Rule reasoning-based comprehensive management method for hydropower station year-by-year repair process

Abstract

The invention, which belongs to the technical field of hydropower station management, relates to a rule-reasoning-based comprehensive management method for an age repair process of a hydropower station, and the method comprises the following steps: step 1, establishing a graph modeling standard age repair process library based on semantic recognition; comprising the steps of graph modeling rule construction and work ticket semantic recognition matching based on a UIE framework; and Step 2, establishing an age repair process management and control based on semantic recognition automatic discrimination. Through combination of a graph modeling technology and a semantic recognition technology, custom configuration and intelligent matching of an age repair process and lean management and control of key links are realized, so that the efficiency and accuracy of age repair work are improved, human errors are reduced, and safe operation of hydropower station equipment is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of hydropower station operation, and in particular to a comprehensive management method for annual maintenance procedures of hydropower stations based on rule reasoning. Background Art

[0002] Annual maintenance of hydropower stations is a key link in ensuring safe and efficient operation of hydropower stations. Traditional annual maintenance process management methods often rely on paper documents and manual operations, which are inefficient and prone to errors. With the development of information technology, annual maintenance process management of hydropower stations has also begun to gradually transform into digitalization and intelligence.

[0003] The existing annual maintenance management includes the following two categories: (1) Workflow-based annual maintenance management system: This type of system manages annual maintenance tasks through preset workflows, realizing automated task allocation and progress tracking. However, this type of system usually lacks flexibility and is difficult to adapt to the complex and changing working environment and needs of hydropower stations. At the same time, when processing unstructured data (such as text descriptions, work records, etc.), they often cannot accurately extract key information, affecting the accuracy and efficiency of the annual maintenance process.

[0004] (2) AI-based maintenance decision-making support systems: These systems use AI technologies, such as natural language processing and machine learning, to optimize the maintenance process and assist in decision-making. They can automatically identify and analyze large amounts of data and provide intelligent advice and support for maintenance work. However, in actual applications, these systems often face problems such as difficulty in data labeling and weak model generalization, which limits their application in actual scenarios.

[0005] Although these traditional hydropower station annual maintenance process management methods and related information systems have improved the efficiency of annual maintenance work to a certain extent, they still have some significant defects: (1) Insufficient flexibility and adaptability: Many existing hydropower station maintenance management systems are based on fixed rules and processes and are difficult to adapt to the complex and changing working environment and needs of hydropower stations. When encountering special circumstances or when the maintenance process needs to be flexibly adjusted, such systems often seem unable to cope.

[0006] (2) Low level of intelligence: Some current systems rely mainly on manual input and fixed rule judgments, and lack intelligent decision support. For example, they may not be able to automatically identify and match standard tickets such as work tickets and operation tickets, nor can they make intelligent recommendations or warnings based on historical data and real-time conditions.

[0007] (3) Limited semantic recognition capabilities: Existing systems are often unable to accurately extract key information when processing unstructured data (such as text descriptions, work records, etc.), resulting in inaccurate or missing information entry, which in turn affects the accuracy and efficiency of the annual maintenance process.

[0008] (4) Poor user experience: The user interface design of some systems is not user-friendly and the operation is cumbersome, which makes it difficult to meet the user's intuitive and convenient use needs. At the same time, the system lacks an intelligent feedback mechanism, and problems encountered by users during use are difficult to be solved in a timely manner.

[0009] (5) Insufficient data integration and analysis capabilities: Existing systems are often unable to effectively integrate and analyze data from different sources and types to provide comprehensive operational insights and decision support. This limits the room for optimization and improvement of annual maintenance work at hydropower stations. Summary of the invention

[0010] The technical problem to be solved by the present invention is to provide a comprehensive management method for the annual maintenance process of a hydropower station based on rule reasoning, aiming to solve the problems of low efficiency, easy errors and low degree of informatization in the annual maintenance process of traditional hydropower stations. By combining graphical modeling technology and semantic recognition technology, the present invention realizes the customized configuration, intelligent matching and lean management and control of key links of the annual maintenance process, thereby improving the efficiency and accuracy of the annual maintenance work, reducing human errors and ensuring the safe operation of hydropower station equipment.

[0011] In order to solve the above technical problems, the technical solution adopted by the present invention is: A comprehensive management method for annual maintenance process of hydropower stations based on rule reasoning, including: Step 1: Establish a graphical standard annual maintenance process library based on semantic recognition, including graphical rule construction and semantic recognition and matching of work tickets based on UIE framework; Step 2: Establish annual maintenance process control based on automatic identification of semantics.

[0012] The graphical rule construction in Step 1 above specifically includes: Draw a process node diagram by dragging and dropping to configure the process nodes, associate the pre- and post-conditions, and configure the corresponding node properties. The process nodes include start, judgment, and event. The semantic recognition technology is used to identify and match the standard work tickets, operation tickets, and test tickets in the standard two-ticket library. During the configuration process, set the nodes that need to be associated with the work tickets, operation tickets, and tests to form a structured standard annual maintenance process.

