Nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction
The experience feedback map of nuclear power construction is constructed through the multi-dimensional cascaded knowledge extraction method, which solves the problems of weak timeliness of experience feedback, inconsistent cognition and low reuse rate in nuclear power construction, and realizes the efficient utilization and rapidity of experience knowledge in nuclear power construction. Get it.
Patent Information
- Application Number
- CN202410858829.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-28
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-06-28
AI Technical Summary
During the construction of nuclear power, there is a problem of weak timeliness of experience feedback, inconsistent understanding of experience knowledge among all participants, and low reuse of experience feedback knowledge.
Using a method based on multi-dimensional cascade knowledge extraction, the quality event records in the entire life cycle of nuclear power construction are obtained, the characteristics of empirical feedback information are analyzed, the key knowledge is layered and the ontology model of the empirical feedback map is constructed, and natural language processing technology is used to perform the Intelligent extraction and storage to realize knowledgeization.
Improve the timeliness and availability of experience feedback information, ensure the unified understanding of experience knowledge among all participants, improve the reuse rate of experience feedback knowledge, and realize the rapid and accurate knowledge acquisition and analysis and processing during nuclear power construction. .
Smart Images

Figure CN118939809A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of nuclear power construction experience feedback, and in particular relates to a nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction. Background Art
[0002] Nuclear power construction is a knowledge-intensive industry involving multiple disciplines, with the characteristics of high safety, long service life and huge systems. Due to the high cost of quality accidents, all parties are doing their best to avoid nuclear accidents. Therefore, nuclear facility operators should establish a nuclear safety experience feedback system. Experience feedback refers to the collection, screening, evaluation, analysis, processing and distribution of information such as nuclear facility incidents, quality problems and good practices, with the aim of summarizing and promoting good practice experience and preventing similar quality incidents and problems from recurring.
[0003] With the development of nuclear power engineering and the improvement of intelligent manufacturing technology, complex and changing application scenarios have put forward higher requirements for the collaborative quality control of nuclear power construction. Nuclear power companies have begun to study how to further utilize and transmit experience feedback information, and shift from the traditional document-centered to knowledge-centered. To this end, Zhao Dong et al. developed an experience feedback management system to provide a unified platform for the archiving, summarization, release and reuse of nuclear power engineering design experience; Zou Xiang et al. proposed an AHP-based nuclear power plant safety-related experience feedback information importance evaluation method, which evaluates the relative importance of information to obtain feedback value to identify important experience feedback information. These platforms and methods have accelerated information sharing among various departments and professional offices to a certain extent, but the granularity of the transmitted experience feedback information is relatively coarse, and the implicit key elements still need to be further analyzed and processed by professionals, and the experience feedback information is difficult to be directly and effectively used. The nuclear power field urgently needs to use intelligent technology to analyze and extract experience feedback information and obtain important content, sort out and integrate a large number of quality event records generated during the construction process into structured experience feedback knowledge, assist relevant personnel to quickly analyze key elements, and realize the efficient cross-enterprise circulation and reuse of nuclear power construction quality information.
[0004] In recent years, machine learning has achieved remarkable results in knowledge mining, which can be introduced into the research of experience feedback in nuclear power construction. Natural language processing (NLP) can use computers to process, understand and standardize the expression of human language text, and can intelligently parse the important knowledge in experience feedback information, thereby improving its usability. The knowledge graph proposed by SINGHAL et al. of Google can visualize the important knowledge in the key domain of experience feedback in a unified form, which is conducive to the efficient use and transmission of large-scale experience feedback information. The knowledge graph based on NLP technology can effectively reduce the dependence of nuclear power experience feedback information processing on artificial prior knowledge, reduce human errors to a certain extent, further enhance the empowerment of data elements, and meet the needs of cross-enterprise intelligent transmission. Therefore, some scholars have applied natural language processing technology to the field of nuclear power: Choi et al. extracted grammatical and semantic information from the operating procedures of nuclear power plants by adopting rule-based information extraction methods; Deng Zhiguang et al. extracted fault features from the field data of reactor outlet temperature of nuclear power plants by improving long and short-term memory neural networks.
