Intelligent nuclear power system health assessment method based on large model technology

Abstract

This application relates to the field of fault prediction and health management technology, specifically to a health assessment method for intelligent nuclear power systems based on large model technology. The method includes: extracting knowledge from historical nuclear power data to obtain triplet data; filtering similar triplet data from the remaining triplet data and constructing an undirected graph of the triplet data; obtaining path association parameters from entities in the triplet data to each node in the undirected graph; obtaining pre-filtered triplet data for each triplet data; obtaining the association redundancy of the pre-filtered triplet data; and then filtering out redundant triplet data from all triplet data extracted from the knowledge, constructing a nuclear power knowledge graph, and assessing the health status of the nuclear power system. This application aims to make the nuclear power knowledge graph more accurate and concise by filtering out redundant parts of the triplet data, thereby achieving a more precise health assessment of the nuclear power system.

Description

Technical Field

[0001] This application relates to the field of fault prediction and health management technology, specifically to a health assessment method for intelligent nuclear power systems based on large model technology. Background Technology

[0002] Health assessments of intelligent nuclear power systems have a significant impact on the safety and economics of nuclear power plants. In recent years, with the development of large-scale modeling technology, health assessments of nuclear power systems have become increasingly intelligent. Specifically, this involves using large-scale modeling technology to integrate expert experience and historical data to form a knowledge base, and combining this with multimodal data from nuclear power plant operations to achieve more complex and accurate health assessment tasks. Because nuclear power equipment is exposed to extreme environments such as high temperatures and high radiation for extended periods, it is highly susceptible to corrosion and cracking. Therefore, accurately implementing health assessments of intelligent nuclear power systems based on large-scale modeling technology is of great importance.

[0003] Due to the complexity and diversity of nuclear power system data, the use of large model technology for nuclear power system health assessment is prone to the "illusion" problem. Although some studies have combined knowledge graphs to alleviate the illusion problem of large model technology, the knowledge in nuclear power systems is usually accompanied by knowledge redundancy, with a large number of redundant triples. This dilutes the useful information in the knowledge graph, resulting in a lack of accuracy in the constructed knowledge graph. Consequently, when using large models for nuclear power system health assessment, it is difficult to distinguish the relative importance between nuclear power failure data and nuclear power normal data, leading to poor final health assessment results. Summary of the Invention

[0004] In light of the above, it is necessary to provide a health assessment method for intelligent nuclear power systems based on large model technology. Compared with traditional health assessment methods based on large model technology, this method filters out redundant parts in the triplet data, making the nuclear power knowledge graph more accurate and concise, thereby achieving a more precise health assessment of the nuclear power system.

[0005] The health assessment method for intelligent nuclear power systems based on large model technology in this application adopts the following technical solution:

[0006] One embodiment of this application provides a health assessment method for intelligent nuclear power systems based on large model technology, the method comprising the following steps:

[0007] Knowledge extraction is performed on historical nuclear power data to obtain data for each triplet, and word vectors for each word in each triplet are obtained.

[0008] By analyzing the similarity between word vectors in each triplet and those in all other triplets, similar triplet data is filtered from the remaining triplet data, and an undirected graph of the triplet data is constructed. The path association parameters from entities in each triplet to each node in the undirected graph are obtained by considering the number of paths and shortest path lengths from the entities in each triplet, as well as the number of triplet data used to construct the undirected graph. Based on the distribution of these path association parameters, similar triplet data is filtered from each triplet. Redundant paths in the undirected graph starting from entities in the triplet data are obtained, and pre-filtered triplet data are obtained based on the redundant paths. By analyzing the distribution of path association parameters from entities in the filtered triplet data to nodes in the undirected graph, and combining the information contained in each pre-filtered triplet data, the association redundancy of each pre-filtered triplet data is obtained. Redundant triplet data is then filtered out from all triplet data extracted from knowledge. A nuclear power knowledge graph is constructed based on the filtering results to assess the health status of the nuclear power system.

[0009] In one embodiment, the step of filtering similar triplet data from the remaining triplet data and constructing an undirected graph of the triplet data includes:

[0010] The cumulative similarity coefficient between each triplet and each of the other triplets is obtained by measuring the similarity between the word vectors of each word in each triplet and each word in the other triplets.