[0013] The semantic recognition and matching of work tickets based on the UIE framework in the above Step 1 includes: establishing the recognition process, the UIE information extraction process and the BERT model.

[0014] The above identification process is established as follows: ① Data preparation: Collect text data on the maintenance work content of the hydropower station, including historical work tickets, maintenance records, and equipment failure reports, clean, annotate, and format the data to form an annotated data set for model training; ②Construct UIE framework: According to the characteristics of maintenance work of hydropower stations, define the types of information that need to be extracted, including equipment name, maintenance items, required tools and working hours, design a user interaction interface, allow users to enter maintenance work content, and display the results of information extraction and work ticket matching; ③ Training the BERT model: Use the labeled data set to adjust the BERT model. Through training, the BERT model can understand the semantics and context of the maintenance work content; ④Semantic recognition and information extraction: Users input maintenance work content through the UIE framework. The BERT model performs semantic recognition on the input text, understands its meaning and context, and extracts relevant information from the input text, including equipment name and maintenance items, according to the information extraction structure defined by the UIE framework; ⑤ Work ticket matching: sort out and standardize the standard work ticket sheets of the hydropower station, and establish a database. Use the BERT model to calculate the semantic similarity between the input text and the standard work ticket in the database, and select the most matching standard work ticket; ⑥ Work ticket matching and model iteration: The information extraction results and the most matching standard work ticket are displayed through the UIE framework, allowing users to confirm or modify the matching results to improve the accuracy and adaptability of the system, and collect user feedback during use for continuous optimization and updating of the model.

[0015] The above UIE information extraction process includes: Input stage: The user inputs the text data to be processed through the UIE framework. The text data is sentences, paragraphs or entire articles generated according to the requirements of the information extraction task; Encoding stage: The structured extraction language encodes the input text and converts it into a format that the model can understand; Information extraction stage: The model uses the IE capabilities obtained from structured pattern prompts and large-scale data training to extract information from the encoded text, which includes identifying entities, extracting relationships, and detecting events. Structured output stage: The model organizes and formats the extracted information according to a predefined structured pattern to generate a unified structured output; Post-processing and presentation stage: Perform necessary post-processing on the generated structured data, including cleaning, verification, and integration.

[0016] The above BERT model building process is: Data collection and preprocessing: Collect unlabeled data for preprocessing, including word segmentation and stop word removal, and then convert the text into a format that the model can process; Pre-training: Use unlabeled data to pre-train the BERT model. The pre-training tasks include "cloze test", namely Masked Language Model, MLM; and "Next Sentence Prediction", namely Next Sentence Prediction, NSP, so that the model can learn the deep representation of language; Fine-tuning: Further training is performed on the basis of the pre-trained model to adapt it to a specific NLP task; the fine-tuning process usually requires a small amount of labeled data and is adjusted according to the specific needs of the task, including using BERT as a feature extractor in sentiment analysis tasks, and then inputting the extracted features into the classifier for sentiment classification.

[0017] The above-mentioned Step 2 specifically includes: According to the annual maintenance work plan of the hydropower station, a maintenance plan is formulated and transferred to the annual maintenance process to be executed library according to the time node configured in the maintenance plan; maintenance personnel formulate annual maintenance tasks according to the maintenance plan content by associating the corresponding annual maintenance process template. After the task is issued, the annual maintenance process execution progress is updated based on the automatic judgment of semantic recognition; The annual maintenance process is completed through the cooperation of the system web and APP. The web side includes comprehensive management of maintenance plans, comprehensive management of annual maintenance processes, comprehensive management of two tickets, test process management and process management, which is used to manage the construction of annual maintenance processes, task issuance and information management; the APP side includes comprehensive management of annual maintenance processes, two-ticket execution management and task supervision, which is used to record the content of work tickets, execution results and feedback results.

[0018] The above annual maintenance process includes: ① According to the maintenance plan, select and match the specific equipment or parts that need to be repaired, submit the maintenance application form, and the system background will automatically match the annual maintenance process that needs to be executed in the standard annual maintenance process library, and generate a waiting annual maintenance task list. According to the node logical sequence of the annual maintenance process, the system will automatically issue the node maintenance task package. The node maintenance task package contains all the configured two tickets. Each work ticket is issued to the designated maintenance personnel APP end according to the configured attributes; ② The maintenance personnel confirm and receive the two tasks on the APP, perform the work tasks according to the standards, and record the status information of the two tickets. After the tasks are completed, the tickets are transferred to the supervisor's system. After the supervisor confirms the results of the work tasks on site, he / she will feedback the inspection results to the maintenance personnel. If it fails, the maintenance personnel system will receive the rectification task. The maintenance personnel will improve the work content according to the supervisor's guidance information and complete the submission for review. If it passes, the system will automatically close the work task and enter the next work task. ③ After the work ticket is executed, the system will use multi-dimensional data access and unstructured data semantic recognition technology, combined with automatic identification of process nodes, to identify the status of the process node maintenance task package. After determining that all work tickets are closed, the annual maintenance process node will be closed and the closed status of the process node will be displayed by lighting up, and then the next annual maintenance process node task will be entered; ④ After ② and ③ are completed according to the annual maintenance process node logic, close the annual maintenance process task, save all work information, and transfer it to the historical case library for backup.

