A method, system and medium for constructing equipment operation and maintenance knowledge graph
Through multi-modal large model and a small sample prompt algorithm, the strategy network is optimized, and the problem of insufficient data extraction capabilities in the equipment operation and maintenance knowledge graph is solved, efficient and accurate information extraction is achieved, and the construction quality of the equipment maintenance and maintenance graph is improved.
Patent Information
- Application Number
- CN202411068774.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-06
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-08-06
AI Technical Summary
In the construction of equipment operation and maintenance knowledge graphs, the data extraction capability is insufficient and the documents cannot be effectively analyzed, resulting in the omission or misreading of key equipment information. Multiple models are required for different information extraction tasks, and the application flexibility is poor.
Using a multimodal large model to convert non-text format data into text data, obtain short text fragments with semantic integrity through data cleaning and segmentation, optimize the strategy network using a small sample prompt algorithm and reinforcement learning algorithm, and generate trained multimodal Large model to realize the construction of equipment operation and maintenance knowledge graph.
It improves the accuracy of equipment maintenance document information identification, enhances the ability to extract structured information, ensures the accuracy and completeness of the map, reduces dependence on the quality and diversity of training data, and improves the efficiency and accuracy of information extraction.
Smart Images

Figure CN118586483B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of equipment operation and maintenance knowledge graphs, and in particular to a method, system and medium for constructing an equipment operation and maintenance knowledge graph. Background Art
[0002] Enterprise equipment maintenance information is often scattered in different databases, electronic files and paper documents, resulting in serious information islands and difficulty in achieving effective data integration and knowledge sharing. With the expansion of enterprise scale and the increase in equipment complexity, the demand for knowledge management of equipment operation and maintenance is growing, requiring systematic and structured management and application of equipment operation and maintenance knowledge.
[0003] The construction of equipment operation and maintenance knowledge graph is an important means to respond to the needs of intelligent equipment operation and maintenance, solve the problem of information islands, improve operation and maintenance efficiency and reduce operation and maintenance costs. By constructing equipment operation and maintenance knowledge graph, centralized management and correlation analysis of enterprise operation and maintenance knowledge can be achieved. On the one hand, it can help operation and maintenance personnel quickly acquire equipment-related knowledge and improve the efficiency and accuracy of operation and maintenance work; on the other hand, it can assist in the realization of intelligent fault diagnosis of equipment, quickly locate problems and causes through the correlation relationships in the graph, thereby reducing the overall operation and maintenance costs.
[0004] The main workload of the equipment operation and maintenance knowledge graph construction process lies in data processing and the extraction of triple information such as objects, relationships, and events. Traditional data processing and information extraction are generally completed with the help of pre-trained models such as OCR (optical character recognition) and Bert. First of all, for different styles of equipment maintenance documents, OCR technology needs to be tuned for specific formats, but the data extraction capability is insufficient and the documents cannot be effectively analyzed, which may lead to the omission or misreading of key equipment information; the performance of pre-trained models such as Bert also depends to a large extent on the quality and diversity of training data. For different information extraction tasks such as objects, relationships, and events, it is usually necessary to train multiple models, and the application flexibility is poor. Summary of the invention
[0005] In response to the above-mentioned deficiencies in the prior art, the present application provides a method, system and medium for constructing an equipment operation and maintenance knowledge graph to solve the problems that the existing solutions are insufficient in data extraction capabilities, cannot effectively analyze documents, may lead to omissions or misinterpretations of key equipment information, rely on the quality and diversity of training data, usually require training multiple models for different information extraction tasks such as objects, relationships and events, and have poor application flexibility.