[0011] The segmentation threshold is obtained by calculating the cumulative similarity coefficient between each triplet data and all other triplet data. The cumulative similarity coefficients that are greater than or equal to the segmentation threshold correspond to each triplet data and are used as each similar triplet data of each triplet data.

[0012] Construct an undirected graph of the triplet data using each triplet data and its similar triplet data.

[0013] In one embodiment, the process of obtaining the cumulative similarity coefficient is as follows:

[0014] Calculate the mean similarity between each word in each triplet and all words in each of the remaining triplets;

[0015] The cumulative similarity coefficient is the sum of the mean values ​​between all words in each triplet and the mean values ​​between each of the other triplet data.

[0016] In one embodiment, the process of obtaining the path association parameters is as follows:

[0017] For each triplet data, the association rate of each triplet data is obtained by comparing the number of similar triplet data participating in the construction of the undirected graph with the total number of similar triplet data.

[0018] Calculate the product of the correlation rate and the number of paths;

[0019] The path association parameter is directly proportional to the product and inversely proportional to the shortest path length.

[0020] In one embodiment, the association rate is the percentage of similar triplet data that participate in constructing the undirected graph among all similar triplet data.

[0021] In one embodiment, the path association parameter is the ratio of the product to the shortest path length.

[0022] In one embodiment, the process of obtaining the pre-filtered triplet data is as follows:

[0023] Obtain the segmentation threshold of the path association parameters from the entity in each triplet data to all nodes in the undirected graph, and select the nodes in the undirected graph of each triplet data whose path association parameters are greater than the segmentation threshold.

[0024] Each path from the entity in each triplet data to each selected node in the undirected graph, excluding the shortest path, is taken as a redundant path; each triplet data consisting of every two adjacent nodes and the edge between adjacent nodes on each redundant path is taken as a pre-filtered triplet data.

[0025] In one embodiment, the process of obtaining the associated redundancy is as follows:

[0026] Calculate the average value of the path association parameters from the entities in each of the pre-filtered triplet data to all other nodes in the undirected graph;

[0027] The entropy of each word in each pre-filtered triplet data is calculated by the probability of each word appearing in the words at the same position in all pre-filtered triplet data of each triplet data, and the maximum value of the entropy of all words in each pre-filtered triplet data is counted.

[0028] The correlation redundancy is directly proportional to the average value and inversely proportional to the maximum value.

[0029] In one embodiment, the correlation redundancy is the ratio of the average value to the maximum value.

[0030] In one embodiment, the process of filtering out redundant triplet data is as follows:

[0031] Obtain the threshold for the association redundancy of all pre-filtered triplet data for each triplet data, and filter out the pre-filtered triplet data whose association redundancy is greater than or equal to the threshold as redundant triplet data.

[0032] This application has at least the following beneficial effects:

[0033] This application identifies other triplet data that are semantically similar to each triplet data by calculating word vector similarity. Then, it constructs an undirected graph by combining each triplet data with its similar triplet data, which can intuitively display the relationship between triplet data and present complex knowledge structures in graphical form. This allows for subsequent path analysis based on the undirected graph. Features such as the number of paths and the shortest path length are quantified as path association parameters, which can measure the degree of association between triplet data and provide a reference for subsequent screening of redundant triplet data, thus helping to improve the accuracy of subsequent judgment of redundant triplet data.

[0034] Furthermore, by using path association parameters, pre-filtering triplet data with high correlation and potential redundancy is performed. The correlation redundancy of the pre-filtered triplet data is then calculated based on the average correlation degree and information content of the pre-filtered triplet data. This comprehensive consideration of the correlation between triplet data and the information content it carries makes the judgment of redundant triplet data more comprehensive and accurate. Redundant triplet data is then filtered out based on correlation redundancy, retaining useful information. A nuclear power knowledge graph is constructed using the filtered redundant triplet data, making the nuclear power knowledge graph more accurate and concise. By combining the nuclear power knowledge graph with large-scale model technology, a health assessment of the nuclear power system is conducted, thereby achieving a more precise health assessment of the nuclear power system. Attached Figure Description

[0035] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 A flowchart illustrating the steps of the health assessment method for intelligent nuclear power systems based on large model technology provided in this application;

[0037] Figure 2 This is a schematic diagram of the process of filtering out redundant triplet data. Detailed Implementation

[0038] In the description of the embodiments in this application, the words "exemplary," "or," and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary," "or," and "for example" is intended to present the relevant concepts in a specific manner.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. It should be understood that, unless otherwise stated, " / " in this application means "or".