[0005] In summary, at this stage, some scholars have made some attempts to mine nuclear power knowledge using methods such as machine learning. However, most of the research relies on manual and rule templates or a single algorithm. The generalization ability of the model is weak and the application support capability is insufficient. In particular, there is still a certain gap in the research on multi-dimensional correlation mining of nuclear power construction experience feedback information.
[0006] Therefore, how to effectively solve the problems of weak timeliness of experience feedback in the nuclear power construction process, inconsistent understanding of experience knowledge among various participants, and low reuse rate of experience feedback knowledge has become a problem that needs to be solved urgently. Summary of the invention
[0007] In view of the above-mentioned deficiencies in the prior art, the present invention provides a nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction, which can effectively solve the problems of weak timeliness of experience feedback in the nuclear power construction process, inconsistent cognition of experience knowledge among various participants, and low reuse rate of experience feedback knowledge.
[0008] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0009] A nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction includes the following steps:
[0010] S1. Obtaining experience feedback information on quality incident records generated internally during the entire life cycle of nuclear power construction;
[0011] S2. Analyze the characteristics of experience feedback information to obtain key knowledge categories, hierarchically associate and couple various key knowledge categories, define the organizational structure of the experience feedback key knowledge domain, and obtain the ontology model of the experience feedback map;
[0012] S3, based on the model architecture of the ontology model in S2, construct a multi-dimensional cascade knowledge extraction model to obtain filling data of the ontology model from the experience feedback information;
[0013] The multi-dimensional cascade knowledge extraction model includes a knowledge mining model and a relationship extraction model; the knowledge mining model is used to perceive and capture each category of key knowledge in the experience feedback information; the relationship extraction model is used to extract the relationship characteristics of each key knowledge in the experience feedback information;
[0014] S4, using the multi-dimensional knowledge extraction model built by S3, extracts structured filling data from the experience feedback information obtained by S1, and maps the extracted filling data to the ontology model of S2, generates the corresponding experience feedback map, and stores it in the knowledge base;
[0015] S5. When a quality incident occurs during the nuclear power construction process, the knowledge base is searched to obtain the required knowledge to assist nuclear power professional and / or management personnel in analysis and processing.
[0016] Compared with the prior art, the present invention has the following beneficial effects:
[0017] 1. This method can transform the experience feedback information of the quality event records of the construction and management into knowledge, and construct a unified experience feedback map through multi-dimensional cascade knowledge mining. Specifically, after obtaining the experience feedback information, analyze the characteristics of the experience feedback information, hierarchically associate and couple multiple types of high-value knowledge, define the organizational structure of the key knowledge domain of the experience feedback, and thus construct the ontology model (i.e., framework model) of the experience feedback map; then, construct a multi-dimensional cascade knowledge extraction model based on the model architecture of the ontology model, obtain each category of key knowledge from the experience feedback information, extract the relationship characteristics of each key knowledge, and map it to the ontology model to obtain the corresponding experience feedback map, and store it in the knowledge base.
[0018] This method breaks the traditional text recording method and uses the knowledge graph carrier to present the knowledge-based results of experience feedback. Construct an experience feedback graph of unified data, and under the guidance of the key knowledge domain of experience feedback, use natural language processing technology to intelligently extract experience knowledge, and store structured knowledge through a graph database to realize the knowledge-based experience feedback. This method realizes the transformation of nuclear power construction experience feedback information from document-centric to knowledge-centric, and stores experience feedback knowledge in a unified and clear way in the knowledge base. In this way, when a quality incident occurs during the nuclear power construction process, the required knowledge can be quickly and accurately obtained through retrieval in the knowledge base to assist nuclear power professional and technical and / or management personnel in analysis and processing.
[0019] 2. When constructing the experience feedback map, this method adopts a top-down and bottom-up hybrid construction method to sort out and reorganize a large amount of complex and lengthy quality information, and use the key knowledge domain of experience feedback as the map model layer; at the same time, according to the characteristics of nuclear power experience feedback text, natural language processing technology is used to specifically design named entity recognition (i.e., knowledge mining model) and relationship extraction methods, from which key knowledge groups in structured form are extracted as data layers and mapped into the ontology model to update the knowledge structure. In this way, the high efficiency and effectiveness of the obtained experience feedback map can be guaranteed.
