Power generation field knowledge retrieval method and system based on semantic classification

Abstract

The invention relates to the field of power generation and the technical field of artificial intelligence, in particular to a power generation field knowledge retrieval method and system based on semantic classification. The method comprises the steps that input questions of a user are obtained, context sensing is conducted on the input questions through a trained semantic classification model, the input questions are dynamically classified according to a threshold value, and a semantic classification result is obtained; and the big language model selects a corresponding retrieval source from the power generation field knowledge base according to the semantic classification result to perform knowledge retrieval, and generates an answer for the user question in combination with a knowledge retrieval result. The semantic classification model adopts a fine-tuned BERT model, so that the intention and the demand of the user question can be identified more accurately.

Description

Technical Field

[0001] The present invention relates to the fields of power generation and artificial intelligence technology, and specifically relates to a method and system for knowledge retrieval in the power generation field based on semantic classification. Background Art

[0002] At present, large language models (LLMs) combined with retrieval-augmented generation (RAG) technology have shown significant advantages in the field of knowledge Q&A. On the one hand, researchers supplement the model knowledge by real-time retrieving external databases, solving the problem of the lag in updating static knowledge bases; on the other hand, RAG technology can customize retrieval sources for specific industries (such as power generation) to improve the response ability of large language models in vertical fields.

[0003] However, the keyword matching and shallow semantic analysis relied on by RAG technology are difficult to capture complex contexts. Especially when dealing with professional terms in the power generation industry, the text is not fully modeled, affecting the retrieval accuracy; in addition, traditional RAG lacks a fine-grained division of user question types, resulting in redundant or missing retrieval scopes. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for knowledge retrieval in the power generation field based on semantic classification to solve the technical problem of inaccurate retrieval results caused by redundant or missing knowledge retrieval scopes in the current power generation field.

[0005] The object of the present invention is achieved by the following technical solutions: In a first aspect, the present invention provides a method for knowledge retrieval in the power generation field based on semantic classification, including: Obtain the question input by the user, perform context awareness on the input question using a trained semantic classification model, and dynamically classify the input question according to a threshold to obtain a semantic classification result; The large language model selects corresponding retrieval sources from the knowledge base in the power generation field according to the semantic classification result for knowledge retrieval, and generates an answer to the user's question in combination with the knowledge retrieval result; The semantic classification model uses a fine-tuned BERT model.

[0006] As a further improvement of the present invention, the dynamically classifying the input question according to a threshold to obtain a semantic classification result specifically includes: labeling the input question according to a threshold, and the question labels include technical questions, policy questions, and general questions; The threshold is set as: the threshold for technical questions is 0.4 ≤ S ≤ 0.7, the threshold for policy questions is S > 0.7, and the threshold for general questions is S < 0.4, where S is the classification threshold.

[0007] As a further improvement of the present invention, according to the semantic classification results, corresponding retrieval sources are selected from the knowledge base in the power generation field for knowledge retrieval, specifically including: For technical problems, vector retrieval is adopted and matched by the cosine similarity method, and the most relevant technical documents are used as references for the answers of the large language model; For policy problems, the latest regulations are preferentially retrieved in chronological order; For general problems, vectorized retrieval is adopted to improve the relevance of retrieval results.

[0008] As a further improvement of the present invention, the construction of the knowledge base in the power generation field specifically includes: Segment the unstructured documents, extract entities, relationships and corresponding attributes in the data; form entity-relationship-attribute triples, and use the triples as the identification keywords for RAG retrieval; Conduct a correlation analysis of fault cases on the structured data, and establish a mapping relationship library between the parameters of power generation equipment and the corresponding solutions.

[0009] In the second aspect, the present invention provides a power generation field knowledge retrieval system based on semantic classification, including: A semantic classification module that uses a fine-tuned BERT model to perform semantic classification on user questions and outputs question labels of technical, policy or general types; A RAG enhancement module for retrieving relevant content from the knowledge base in the power generation field according to the semantic classification results. The knowledge base in the power generation field is built by integrating Elastic Search and includes structured data indexing and unstructured data entity extraction; A collaborative processing module for dynamically selecting retrieval strategies according to the classified question labels and generating corresponding answers through a generative model in combination with the retrieval results.