[0006] In a first aspect, the present application provides a method for constructing an equipment operation and maintenance knowledge graph, the method comprising: obtaining an operation and maintenance knowledge graph schema database; wherein the operation and maintenance knowledge graph schema database is composed of a number of triple information, and any triple information includes an object, a relationship between objects, and a trigger event corresponding to the object; obtaining multi-source data to be processed, and converting the non-text formatted data to be processed into text data through a multimodal large model to obtain multi-source text data to be processed; performing data cleaning and data segmentation on the multi-source text data to be processed through the multimodal large model to obtain a number of short text fragments with semantic integrity; extracting a preset number of short text fragments from the several short text fragments as training data, and using the remaining parts as Test data; send the training data to the preset annotation terminal to obtain the correspondence between the returned training data and the triple information; use the training data + triple information as a set of final training data; use the final training data as the input data of the policy network in the multimodal large model to obtain the trained policy network, and then obtain the trained multimodal large model; obtain the text extraction prompt words as the prompt corpus of the few-sample prompt algorithm; use the few-sample prompt algorithm and the trained multimodal large model to obtain the output triple information corresponding to the test data; based on the multi-source text data to be processed and the operation and maintenance knowledge graph schema database, determine the final output triple information from the output triple information, and then combine to form the equipment operation and maintenance knowledge graph.
[0007] Furthermore, multi-source data to be processed is obtained, and the data to be processed in non-text format is converted into text data through a multimodal large model to obtain multi-source text data to be processed, specifically including: obtaining multi-source data to be processed corresponding to equipment maintenance; wherein the multi-source data to be processed includes at least PDF documents, pictures, and Word documents; using the Qwen-VL multimodal large model to extract text data from PDF documents and pictures; and combining the text data and the Word document together as multi-source text data to be processed.
[0008] Furthermore, the multi-source text data to be processed is cleaned and segmented through the multi-modal big model to obtain a number of short text fragments with semantic integrity, specifically including: using the Qwen-VL multi-modal big model to perform data preprocessing operations on the multi-source text data to be processed to remove preset irrelevant characters and preset format errors; using the Qwen-VL multi-modal big model to identify preset logical breakpoints in the multi-source text data to be processed after preprocessing to perform text segmentation to obtain multiple short text fragments with semantic integrity.
[0009] Furthermore, before using the final training data as input data of the policy network in the multimodal large model to obtain the trained policy network, the method also includes: obtaining the current weight matrix of the multimodal large model, decomposing the current weight matrix into a first sub-matrix and a second sub-matrix to perform rank reduction processing on the multimodal large model; wherein, .
[0010] Furthermore, the final training data is used as input data of the policy network in the multimodal large model to obtain a trained policy network, specifically including: using the final training data as input data of the policy network in the multimodal large model, obtaining the accuracy of the output data, and then using the accuracy as input data of the value network in the multimodal large model, using the output data of the value network as the reward signal of the policy network, and then using the PPO reinforcement learning algorithm to optimize the policy network, and through continuous iteration, a trained policy network is obtained.
[0011] Furthermore, based on the multi-source text data to be processed and the operation and maintenance knowledge graph schema database, the final output triple information is determined from the output triple information, specifically including: checking whether the output triple information exists in the multi-source text data to be processed and the operation and maintenance knowledge graph schema database through a text matching algorithm; taking the triple information whose verification result is that it exists in the multi-source text data to be processed and the operation and maintenance knowledge graph schema database as the final output triple information; after determining the final output triple information from the output triple information based on the multi-source text data to be processed and the operation and maintenance knowledge graph schema database, the method also includes: obtaining a sampling result of whether the output triple information exists in the multi-source text data to be processed through a preset sampling interface; obtaining a sampling result of whether the output triple information exists in the operation and maintenance knowledge graph schema database through a preset sampling interface; when the sampling result is that it does not exist in the multi-source text data to be processed or the operation and maintenance knowledge graph schema database, deleting the corresponding triple information from the final output triple information.