[0020] In summary, this method can effectively solve the problems of weak timeliness of experience feedback in the nuclear power construction process, inconsistent understanding of experience knowledge among various participants, and low reuse rate of experience feedback knowledge.
[0021] Preferably, S2 includes:
[0022] Analyze the characteristics of experience feedback information to obtain key knowledge categories related to fault manifestations;
[0023] Each knowledge category is formed into a knowledge layer of the key knowledge domain of experience feedback, and the domain knowledge spectrum is obtained; at the same time, the association coupling is realized through the interactive relationship between each knowledge layer, and the spatial knowledge structure with semantic association and logical reasoning ability is obtained as the organizational structure of the key knowledge domain of experience feedback;
[0024] Afterwards, the class and object attributes are constructed for the different levels of knowledge in the key knowledge domain of experience feedback, as well as the associations between the layers, so as to establish the semantic relationships between node types and concepts or instances, obtain the knowledge form with the <entity-relationship-entity> structure, and explicitly represent the key knowledge domain of experience feedback to form an ontology model of the experience feedback map.
[0025] In this way, by defining unified data specifications for key knowledge domains of experience feedback, we can clarify goals for knowledge mining and provide model support for the construction of subsequent experience feedback maps.
[0026] Preferably, the knowledge categories include equipment information, causes, countermeasures, and time of occurrence, stage, related equipment, unit, main responsible personnel, and enterprise departments of the fault; the association relationships include causal relationships, dependency relationships, hierarchical relationships, and time relationships.
[0027] Such a setting can ensure the comprehensive validity of each key knowledge in the ontology model, thereby ensuring the validity of the subsequent experience feedback map.
[0028] Preferably, in S3, the knowledge mining model is a multi-layer neural network model integrating an attention mechanism built using a cascade strategy; the knowledge mining model includes a word embedding layer, a knowledge perception layer, and a knowledge prediction layer arranged in sequence.
[0029] In view of the problem that the text content of nuclear power construction experience feedback information is complex and has many relationship categories, a single network model cannot well identify and learn key features, and experience feedback knowledge is not easy to mine, organize and represent. This setting can perform multi-dimensional feature mining on experience feedback information in the nuclear power field, efficiently extract key field knowledge, and thus improve the mining effect in the niche field of nuclear power.
[0030] Preferably, the word embedding layer is designed based on Transformer's bidirectional encoder representation technology BERT; the working process of the word embedding layer includes: the input has three different embedding layers, which respectively represent the word information, sentence paragraph information and sentence position information of the input quality event text, and the three-layer information vector is spliced as the input of BERT, and the output vector containing semantic features of multiple dimensions is obtained through multi-layer Transformer.
[0031] In this way, the computer model can recognize texts in the nuclear power field and obtain output vectors containing semantic features in multiple dimensions, providing numerical data input for subsequent models.
[0032] Preferably, the knowledge perception layer is constructed based on a bidirectional long short-term memory network BiLSTM fused with an attention mechanism Attention; the bidirectional long short-term memory network BiLSTM is used to bidirectionally mine historical information and future information in quality event text segments, and comprehensively link long-term features; the attention mechanism Attention is used to calculate the degree of correlation between word features in the experience feedback information, so that the knowledge perception layer focuses on areas related to nuclear power when processing information, so as to obtain more effective semantic information.
[0033] The text of quality events generated in nuclear power construction is usually long, and the model is prone to forgetting past information during the learning and recognition process. This setting can obtain more effective semantic information.
[0034] Preferably, the knowledge prediction layer is constructed based on conditional random fields to optimize the output results of the knowledge perception layer; when predicting, the emission matrix and the transfer matrix are combined, and the state score and transfer score output by the upper layer are used to calculate the annotation sequence with the highest score corresponding to the sentence label as the corresponding key knowledge.
[0035] The result of the knowledge perception layer only learns each word in the nuclear power quality event text, and does not consider the constraint relationship between labels. For example, "pressure vessel" is a fixed professional term in the nuclear power field, and the order between words cannot be adjusted, so the connection order of labels needs to be constrained. Introducing conditional random fields (CRF) to optimize the output results of the upper layer can ensure that the final predicted experience feedback knowledge is reasonable and effective.