[0010] As a further improvement of the present invention, the semantic classification module specifically includes: Construct a BERT vocabulary containing terms in the power generation field, and fine-tune the BERT model parameters through domain data; Adopt a multi-task learning framework to synchronously optimize the question classification and intent recognition tasks; Perform temporal modeling on user questions and adjust the classification threshold. As a further improvement of the present invention, the RAG enhancement module specifically includes: For technical problems, vector retrieval is adopted and matched by the cosine similarity method; For policy problems, the latest regulations are preferentially retrieved in chronological order; For general problems, vectorized retrieval is adopted to improve the semantic relevance of retrieval results.

[0011] As a further improvement of the present invention, the collaborative processing module specifically includes: A context fusion unit for splicing the retrieval result and the user question and then inputting them into a generation model; A multi-round dialogue management component for clarifying complex or ambiguous question intents.

[0012] As a further improvement of the present invention, it further includes a personalized recommendation module and a multi-modal fusion module. The personalized recommendation module specifically includes: a graph neural network analyzes the user's search records to construct an interest association graph; uses a reinforcement learning algorithm to dynamically adjust the recommendation strategy and push relevant knowledge in real time; The multi-modal fusion module is used to extract text, image, and audio features according to a multi-modal pre-training model; establish cross-modal data associations using knowledge graph technology; and comprehensively generate intuitive answers according to multi-modal information.

[0013] In a third aspect, the present invention provides a computer-readable storage medium storing one or more programs. The one or more programs include instructions that, when executed by a computing device, cause the computing device to execute the above-mentioned method for retrieving knowledge in the power generation field based on semantic classification.

[0014] In a fourth aspect, the present invention provides a computing device, including: One or more processors, a memory, and one or more programs, where the one or more programs are stored in the memory and configured to be executed by the one or more processors. The one or more programs include steps for executing the above-mentioned method for retrieving knowledge in the power generation field based on semantic classification.

[0015] The beneficial effects of the present invention are as follows: The method for retrieving knowledge in the power generation field based on semantic classification provided by the present invention can better understand the semantics and context information of user questions by using a fine-tuned BERT model. The system can more accurately identify the intents and needs of user questions. The RAG enhancement processing enables the model to not only rely on pre-trained knowledge but also retrieve relevant information from external knowledge bases. It improves the accuracy and comprehensiveness of knowledge retrieval, making the generated answers more reliable and authoritative. The use of a dynamic classification method can flexibly adjust the classification strategy according to different types and complexities of questions. This method improves the accuracy and adaptability of classification, enabling the system to handle a wider range of question types. Generating answers by combining knowledge retrieval results makes the generated answers more accurate and comprehensive. By retrieving relevant knowledge, the model can provide more specific solutions and more in-depth explanations.

[0016] Furthermore, by setting clear classification thresholds, the system can more accurately identify and classify user input questions, thereby reducing the possibility of misclassification and improving the accuracy of classification.

[0017] Through classification, the system can retrieve information from the corresponding knowledge base more effectively. For example, technical questions can be retrieved from technical documents or professional databases; policy questions can be retrieved from laws, regulations or policy documents; general questions can be retrieved from common sense databases or encyclopedias. This method improves the efficiency and accuracy of knowledge retrieval. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 is a schematic flow chart of a method for retrieving knowledge in the power generation field based on semantic classification according to an embodiment of the present invention; Figure 2 is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] In order to make the purpose and technical solutions of the present invention clearer and easier to understand. The following further details the present invention in conjunction with the drawings and embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0021] The technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings and specific embodiments. Among them, the described embodiments are only some embodiments of the present invention, not all embodiments.

[0022] Embodiment 1 As Figure 1 shown, this embodiment provides a method for retrieving knowledge in the power generation field based on semantic classification. The following are the specific implementation manners.

[0023] First, obtain the input question of the user, perform context awareness on the input question through a trained semantic classification model, and dynamically classify the input question according to a threshold to obtain a semantic classification result.