[0019] The present invention provides a comprehensive management method for the annual maintenance process of a hydropower station based on rule reasoning, which has the following beneficial effects: (1) Significantly improved efficiency: Through graphical modeling technology, the annual maintenance process can be customized, making the creation and modification of the process more flexible and efficient. Compared with the traditional method of manually drawing and updating flowcharts, graphical modeling rule construction can greatly improve the efficiency of process design. At the same time, automatic judgment based on semantic recognition can update the execution progress of the annual maintenance process in real time, reducing the workload of manual tracking and updating, and further improving management efficiency.

[0020] (2) Accuracy is greatly improved: The semantic recognition method proposed in this paper, which combines the UIE framework and the BERT model, can automatically extract key information from the maintenance work content and accurately match the corresponding standard work ticket of the hydropower station. This automated information extraction and matching mechanism not only reduces human errors, but also greatly improves the accuracy of work ticket matching.

[0021] (3) Optimal utilization of resources: Through intelligent annual maintenance process control, maintenance resources and human resources can be allocated more reasonably. The system automatically sends node maintenance task packages to the designated maintenance personnel APP terminal based on the actual situation of the annual maintenance process and tasks, ensuring the effective utilization of resources and timely completion of tasks.

[0022] (4) Management transparency: Through the cooperation of the system web and APP, the annual maintenance process is electronically managed and updated in real time. Managers at all levels can check the execution status of the annual maintenance process and the completion status of work tickets at any time, which improves the transparency and traceability of management.

[0023] (5) Enhanced decision support: The standard annual maintenance process library and rich historical data constructed by the present invention provide powerful decision support for the management of hydropower stations. By analyzing and mining historical data, it can help the management to find bottlenecks and problems in the annual maintenance process, so as to formulate more scientific and reasonable annual maintenance plans and strategies.

[0024] (6) Dynamic strength: Automated annual maintenance process control greatly reduces the workload of maintenance personnel and supervisors. Maintenance personnel only need to confirm and receive tasks and record work ticket status information on the APP to complete their work tasks; supervisors can view work progress and results in real time through the system, reducing the workload of on-site inspections and acceptance.

[0025] (7) Environmental protection and energy saving: Intelligent annual maintenance process management can reduce unnecessary maintenance activities and resource waste, thereby reducing the impact on the environment and energy consumption. At the same time, electronic management methods also reduce the use of paper documents and storage costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 Constructing a flow chart for the graphical modeling rules of the present invention; Figure 2 This is a structural diagram of the comprehensive management system for the annual maintenance process of a hydropower station according to the present invention; Figure 3 This is a business flow chart of annual maintenance process management and control based on automatic identification of semantics according to the present invention. DETAILED DESCRIPTION

[0027] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and embodiments.

[0028] A comprehensive management method for annual maintenance process of hydropower stations based on rule reasoning, including: Step 1: Establish a graphical standard annual maintenance process library based on semantic recognition, including graphical rule construction and semantic recognition and matching of work tickets based on UIE framework; Step 2: Establish annual maintenance process control based on automatic identification of semantics.

[0029] The graphical rule construction in Step 1 above specifically includes: Draw a process node diagram by dragging and dropping to configure the process nodes, associate the pre- and post-conditions, and configure the corresponding node properties. The process nodes include start, judgment, and event. The semantic recognition technology is used to identify and match the standard work tickets, operation tickets, and test tickets in the standard two-ticket library. During the configuration process, set the nodes that need to be associated with the work tickets, operation tickets, and tests to form a structured standard annual maintenance process.

[0030] The "two tickets" in the standard two-ticket library include operation tickets and work tickets.

[0031] Operation method and process: 1) Graphical rules to build a workflow template library Based on graphical modeling technology, customized configuration of different annual maintenance type process templates is realized. By dragging and dropping the process nodes (including start, judgment, event, etc.), the process node diagram is drawn, the pre- and post-conditions are associated, and the corresponding node attributes and the key information of the two general tickets (operation ticket, work ticket) (work ticket, operation ticket) are configured to form a structured standard process template to support the template selection during the start-up process of the front-end job application form.

[0032] 2) Construction and execution of job application forms (1) Selection of workflow template Connect to the job application form of the industrial Internet platform, identify and hang it under the equipment structure tree, select a standard job template, or customize and rebuild the job process, or modify the standard job template to complete the process matching of the job form.