[0012] In the second aspect, the present application provides a system for constructing an equipment operation and maintenance knowledge graph, and the system includes: an acquisition module, which is used to acquire an operation and maintenance knowledge graph schema database; wherein the operation and maintenance knowledge graph schema database is composed of a number of triple information, and any triple information includes an object, a relationship between objects, and a trigger event corresponding to the object; an acquisition module, which is used to acquire multi-source data to be processed, and convert the non-text formatted data to be processed into text data through a multimodal large model to obtain multi-source text data to be processed; the multi-source text data to be processed is cleaned and segmented through the multimodal large model to obtain a number of short text fragments with semantic integrity; a data acquisition module, which is used to extract a preset number of short text fragments from a number of short text fragments as training data, and the remaining parts as test data; sending the training data to the preset annotation terminal to obtain the correspondence between the returned training data and the triple information; taking the training data + triple information as a set of final training data; a model acquisition module, used to use the final training data as the input data of the policy network in the multimodal large model, to obtain the trained policy network, and then to obtain the trained multimodal large model; a graph acquisition module, used to obtain text extraction prompt words as the prompt corpus of the few-sample prompt algorithm; using the few-sample prompt algorithm and the trained multimodal large model to obtain the output triple information corresponding to the test data; based on multi-source text data to be processed and the operation and maintenance knowledge graph schema database, determining the final output triple information from the output triple information, and then combining to form the equipment operation and maintenance knowledge graph.
[0013] Furthermore, the acquisition module includes a text processing unit, which is used to use the Qwen-VL multimodal large model to perform data preprocessing operations on multi-source text data to be processed to remove preset irrelevant characters and preset format errors; use the Qwen-VL multimodal large model to identify preset logical breakpoints in the preprocessed multi-source text data to perform text segmentation, and obtain multiple short text fragments with semantic integrity.
[0014] Furthermore, the model acquisition module includes a network training unit, which is used to use the final training data as input data of the policy network in the multimodal large model, obtain the accuracy of the output data, and then use the accuracy as input data of the value network in the multimodal large model, and use the output data of the value network as a reward signal of the policy network, and then use the PPO reinforcement learning algorithm to optimize the policy network, and obtain a trained policy network through continuous iteration.
[0015] In a third aspect, the present application provides a non-volatile computer storage medium on which computer instructions are stored. When the computer instructions are executed, they implement a method for constructing a device operation and maintenance knowledge graph as described in any of the above items.
[0016] Those skilled in the art can understand that the present application has at least the following beneficial effects:
[0017] In order to improve the accuracy of information recognition in equipment maintenance paper documents, this application uses a multimodal large model algorithm to replace traditional OCR recognition technology, which can automatically identify and extract text information in equipment operation and maintenance paper documents of different formats.
[0018] In order to enhance the ability of the big model to extract structured information in equipment maintenance scenarios: this application uses a small amount (preset amount) of manually annotated final training data to tune the basic multimodal big model, so that the multimodal big model can better adapt to the extraction of structured information such as objects (objects (or entities)), relationships and events in equipment maintenance scenarios. Combined with the big model's own text understanding and autonomous thinking ability, it can automatically extract and verify structured information in short text fragment corpus, and generate extraction results that conform to the equipment maintenance graph schema, thereby improving the efficiency of information extraction while ensuring the accuracy and completeness of the graph.
[0019] In addition, the present application obtains the final output triplet information by adding at least a sample prompt algorithm to a series of task prompt words (text extraction prompt words), thereby prompting the trained multimodal large model, thereby improving the efficiency and accuracy of obtaining triplet information from test data; it solves the problem that the existing solution has insufficient data extraction capabilities and cannot effectively analyze documents, which may lead to omissions or misinterpretation of key equipment information.
[0020] In addition, the present application uses a few-sample prompt algorithm to extract data corresponding to text extraction prompt words. Since the text extraction prompt words can preset specific content and types, the accuracy and type of data extraction can be improved, which solves the problem that the existing solutions are too dependent on the quality and diversity of training data, and usually require training multiple models for different information extraction tasks such as objects, relationships and events, resulting in poor application flexibility. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Some embodiments of the present disclosure are described below with reference to the accompanying drawings, in which:
[0022] Figure 1 It is a flow chart of a method for constructing an equipment operation and maintenance knowledge graph provided in an embodiment of the present application.