[0024] Among them, the semantic classification model uses a fine-tuned BERT model The semantic classification model dynamically classifies the input question according to a threshold to obtain a semantic classification result, which specifically includes: labeling the input question according to the threshold, and the question labels include technical questions, policy questions, and general questions; The threshold is set as follows: for technical problems, the threshold is 0.4 ≤ S ≤ 0.7; for policy problems, S > 0.7; for general problems, S < 0.4, where S is the classification threshold.

[0025] Technical problems in this embodiment usually involve professional knowledge, technical details, or specific operation methods. Policy problems usually involve laws, regulations, policy interpretations, or administrative decisions. General problems usually involve daily life, common sense, or general topics.

[0026] Different types of problems correspond to different processing strategies. For example, for technical problems, information needs to be retrieved from technical documents or professional databases; for policy problems, answers need to be obtained from laws, regulations, or policy documents; for general problems, information may be obtained from common sense databases or encyclopedias. Through classification, the system can adopt different retrieval and generation strategies for different types of problems, improving the pertinence and accuracy of the answers.

[0027] The training process of the semantic classification model specifically includes: First, construct a knowledge base for the power generation field; specifically include: segmenting unstructured documents, extracting entities, relationships, and corresponding attributes in the data; forming entity-relationship-attribute triples, and using the triples as the identification keywords for RAG retrieval; performing fault case correlation analysis on structured data to establish a mapping relationship library between power generation equipment parameters and corresponding solutions. For structured data, in this embodiment, Elastic Search indexing storage is used, such as files like equipment parameter tables and ISO standard documents.

[0028] In the power generation field knowledge base, there are 100,000 QA pairs in the power generation field, covering scenarios such as equipment operation and maintenance (such as "treatment of abnormal steam turbine vibration") and policy interpretation (such as "carbon emission quota calculation standard in 2024"), and classification labels and classification threshold S are marked.

[0029] At the same time, 512 power generation industry terms (such as "backpressure steam turbine" and "SCR denitration efficiency") are added to the BERT vocabulary.

[0030] Use the power generation field knowledge base to train the semantic classification model. During the training process, fine-tuning is performed using the RAG enhancement processing method. Specifically, the newly constructed BERT vocabulary is used for vector encoding to achieve parameter fine-tuning of the BERT model.

[0031] Secondly, the large language model selects the corresponding retrieval source from the power generation field knowledge base according to the semantic classification result, and generates an answer to the user's question in combination with the knowledge retrieval result; Adopt different retrieval methods for different categories of knowledge bases. For technical questions, vector retrieval is used, and the most relevant technical documents are used as references for the answers of the large language model by matching through the cosine similarity method; for policy questions, the latest regulations are retrieved first in chronological order; for general questions, vectorized retrieval is used to improve the semantic relevance of retrieval results.

[0032] According to the analysis results of the semantic classification model, select the corresponding RAG retrieval strategy and knowledge base. Through the retrieval of the user's question, obtain the domain knowledge highly relevant to the user's question in the knowledge base, and input the results obtained by the retrieval module as context information into the large language model to reply to the user's question. Through the above method, the semantic classification model in the power generation field is combined with the retrieval enhanced generation of the large language model to provide users with more reasonable and timely answers, thereby improving the user experience. The large language model in this embodiment is the Qwen2.5-32B-Instruct model.

[0033] Embodiment 2 This embodiment provides a power generation field knowledge retrieval system based on semantic classification, which specifically includes: a semantic classification module, a RAG enhancement module, and a collaborative processing module.

[0034] Semantic classification module: Construct a question type classifier based on the BERT model, and classify the questions in the power generation field into three categories, namely: technical category (0.4 ≤ S ≤ 0.7), policy category (S > 0.7), and general category (S < 0.4). Before applying the BERT model to inference, fine-tune the model parameters using power generation field data, and dynamically adjust the classification threshold for each piece of data. For example, questions such as "nuclear waste treatment standards" should be set with a higher confidence level. The semantic classification module includes a semantic classification model, which uses the fine-tuned BERT model.

