Power generation equipment maintenance knowledge question-answering system and method based on multi-modal large model

Abstract

The invention discloses a power generation equipment maintenance knowledge question-answering system and method based on a multi-modal large model. The system comprises an equipment question-answering robot module, a data lake module, a large data model module and the like. A data lake knowledge base is constructed; receiving user demand questions for semantic analysis and classification, and determining data types in the demand questions; according to the data type, selecting an appropriate trained RAG large model to carry out comprehensive processing and reasoning, and generating a question answer; and presenting the question answer to the user as a reply. According to the invention, data intercommunication between equipment knowledge questions and answers and maintenance questions and answers is realized through the data sharing module, so that the accuracy of fault diagnosis and the maintenance efficiency are improved; the application of the multi-modal large model and the data management module provides powerful data support for equipment maintenance; and through the real-time monitoring function of the fault alarm module, the equipment fault can be found in time, the downtime is reduced, and the production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power generation equipment maintenance knowledge question and answer, and in particular to a power generation equipment maintenance knowledge question and answer system and method based on a multi-modal large model. Background Art

[0002] In recent years, with the development of industry, the demand for electricity has continued to rise, which has brought considerable pressure to power grid companies; traditional thermal power stations rely on a large amount of coal resources, which not only has high power generation costs but also pollutes the environment. To meet this challenge, hydropower generation technology has been widely used, and hydropower stations have gradually replaced some thermal power stations.

[0003] To build a high-quality hydropower station, it is necessary to pay attention to the operation, maintenance and troubleshooting of electromechanical equipment. By formulating a reasonable maintenance and repair plan and performing daily maintenance work, electromechanical equipment failures can be effectively eliminated to ensure the stable operation of the hydropower station. Therefore, it is necessary to develop a reasonable and intelligent hydropower equipment maintenance decision-making method and the demand for intelligent systems is also increasing.

[0004] At present, there are still some problems in the operation, maintenance and troubleshooting of electromechanical equipment in hydropower stations, which are mainly reflected in the following two aspects: 1) The equipment maintenance inspection system is not sound enough. Due to the various types of electromechanical equipment and technologies used in hydropower stations, the factors involved in the operation process are complex and diverse. In order to ensure the stable operation of hydropower stations and the efficient development of maintenance work, it is necessary to build a complete equipment inspection system according to the actual situation of hydropower stations. However, many hydropower stations have not fully realized the importance of establishing a complete inspection system and still use traditional maintenance methods. This practice is difficult to achieve comprehensive inspections and is not conducive to the stable operation of hydropower stations. 2) Equipment safety management and intelligent maintenance management system need to be improved. In the operation, maintenance and troubleshooting of electromechanical equipment in hydropower stations, equipment safety management is particularly critical. In order to ensure the level of safety management, it is necessary to establish a sound safety management system to create a safe production environment for employees and ensure the safety of employees. However, many hydropower stations currently lack such equipment safety management systems and intelligent equipment maintenance management systems, which undoubtedly affects the quality of safety management and has an adverse impact on the development of hydropower stations.

[0005] The invention application with application number 202410568353.5 discloses a wind power equipment maintenance method, system, equipment and medium, and relates to the field of wind power technology. It includes a maintenance knowledge base module, which is used to establish a maintenance knowledge base and train a question-and-answer model based on the data in the maintenance knowledge base to generate a question-and-answer robot to provide corresponding fault handling measures; a scheduling module, which is used to calculate the tasks in the task pool using preset influencing factors to determine the priority order of task execution, and then set the corresponding task execution time for the task. Using the technical solution of this application, intelligent wind power maintenance can be achieved based on multidimensional data. However, it also has the following problems: it is unable to provide equipment query services and knowledge answers; it is unable to provide fault diagnosis services and maintenance suggestions; it is unable to monitor equipment operation failures online, and so on. Summary of the invention

[0006] In view of the above-mentioned problems, the purpose of the present invention is to provide a knowledge question-answering system and method for power generation equipment maintenance based on a multimodal large model, which can provide equipment query services, knowledge answers, fault diagnosis, maintenance suggestion services, etc., and realize online monitoring of equipment operation failures.

[0007] The embodiment of the present invention provides a knowledge question-answering system and method for power generation equipment maintenance based on a multi-modal large model.