[0040] It should also be noted that the terms "first" and "second" in this application are used to distinguish similar objects, rather than to describe a specific order or sequence.

[0041] The following section, in conjunction with the accompanying drawings, details the specific scheme of the health assessment method for intelligent nuclear power systems based on large model technology provided in this application.

[0042] This application provides an embodiment of a health assessment method for intelligent nuclear power systems based on large model technology. Specifically, the method is described below. Please refer to [link to relevant documentation]. Figure 1 The method includes the following steps:

[0043] Step 1: Extract knowledge from historical nuclear power data to obtain data for each triplet.

[0044] Historical nuclear power data is retrieved from the nuclear power system database. This data includes the name, location, operating status, historical fault causes, solution data, and operating parameters of the nuclear power equipment. Operating status refers to the different states of the nuclear power equipment during operation, including: normal operation, fault status, standby status, and shutdown status. Operating parameters include: temperature, vibration, and pressure.

[0045] Knowledge extraction is performed on historical nuclear power data using an open-source knowledge extraction tool configured on the intelligent nuclear power system. Knowledge extraction includes named entity recognition, relation extraction, and attribute extraction. The identified entities, relations, and attributes are then grouped into triples, and all resulting triples are combined into a dataset. In this embodiment, DeepKE is used for knowledge extraction. DeepKE is a well-known technology and will not be described further in this application.

[0046] Step 2: Obtain the word vectors of each word in each triplet data; filter the similar triplet data of each triplet data from the remaining triplet data; and construct an undirected graph of each triplet data; obtain the path association parameters from the entity in each triplet data to each node in the undirected graph; obtain the pre-filtered triplet data of each triplet data; obtain the association redundancy of each pre-filtered triplet data; and then filter out redundant triplet data from all triplet data extracted from knowledge.

[0047] Step 2.1: Based on the similarity of word vectors between each word in each triplet data and each word in the remaining triplet data, filter out the similar triplet data of each triplet data, and construct an undirected graph of each triplet data; obtain the path association parameters from the entity in each triplet data to each node in the undirected graph by the number of paths and the shortest path length from the entity in each triplet data to each node in the undirected graph, as well as the number of triplet data in the constructed undirected graph.

[0048] During nuclear power plant operation, nuclear power data is typically complex and diverse. For example, there might be three sets of data: (main pump, connection, coolant piping), (coolant piping, connection, steam generator), and (main pump, connection, steam generator). The third set of data can be derived from the first two sets, indicating path redundancy. If there is significant knowledge redundancy in the knowledge base, it may dilute the useful information, leading to inaccurate knowledge graphs and affecting the effectiveness of large-scale modeling techniques in training and learning from historical nuclear power data. This, in turn, impacts the accuracy of health assessments of intelligent nuclear power systems.

[0049] To analyze the similarity of semantic information between different triplet data, each word in the triplet data of historical nuclear power data is used as input, and the word vector model is used to output the word vector of each word in the triplet data.

[0050] In this embodiment, the Word2Vec word vector model is used to obtain word vectors. The Word2Vec word vector model is a well-known technology and will not be described in detail in this application. Implementers can choose other existing feasible word vector models.

[0051] Taking the i-th triplet as an example, calculate the similarity between the word vectors of each word in the i-th triplet and the word vectors of each of the remaining triplets in the dataset. Calculate the mean of the similarity between each word in the i-th triplet and all words in the remaining triplets. Use the sum of these mean values ​​as the cumulative similarity coefficient between the i-th triplet and all the remaining triplets. The larger the cumulative similarity coefficient, the higher the similarity between the i-th triplet and all the remaining triplets, and the greater the possibility of redundancy.