[0033] (2) Two-ticket (operation ticket, work ticket) configuration The two tickets (operation tickets, work tickets) include work tickets and operation tickets, and the data information is connected to the standard two-ticket (operation ticket, work ticket) work order database of the industrial Internet platform; according to the two-ticket (operation ticket, work ticket) information [keywords] in the standard operating process, the system will [automatically] identify the two-ticket (operation ticket, work ticket) keyword group in the standard operating process template through semantic recognition technology in the background, decompose it for semantic recognition, and match the standard two-ticket (operation ticket, work ticket) standard ticket sheets of work tickets and operation tickets in the standard two-ticket (operation ticket, work ticket) library for automatic extraction and pairing; for those without standard information reminders, you can [manually] search for keywords in the two-ticket (operation ticket, work ticket) library to customize the matching of two-ticket (operation ticket, work ticket) work orders according to actual operating needs.

[0034] The semantic recognition and matching of work tickets based on the UIE framework in the above Step 1 includes: establishing the recognition process, the UIE information extraction process and the BERT model.

[0035] The above identification process is established as follows: ① Data preparation: Collect text data on the maintenance work content of the hydropower station, including historical work tickets, maintenance records, and equipment failure reports, clean, annotate, and format the data to form an annotated data set for model training; ②Construct UIE framework: According to the characteristics of maintenance work of hydropower stations, define the types of information that need to be extracted, including equipment name, maintenance items, required tools and working hours, design a user interaction interface, allow users to enter maintenance work content, and display the results of information extraction and work ticket matching; The main characteristics of the maintenance work of a hydropower station are planned maintenance events, fixed maintenance main processes, and different on-site requirements. An operation process template library can be constructed based on the planned maintenance events and fixed maintenance main processes, and the power station personnel can directly extract and screen when establishing a maintenance application; different on-site requirements mainly refer to the different operation tickets and work ticket contents included in the maintenance plan. According to this feature, the power station personnel in the present invention can self-match different two-ticket (operation ticket, work ticket) information in the operation process template to customize the specific execution content of the process node.

[0036] ③ Training the BERT model: Use the labeled data set to adjust the BERT model to better adapt it to the semantic recognition of the maintenance work content of the hydropower station. Through training, the BERT model can understand the semantics and context of the maintenance work content; ④Semantic recognition and information extraction: Users input maintenance work content through the UIE framework. The BERT model performs semantic recognition on the input text, understands its meaning and context, and extracts relevant information from the input text, including equipment name and maintenance items, according to the information extraction structure defined by the UIE framework; ⑤ Work ticket matching: sort out and standardize the standard work ticket sheets of the hydropower station, and establish a database. Use the BERT model to calculate the semantic similarity between the input text and the standard work ticket in the database, and select the most matching standard work ticket; ⑥ Work ticket matching and model iteration: The information extraction results and the most matching standard work ticket are displayed through the UIE framework, allowing users to confirm or modify the matching results to improve the accuracy and adaptability of the system, and collect user feedback during use for continuous optimization and updating of the model.

[0037] The above UIE information extraction process includes: Input stage: The user inputs the text data to be processed through the UIE framework. The text data is sentences, paragraphs or entire articles generated according to the requirements of the information extraction task; Encoding stage: The structured extraction language encodes the input text and converts it into a format that the model can understand. This step is the key to ensure that the model can accurately identify the information points in the text and perform subsequent processing; Information extraction stage: The model uses the IE capabilities obtained from structured pattern prompts and large-scale data training to extract information from the encoded text, which includes identifying entities, extracting relationships, and detecting events. Structured output stage: The model organizes and formats the extracted information according to a predefined structured pattern to generate a unified structured output, which is convenient for subsequent data analysis and application; Post-processing and presentation stage: Perform necessary post-processing on the generated structured data, including cleaning, verification, and integration.

[0038] The above BERT model building process is as follows: Data collection and preprocessing: Collect unlabeled data for preprocessing, including word segmentation and stop word removal, and then convert the text into a format that the model can process; Pre-training: Use unlabeled data to pre-train the BERT model. The pre-training tasks include "cloze test", namely Masked Language Model, MLM; and "Next Sentence Prediction", namely Next Sentence Prediction, NSP, so that the model can learn the deep representation of language; Fine-tuning: Further training is performed on the basis of the pre-trained model to adapt it to a specific NLP task. The fine-tuning process usually requires a small amount of labeled data and is adjusted according to the specific needs of the task, including in the sentiment analysis task, using BERT as a feature extractor and then inputting the extracted features into the classifier for sentiment classification.