[0035] RAG enhancement module: Integrate Elastic Search to build a knowledge graph of the power generation industry, and the indexed content includes knowledge in aspects such as equipment manuals, industry standards, and fault case libraries. Through the fine-tuned BERT model, implement a domain adaptive retrieval strategy to obtain more accurate retrieval results.

[0036] Collaborative processing module: Dynamically select the retrieval source according to the semantic classification results, and call the professional knowledge base to which the user's question belongs for knowledge retrieval. Then, use the generative model Qwen2.5-32B-Instruct to answer the user's question in combination with the retrieved content.

[0037] The BERT model is used to deeply understand the semantic features of user questions, and on this basis, knowledge enhancement is carried out through retrieval. Different from simply indexing and matching user questions through retrieval, the BERT model can fully analyze the semantic information of the text, detect the deep meaning and complex context of user questions, and the model converts the extracted semantic features into classification thresholds, so as to retrieve content highly relevant to user questions in the corresponding knowledge base. The BERT model fine-tuned with knowledge in the power generation field accurately classifies user questions, improving the accuracy of knowledge retrieval by 30% compared with traditional RAG, effectively improving the retrieval results of domain knowledge and avoiding the situation of "answering irrelevant questions" by large language models.

[0038] The training process of the semantic classification model specifically includes: Use a large language model to extract knowledge from power generation field documents, construct 100,000 QA pairs, and label and classify the question types. Set the classification threshold (S) of the QA pairs. 0.4 ≤ S ≤ 0.7 belongs to the technical category, S > 0.7 belongs to the policy category, and S < 0.4 belongs to the general category. In addition, add a power generation industry term list to the BERT vocabulary to enhance the word embedding representation of the BERT model.

[0039] Add power generation field data to the general corpus, and use the newly constructed BERT vocabulary for vector encoding to achieve parameter fine-tuning of the BERT model.

[0040] For structured data in the power generation field, use Elastic Search index storage, such as device parameter tables, ISO standard documents and other files; for unstructured data, use a large language model to segment the documents and extract the key entities in the data as the identification keywords for RAG retrieval.

[0041] Adopt different retrieval methods for different categories of knowledge bases. For technical questions, adopt the vector knowledge retrieval method, use cosine similarity calculation, and preferentially match the most relevant technical documents as the reference for the large language model to answer; for policy questions, the timeliness of the retrieved content is particularly important, so sort the retrieval results in reverse chronological order to ensure the citation of the latest regulations. For general questions, also adopt the vector knowledge retrieval method to make the retrieval results more accurate.

[0042] According to the analysis results of the semantic classification model, select the corresponding RAG retrieval strategy and knowledge base. Through the retrieval of the user's question, obtain the domain knowledge highly relevant to the user's question in the knowledge base, and input the results obtained by the retrieval module as context information into the Qwen2.5-32B-Instruct model to reply to the user's question. Through the above method, combine the semantic classification model in the power generation field with the retrieval-enhanced generation of the large language model to provide users with more reasonable and timely answers, thereby improving the user experience.

[0043] In addition, the system in this embodiment further includes a personalized recommendation module and a multimodal fusion module. The personalized recommendation module specifically includes: a graph neural network analyzes the user's search records to construct an interest association graph; uses a reinforcement learning algorithm to dynamically adjust the recommendation strategy and push relevant knowledge in real time.

[0044] The multimodal fusion module is used to extract text, image, and audio features according to the multimodal pre-training model; establish cross-modal data associations using knowledge graph technology; and comprehensively generate intuitive answers according to multimodal information.

[0045] Among them, the multimodal fusion module adopts a multimodal pre-training model, and the input layer of this model is text token embedding + image region feature + audio segment embedding. Cross-modal attention mainly adopts the visualizability of the text→image cross-attention weight interpretability. The output layer supports Markdown format typesetting and automatically inserts associated picture / audio links. For example, when the user uploads a recording and the text description "periodic abnormal noise in the bearing part", after this system, the system responds as follows: the audio feature matches the typical frequency spectrum of "lack of oil and wear of rolling bearings" (similarity 0.91), the associated knowledge graph returns the maintenance plan document and 3D disassembly animation, and generates an answer containing a playable comparison segment of the same type of fault audio.