[0052] In this embodiment, the similarity between word vectors is calculated using the Jaccard similarity coefficient. The Jaccard similarity coefficient is a well-known technique and will not be described in detail here. As other implementation methods, based on the ability to calculate the similarity between word vectors, implementers may use other existing techniques, such as cosine similarity, etc. This application does not impose any special restrictions.

[0053] Furthermore, obtain the segmentation threshold of the cumulative similarity coefficient between the i-th triplet data and all other triplet data, and take each triplet data corresponding to the cumulative similarity coefficient greater than or equal to the segmentation threshold as each similar triplet data of the i-th triplet data.

[0054] In this embodiment, the Otsu threshold segmentation algorithm is used to obtain the segmentation threshold of the cumulative similarity coefficient. The Otsu threshold segmentation algorithm is a well-known technology and will not be described in detail in this application. As other implementation methods, based on the ability to obtain the segmentation threshold of the cumulative similarity coefficient, the implementer may use other existing technologies, such as global threshold segmentation, iterative threshold segmentation, etc. This application does not impose any special restrictions.

[0055] To further analyze the association between the i-th triplet and its similar triplets, an undirected graph of the i-th triplet is constructed using the i-th triplet and its similar triplets. Specifically, entities and attributes in the i-th triplet and its similar triplets are used as nodes, and relationships are used as edges to construct the undirected graph. The proportion of similar triplets participating in the construction of the undirected graph among all similar triplets is calculated and used as the association rate of the i-th triplet. The higher the calculated association rate, the more triplets among the similar triplets of the i-th triplet are associated with the i-th triplet.

[0056] Furthermore, starting from the first-dimensional entity in the i-th triplet data, for example, if the triplet data is (entity 1, relation, entity 2), start from entity 1; if the triplet data is (entity 3, relation, attribute), start from entity 3. Iterate through all paths to other nodes in the undirected graph in turn. The more paths there are, the greater the possibility of path redundancy. And obtain the shortest path from the starting point to each other node. The shorter the shortest path, the fewer triplet data it contains, and the smaller the possibility of redundancy.

[0057] In this embodiment, Dijkstra's algorithm is used to obtain the shortest path. Dijkstra's algorithm is a well-known technology and will not be described in detail in this application. As other implementation methods, implementers may adopt other existing feasible technologies based on the ability to obtain the shortest path, and this application does not impose any special restrictions.

[0058] Based on the above analysis, by combining the association rate of the i-th triplet data with the number of paths and the shortest path length from the first-dimensional entity in the i-th triplet data to each node in the undirected graph, the path association parameters from the first-dimensional entity in the i-th triplet data to each node in the undirected graph are obtained. The expression is:

[0059] In the formula, This represents the path association parameter from the first-dimensional entity in the i-th triplet data to the j-th node in the undirected graph; This represents the association rate of the i-th triplet data; This represents the number of paths from the first-dimensional entity in the i-th triplet to the j-th node in the undirected graph. This represents the shortest path from the first-dimensional entity in the i-th triplet data to the j-th node in the undirected graph; This represents the function that takes the shortest path length.

[0060] It should be noted that the more similar triplet data involved in constructing the undirected graph, the more paths there are in the undirected graph, and the more paths between nodes in the undirected graph. If the shortest path length is also smaller, it indicates a greater path association between the first-dimensional entity in the i-th triplet data and the j-th node in the undirected graph, i.e., the path association parameter. The larger the value.

[0061] Step 2.2: Based on the distribution of the path association parameters, filter the redundant paths in the undirected graph that start from the entities in each triplet data, and then obtain each pre-filtered triplet data of each triplet data according to the redundant paths.

[0062] If we directly take the first-dimensional entity in the i-th triplet as the starting point, find the nodes with high path association parameters between the starting point and the starting point in the undirected graph, and treat all paths between the starting point and these nodes except for the shortest path as redundant paths, then filtering out the triplet data contained in the redundant paths may result in the filtering out of triplet data containing useful information. This is because the triplet data contained in the redundant paths may only be redundant relative to the i-th triplet. If the triplet data contained in the redundant paths occurs with low frequency or contains new attributes, and may belong to the shortest path between other nodes, then the triplet data cannot be filtered out as redundancy.