[0039] The above-mentioned Step 2 specifically includes: According to the annual maintenance work plan of the hydropower station, a maintenance plan is formulated and transferred to the annual maintenance process to be executed library according to the time node configured in the maintenance plan; maintenance personnel formulate annual maintenance tasks according to the maintenance plan content by associating the corresponding annual maintenance process template. After the task is issued, the annual maintenance process execution progress is updated based on the automatic judgment of semantic recognition; The annual maintenance process is completed through the cooperation of the system web and APP. The web side includes comprehensive management of maintenance plans, comprehensive management of annual maintenance processes, comprehensive management of two tickets, test process management and process management, which is used to manage the construction of annual maintenance processes, task issuance and information management; the APP side includes comprehensive management of annual maintenance processes, two-ticket execution management and task supervision, which is used to record the content of work tickets, execution results and feedback results.

[0040] The above annual maintenance process includes: ① According to the maintenance plan, select and match the specific equipment or parts that need to be repaired, submit the maintenance application form, and the system background will automatically match the annual maintenance process that needs to be executed in the standard annual maintenance process library, and generate a waiting annual maintenance task order. According to the node logical sequence of the annual maintenance process, the system will automatically issue the node maintenance task package. The node maintenance task package contains all the configured two tickets. Each work ticket is issued to the designated maintenance personnel APP end according to the configured attributes. The attributes include the work order number, work order status, related equipment, related maintenance personnel and supervisory personnel information, and work content; ② The maintenance personnel confirm and receive the two tasks on the APP, perform the work tasks according to the standards, and record the status information of the two tickets. The status information includes the task status, ticket number, execution time, and closing time. After the task is completed, the ticket is transferred to the supervisor's system. After the supervisor confirms the results of the work task on site, he will feedback the inspection results to the maintenance personnel. If it fails, the maintenance personnel system will receive the rectification task. The maintenance personnel will improve the work content according to the guidance information of the supervisor and complete the submission for review. If it passes, the system will automatically close the work task and enter the next work task. ③ After the work ticket is executed, the system will use multi-dimensional data access and unstructured data semantic recognition technology, combined with automatic identification of process nodes, to identify the status of the process node maintenance task package. After determining that all work tickets are closed, the annual maintenance process node will be closed and the closed status of the process node will be displayed by lighting up. Then, the next annual maintenance process node task will be entered to achieve lean management and control of the key links of the annual maintenance process. ④ After ② and ③ are completed according to the annual maintenance process node logic, close the annual maintenance process task, save all work information, and transfer it to the historical case library for backup.

[0041] Example: Establish a comprehensive management method for the annual maintenance process of hydropower stations based on rule reasoning, including: 1. Construction of a graphical standard annual maintenance process library based on semantic recognition 1) Graphical rule construction Based on the graphical modeling technology, the customized configuration of different annual maintenance types of processes is realized. The process node diagram is drawn by dragging and dropping the process nodes. The process nodes include start, judgment, event, etc., and the pre- and post-conditions are associated. The corresponding node attributes are configured. The semantic recognition technology is used to identify and match the standard tickets such as work tickets, operation tickets, and tests in the standard two-ticket library. During the configuration process, the nodes that need to be associated with work tickets, operation tickets, and tests are set to form a structured standard annual maintenance process. The graphical modeling rule construction is shown in the attached figure. Figure 1 shown.

[0042] 2) Semantic recognition and matching method of work tickets based on UIE framework A semantic recognition method combining the UIE framework and the BERT model is proposed to automatically extract key information in the maintenance work content and match the corresponding standard work ticket of the hydropower station.

[0043] (1) Identification process ①Data preparation Collect text data on the maintenance work content of the hydropower station, including historical work tickets, maintenance records, equipment failure reports, etc. Clean, annotate and format the data to form an annotated data set for model training.

[0044] ②Build UIE framework According to the characteristics of the maintenance work of the hydropower station, define the types of information that need to be extracted, such as equipment name, maintenance items, required tools, working hours, etc. Design a user-friendly interface that allows users to enter the maintenance work content and displays the results of information extraction and work ticket matching.

[0045] ③Training the BERT model The BERT model is fine-tuned using the annotated dataset to better adapt to the semantic recognition of maintenance work content in hydropower stations. Through training, the BERT model can understand the semantics and context of maintenance work content.

[0046] ④Semantic recognition and information extraction The user inputs the maintenance work content through the UIE framework, and the BERT model performs semantic recognition on the input text to understand its meaning and context. According to the information extraction structure defined by the UIE framework, relevant information such as equipment name and maintenance items are extracted from the input text.

[0047] ⑤ Work ticket matching Organize and standardize the standard work tickets of hydropower stations and build a database. Use the BERT model to calculate the semantic similarity between the input text and the standard work tickets in the database, and select the most matching standard work ticket.

[0048] ⑥ Work ticket matching and model iteration The information extraction results and the most matching standard work ticket are displayed through the UIE framework. Users are allowed to confirm or modify the matching results to improve the accuracy and adaptability of the system. User feedback during use is collected for continuous optimization and updating of the model.

[0049] (2) UIE information extraction process Input stage: Users input text data to be processed through the UIE framework. These data can be sentences, paragraphs, or entire articles, depending on the requirements of the information extraction task.