[0046] This embodiment is also described in detail with a specific example. The specific steps are as follows: First, the user inputs "How to reduce NOx emissions from coal-fired units?". The semantic classification module identifies it as a technical question (i.e., the classification value S = 0.65). Finally, the collaborative processing module synthesizes the retrieval results and outputs a step-by-step solution and a policy compliance description.

[0047] Embodiment 3 In an embodiment of the present invention, a computer-readable storage medium is provided. This medium belongs to the memory device of a terminal device and is mainly used to store programs and data. The computer-readable storage medium includes both the storage medium built into the terminal and the extended storage medium supported by the terminal. Specifically, any tangible medium that can store a program and be used by an instruction execution system, device, or component belongs to this category. This storage medium provides storage space for storing the terminal operating system and instructions (including one or more computer programs and their codes) that can be loaded and executed by a processor. Examples include electrically connected devices, portable disks, hard disks, RAM, ROM, EPROM / flash memory, optical fibers, CD-ROMs, optical storage devices, magnetic storage devices, etc., and combinations thereof.

[0048] In addition, the computer-readable storage medium also relates to data signals propagated in a baseband or as a carrier wave. These signals carry readable program codes and can be in the form of electromagnetic signals, optical signals, etc. The readable storage medium is not limited to the above types and also includes other media that can send, propagate, or transmit a program for use by an instruction execution system, device, or component. The program code can be transmitted wirelessly, wired, via optical cable, RF, etc.

[0049] The program code can be written in various programming languages, such as object-oriented languages (Python, Java, C++, etc.) and procedural languages (C language, etc.). The code can be executed fully or partially on a user device, or can be used as an independent software package, or be executed partially / fully on a remote device. The remote device is connected to the user device via a LAN, WAN, or an Internet service provider.

[0050] One or more instructions stored in the computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the method for retrieving knowledge in the power generation field based on semantic classification in the above embodiment; Obtain the input question of the user, perform context awareness on the input question through a trained semantic classification model, and dynamically classify the input question according to a threshold to obtain a semantic classification result; The large language model selects the corresponding retrieval source from the knowledge base in the power generation field according to the semantic classification result for knowledge retrieval, and generates an answer to the user's question in combination with the knowledge retrieval result; The semantic classification model uses a fine-tuned BERT model.

[0051] Embodiment 4 Figure 2 It is a block diagram of an electronic device provided by the present invention according to an embodiment.

[0052] Please refer to Figure 2, the terminal device 600 is an electronic device, which is presented in the form of a general computing device. The components of the electronic device may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including the storage unit 620 and the processing unit 610), a display unit 640, etc.

[0053] Among them, the storage unit stores program code, which can be executed by the processing unit 610, so that the processing unit 610 executes the steps according to various exemplary embodiments of the present invention described in the above method part of this specification. For example, the processing unit 610 can execute the steps as shown in Figure 1 shown.

[0054] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 6201 and / or a cache storage unit 6202, and may further include a read-only storage unit (ROM) 6203.

[0055] The storage unit 620 may also include a program / utility 6204 having a set (at least one) of program modules 6205. Such program modules 6205 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include the implementation of a network environment.

[0056] The bus 630 may represent one or more of several types of bus structures, including a storage unit bus or a storage unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the multiple bus structures.

[0057] The electronic device 600 can also communicate with one or more external devices 700 (such as a keyboard, a pointing device, a Bluetooth device, etc.), and can also communicate with one or more devices that enable a user to interact with the electronic device 600, and / or communicate with any device that enables the electronic device 600 to communicate with one or more other computing devices (such as a router, a modem, etc.). Such communication can be carried out through an input / output (I / O) interface 650. And, the electronic device 600 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 660. The network adapter 660 can communicate with other modules of the electronic device 600 through the bus 630. It should be understood that although not shown in the figure, other hardware and / or software modules can be used in combination with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms, etc.

