

[0028] 1) Obtain the design specifications, budget estimates and other documents of the design results, and export sample files annotated with the corresponding indicators to be identified according to each divided model;

[0029] 2) The labeling personnel create engineering projects corresponding to the model and add corresponding labels to each project;

[0030] 3) After the labelers import and label the sample files of all models in the doccano labeling platform and review the labeling results, they export the sample data sets of all models respectively;

[0031] 4) Synchronously build the PaddleNLP environment required for UIE model training and debug it to meet the training conditions;

[0032] 5) Do not split different model samples into training, validation set, and test set through Python code for training;

[0033] 6) After training is completed, you can use commands to perform reasoning and testing. If there are accuracy issues with the indicators, you can debug them through debugging commands;

[0034] 7) After inference is completed, export the model and deploy it using paddleserving;

[0035] 8) Write the corresponding business implementation code and the client httpclient calling code that calls the deployment service, perform joint debugging, and finally extract the results of the indicators, perform structured storage, and store them in the database table.

[0036] Preferably, the specific method for automatically generating the review opinions that conform to the format in step 5 is:

[0037] 1) Obtain the index data stored in the structured database;

[0038] 2) Prepare initial templates for various types of opinions, adjust the basic content and format of the templates, set different templates according to different project types, and prompt an error if the corresponding template is not found during the generation process;

[0039] 3) Set the document bookmark for the dynamically generated part. The bookmark is used as a placeholder to replace the content at runtime. During this period, ensure that the bookmark name is unique in the document to avoid conflicts;

[0040] 4) Use aspose-words to automatically generate opinions. First, use the Document class to load the template document, ensure that the paths of the template file and the output file are correct, and copy a copy of the template file as the opinion document for the current project;

[0041] 5) Finally, save the automatically generated complete opinion file, save the document with replaced content into a new file, name the file according to different types of projects, and store the file in the unstructured service platform.

[0042] Working principle: Through PaddleNLP natural language processing technology, key indicators are extracted from the approved design specifications and budget estimates, and a review opinion template library is established to achieve custom configuration and permission settings for the review opinion catalog chapters to meet different types of opinion templates for feasibility studies and preliminary designs. By adopting online editing controls, collaborative online editing of professional opinions is achieved, and the process is "traceable and traceable". Opinion writers only care about the content of the opinions and do not need to worry about issues such as typesetting. Each profession is only responsible for writing the content of its own profession; to avoid conflicts, opinions are eventually merged automatically, reducing the trouble of manual merging.

[0043] (III) Beneficial effects

[0044] The present invention provides an intelligent auxiliary generation method for power grid project review opinions based on information extraction.

[0045] It has the following beneficial effects:

[0046] 1. This method obtains key indicators from design specifications, budget estimates and other documents to achieve intelligent assisted generation of review opinions.

[0047] 2. Use online editing controls to achieve multi-professional and online collaborative editing of review opinions, standardize the format of documents, and make every item in the process "traceable and traceable", thereby improving the quality and efficiency of review opinion issuance and greatly shortening the time for issuing opinions. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a method for intelligently assisting generation of power grid project review opinions based on information extraction proposed by the present invention;

[0049] Figure 2 This is an indicator extraction flow chart of an intelligent auxiliary generation method for power grid project review opinions based on information extraction proposed by the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] Example

[0052] like Figure 1-2 As shown, an embodiment of the present invention provides a method for intelligently assisting generation of power grid project review opinions based on information extraction, comprising the following steps:

[0053] S1. Collect material files, sort out index extraction rules;

[0054] S2. Based on the universal information extraction model UIE, the indicators in the unstructured files are extracted to form an engineering indicator library;

[0055] S3. Build a template library for power grid feasibility study and power grid preliminary design review opinions, and customize the configuration catalog chapter;

[0056] S4. According to different individual projects, form the secondary and tertiary directories of the main technical solution chapters;

[0057] S5. Combine the automatically extracted indicators to automatically generate review comments in a format that meets the requirements;

[0058] S6. Use the online editing controls to complete the editing of review comments.

[0059] In step 2, the specific extraction method of indicators is as follows:

[0060] 1) Obtain the design specifications, budget estimates and other documents of the design results, and export sample files annotated with the corresponding indicators to be identified according to each divided model;

[0061] First, the design manual and other files are parsed through Java's aspose-pdf, tabula and other technologies, and the final parsing results are assembled into the data structure of the entire document content according to the chapter structure. The parsing uses aspose-pdf to parse the design manual to obtain the text content of the PDF file, and the tabula technology to parse the PDF file to extract the table and obtain the table content data.