[0023] Figure 2 It is a schematic diagram of the internal structure of a system for constructing an equipment operation and maintenance knowledge graph provided in an embodiment of the present application. DETAILED DESCRIPTION
[0024] It should be understood by those skilled in the art that the embodiments described below are only preferred embodiments of the present disclosure, and do not mean that the present disclosure can only be implemented through the preferred embodiments. The preferred embodiments are only used to explain the technical principles of the present disclosure, and are not used to limit the protection scope of the present disclosure. Based on the preferred embodiments provided by the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work should still fall within the protection scope of the present disclosure.
[0025] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0026] The technical solution proposed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.
[0027] The present application embodiment provides a method for constructing a device operation and maintenance knowledge graph, such as Figure 1 As shown, the method provided in the embodiment of the present application mainly includes the following steps:
[0028] Step 110: Obtain the operation and maintenance knowledge graph schema database.
[0029] It should be noted that the operation and maintenance knowledge graph schema database is composed of several triple information, and the acquisition process can be obtained through a preset interface. And the schema database contains all triple information. Any triple information includes objects, relationships between objects, and trigger events corresponding to objects. As an example, in the equipment operation and maintenance knowledge graph, objects (entities) are the basic units in the knowledge graph, representing objects with unique identifiers, such as equipment, components, states, measures, locations, personnel, etc.; relationships are the links between objects, used to express the association and interaction between objects, such as the manufacturer of the equipment, the person in charge of the workshop, the phenomenon of the fault, the repair measures for the fault, etc.; trigger events refer to a type of things that occur at a certain time and environment, involving several roles and showing several action characteristics, and usually have their own unique trigger words (used for event identification and extraction) and event elements, including equipment failure events, equipment maintenance events, equipment maintenance events, equipment replacement events, etc. For example, equipment failure events mainly include the time, location, type of failure, scope of impact, etc. of the equipment failure, and trigger words may include "failure", "failure", etc.
[0030] Step 120: Obtain multi-source data to be processed, convert the data to be processed in non-text format into text data through a multimodal large model, and obtain multi-source text data to be processed.
[0031] This step can be specifically as follows: obtaining multi-source data to be processed corresponding to equipment maintenance; wherein the multi-source data to be processed includes at least PDF documents, pictures, and Word documents; using the Qwen-VL multimodal large model to extract text data from PDF documents and pictures; and combining the text data and Word documents together as multi-source text data to be processed.
[0032] Step 130: Use a multimodal large model to perform data cleaning and data segmentation on the multi-source text data to be processed, and obtain a number of short text segments with semantic integrity.
[0033] This step can be specifically as follows: using the Qwen-VL multimodal large model to perform data preprocessing operations on the multi-source text data to be processed to remove preset irrelevant characters and preset format errors; using the Qwen-VL multimodal large model to identify preset logical breakpoints in the preprocessed multi-source text data to be processed to perform text segmentation, and obtain multiple short text fragments with semantic integrity.
[0034] Step 140: extract a preset number of short text segments from a number of short text segments as training data, and use the remaining part as test data; send the training data to a preset annotation terminal, and then obtain the corresponding relationship between the returned training data and the triple information; use the training data + triple information as a set of final training data.
[0035] Step 150: Use the final training data as input data of the policy network in the multimodal large model to obtain a trained policy network, and then obtain a trained multimodal large model.
[0036] This step can be specifically as follows:
[0037] The final training data is used as the input data of the policy network in the multimodal large model to obtain the accuracy of the output data. The accuracy is then used as the input data of the value network in the multimodal large model. The output data of the value network is used as the reward signal of the policy network. The PPO reinforcement learning algorithm is then used to optimize the policy network. Through continuous iteration, a trained policy network is obtained.
[0038] In addition, in order to further improve the model's understanding and generalization capabilities.
[0039] The present application obtains the current weight matrix of the multimodal large model and decomposes the current weight matrix into a first sub-matrix and a second sub-matrix to perform rank reduction processing on the multimodal large model.