[0050] Encoding stage: The structured extraction language encodes the input text and converts it into a format that the model can understand. This step is the key to ensure that the model can accurately identify the information points in the text and perform subsequent processing.

[0051] Information extraction stage: The model uses structured pattern prompters and IE capabilities obtained through large-scale data training to extract information from the encoded text. This includes tasks such as identifying entities, extracting relationships, and detecting events.

[0052] Structured output stage: The model organizes and formats the extracted information according to a predefined structured pattern to generate a unified structured output. This output method facilitates subsequent data analysis and application.

[0053] Post-processing and presentation stage: Perform necessary post-processing on the generated structured data, such as cleaning, verification, and integration.

[0054] (3) BERT model Data collection and preprocessing: Collect large amounts of unlabeled data for preprocessing, including operations such as word segmentation and stop word removal, and then convert the text into a format that the model can process.

[0055] Pre-training: Use large-scale unlabeled data to pre-train the BERT model. Pre-training tasks include "cloze", namely Masked Language Model, MLM, and "Next Sentence Prediction", namely Next Sentence Prediction, NSP. Through these tasks, the model can learn the deep representation of language.

[0056] Fine-tuning: Further training is performed on the basis of the pre-trained model to adapt it to a specific NLP task. The fine-tuning process usually requires a small amount of labeled data and is adjusted according to the specific needs of the task. For example, in a sentiment analysis task, BERT can be used as a feature extractor, and the extracted features are then input into a classifier for sentiment classification.

[0057] 2. Annual maintenance process control based on automatic identification of semantics According to the annual maintenance work plan of the hydropower station, a maintenance plan is formulated, and according to the time node configured in the maintenance plan, it is directly transferred to the annual maintenance process to be executed library. The maintenance personnel formulate the annual maintenance tasks according to the maintenance plan content by associating the corresponding annual maintenance process template. After the task is issued, the annual maintenance process execution progress is updated based on the automatic judgment of semantic recognition. The annual maintenance process control business process is as follows Figure 2 shown.

[0058] This business process is mainly completed in the cooperation of the system web and APP. The functional topology of the hydropower station annual maintenance process integrated management system is as shown in the attached figure. Figure 3 The Web end includes comprehensive management of maintenance plans, comprehensive management of annual maintenance processes, comprehensive management of two tickets, test process management, process management, etc., which are used to manage the construction of annual maintenance processes, task distribution, information management, etc.; the APP end includes comprehensive management of annual maintenance processes, two ticket execution management, and task supervision, etc., which are used to record the content of work tickets, execution results, and feedback results.

[0059] The detailed business process is as follows: ① According to the maintenance plan, select and match the specific equipment or parts that need to be repaired, submit the maintenance application form, and the system background will automatically match the annual maintenance process that needs to be executed in the standard annual maintenance process library, and generate a waiting annual maintenance task order. According to the logical sequence of the nodes in the annual maintenance process, the system will automatically issue the node maintenance task package. The maintenance task package contains all the two tickets that have been configured. Each work ticket is issued to the designated maintenance personnel APP end according to the configured attributes. The attributes include work order number, work order status, related equipment, related maintenance personnel and supervisory personnel information, work content, etc.

[0060] ② The maintenance personnel confirm and receive the two-ticket task on the APP, perform the work tasks according to the standards, and record the two-ticket status information, which includes the task status, ticket number, execution time, and closing time. After the task is completed, the ticket will be transferred to the supervisor system. After the supervisor confirms the results of the work task on site, he will feedback the inspection results to the maintenance personnel; if it fails, the maintenance personnel system will receive the rectification task, and the maintenance personnel will improve the work content according to the supervisor's guidance information and complete the submission for review; if it passes, the system will automatically close the work task and enter the next work task.

[0061] ③After the work ticket is executed, the system will access multi-dimensional data and semantic recognition technology of unstructured data, combined with automatic identification of process nodes, to realize the status identification of process node maintenance task packages. After determining that the status of all work tickets is closed, the annual maintenance process node will be closed, and the closed status of the process node will be displayed by lighting up. Then, the system will directly enter the next annual maintenance process node task to realize lean management and control of key links in the annual maintenance process.

[0062] ④ After ② and ③ are completed according to the annual maintenance process node logic, close the annual maintenance process task, save all work information, and transfer it to the historical case library for backup.

[0063] The key points of the present invention are: (1) Graphical rule construction: The present invention uses graphical technology to achieve custom configuration of the annual maintenance process. By dragging and dropping the process nodes, a process node diagram is drawn, and the pre- and post-conditions can be associated to configure the corresponding node attributes. This graphical method makes the annual maintenance process more intuitive, easy to understand and operate.

[0064] (2) Application of semantic recognition technology: This paper combines the UIE framework and the BERT model to automatically extract key information from maintenance work content through semantic recognition technology and match the corresponding standard work tickets for hydropower stations. The application of this technology greatly improves the accuracy and efficiency of work ticket matching.