[0062] The file text is first processed with special treatment such as page headers and spaces, and then the chapters and chapter numbers are identified by regular expressions. The text of each page is looped through, and the text content is filled and encapsulated into the corresponding chapter, and the chapter records the page number index.

[0063] During the parsing process, the parsing algorithm needs to be compatible with the chapter number format in various scenarios. After the text paragraphs are extracted, a data structure with chapter dimensions will be formed, and then tabul a The extracted table data is matched according to the page number and content and placed in the corresponding chapter.

[0064] After parsing the file to obtain the text and table content, the divided model design and the determined extraction and positioning rules are implemented through code to extract text or tables from all sample files to be extracted, and extract the text within the corresponding range of the indicators.

[0065] Loop through each chapter of the parsed file object data structure, and perform index matching according to different model index configurations. Prioritize whether it belongs to the current chapter, and then match the paragraph and table content.

[0066] The extracted matching positioning rules are configured in a database table.

[0067] By querying the configuration data in the data table, and then through Java a The Aviator expression engine technology matches the parsed content of the file by chapter level by level.

[0068] After matching the corresponding content, the text is intercepted according to certain rules. The text matched by each indicator is spliced ​​into a large string. If a paragraph or table is matched multiple times, it is deduplicated. After the splicing is completed, the string is processed with special characters as a whole, and then each item is extracted according to each model and a sample txt file to be identified and annotated is generated and exported to the specified directory location.

[0069] 2) The labeling personnel create engineering projects corresponding to the model and add corresponding labels to each project;

[0070] 3) After the labelers import and label the sample files of all models in the doccano labeling platform and review the labeling results, they export the sample data sets of all models respectively;

[0071] 4) Synchronously build the PaddleNLP environment required for UIE model training and debug it to meet the training conditions;

[0072] 5) Do not split different model samples into training, validation set, and test set through Python code for training;

[0073] 6) After training is completed, you can use commands to perform reasoning and testing. If there are accuracy issues with the indicators, you can debug them through debugging commands;

[0074] 7) After inference is completed, export the model and deploy it using paddleserving;

[0075] 8) Write the corresponding business implementation code and the client httpclient calling code that calls the deployment service, perform joint debugging, and finally extract the results of the indicators, perform structured storage, and store them in the database table.

[0076] In step 2, based on the universal information extraction model UIE, the indicators in the unstructured files are extracted to form an engineering indicator library. The specific contents are as follows:

[0077] PaddleNLP natural language processing technology is selected to realize the extraction of indicators from unstructured files based on the universal information extraction model UIE. In the data preprocessing stage, Aspose is first used to convert the editable PDF version of the design manual and the budget estimate into txt format, and then input it into the UIE model. Sample training and data labeling are carried out, and the UIE model is tuned to realize the automatic extraction of indicators and form an engineering indicator library.

[0078] In step 1, we collect material files to sort out the index extraction rules. The specific contents are as follows:

[0079] Collect historical power grid feasibility studies and preliminary design review opinions, sort out the technical and economic indicators involved in the review opinions, and form preliminary design opinion templates, 330 kV and above feasibility study opinion templates, and 220 kV and below feasibility study opinion templates; combine design results documents, analyze the data sources of engineering indicators such as substations, overhead cables, and cables, and sort out indicator extraction rules from design specifications, budget estimates, and equipment and material inventory documents.

[0080] In step 4, according to different individual projects, the secondary and tertiary directories of the main technical solution chapters are formed. The specific contents are as follows:

[0081] The first-level title "Main Technical Solution" chapter catalog consists of two parts: non-fixed and fixed. For the non-fixed part, create the name and type of individual projects, including 9 categories: new substation construction, main transformer expansion, interval modification, new line construction, line modification, interval modification, protection modification, cable line, and safety control system. Set the corresponding third-level titles and editing permissions in the configuration file; the fixed part includes the system and secondary part, and the communication part, which are configured in the opinion template library.

[0082] In step 5, the automatically extracted indicators are combined to automatically generate review opinions in a format that conforms to the format. The specific contents are as follows:

[0083] Set the indicator bookmark in the opinion template to realize automatic filling and display of the indicator in the opinion, and according to the document format requirements, the first line is indented by 2 characters, the line spacing is 28.5, the first-level title is in Fangzheng bold GBK, the font size is No. 3, the second-level title is in Fangzheng Kaiti GBK, the font size is No. 3 and it should be bold, the main text is in Fangzheng Fangsong_GBK, the font size is No. 3, and the review opinions that conform to the format are automatically generated.