[0008] A first aspect: A knowledge question-answering system for power generation equipment maintenance based on a multi-modal large model, comprising:

[0009] The device question-answering robot module receives user questions and feeds them back to the data lake module;

[0010] The data lake module stores a large amount of device-related data and calls the big data model module based on user questions;

[0011] The big data model module processes and analyzes the massive device-related data stored in the data lake module based on user questions, obtains the results of user questions, and feeds back to the user through the device question-answering robot module.

[0012] Furthermore, it also includes:

[0013] Equipment operation management module, used to collect and obtain equipment-related data;

[0014] Data governance and control module, used to clean, integrate and standardize collected data, and manage and control it;

[0015] Equipment fault alarm module monitors the real-time collected equipment-related data and issues an alarm when a fault is detected;

[0016] Data sharing module, used for data sharing between system modules.

[0017] A second aspect: A knowledge question-answering method for power generation equipment maintenance based on the system of the first aspect, comprising the steps of:

[0018] S1. Collect equipment-related data, perform preprocessing, and build a data lake knowledge base;

[0019] S2. Receive user demand questions using the knowledge question answering system, perform semantic analysis and classification on the user demand questions, and determine the data type in the demand questions;

[0020] S3. Select the appropriate trained RAG model based on the data type in the requirement question;

[0021] S4. Use the selected RAG big model to perform comprehensive processing and reasoning based on the data lake knowledge base to generate problem answers;

[0022] S5. Present the answer to the question to the user as a response to the requirement question.

[0023] Furthermore, the RAG large model includes: a multimodal large model and a large language model.

[0024] Furthermore, the data type of the demand question in S3 includes text, document or picture, and the processing steps include:

[0025] S31. If the data type of the requirement question is text or document, call the large language model to perform comprehensive processing and reasoning to generate a question answer;

[0026] S32. If the data type of the required question is a picture, the multimodal large model is called to perform comprehensive processing and reasoning to generate an answer to the question.

[0027] Furthermore, the S1 includes:

[0028] S11. Collect equipment real-time operation data, Internet equipment knowledge, equipment manuals, equipment maintenance reports, group equipment standards, industry equipment standards, enterprise equipment standards and national equipment standards as equipment-related data;

[0029] S12. De-duplication, screening, compression, formatting, cleaning and standardization of device-related data;

[0030] S13. Convert into structured data, semi-structured data and unstructured data for storage, and establish a vectorized data lake database.

[0031] Furthermore, in S31, the large language model is called to perform comprehensive processing and reasoning, and the steps include:

[0032] S31a, introduce the Embedding model to convert text or documents into vector form;

[0033] S31b, based on the vector form converted from text or document, perform vector query on the data lake database to retrieve relevant knowledge;

[0034] S31c, sort the relevant knowledge using the Re-ranking model;

[0035] S31d. The large language model generates answers to questions based on the sorting and question prompt words.

[0036] Furthermore, when the RAG large model is trained in S4, the African bee optimization algorithm ABOA is used to perform prompt-tuning fine-tuning, LoRA fine-tuning and Prefix-tuning fine-tuning optimization on the RAG large model;

[0037] African bee optimization algorithm ABOA, the formula is expressed as:

[0038] x k (t+1)=x k (t)+ηe v k (t) (1)

[0039]

[0040] x i (t+1)=x i (t)+v i (t+1) (3)

[0041] Among them, v i (t) is the original speed of each individual i in the swarm, v i (t+1) is the update speed, x k (t) is the original position of the queen bee, x i (t+1) is the updated position of the queen bee.

[0042] A third aspect: An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method provided in the second aspect when executing the program.

[0043] A fourth aspect: A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in the second aspect.

[0044] Beneficial effects of the present invention:

[0045] 1. The present invention realizes data intercommunication between equipment knowledge questions and answers and maintenance questions and answers through the data sharing module, thereby improving the accuracy of fault diagnosis and maintenance efficiency; at the same time, the application of multimodal large models and data governance modules enables the system to effectively manage and analyze massive data, providing strong data support for equipment maintenance; through the real-time monitoring function of the fault alarm module, equipment failures can be discovered in time, downtime can be reduced, and production efficiency can be improved.