[0036] Preferably, the relation extraction model is optimized by a piecewise convolutional neural network (PCNN) based on multi-instance learning, including:
[0037] A vector expression layer, used to convert the input key knowledge into an embedding vector expression matrix; the embedding vector expression matrix includes a word embedding vector and a position embedding vector, the word embedding vector is used to indicate the text words in the key knowledge, and the position embedding vector is used to indicate the position information of each text word in the key knowledge;
[0038] The convolution layer is used to extract local features of the embedded vector expression matrix by region to obtain local feature maps of different regions;
[0039] The segmented maximum pooling layer is used to perform segmented maximum pooling operations on the local feature maps of different regions to obtain the pooled maximum value feature maps corresponding to different regions;
[0040] The fully connected layer is used to concatenate the pooled maximum feature maps corresponding to different regions, and then use the Gaussian error linear unit GELU as the activation function to obtain the fully connected feature map of key knowledge;
[0041] The classification layer is used to make classification decisions and extract relationships on the fully connected feature graph of key knowledge, obtain the relationship features presented in the text information in the key knowledge, and output them.
[0042] Since the nuclear power experience feedback information expressing quality events is long, the traditional method of extracting relations based on lexical and syntactic features has low accuracy and is prone to introduce noise data. This setting can improve the generalization ability of the experience feedback information relation extraction model and avoid the gradient vanishing problem.
[0043] Preferably, the activation function expression of the Gaussian error linear unit GELU is:
[0044]
[0045] Among them, x represents the input of the activation function, and π is the ratio of pi.
[0046] Preferably, in S5, a search is performed in the knowledge base through an inference engine to obtain required knowledge.
[0047] In this way, the convenience and effectiveness of knowledge acquisition can be guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] In order to make the purpose, technical solution and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:
[0049] Figure 1 is a schematic diagram of the process of this method;
[0050] Figure 2 This is a schematic diagram of the organizational structure of the key knowledge domain of experience feedback in Example 1;
[0051] Figure 3 This is a schematic diagram of the ontology model of the experience feedback graph in Example 1;
[0052] Figure 4 This is a schematic diagram of the knowledge mining model in Example 1;
[0053] Figure 5 This is a schematic diagram of the PCNN network model structure in Example 1;
[0054] Figure 6 It is a schematic diagram of the feedback spectrum of nuclear power construction experience in Example 1;
[0055] Figure 7 This is a schematic diagram of the experience feedback knowledge application framework in Example 1;
[0056] Figure 8 This is a schematic diagram of the performance comparison results of the knowledge mining model in Example 2;
[0057] Fig. 9 This is a schematic diagram of an example of quality problem association in Example 3;
[0058] Fig.10 This is a schematic diagram of entity association knowledge in Example 3;
[0059] Fig.11 This is a schematic diagram of the fault knowledge spectrum in Example 3;
[0060] Fig.12 This is a schematic diagram of RPV-related knowledge retrieval in Example 3;
[0061] Fig.13 This is a schematic diagram of the application process of the experience feedback map in Example 3. DETAILED DESCRIPTION
[0062] The following is a further detailed description through specific implementation methods:
[0063] Embodiment 1
[0064] like Figure 1 As shown, this embodiment discloses a nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction, comprising the following steps:
[0065] S1. Obtain experience feedback information on quality incident records generated internally during the entire life cycle of nuclear power construction.
[0066] The sources of experience feedback information for nuclear power construction are wide-ranging, mainly including internal and external feedback. External experience mainly comes from the Major Event Report (SER) and Major Operational Event Report (SOER) of the World Association of Nuclear Owners (WANO). On the one hand, root cause analysis is conducted on problems and events that have occurred in this power plant and its peers in the world to prevent the recurrence of similar problems and events in the future. For example, after the Fukushima nuclear accident, nuclear power companies have learned lessons and incorporated safety improvements into design standards, put forward clearer requirements for prevention and mitigation measures for severe accidents, and increased research on the probabilistic safety assessment (PSA) technical system for nuclear power construction. On the other hand, good practices of peer power plants around the world are absorbed, internal comparative inspections are carried out, improvement action plans are proposed and followed up.