[0065] (3) Work order matching and model iteration: The present invention establishes a standard work order database and uses the BERT model to calculate the semantic similarity between the input text and the standard work order in the database, and selects the most matching standard work order. At the same time, the system allows users to confirm or modify the matching results, thereby continuously collecting and feeding back user data for continuous optimization and updating of the model, thereby improving the accuracy and adaptability of the system.

[0066] (4) Automated control of annual maintenance processes: The present invention realizes automated control of annual maintenance processes. The system automatically transfers tasks to the annual maintenance process to be executed based on the configured time nodes of the maintenance plan. During task execution, the system automatically identifies and updates the status of maintenance task packages at process nodes through the access of multidimensional data and semantic recognition technology of unstructured data, combined with automatic identification of process nodes. This automated control method greatly improves the efficiency and accuracy of the annual maintenance process.

[0067] (5) Multi-terminal collaborative work: The present invention realizes the electronic management of maintenance plans, annual maintenance processes, work tickets and other information through the cooperation of the system web terminal and the APP terminal. The web terminal is mainly used to manage the construction of the annual maintenance process, task issuance, information management and other tasks; while the APP terminal is mainly used to record the content of the work ticket, execution results and feedback results. This multi-terminal collaborative work method improves work efficiency and convenience.

[0068] The objects of the present invention are: (1) Custom configuration of annual maintenance process: Based on graphical modeling technology, users are allowed to customize different types of annual maintenance processes according to the actual needs and working environment of the hydropower station. By dragging and dropping process nodes, drawing process node diagrams, associating the pre- and post-conditions, and configuring the corresponding node properties, a structured standard annual maintenance process is formed.

[0069] (2) Improve the accuracy and efficiency of work ticket matching: By combining the UIE framework and the semantic recognition method of the BERT model, key information in the maintenance work content is automatically extracted and matched with the corresponding standard work tickets of the hydropower station. This method can greatly improve the accuracy and efficiency of work ticket matching and reduce the possibility of manual intervention and errors.

[0070] (3) Realize intelligent control of the annual maintenance process: During the annual maintenance process, through the access of multi-dimensional data and the semantic recognition technology of unstructured data, combined with the automatic identification of process nodes, the status of the maintenance task package at the process node can be automatically identified and updated. This can ensure that the annual maintenance process proceeds smoothly according to the preset logical sequence, while realizing lean control of key links.

[0071] (4) Improving the level of informatization of annual maintenance work of hydropower stations: Through the comprehensive management method provided by the present invention, the level of informatization of annual maintenance work of hydropower stations will be significantly improved. Maintenance plans, annual maintenance processes, work tickets and other information will be managed and stored electronically, which is convenient for query, analysis and optimization. At the same time, through the cooperation of the system web and APP, maintenance personnel and supervisors can communicate and collaborate more conveniently.

Claims

1. A comprehensive management method for annual maintenance process of hydropower station based on rule reasoning, characterized in that: include: Step 1: Establish a graphical standard annual maintenance process library based on semantic recognition; Including graphical modeling rule construction and work ticket semantic recognition and matching based on UIE framework; Step 2: Establish annual maintenance process control based on automatic identification of semantics.

2. According to claim 1, a comprehensive management method for annual maintenance process of a hydropower station based on rule reasoning is characterized in that: The graphical modeling rule construction in Step 1 specifically includes: Draw a process node diagram by dragging and dropping to configure the process nodes, associate the pre- and post-conditions, and configure the corresponding node properties. The process nodes include start, judgment, and event. The semantic recognition technology is used to identify and match the standard work tickets, operation tickets, and test tickets in the standard two-ticket library. During the configuration process, set the nodes that need to be associated with the work tickets, operation tickets, and tests to form a structured standard annual maintenance process.

3. According to claim 2, a comprehensive management method for annual maintenance process of a hydropower station based on rule reasoning is characterized in that: The semantic recognition and matching of work tickets based on the UIE framework in Step 1 includes: establishing a recognition process, a UIE information extraction process, and a BERT model.

4. A comprehensive management method for annual maintenance process of a hydropower station based on rule reasoning according to claim 3, characterized in that: The identification process is established as follows: ① Data preparation: Collect text data on the maintenance work content of the hydropower station, including historical work tickets, maintenance records, and equipment failure reports, clean, annotate, and format the data to form an annotated data set for model training; ②Construct UIE framework: According to the characteristics of maintenance work of hydropower stations, define the types of information that need to be extracted, including equipment name, maintenance items, required tools and working hours, design a user interaction interface, allow users to enter maintenance work content, and display the results of information extraction and work ticket matching; ③ Training the BERT model: Use the labeled data set to adjust the BERT model. Through training, the BERT model can understand the semantics and context of the maintenance work content; ④Semantic recognition and information extraction: Users input maintenance work content through the UIE framework. The BERT model performs semantic recognition on the input text, understands its meaning and context, and extracts relevant information from the input text, including equipment name and maintenance items, according to the information extraction structure defined by the UIE framework; ⑤ Work ticket matching: sort out and standardize the standard work ticket sheets of the hydropower station, and establish a database. Use the BERT model to calculate the semantic similarity between the input text and the standard work ticket in the database, and select the most matching standard work ticket; ⑥ Work ticket matching and model iteration: The information extraction results and the most matching standard work ticket are displayed through the UIE framework, allowing users to confirm or modify the matching results to improve the accuracy and adaptability of the system, and collect user feedback during use for continuous optimization and updating of the model.