[0046] 2. The present invention monitors the status of equipment in real time and combines historical data analysis to predict potential failures and provide timely maintenance suggestions, helping technicians to quickly evaluate the pros and cons of different maintenance solutions and achieve optimal data-driven decisions; adopts a large-scale pre-trained model trained with deep learning technology to enable it to understand industry-specific terminology and technical issues and provide accurate answers; integrates multimodal data such as image recognition and text documents, and combines advanced technologies to process various types of data files to achieve comprehensive information extraction and classification functions; uses retrieval enhancement generation technology to search for relevant documents as support for contextual information when answering user questions, thereby improving the quality and credibility of answers; adopts fine-tuning and customized services to fine-tune and optimize the model according to the business needs of specific enterprises to ensure that the system can adapt to diverse application scenarios.

[0047] 3. When governing data in a data lake, the present invention can share data according to the needs of different businesses and users. By building an efficient data sharing mechanism, it breaks down departmental barriers and information islands, maximizes the use of data resources, improves work efficiency, and promotes collaboration and innovation between businesses. At the same time, it constructs knowledge graphs for the collected data. Based on these knowledge graphs, it uses RAG large model technology to learn and train multimodal text, documents, pictures and other data, and implements intelligent applications such as equipment maintenance question-and-answer robots and equipment knowledge question-and-answer robots. These robots can provide users with instant and accurate equipment maintenance guidance and knowledge answer services to help users quickly solve problems and improve work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the structure of the knowledge question answering system of the present invention;

[0049] Figure 2 Schematic diagram of the process of the knowledge question answering method of the present invention;

[0050] Figure 3 It is a principle flow chart of the knowledge question answering method of the present invention;

[0051] Figure 4 This is a schematic diagram of the data structure of the data lake of the present invention;

[0052] Figure 5This is a principle flow chart of the RAG large model of the present invention.

[0053] Figure 6 It is a schematic diagram of the structure of the electronic device of the present invention. DETAILED DESCRIPTION

[0054] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar symbols throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0055] In the operation, maintenance and troubleshooting of existing hydropower station electromechanical equipment, there are problems such as the inability to provide equipment query services and knowledge answers, the inability to provide fault diagnosis services and maintenance suggestions, and the inability to monitor equipment operation failures online.

[0056] In view of the above problems, the present invention provides a knowledge question-answering system for power generation equipment maintenance based on a multi-modal large model. Figure 1 This is a structural diagram of the knowledge question-answering system, including: device question-answering robot module, data lake module and big data model module.

[0057] Among them, the device question and answer robot module receives user questions and feeds back to the data lake module. The data lake module stores a large amount of device-related data. The big data model module is called according to the user questions. The big data model module processes and analyzes the massive device-related data stored in the data lake module according to the user questions, obtains the results of the user questions, and feeds back to the user through the device question and answer robot module.

[0058] The equipment question and answer robot module can include a hydropower equipment knowledge question and answer robot and a hydropower equipment maintenance question and answer robot. Through the hydropower equipment knowledge question and answer robot, the system can provide equipment knowledge query and answers; through the hydropower equipment maintenance question and answer robot, the system can provide equipment fault diagnosis answers and maintenance suggestions.

[0059] The knowledge question and answer system also includes data governance and control modules, equipment operation management modules, equipment fault alarm modules and data sharing modules.

[0060] The equipment operation management module can be used to obtain equipment information in real time, and the data governance and control module can be used to clean, integrate and standardize the collected data, and manage and control the data.

[0061] The equipment fault alarm module can monitor the real-time operation data of the equipment in real time based on the collected real-time operation data of the equipment through the set thresholds and algorithms. Once an abnormality is found, the technical staff will be notified in time through sound and light alarms or message push.

[0062] The data sharing module is used to share data between system modules. The data sharing module includes functional modules such as data interface, data transmission protocol and data processing logic. Among them, the data interface function is used for data interaction between different modules, and the data transmission protocol is used to ensure the security and reliability of data transmission between different modules; the data processing logic is used to process and convert the shared data between different modules.

[0063] In actual applications, the system uses hydropower equipment knowledge question-and-answer robots and hydropower equipment maintenance question-and-answer robots to provide equipment knowledge query and answers. The RAG large model module provides calculation and answer services for the hydropower large language model and the hydropower multimodal large model; the data governance and control module provides data cleaning, management and control functions; the data lake module is used to store data; the equipment operation management module provides the function of obtaining real-time equipment information, and the equipment fault alarm module provides real-time equipment monitoring and alarm functions.