[0067] Compared with external feedback, the more timely and effective feedback from internal experience is during the nuclear power construction process. A large number of records such as DEN, UES, NCR and FCR are generated and accumulated in real time during the construction of nuclear power plants. The content covers the quality information of all aspects of the nuclear power project life cycle, including design, procurement and manufacturing, construction and installation, commissioning and operation and maintenance. For example, if the design drawings do not meet the requirements and the on-site quality witness fails, records with reference and feedback value will be upgraded to form quality events, which will accurately record the event process, causes, specific rectification measures and improvement plans in detail. This type of information is more relevant and available, and needs to be processed and transmitted in a timely manner.
[0068] Therefore, in order to improve the efficiency of experience feedback, the scope of experience feedback knowledge in this method is mainly the quality event records generated internally during the entire life cycle of nuclear power construction.
[0069] S2. Analyze the characteristics of experience feedback information to obtain key knowledge categories, hierarchically associate and couple various key knowledge categories, define the organizational structure of the experience feedback key knowledge domain, and obtain the ontology model of the experience feedback map.
[0070] In specific implementation, S2 includes:
[0071] Analyze the characteristics of experience feedback information to obtain key knowledge categories related to fault manifestations;
[0072] Each knowledge category is formed into a knowledge layer of the key knowledge domain of experience feedback, and the domain knowledge spectrum is obtained; at the same time, the association coupling is realized through the interactive relationship between each knowledge layer, and the spatial knowledge structure with semantic association and logical reasoning ability is obtained as the organizational structure of the key knowledge domain of experience feedback;
[0073] Afterwards, the class and object attributes are constructed for the different levels of knowledge in the key knowledge domain of experience feedback, as well as the associations between the layers, so as to establish the semantic relationships between node types and concepts or instances, obtain the knowledge form with the <entity-relationship-entity> structure, and explicitly represent the key knowledge domain of experience feedback to form an ontology model of the experience feedback map.
[0074] Among them, the knowledge categories include the equipment information, causes, countermeasures, time of the fault, stage, related equipment, unit, main responsible personnel, and enterprise departments; the association relationships include cause-and-effect relationships, dependency relationships, hierarchical relationships, and time relationships.
[0075] The quality event records generated internally during the entire life cycle of nuclear power construction contain a wide range of content and categories, but not all information has transferable significance. In order to realize the knowledge-based experience feedback, it is necessary to screen and define the elements with feedback value, form a unified and standardized knowledge domain, and ensure the consistency and interpretability of the information.
[0076] The key knowledge domain of experience feedback is a collection of key knowledge in experience feedback information, and is a hierarchical description of quality events in the construction process. Its content mainly includes key domain entities and domain relationships. Key domain entities refer to entities with important significance and role in the field of nuclear power construction, including events, objects, locations, causes, actions, personnel, etc.; domain relationships describe the connection, dependency or influence relationship between them, which can be causal relationship, dependency relationship, hierarchical relationship, time relationship, etc.
[0077] Statistical analysis of the specific content of experience feedback information in nuclear power enterprises shows that the descriptive information about fault problems in the quality events recorded in the archives has a high frequency of occurrence and has extremely high data value. It is the key data element that all participants need to master during the construction process and is also the key to affecting the efficiency and quality of nuclear power construction. Based on this analysis result and combined with the opinions of experts in related fields, this paper preliminarily defines the key knowledge domain of experience feedback, which revolves around the fault problem as the core to form a fault manifestation layer. Further analysis and evaluation of various key information closely related to fault manifestations show that collecting equipment information of faults is conducive to screening key focus objects and continuously tracking them in subsequent stages, while horizontally comparing other units of the project to prevent quality problems from occurring again in the same type of equipment; extracting the cause information of the fault can help enterprises or departments deal with problems more effectively, prevent potential problems from the root, and reduce the number of quality events; centralized fault corresponding measures information facilitates relevant personnel to quickly query and solve the same problem after it occurs, thereby improving the efficiency of problem solving. In addition, the time of fault occurrence, the stage to which it belongs, the relevant equipment, the unit to which it belongs, the main responsible personnel, and the enterprise department are all necessary information with transmission value, as shown in Table 1.
[0078] Table 1 Knowledge categories and functions
[0079]
[0080] The key knowledge elements in different dimensions sorted out above are respectively formed into knowledge layers such as fault equipment layer, fault cause layer, and corresponding measures layer of the key knowledge domain of experience feedback, and they are used as the domain knowledge spectrum. At the same time, the association coupling is realized through the interaction relationship between layers, and it is transformed into a spatial knowledge structure with semantic association and logical reasoning capabilities, integrating the experience feedback resources of nuclear power construction. The specific organizational structure is as follows: Figure 2 shown.