[0084] In step 3, a template library for power grid feasibility study and power grid preliminary design review opinions is constructed, and the directory chapter is customized. The specific contents are as follows:

[0085] In accordance with the document publication serial number requirements, establish a multi-level title serial number dictionary, and set the directory chapter compilation authority according to [one, to twenty,], [(one) to (twenty)], [1. to 20.], [(1) to (20)]. Automatically generate first-level and second-level directory chapters based on the review type and voltage level.

[0086] In step 6, the online editing control is used to complete the review comments. The specific contents are as follows:

[0087] By adopting online editing controls, coordinated editing by multiple disciplines is achieved, and different writing areas are controlled by editing permissions of catalog chapters. Review experts improve the content and adopt revision models to record the improved content, leaving traces of the process. After the professional review experts complete the preparation of professional opinions, the review opinions are edited after verification by professional reviewers, summary by assistants, and review by project managers.

[0088] The specific method for automatically generating review comments that conform to the format in step 5 is:

[0089] 1) Obtain the index data stored in the structured database;

[0090] 2) Prepare initial templates for various types of opinions, adjust the basic content and format of the templates, set different templates according to different project types, and prompt an error if the corresponding template is not found during the generation process;

[0091] 3) Set the document bookmark for the dynamically generated part. The bookmark is used as a placeholder to replace the content at runtime. During this period, ensure that the bookmark name is unique in the document to avoid conflicts;

[0092] 4) Use aspose-words to automatically generate opinions. First, use the Document class to load the template document, ensure that the paths of the template file and the output file are correct, and copy a copy of the template file as the opinion document for the current project;

[0093] Through the title relationships maintained by different project types, the document title is generated dynamically according to the bookmark position of the title in the document.

[0094] Loop through each tag in the document, and match and replace the indicator data queried from the database based on the relationship between the tag and the engineering business data. Replace text type data first, then replace table type data.

[0095] In the prepared template, the extracted indicator structured data is set in the corresponding bookmark to achieve the effect of automatically generating a complete opinion file.

[0096] After the content is generated, the titles and contents of different chapters are formatted step by step, including font, font size, bold and other formats.

[0097] 5) Finally, save the automatically generated complete opinion file, save the document with replaced content into a new file, name the file according to different types of projects, and store the file in the unstructured service platform.

[0098] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligently assisting generation of power grid project review opinions based on information extraction, characterized in that: The following steps are involved: S1. Collect material files, sort out index extraction rules; S2. Based on the universal information extraction model UIE, the indicators in the unstructured files are extracted to form an engineering indicator library; S3. Build a template library for power grid feasibility study and power grid preliminary design review opinions, and customize the configuration catalog chapter; S4. According to different individual projects, form the secondary and tertiary directories of the main technical solution chapters; S5. Combine the automatically extracted indicators to automatically generate review comments in a format that meets the requirements; S6. Use the online editing controls to complete the editing of review comments.

2. The method for intelligently assisting generation of power grid project review opinions based on information extraction according to claim 1 is characterized in that: In step 1, the material files are collected to sort out the index extraction rules, and the specific contents are as follows: Collect historical power grid feasibility studies and preliminary design review opinions, sort out the technical and economic indicators involved in the review opinions, and form preliminary design opinion templates, 330 kV and above feasibility study opinion templates, and 220 kV and below feasibility study opinion templates; combine design results documents, analyze the data sources of engineering indicators such as substations, overhead cables, and cables, and sort out indicator extraction rules from design specifications, budget estimates, and equipment and material inventory documents.

3. The method for intelligently assisting generation of power grid project review opinions based on information extraction according to claim 1 is characterized in that: In step 2, based on the universal information extraction model UIE, the index extraction in the unstructured file is realized to form an engineering index library, the specific contents of which are as follows: PaddleNLP natural language processing technology is selected to realize the extraction of indicators from unstructured files based on the universal information extraction model UIE. In the data preprocessing stage, Aspose is first used to convert the editable PDF version of the design manual and the budget estimate into txt format, and then input it into the UIE model. Sample training and data labeling are carried out, and the UIE model is tuned to realize the automatic extraction of indicators and form an engineering indicator library.