[0040] in, .
[0041] Step 160: Obtain text extraction prompt words as prompt corpus for the few-sample prompt algorithm; use the few-sample prompt algorithm and the trained multimodal large model to obtain output triplet information corresponding to the test data.
[0042] It should be noted that the text extraction prompt words can be obtained through a preset backend interface.
[0043] Using the few-sample prompting algorithm and the trained multimodal large model, the output triplet information corresponding to the test data is obtained, specifically:
[0044] The test data is input into the few-sample prompt algorithm, so that the few-sample prompt algorithm extracts instance data similar to the text extraction prompt word as the prompt data in the test data; the test data with the prompt data is then input into the multimodal large model to obtain the output triplet information corresponding to the test data.
[0045] Step 170: Based on the multi-source text data to be processed and the operation and maintenance knowledge graph schema database, the final output triple information is determined from the output triple information, and then combined to form the equipment operation and maintenance knowledge graph.
[0046] It should be noted that the scheme of combining into a graph can be implemented by existing schemes, and this application does not limit this.
[0047] This step can be specifically as follows:
[0048] A text matching algorithm is used to verify whether the output triple information exists in the multi-source text data to be processed and the operation and maintenance knowledge graph schema database; the triple information whose verification result is that it exists in the multi-source text data to be processed and the operation and maintenance knowledge graph schema database is taken as the final output triple information.
[0049] After determining the final output triple information from the output triple information based on the multi-source text data to be processed and the operation and maintenance knowledge graph schema database, the method may further include:
[0050] The sampling result of whether the output triple information exists in the multi-source text data to be processed is obtained through the preset sampling interface; the sampling result of whether the output triple information exists in the operation and maintenance knowledge graph schema database is obtained through the preset sampling interface; when the sampling result is that it does not exist in the multi-source text data to be processed or the operation and maintenance knowledge graph schema database, the corresponding triple information will be deleted from the final output triple information.
[0051] besides, Figure 2 A system for constructing a device operation and maintenance knowledge graph is provided in an embodiment of the present application. Figure 2 As shown, the system provided in the embodiment of the present application mainly includes:
[0052] The acquisition module 210 is used to obtain the operation and maintenance knowledge graph schema database.
[0053] Among them, the operation and maintenance knowledge graph schema database is composed of several triple information, and any triple information includes an object, the relationship between objects, and the trigger event corresponding to the object.
[0054] The acquisition module 220 is used to obtain multi-source data to be processed, and convert the non-text format data to be processed into text data through a multimodal large model to obtain multi-source text data to be processed; the multi-source text data to be processed is cleaned and segmented through the multimodal large model to obtain a number of short text fragments with semantic integrity.
[0055] The acquisition module 220 includes a text processing unit, which is used to use the Qwen-VL multimodal large model to perform data preprocessing operations on multi-source text data to be processed to remove preset irrelevant characters and preset format errors; use the Qwen-VL multimodal large model to identify preset logical breakpoints in the preprocessed multi-source text data to perform text segmentation, and obtain multiple short text fragments with semantic integrity.
[0056] The data acquisition module 230 is used to extract a preset number of short text segments from a number of short text segments as training data, and use the remaining part as test data; send the training data to a preset annotation terminal, and then obtain the corresponding relationship between the returned training data and the triple information; and use the training data + triple information as a set of final training data.
[0057] The model acquisition module 240 is used to use the final training data as input data of the policy network in the multimodal large model to obtain the trained policy network, and then obtain the trained multimodal large model.
[0058] The model acquisition module 240 includes a network training unit, which is used to use the final training data as the input data of the policy network in the multimodal large model, obtain the accuracy of the output data, and then use the accuracy as the input data of the value network in the multimodal large model, and use the output data of the value network as the reward signal of the policy network, and then use the PPO reinforcement learning algorithm to optimize the policy network, and obtain a trained policy network through continuous iteration.