5. A comprehensive management method for annual maintenance process of a hydropower station based on rule reasoning according to claim 4, characterized in that: The UIE information extraction process includes: Input stage: The user inputs the text data to be processed through the UIE framework. The text data is sentences, paragraphs or entire articles generated according to the requirements of the information extraction task; Encoding stage: The structured extraction language encodes the input text and converts it into a format that the model can understand; Information extraction stage: The model uses the IE capabilities obtained from structured pattern prompts and large-scale data training to extract information from the encoded text, which includes identifying entities, extracting relationships, and detecting events. Structured output stage: The model organizes and formats the extracted information according to a predefined structured pattern to generate a unified structured output; Post-processing and presentation stage: Perform necessary post-processing on the generated structured data, including cleaning, verification, and integration.

6. A comprehensive management method for annual maintenance process of a hydropower station based on rule reasoning according to claim 5, characterized in that: The BERT model building process is as follows: Data collection and preprocessing: Collect unlabeled data for preprocessing, including word segmentation and stop word removal, and then convert the text into a format that the model can process; Pre-training: Use unlabeled data to pre-train the BERT model. The pre-training tasks include "cloze test", namely Masked Language Model, MLM; and "Next Sentence Prediction", namely Next Sentence Prediction, NSP, so that the model can learn the deep representation of language; Fine-tuning: Further training is performed on the basis of the pre-trained model to adapt it to a specific NLP task; the fine-tuning process usually requires a small amount of labeled data and is adjusted according to the specific needs of the task, including using BERT as a feature extractor in sentiment analysis tasks, and then inputting the extracted features into the classifier for sentiment classification.

7. A comprehensive management method for annual maintenance process of a hydropower station based on rule reasoning according to claim 6, characterized in that: The Step 2 specifically includes: According to the annual maintenance work plan of the hydropower station, a maintenance plan is formulated and transferred to the annual maintenance process to be executed library according to the time node configured in the maintenance plan; maintenance personnel formulate annual maintenance tasks according to the maintenance plan content by associating the corresponding annual maintenance process template. After the task is issued, the annual maintenance process execution progress is updated based on the automatic judgment of semantic recognition; The annual maintenance process is completed through the cooperation of the system web and APP. The web side includes comprehensive management of maintenance plans, comprehensive management of annual maintenance processes, comprehensive management of two tickets, test process management and process management, which is used to manage the construction of annual maintenance processes, task issuance and information management; the APP side includes comprehensive management of annual maintenance processes, two-ticket execution management and task supervision, which is used to record the content of work tickets, execution results and feedback results.

8. A comprehensive management method for annual maintenance process of a hydropower station based on rule reasoning according to claim 7, characterized in that: The annual maintenance process includes: ① According to the maintenance plan, select and match the specific equipment or parts that need to be repaired, submit the maintenance application form, and the system background will automatically match the annual maintenance process that needs to be executed in the standard annual maintenance process library, and generate a waiting annual maintenance task list. According to the node logical sequence of the annual maintenance process, the system will automatically issue the node maintenance task package. The node maintenance task package contains all the configured two tickets. Each work ticket is issued to the designated maintenance personnel APP end according to the configured attributes; ② The maintenance personnel confirm and receive the two tasks on the APP, perform the work tasks according to the standards, and record the status information of the two tickets. After the tasks are completed, the tickets are transferred to the supervisor's system. After the supervisor confirms the results of the work tasks on site, he / she will feedback the inspection results to the maintenance personnel. If it fails, the maintenance personnel system will receive the rectification task. The maintenance personnel will improve the work content according to the supervisor's guidance information and complete the submission for review. If it passes, the system will automatically close the work task and enter the next work task. ③ After the work ticket is executed, the system will use multi-dimensional data access and unstructured data semantic recognition technology, combined with automatic identification of process nodes, to identify the status of the process node maintenance task package. After determining that all work tickets are closed, the annual maintenance process node will be closed and the closed status of the process node will be displayed by lighting up, and then the next annual maintenance process node task will be entered; ④ After ② and ③ are completed according to the annual maintenance process node logic, close the annual maintenance process task, save all work information, and transfer it to the historical case library for backup.

Images