Claims

1. A knowledge retrieval method in the power generation field based on semantic classification, characterized in that, It includes: Obtain the question input by the user, perform context awareness on the input question through a trained semantic classification model, and dynamically classify the input question according to a threshold to obtain a semantic classification result; The large language model selects the corresponding retrieval source from the power generation domain knowledge base according to the semantic classification result for knowledge retrieval, and generates an answer to the corresponding question in combination with the knowledge retrieval result; The semantic classification model uses a fine-tuned BERT model.

2. The method for retrieving power generation field knowledge based on semantic classification according to claim 1, wherein The dynamically classifying the input question according to the threshold to obtain a semantic classification result specifically includes: labeling the input question according to the threshold, and the question labels include technical questions, policy questions, and general questions; The threshold is set as: the threshold for technical questions is 0.4 ≤ S ≤ 0.7, the policy question is S > 0.7, and the general question is S < 0.4, where S is the classification threshold.

3. The method for retrieving power generation field knowledge based on semantic classification according to claim 2, characterized in that Selecting the corresponding retrieval source from the power generation domain knowledge base for knowledge retrieval according to the semantic classification result specifically includes: For technical questions, vector retrieval is used and matched by the cosine similarity method, and the most relevant technical documents are used as references for the large language model to answer; For policy questions, the latest regulations are preferentially retrieved in chronological order; For general questions, vectorized retrieval is used to improve the semantic relevance of the retrieval results.

4. The method for retrieving power generation field knowledge based on semantic classification according to claim 3, wherein The construction of the power generation domain knowledge base specifically includes: Segment the unstructured documents, extract entities, relationships, and corresponding attributes in the data; form entity-relationship-attribute triples, and use the triples as the identification keywords for RAG retrieval; Conduct a fault case correlation analysis on the structured data, and establish a mapping relationship library between power generation equipment parameters and corresponding solutions.

5. A knowledge retrieval system for the power generation field based on semantic classification, characterized in that, It includes: A semantic classification module, which is used to perform semantic classification on the user's question based on the fine-tuned BERT model and output question labels of technical, policy, or general types; A RAG enhancement module, which is used to retrieve relevant content from the power generation domain knowledge base according to the semantic classification result. The power generation domain knowledge base is built by integrating ElasticSearch and includes structured data indexing and unstructured data entity extraction; A collaborative processing module, which is used to dynamically select a retrieval strategy according to the classified question labels and generate corresponding answers through a generative model in combination with the retrieval results.

6. The knowledge retrieval system for the power generation field based on semantic classification according to claim 5, characterized in that The semantic classification module specifically includes: Construct a BERT vocabulary containing power generation domain terms, and fine-tune the BERT model parameters through domain data; Adopt a multi-task learning framework to synchronously optimize the question classification and intent recognition tasks; Perform temporal modeling on the user's question and adjust the classification threshold.

7. The knowledge retrieval system for the power generation field based on semantic classification according to claim 5, characterized in that, In the RAG enhancement module, it specifically includes: For technical questions, vector retrieval is used and matched by the cosine similarity method; For policy questions, the latest regulations are preferentially retrieved in chronological order; For general questions, vectorized retrieval is used to improve the semantic relevance of the retrieval results.

8. The knowledge retrieval system for the power generation field based on semantic classification according to claim 5, characterized in that, The collaborative processing module specifically includes: A context fusion unit, which is used to splice the retrieval result and the user's question and input them into the generation model; A multi-round dialogue management component, which is used to clarify complex or ambiguous question intents.

9. The knowledge retrieval system for the power generation field based on semantic classification according to claim 5, wherein It further includes a personalized recommendation module and a multimodal fusion module. The personalized recommendation module specifically includes: a graph neural network analyzes user search records to construct an interest association graph; a reinforcement learning algorithm is used to dynamically adjust the recommendation strategy and push relevant knowledge in real time. The multimodal fusion module is used to extract text, image, and audio features according to a multimodal pre-training model; establish cross-modal data associations using knowledge graph technology; and comprehensively generate intuitive answers based on multimodal information.

10. A computing device, characterized in that, It includes: One or more processors, a memory, and one or more programs, where the one or more programs are stored in the memory and configured to be executed by the one or more processors. The one or more programs include steps for executing the method for retrieving knowledge in the power generation field based on semantic classification according to any one of claims 1 to 4.

Images