[0064] Based on the above knowledge question answering system, such as Figure 2 and Figure 3 As shown, the present invention also discloses a method for power generation equipment maintenance knowledge question and answer based on a multi-modal large model, comprising the steps of:

[0065] S1. Collect equipment-related data, perform preprocessing, and build a data lake knowledge base.

[0066] Big data technology is used to widely collect real-time operation data, equipment manuals, equipment maintenance reports and various standards (including national standards, industry standards, group standards, enterprise standards, etc.) of various equipment in the hydropower station, as well as diversified information sources such as industry equipment knowledge, Internet search knowledge, and internal documents of group enterprises. This information provides rich materials and basis for subsequent data analysis and RAG large model training and learning applications.

[0067] The massive amount of collected data is pre-processed, including deduplication, screening, compression, formatting, data cleaning and standardization. Through the above processing, special characters, paragraphs, sentences, etc. can be removed from the data for subsequent applications.

[0068] like Figure 4 As shown, the processed data is converted into structured data, semi-structured data and unstructured data for storage, and a vectorized data lake database is established. At the same time, data aggregation technology can also be used to reasonably manage and govern the data stored in the data lake to ensure the accuracy, completeness and consistency of the data, laying a solid foundation for subsequent data analysis and mining.

[0069] S2. Use the knowledge question answering system to receive user demand questions. The system performs semantic analysis and classification on the user demand questions to determine the data type in the demand questions.

[0070] The data types of the demand questions input by the user include text, document or picture. Different RAG large models are used to process different data types.

[0071] S4. Select the appropriate trained RAG model based on the data type in the requirement.

[0072] The RAG big model used in the present invention includes a multimodal big model (e.g., Qwen-VL) and a big language model (e.g., GLM4). When the data type of the user input demand question is text or document, the big language model is called to perform comprehensive processing and reasoning to generate a question answer. When the data type of the user demand question is a picture, the multimodal big model is called to perform comprehensive processing and reasoning to generate a question answer.

[0073] Use the training set and test set to train the RAG large model, and fine-tune the parameters of the RAG large model (such as Qwen-VL, GLM4), such as prompt-tuning fine-tuning, LoRA fine-tuning, and Prefix-tuning fine-tuning of the RAG large model, to optimize the collaborative effect of the retrieval and generation modules and improve the overall accuracy of the RAG large model.

[0074] Specifically, the African bee optimization algorithm (ABOA) can be used. The formula of the African bee optimization algorithm is expressed as:

[0075] x k (t+1)=x k (t)+ηe v k (t) (1)

[0076]

[0077] x i (t+1)=x i (t)+v i (t+1) (3)

[0078] Among them, v i (t) is the original speed of each individual i in the swarm, x k (t) is the original position of the queen bee, v i (t+1) is the update speed, x i (t+1) is the updated position of the queen bee.

[0079] η⊙v k (t) represents the speed of the queen bee at time t, where η is expressed as η = {a1, a2, ..., a m}, a1 = 0.5 (MaxIter-t) / MaxIter, pheromone retention upper limit coefficient am The normal value of is 1, and a2, a3, ..., a m－1 Take a1~a m The equal value is used (to achieve the purpose of reducing the weight of historical pheromones as time goes by), and MaxIter is the maximum number of iterations.

[0080] Then according to the current position x i (t), calculate the threat level (fitness) value f i (x i (t)), and compare the threat level found with the global threat level (global optimal) value f Best (t-1) for comparison, update the global threat level f Best (t), and update the global threat position g Best (t).

[0081] The specific implementation process of the ABOA algorithm includes:

[0082] 1) Initialization parameters: including the maximum number of iterations MaxIter, queen bee lifespan m, and African bee population number N p 、Pheromone retention upper limit coefficient a m wait;

[0083] 2) Population initialization: Randomly initialize the positions of all individuals in the swarm and calculate the threat level f of their positions i (x i (t)) and speed v i (t);

[0084] 3) Check if the queen bee exists. If not, initialize the queen bee position x k is the current highest threat value point (optimal value point), and initializes the speed v k and pheromone vector M; check the queen bee lifespan, and if it reaches m, destroy it;

[0085] 4) The queen bee enters the current global highest threat level point into the pheromone vector M, and obtains the queen bee speed and updates the queen bee position x according to equations (2) and (1). k ;

[0086] 5) The individual bee colony updates its own position according to formula (3), that is, it obtains its new speed and updates its position based on its original speed and the queen bee position.