[0063] Specifically, obtain the segmentation threshold of the path association parameters from the first-dimensional entity in the i-th triplet data to all nodes in the undirected graph. Select the nodes in the undirected graph of the i-th triplet data whose path association parameters are greater than the segmentation threshold. Take each path from the first-dimensional entity in the i-th triplet data to each selected node, except for the shortest path, as each redundant path. The redundant path contains multiple entities, relations and attributes. Take each pair of adjacent nodes on the redundant path and the edge between adjacent nodes to form each triplet data, and use it as each pre-filtered triplet data of the i-th triplet data.

[0064] In this embodiment, the Otsu threshold segmentation algorithm is used to obtain the segmentation threshold of the path association parameters. The Otsu threshold segmentation algorithm is a well-known technology and will not be described in detail in this application. As other implementation methods, based on the ability to obtain the segmentation threshold of the path association parameters, implementers may use other existing technologies, such as global threshold segmentation, iterative threshold segmentation, etc. This application does not impose any special restrictions.

[0065] Step 2.3: By analyzing the distribution of path association parameters from entities in each filtered triplet data to nodes in the undirected graph, and combining this with the information content contained in each pre-filtered triplet data, the association redundancy of each pre-filtered triplet data is obtained, thereby filtering out redundant triplet data from all triplet data extracted from knowledge.

[0066] Furthermore, taking the a-th pre-filtered triplet data as an example, following the calculation method of the path association parameters from the first-dimensional entity in the i-th triplet data to each node in the undirected graph, the path association parameters from the first-dimensional entity in the a-th pre-filtered triplet data to each node in the undirected graph are calculated. The average value of the path association parameters from the first-dimensional entity in the a-th pre-filtered triplet data to all nodes in the undirected graph is taken as the average association degree of the a-th pre-filtered triplet data. The larger the calculated average association degree, the more frequent the association between the a-th pre-filtered triplet data and other triplet data in the historical nuclear power data, and the greater the possibility of redundancy in the a-th pre-filtered triplet data. If the information contained in the a-th pre-filtered triplet data, namely entity, relation, and attribute information, appears more frequently, it indicates that the amount of new information carried in the a-th pre-filtered triplet data may be less, and the a-th pre-filtered triplet data is redundant. The greater the likelihood of redundancy, the more likely the a-th pre-filtered triplet data consists of entities, relations, and attributes, each carrying different information. An information entropy is calculated based on the probability of the entity in the a-th pre-filtered triplet data appearing among all pre-filtered triplet data entities, and this entropy is used as the information entropy of the entity in the a-th pre-filtered triplet data. Following the same method for calculating the information entropy of entities in the a-th pre-filtered triplet data, the information entropy of relations and attributes in the a-th pre-filtered triplet data is calculated separately. The maximum value is then extracted from the information entropy of entities, relations, and attributes in the a-th pre-filtered triplet data, and this maximum value is used as the maximum information entropy of the a-th pre-filtered triplet data. Extracting the maximum value of the information entropy reflects the most uncertain part of the a-th pre-filtered triplet data. The smaller the calculated maximum information entropy, the less useful information the a-th pre-filtered triplet data contains, and the greater the likelihood that the a-th pre-filtered triplet data is redundant. The calculation of information entropy is a well-known technique and will not be elaborated upon in this application.

[0067] Based on the above analysis, the association redundancy of the a-th pre-filtered triplet data is obtained by using the average association degree and maximum information entropy of the a-th pre-filtered triplet data. This redundancy is used to characterize the probability that the a-th pre-filtered triplet data is redundant, and the expression is:

[0068] In the formula, This indicates the correlation redundancy of the a-th pre-filtered triplet data; This represents the average correlation of the a-th pre-filtered triplet data; This represents the maximum information entropy of the a-th pre-filtered triplet data.

[0069] It should be noted that the greater the calculated association redundancy, the greater the probability that the a-th pre-filtered triplet is a redundant triplet.

[0070] Calculate the association redundancy of each pre-filtered triplet in the i-th triplet according to the method for calculating the association redundancy of the a-th pre-filtered triplet.