4. The method for intelligently assisting generation of power grid project review opinions based on information extraction according to claim 1 is characterized in that: In step 3, a template library for power grid feasibility study and preliminary design review opinions is constructed, and a custom configuration directory chapter is configured. The specific contents are as follows: In accordance with the document publication serial number requirements, establish a multi-level title serial number dictionary, and set the directory chapter compilation authority according to [one, to twenty,], [(one) to (twenty)], [1. to 20.], [(1) to (20)]. Automatically generate first-level and second-level directory chapters based on the review type and voltage level.

5. The method for intelligently assisting generation of power grid project review opinions based on information extraction according to claim 1 is characterized in that: In step 4, the secondary and tertiary directories of the main technical solution chapters are formed according to different individual projects. The specific contents are as follows: The first-level title "Main Technical Solution" chapter catalog consists of two parts: non-fixed and fixed. For the non-fixed part, create the name and type of individual projects, including 9 categories: new substation construction, main transformer expansion, interval modification, new line construction, line modification, interval modification, protection modification, cable line, and safety control system. Set the corresponding third-level titles and editing permissions in the configuration file; the fixed part includes the system and secondary part, and the communication part, which are configured in the opinion template library.

6. The method for intelligently assisting generation of power grid project review opinions based on information extraction according to claim 1 is characterized in that: In step 5, the automatically extracted indicators are combined to automatically generate review opinions that conform to the format. The specific contents are as follows: Set the indicator bookmark in the opinion template to realize automatic filling and display of the indicator in the opinion, and according to the document format requirements, the first line is indented by 2 characters, the line spacing is 28.5, the first-level title is in Fangzheng bold GBK, the font size is No. 3, the second-level title is in Fangzheng Kaiti GBK, the font size is No. 3 and it should be bold, the main text is in Fangzheng Fangsong_GBK, the font size is No. 3, and the review opinions that conform to the format are automatically generated.

7. The method for intelligently assisting generation of power grid project review opinions based on information extraction according to claim 1 is characterized in that: In step 6, an online editing control is used to complete the review and editing of the review opinions. The specific contents are as follows: By adopting online editing controls, coordinated editing by multiple disciplines is achieved, and different writing areas are controlled by editing permissions of catalog chapters. Review experts improve the content and adopt revision models to record the improved content, leaving traces of the process. After the professional review experts complete the preparation of professional opinions, the review opinions are edited after verification by professional reviewers, summary by assistants, and review by project managers.

8. The method for intelligently assisting generation of power grid project review opinions based on information extraction according to claim 1 is characterized in that: The specific extraction method of the index in step 2 is as follows: 1) Obtain the design specifications, budget estimates and other documents of the design results, and export sample files annotated with the corresponding indicators to be identified according to each divided model; 2) The labeling personnel create engineering projects corresponding to the model and add corresponding labels to each project; 3) After the labelers import and label the sample files of all models in the doccano labeling platform and review the labeling results, they export the sample data sets of all models respectively; 4) Synchronously build the PaddleNLP environment required for UIE model training and debug it to meet the training conditions; 5) Do not split different model samples into training, validation set, and test set through Python code for training; 6) After training is completed, you can use commands to perform reasoning and testing. If there are accuracy issues with the indicators, you can debug them through debugging commands; 7) After inference is completed, export the model and deploy it using paddleserving; 8) Write the corresponding business implementation code and the client httpclient calling code that calls the deployment service, perform joint debugging, and finally extract the results of the indicators, perform structured storage, and store them in the database table.

9. The method for intelligently assisting generation of power grid project review opinions based on information extraction according to claim 1, characterized in that: The specific method for automatically generating the review opinions that conform to the format in step 5 is: 1) Obtain the index data stored in the structured database; 2) Prepare initial templates for various types of opinions, adjust the basic content and format of the templates, set different templates according to different project types, and prompt an error if the corresponding template is not found during the generation process; 3) Set the document bookmark for the dynamically generated part. The bookmark is used as a placeholder to replace the content at runtime. During this period, ensure that the bookmark name is unique in the document to avoid conflicts; 4) Use aspose-words to automatically generate opinions. First, use the Document class to load the template document, ensure that the paths of the template file and the output file are correct, and copy a copy of the template file as the opinion document for the current project; 5) Finally, save the automatically generated complete opinion file, save the document with replaced content into a new file, name the file according to different types of projects, and store the file in the unstructured service platform.

Images