[0081] Based on the above classification and associative organization of experience feedback knowledge, the key knowledge domain of experience feedback is used as the experience feedback graph model layer, which is explicitly represented through ontology modeling method to provide a framework for knowledge extraction.
[0082] Protégé software is used to construct classes (owl: Class) and object properties (owl: Object property) for different levels of domain knowledge in the key knowledge domain of experience feedback, including unit level, equipment level, fault level, cause level, measure level, stage level, time level, enterprise level, personnel level, and the association relationship between each level, so as to establish the semantic relationship between node types and concepts or instances, obtain the knowledge form with the structure of <entity-relationship-entity>, and explicitly represent the key knowledge domain of experience feedback to form the ontology model of the experience feedback map, such as Figure 3 shown.
[0083] By defining unified data specifications for key knowledge domains of experience feedback, we can clarify goals for knowledge mining and provide model support for the construction of subsequent experience feedback maps.
[0084] S3. Based on the model architecture of the ontology model in S2, a multi-dimensional cascade knowledge extraction model is constructed to obtain the filling data of the ontology model from the experience feedback information. Among them, the multi-dimensional cascade knowledge extraction model includes a knowledge mining model and a relationship extraction model; the knowledge mining model is used to perceive and capture each category of key knowledge in the experience feedback information; the relationship extraction model is used to extract the relationship characteristics of each key knowledge in the experience feedback information.
[0085] In order to further construct a complete experience feedback map, under the guidance of the key knowledge domain framework of experience feedback, the knowledge extraction method is used to analyze and mine the important semantic features of quality information in the nuclear power field, and fill data into the constructed experience feedback map ontology model.
[0086] In view of the problem that the text content of nuclear power construction experience feedback information is complex and has many relationship categories, a single network model cannot well identify and learn key features, and experience feedback knowledge is not easy to be mined, organized and represented. In order to improve the mining effect in the niche field of nuclear power, this method proposes a multi-dimensional cascade knowledge extraction method: building a multi-layer neural network model to mine multi-dimensional features of experience feedback information in the nuclear power field, and efficiently extract key field knowledge.
[0087] The key knowledge elements contained in nuclear power quality events need to be automatically identified and extracted from the corpus text through the named entity recognition (NER) method. The methods based on manually written rule templates or feature-based statistical model recognition are more dependent on the establishment of knowledge bases and dictionaries, and are not very portable. There are not many studies on the nuclear power field, and the construction cycle of new dictionaries and rule templates is long. Therefore, an automated entity extraction method is used to embed text into the neural network in the form of vectors, and end-to-end named entity recognition is achieved through deep learning methods. Named entity recognition can be performed on a large amount of text in a short time, saving mining costs and time.
[0088] In view of the complexity of the information content of experience feedback in the nuclear power field, this method uses deep learning methods to construct three-layer functional networks for word embedding, knowledge perception and knowledge prediction, and obtains a knowledge mining model. It perceives and captures the key elements in nuclear power quality events from multiple dimensions, integrates the attention mechanism, strengthens the attention to key words in the nuclear power field, and effectively solves the problem of poor knowledge recognition in the nuclear power field. The structure of the knowledge mining model is as follows: Figure 4 shown.
[0089] The word embedding layer is designed based on Transformer's bidirectional encoder representation technology BERT.
[0090] In order to enable the computer model to recognize texts in the nuclear power field, a word embedding layer based on Transformer's bidirectional encoder representation technology (Bidirectional Encoder Representations from Transformers, BERT) is designed to map vocabulary to real number vectors. BERT is a pre-trained network model with strong generalization ability. The input has three different embedding layers, which represent the word information (token embedding), sentence paragraph information (segment embedding) and sentence position information (position embedding) of the input quality event text. The three-layer information vector is concatenated as the input of BERT, and the output vector containing semantic features of multiple dimensions is obtained through multiple layers of Transformer, which provides numerical data input for subsequent models.
[0091] The knowledge perception layer is constructed based on the bidirectional long short-term memory network BiLSTM fused with the attention mechanism Attention.