[0059] The graph acquisition module 250 is used to obtain text extraction prompt words as prompt corpus for the few-sample prompt algorithm; use the few-sample prompt algorithm and the trained multimodal large model to obtain the output triple information corresponding to the test data; based on the multi-source text data to be processed and the operation and maintenance knowledge graph schema database, determine the final output triple information from the output triple information, and then combine them to form an equipment operation and maintenance knowledge graph.
[0060] In addition, an embodiment of the present application also provides a non-volatile computer storage medium on which executable instructions are stored. When the executable instructions are executed, a method for constructing a device operation and maintenance knowledge graph as described above is implemented.
[0061] So far, the technical solutions of the present disclosure have been described in combination with the above multiple embodiments, but it is easy for those skilled in the art to understand that the protection scope of the present disclosure is not limited to these specific embodiments. Without departing from the technical principles of the present disclosure, those skilled in the art can split and combine the technical solutions in the above-mentioned various embodiments, and can also make equivalent changes or replacements to the relevant technical features. Any changes, equivalent replacements, improvements, etc. made within the technical concept and / or technical principle of the present disclosure will fall within the protection scope of the present disclosure.
Claims
1. A method for constructing a knowledge graph for equipment operation and maintenance. It is characterized in that The method comprises: Obtaining an operation and maintenance knowledge graph schema database; wherein the operation and maintenance knowledge graph schema database is composed of a number of triple information, and any triple information includes an object, a relationship between objects, and a trigger event corresponding to the object; Obtain multi-source data to be processed, and convert the data to be processed in non-text format into text data through a multimodal large model to obtain multi-source text data to be processed; Through the multimodal large model, data cleaning and data segmentation are performed on multi-source text data to be processed, and several short text fragments with semantic integrity are obtained; Extract a preset number of short text segments from a number of short text segments as training data, and use the remaining part as test data; send the training data to a preset annotation terminal, and then obtain the corresponding relationship between the returned training data and the triple information; use the training data + triple information as a set of final training data; The final training data is used as input data of the policy network in the multimodal large model to obtain a trained policy network, and then obtain a trained multimodal large model; Obtain text extraction prompt words as prompt corpus for the few-sample prompt algorithm; use the few-sample prompt algorithm and the trained multimodal large model to obtain the output triple information corresponding to the test data; Based on multi-source text data to be processed and the operation and maintenance knowledge graph schema database, the final output triple information is determined from the output triple information, and then combined to form the equipment operation and maintenance knowledge graph.
2. The method for constructing the equipment operation and maintenance knowledge graph according to claim 1, It is characterized in that Obtain multi-source data to be processed, convert the non-text format data to be processed into text data through a multimodal large model, and obtain multi-source text data to be processed, specifically including: Obtain multi-source data to be processed corresponding to equipment maintenance; wherein the multi-source data to be processed includes at least PDF documents, pictures, and Word documents; Use Qwen-VL multimodal large model to extract text data from PDF documents and images; The text data and the Word document are combined together as multi-source text data to be processed.
3. The method for constructing the equipment operation and maintenance knowledge graph according to claim 1, It is characterized in that Through the multimodal large model, the multi-source text data to be processed is cleaned and segmented to obtain several short text fragments with semantic integrity, including: The Qwen-VL multimodal large model is used to perform data preprocessing operations on multi-source text data to be processed, removing preset irrelevant characters and preset format errors; The Qwen-VL multimodal large model is used to identify the preset logical breakpoints in the multi-source text data to be processed after preprocessing to perform text segmentation and obtain multiple short text fragments with semantic integrity.
4. The method for constructing the equipment operation and maintenance knowledge graph according to claim 1, It is characterized in that Before using the final training data as input data of the policy network in the multimodal large model to obtain a trained policy network, the method further includes: Obtain the current weight matrix of the multimodal large model, and decompose the current weight matrix into a first sub-matrix and a second sub-matrix to perform rank reduction processing on the multimodal large model; wherein, .