[0071] Furthermore, the segmentation threshold for the association redundancy of all pre-filtered triplet data for the i-th triplet data is obtained. Triplet data with association redundancy greater than or equal to the segmentation threshold are identified as redundant triplet data and filtered out. Triplet data with association redundancy less than the segmentation threshold are identified as triplet data containing useful information and retained. A schematic diagram of the redundant triplet data filtering process is shown below. Figure 2 As shown.

[0072] After each triplet is filtered out, the dataset is updated, and subsequent filtering operations are performed based on the updated dataset.

[0073] Step 3: Construct a nuclear power knowledge graph based on the filtering results to assess the health status of the nuclear power system.

[0074] Redundant triples in all triples of historical nuclear power data are filtered out to eliminate the negative impact of knowledge redundancy on the accuracy of the knowledge graph. The Protege open-source software is used to perform knowledge fusion and knowledge graph construction on the filtered historical nuclear power data triples to obtain a nuclear power knowledge graph. The Protege open-source software, knowledge fusion, and knowledge graph construction are all well-known technologies and will not be described in detail in this application.

[0075] By deploying large-scale model algorithms in nuclear power systems, full collaboration between nuclear power knowledge graphs and large-scale model algorithms is achieved. The nuclear power knowledge graph assists the large-scale model in performing professional and accurate knowledge reasoning in the nuclear power field, while the large-scale model algorithms help users more easily access and utilize information from the nuclear power knowledge graph. Using historical nuclear power data and real-time information, the large-scale model algorithms can accurately assess the health of intelligent nuclear power systems. Combining nuclear power knowledge graphs and large-scale model technologies not only enables fault diagnosis and prediction in nuclear power systems but also provides intelligent question-and-answer assistance. Staff can obtain the real-time and accurate health status of nuclear power equipment within the system. For example, by inputting "What is the operational health status of the main pump bearing?", the nuclear power system will quickly retrieve the operating information of the main pump bearing from the nuclear power knowledge graph and provide the health assessment results back to the staff, improving decision-making efficiency and reducing the risk of nuclear power equipment failure. Knowledge graph retrieval is a well-known technology and will not be elaborated upon here.

[0076] In this embodiment, the large model algorithm is LLaMa-3. LLaMa-3 is a well-known technology and will not be described in detail in this application. Implementers may choose other existing feasible large model algorithms, such as Qwen2, etc. This application does not impose any special restrictions.

[0077] In summary, this application identifies other triplet data that are semantically similar to each triplet data by calculating word vector similarity. Then, it constructs an undirected graph using each triplet data and its similar triplet data, which can intuitively display the relationships between triplet data and present complex knowledge structures in graphical form. This allows for subsequent path analysis based on the undirected graph. Furthermore, quantifying features such as the number of paths and the shortest path length into path association parameters can measure the degree of association between triplet data, providing a reference for subsequent screening of redundant triplet data and helping to improve the accuracy of subsequent judgments on redundant triplet data.

[0078] Furthermore, by using path association parameters, pre-filtering triplet data with high correlation and potential redundancy is performed. The correlation redundancy of the pre-filtered triplet data is then calculated based on the average correlation degree and information content of the pre-filtered triplet data. This comprehensive consideration of the correlation between triplet data and the information content it carries makes the judgment of redundant triplet data more comprehensive and accurate. Redundant triplet data is then filtered out based on correlation redundancy, retaining useful information. A nuclear power knowledge graph is constructed using the filtered redundant triplet data, making the nuclear power knowledge graph more accurate and concise. By combining the nuclear power knowledge graph with large-scale model technology, a health assessment of the nuclear power system is conducted, thereby achieving a more precise health assessment of the nuclear power system.

[0079] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0080] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from its essential characteristics. Therefore, the embodiments described above should be considered exemplary and non-limiting in all respects.