[0087] 6) Calculate all individual threat values ​​(fitness values) and update the global threat value (global optimal value). Determine whether the maximum number of iterations MaxIter is reached. If not, return to step 3); otherwise, go to step 7);

[0088] 7) The algorithm terminates and outputs the final population state and target point.

[0089] Among them, step 4) operates the pheromone vector M and enters the global optimal value generated in each iteration into it, so as to achieve the retention of multi-band information.

[0090] The ABOA algorithm can be used to perform prompt-tuning, LoRA fine-tuning, and Prefix-tuning on large RAG models (such as Qwen-VL and GLM4), such as fine-tuning of LoRA_R, LoRA_ALPHA, LoRA_DROPOUT in LoRA, and fine-tuning of input sequence length n, prefix length L, embedding dimension d, key-value dimension dk in Prefix-tuning. The Loss indicator of prompt-tuning, LoRA fine-tuning, and Prefix-tuning is used as the fitness function of the ABOA algorithm. Finally, the optimal hyperparameters are used for model inference to generate a static and persistent deployment of the RAG large model.

[0091] S5. Use the selected RAG big model to perform comprehensive processing and reasoning based on the data lake knowledge base to generate answers to the questions.

[0092] like Figure 5 As shown, a large language model is called to perform comprehensive processing and reasoning on text or documents, including:

[0093] The Embedding model is introduced to convert text or documents into vector form. Based on the vector form of the text or document, the data is vectorized through the Embedding model to establish a vectorized database (knowledge) for efficient retrieval and matching. Vector queries can be performed on the data lake database to retrieve relevant knowledge.

[0094] Based on the large language model after training and fine-tuning, the message question and answer function is realized using RAG and Re-ranking technology.

[0095] According to the fault conditions and classifications of power generation equipment involved in the text or document questions, the key factors are determined, and RAG is applied in combination with the vector database to retrieve relevant documents or fragments in the data lake; the relevance of the context is judged based on the user question and the retrieved relevant documents or fragments, and the Re-ranking model is used to re-rank the retrieval results according to the user question text data to screen out more suitable slices; the slices are combined with the user question and prompt words (prompt) and imported into the fine-tuned large language model. The large language model generates answers to the questions based on the ranking and the question prompt words.

[0096] The principle of calling a large multimodal model (for example, Qwen-VL) is the same as that of a large language model, and will not be introduced in detail here.

[0097] Furthermore, the method of the present invention also includes monitoring the collected real-time operating data of the equipment, issuing an alarm when a fault is detected, and monitoring the real-time operating data of the equipment in real time through set thresholds and algorithms. Once an abnormality is found, the technical staff is promptly notified through sound and light alarms or message push.

[0098] S6. Present the answer to the question to the user as a response to the requirement question.

[0099] The method of the present invention adopts a large-scale pre-trained model trained by deep learning technology, so that it can understand industry-specific terms and technical problems and provide accurate answers; by real-time monitoring of equipment status and combining historical data analysis, potential faults are predicted, and timely maintenance suggestions are provided to help technicians quickly evaluate the pros and cons of different maintenance solutions and achieve data-driven optimal decision-making; by combining advanced technologies such as image recognition and text documents to process various types of data files, comprehensive information extraction and classification functions are achieved; through retrieval enhanced generation (RAG) technology: when answering user questions, relevant documents are searched through search engines as support for contextual information to improve the quality and credibility of answers; through fine-tuning and customized services, the model is fine-tuned and optimized according to the business needs of specific enterprises to ensure that the system can adapt to diverse application scenarios.

[0100] The present invention also provides an electronic device, Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 6 As shown, the electronic device may include: a processor, a communications interface, a memory, and a communications bus, wherein the processor, the communications interface, and the memory communicate with each other via the communications bus. The processor may call the logic instructions in the memory, for example, to execute the following method:

[0101] S1. Collect equipment-related data, perform preprocessing, and build a data lake knowledge base;

[0102] S2. Receive user demand questions using the knowledge question answering system, perform semantic analysis and classification on the user demand questions, and determine the data type in the demand questions;

[0103] S3. Select the appropriate trained RAG model based on the data type in the requirement question;

[0104] S4. Use the selected RAG big model to perform comprehensive processing and reasoning based on the data lake knowledge base to generate problem answers;

[0105] S5. Present the answer to the question to the user as a response to the requirement question.