5. The method for constructing the equipment operation and maintenance knowledge graph according to claim 1, It is characterized in that The final training data is used as the input data of the policy network in the multimodal large model to obtain the trained policy network, which includes: The final training data is used as the input data of the policy network in the multimodal large model to obtain the accuracy of the output data. The accuracy is then used as the input data of the value network in the multimodal large model. The output data of the value network is used as the reward signal of the policy network. The PPO reinforcement learning algorithm is then used to optimize the policy network. Through continuous iteration, a trained policy network is obtained.
6. The method for constructing the equipment operation and maintenance knowledge graph according to claim 1, It is characterized in that Based on the multi-source text data to be processed and the operation and maintenance knowledge graph schema database, the final output triple information is determined from the output triple information, including: Verify the output triple information through text matching algorithm to see if it exists in the multi-source text data to be processed and the operation and maintenance knowledge graph schema database; The triple information whose verification result is present in the multi-source text data to be processed and the operation and maintenance knowledge graph schema database is used as the final output triple information; After determining the final output triple information from the output triple information based on the multi-source text data to be processed and the operation and maintenance knowledge graph schema database, the method further includes: Obtain the sampling results of whether the output triple information exists in the multi-source text data to be processed through the preset sampling interface; obtain the sampling results of whether the output triple information exists in the operation and maintenance knowledge graph schema database through the preset sampling interface; When the sampling result shows that the triple information does not exist in the multi-source text data to be processed or the operation and maintenance knowledge graph schema database, the corresponding triple information will be deleted from the final output triple information.
7. A system for building a knowledge graph for equipment operation and maintenance, It is characterized in that The system comprises: An acquisition module is used to acquire an operation and maintenance knowledge graph schema database; wherein the operation and maintenance knowledge graph schema database is composed of a number of triple information, and any triple information includes an object, a relationship between objects, and a trigger event corresponding to the object; The acquisition module is used to obtain multi-source data to be processed, and convert the non-text format data to be processed into text data through the multi-modal large model to obtain the multi-source text data to be processed; the multi-source text data to be processed is cleaned and segmented through the multi-modal large model to obtain a number of short text fragments with semantic integrity; The data acquisition module is used to extract a preset number of short text segments from a number of short text segments as training data, and use the remaining part as test data; send the training data to a preset annotation terminal, and then obtain the corresponding relationship between the returned training data and the triple information; and use the training data + triple information as a set of final training data; A model acquisition module is used to use the final training data as input data of the policy network in the multimodal large model to obtain the trained policy network, and then obtain the trained multimodal large model; The graph acquisition module is used to obtain text extraction prompt words as the prompt corpus for the few-sample prompt algorithm; the output triplet information corresponding to the test data is obtained by using the few-sample prompt algorithm and the trained multimodal large model; based on the multi-source text data to be processed and the operation and maintenance knowledge graph schema database, the final output triplet information is determined from the output triplet information, and then combined to form the equipment operation and maintenance knowledge graph.
8. The system for constructing a device operation and maintenance knowledge graph according to claim 7, It is characterized in that The acquisition module includes a text processing unit, Used to use the Qwen-VL multimodal large model to perform data preprocessing operations on multi-source text data to be processed to remove preset irrelevant characters and preset format errors; The Qwen-VL multimodal large model is used to identify the preset logical breakpoints in the multi-source text data to be processed after preprocessing to perform text segmentation and obtain multiple short text fragments with semantic integrity.
9. The system for constructing a device operation and maintenance knowledge graph according to claim 7, It is characterized in that The model acquisition module includes a network training unit, It is used to use the final training data as the input data of the policy network in the multimodal large model, obtain the accuracy of the output data, and then use the accuracy as the input data of the value network in the multimodal large model, and use the output data of the value network as the reward signal of the policy network, and then use the PPO reinforcement learning algorithm to optimize the policy network, and obtain a trained policy network through continuous iteration.
10. A non-volatile computer storage medium, It is characterized in that Computer instructions are stored thereon, and when the computer instructions are executed, they implement a method for constructing a device operation and maintenance knowledge graph as described in any one of claims 1-6.