Claims

1. A health assessment method for intelligent nuclear power systems based on large-scale modeling technology, characterized in that, The method includes the following steps: Knowledge extraction is performed on historical nuclear power data to obtain data for each triplet, and word vectors for each word in each triplet are obtained. By analyzing the similarity between word vectors in each triplet and those in all other triplets, similar triplet data is filtered from the remaining triplet data, and an undirected graph of each triplet is constructed. The path association parameters from entities in each triplet to each node in the undirected graph are obtained by analyzing the number of paths and shortest path lengths from the entities in each triplet to each node, as well as the number of triplets used to construct the undirected graph. Based on the distribution of these path association parameters, similar triplet data is filtered from the remaining triplet data. The redundant paths in the undirected graph when the entities in the tuple data are the starting points are obtained, and then the pre-filtered triple data of each triple data is obtained based on the redundant paths; by the distribution of the path association parameters from the entities in each pre-filtered triple data to the nodes in the undirected graph, combined with the information contained in each pre-filtered triple data, the association redundancy of each pre-filtered triple data is obtained, and then redundant triple data is filtered out from all triple data extracted from knowledge, and a nuclear power knowledge graph is constructed based on the filtering results to conduct health status assessment of the nuclear power system; The step of filtering similar triplet data from the remaining triplet data and constructing an undirected graph of the triplet data includes: The cumulative similarity coefficient between each triplet and each of the other triplets is obtained by measuring the similarity between the word vectors of each word in each triplet and each word in the other triplets. The segmentation threshold is obtained by calculating the cumulative similarity coefficient between each triplet data and all other triplet data. The cumulative similarity coefficients that are greater than or equal to the segmentation threshold correspond to each triplet data and are used as each similar triplet data of each triplet data. Construct an undirected graph of the triplet data using each triplet data and its similar triplet data.

2. The health assessment method for intelligent nuclear power systems based on large-scale model technology as described in claim 1, characterized in that, The process of obtaining the cumulative similarity coefficient is as follows: Calculate the mean similarity between each word in each triplet and all words in each of the remaining triplets; The cumulative similarity coefficient is the sum of the mean values ​​between all words in each triplet and the mean values ​​between each of the other triplet data.

3. The health assessment method for intelligent nuclear power systems based on large-scale model technology as described in claim 1, characterized in that, The process of obtaining the path association parameters is as follows: For each triplet data, the association rate of each triplet data is obtained by comparing the number of similar triplet data participating in the construction of the undirected graph with the total number of similar triplet data. Calculate the product of the correlation rate and the number of paths; The path association parameter is directly proportional to the product and inversely proportional to the shortest path length.

4. The health assessment method for intelligent nuclear power systems based on large-scale model technology as described in claim 3, characterized in that, The association rate is the percentage of similar triplet data that participate in the construction of the undirected graph among all similar triplet data.

5. The health assessment method for intelligent nuclear power systems based on large model technology as described in claim 3, characterized in that, The path association parameter is the ratio of the product to the shortest path length.

6. The health assessment method for intelligent nuclear power systems based on large-scale model technology as described in claim 1, characterized in that, The process for obtaining the pre-filtered triplet data is as follows: Obtain the segmentation threshold of the path association parameters from the entity in each triplet data to all nodes in the undirected graph, and select the nodes in the undirected graph of each triplet data whose path association parameters are greater than the segmentation threshold. Each path from the entity in each triplet data to each node selected in the undirected graph, excluding the shortest path, is considered a redundant path. Each triplet, consisting of every two adjacent nodes and the edge between them on each redundant path, is used as the data for each pre-filtered triplet.

7. The health assessment method for intelligent nuclear power systems based on large model technology as described in claim 1, characterized in that, The process for obtaining the correlation redundancy is as follows: Calculate the average value of the path association parameters from the entities in each of the pre-filtered triplet data to all other nodes in the undirected graph; The entropy of each word in each pre-filtered triplet data is calculated by the probability of each word appearing in the words at the same position in all pre-filtered triplet data of each triplet data, and the maximum value of the entropy of all words in each pre-filtered triplet data is counted. The correlation redundancy is directly proportional to the average value and inversely proportional to the maximum value.

8. The health assessment method for intelligent nuclear power systems based on large model technology as described in claim 7, characterized in that, The correlation redundancy is the ratio of the average value to the maximum value.

9. The health assessment method for intelligent nuclear power systems based on large model technology as described in claim 1, characterized in that, The process of filtering out redundant triplet data is as follows: Obtain the threshold for the association redundancy of all pre-filtered triplet data for each triplet data, and filter out the pre-filtered triplet data whose association redundancy is greater than or equal to the threshold as redundant triplet data.

Images