[0106] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0107] An embodiment of the present invention further provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method provided in each of the above embodiments is implemented, for example, including:

[0108] S1. Collect equipment-related data, perform preprocessing, and build a data lake knowledge base;

[0109] S2. Receive user demand questions using the knowledge question answering system, perform semantic analysis and classification on the user demand questions, and determine the data type in the demand questions;

[0110] S3. Select the appropriate trained RAG model based on the data type in the requirement question;

[0111] S4. Use the selected RAG big model to perform comprehensive processing and reasoning based on the data lake knowledge base to generate problem answers;

[0112] S5. Present the answer to the question to the user as a response to the requirement question.

[0113] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0114] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A knowledge question-answering system for power generation equipment maintenance based on a multi-modal large model, characterized in that: include: The device question-answering robot module receives user questions and feeds them back to the data lake module; The data lake module stores a large amount of device-related data and calls the big data model module based on user questions; The big data model module processes and analyzes the massive device-related data stored in the data lake module based on user questions, obtains the results of user questions, and feeds back to the user through the device question-answering robot module.

2. The knowledge question answering system according to claim 1, characterized in that: Also includes: Equipment operation management module, used to collect and obtain equipment-related data; Data governance and control module, used to clean, integrate and standardize collected data, and manage and control it; Equipment fault alarm module monitors the real-time collected equipment-related data and issues an alarm when a fault is detected; Data sharing module, used for data sharing between system modules.

3. A method based on the knowledge question answering system of claim 1 or 2, characterized in that: Includes steps: S1. Collect equipment-related data, perform preprocessing, and build a data lake knowledge base; S2. Receive user demand questions using the knowledge question answering system, perform semantic analysis and classification on the user demand questions, and determine the data type in the demand questions; S3. Select the appropriate trained RAG model based on the data type in the requirement question; S4. Use the selected RAG big model to perform comprehensive processing and reasoning based on the data lake knowledge base to generate problem answers; S5. Present the answer to the question to the user as a response to the requirement question.

4. The method according to claim 3, characterized in that The S1 includes: S11. Collect equipment real-time operation data, Internet equipment knowledge, equipment manuals, equipment maintenance reports, group equipment standards, industry equipment standards, enterprise equipment standards and national equipment standards as equipment-related data; S12. De-duplication, screening, compression, formatting, cleaning and standardization of device-related data; S13. Convert into structured data, semi-structured data and unstructured data for storage, and establish a vectorized data lake database.

5. The method according to claim 3, characterized in that: The S1 also includes: monitoring the collected real-time operation data of the equipment and issuing an alarm when a fault is detected.

6. The method according to claim 3, characterized in that The data type of the demand question in S3 includes text, document or picture, the RAG large model includes a multimodal large model and a large language model, and the processing steps include: S31. If the data type of the requirement question is text or document, call the large language model to perform comprehensive processing and reasoning to generate a question answer; S32. If the data type of the required question is a picture, the multimodal large model is called to perform comprehensive processing and reasoning to generate an answer to the question.

7. The method according to claim 6, characterized in that In S31, the large language model (RAG) is called to perform comprehensive processing and reasoning, and the steps include: S31a, introduce the Embedding model to convert text or documents into vector form; S31b, based on the vector form converted from text or document, perform vector query on the data lake database to retrieve relevant knowledge; S31c, sort the relevant knowledge using the Re-ranking model; S31d. The large language model generates answers to questions based on the sorting and question prompt words.

8. The method according to claim 3, characterized in that When the RAG large model in S4 is trained, the African bee optimization algorithm ABOA is used to perform prompt-tuning fine-tuning, LoRA fine-tuning and Prefix-tuning fine-tuning optimization on the RAG large model; African bee optimization algorithm ABOA, the formula is expressed as: x k (t+1)=x k (t)+ηe v k (t) (1) x i (t+1)=x i (t)+v i (t+1) (3) Among them, v i (t) is the original speed of each individual i in the swarm, v i (t+1) is the update speed, x k (t) is the original position of the queen bee, x i (t+1) is the updated position of the queen bee.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 3 to 8 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 3 to 8 are implemented.

Images