[0092] The text of quality events generated in nuclear power construction is usually long, and the model is prone to forgetting past information during the learning and recognition process. Therefore, a bidirectional long short-term memory network (BiLSTM) is used to bidirectionally mine the historical information and future information in the quality event text to comprehensively link the long-term features. In order to further enhance the sensitivity of the text to nuclear power-related elements, the method integrates the attention mechanism (Attention) to calculate the degree of correlation between the features of words in the nuclear power experience feedback information, so that the model focuses on the areas related to nuclear power when processing information, thereby obtaining more effective semantic information.
[0093] The knowledge prediction layer is built based on conditional random fields (CRF) and is used to optimize the output results of the knowledge perception layer.
[0094] The result of the knowledge perception layer only learns each word in the nuclear power quality event text, and does not consider the constraints between labels. For example, "pressure vessel" is a fixed professional term in the nuclear power field, and the order between words cannot be adjusted, so the connection order of the labels needs to be constrained. The conditional random field (CRF) is introduced to optimize the output results of the upper layer. When predicting, the emission matrix and the transfer matrix are combined, and the state score and transfer score of the upper layer output are used to calculate the highest-scoring annotation sequence corresponding to the sentence label, thereby ensuring that the final predicted experience feedback knowledge is reasonable and effective. The specific calculation process is as follows:
[0095]
[0096] in For label y i Move to label y i+1 The score, S i ,y i Predict the label y for the i-th character in the input text sequence i score.
[0097] The probability of outputting the label sequence Y is calculated using the normalized exponential function Softmax:
[0098]
[0099] where Y X is the set of all possible label sequences, is the true label sequence.
[0100] Finally, the Viterbi algorithm is used to obtain the global optimal label sequence:
[0101]
[0102] The relation extraction model is optimized by the piecewise convolutional neural network PCNN based on multi-instance learning.
[0103] In view of the problem that the relationship of nuclear power construction experience feedback information is complex and difficult to sort out, it is necessary to use relation extraction (RE) technology to extract relational features from nuclear power quality text. Since the nuclear power experience feedback information expressing quality events is long, the traditional method of extracting relationships based on lexical and syntactic features has a low accuracy rate. At the same time, in order to avoid introducing noise data, the piecewise convolutional neural network (Piece-Wise-CNN, PCNN) method based on multi-instance learning is used to complete the relationship extraction. Based on the convolutional neural network (CNN) model, the feature extraction scheme is adjusted, and the relationship extraction is regarded as a multi-instance problem. Only one example closest to the true category in each package is selected as the output. At the same time, in order to reduce the propagation or accumulation of errors, the position vector is introduced in the vector expression of the sentence. According to the position of the key information in the nuclear power experience feedback information, the sentence is divided into three segments, and the maximum pooling operation is performed separately to extract more fine-grained relationship features. Finally, the output result of the segmented pooling layer is sent to the softmax layer to obtain the classification of the nuclear power experience feedback information relationship. The specific structure is as follows Figure 5 shown.
[0104] According to the highly professional characteristics of nuclear power experience feedback information content, the PCNN relationship extraction model is improved by introducing the idea of regularization, and using the Gaussian error linear unit GELU with smoother nonlinear characteristics as the activation function to improve the generalization ability of the experience feedback information relationship extraction model and avoid the gradient vanishing problem.
[0105] Specifically, if Figure 5 As shown, the optimized relation extraction model includes:
[0106] A vector representation layer (Vector Representation), used to convert the input key knowledge into an embedding vector representation matrix; the embedding vector representation matrix includes a word embedding vector and a position embedding vector, the word embedding vector is used to indicate the text words in the key knowledge, and the position embedding vector is used to indicate the position information of each text word in the key knowledge;
[0107] Convolutional layer (Convolutiona), used to extract local features of the embedded vector expression matrix by region, and obtain local feature maps of different regions;
[0108] The piecewise max pooling layer is used to perform piecewise max pooling operations on the local feature maps of different regions to obtain the pooled maximum value feature maps corresponding to different regions;
[0109] The fully connected layer is used to concatenate the pooled maximum feature maps corresponding to different regions, and then use the Gaussian error linear unit GELU as the activation function to obtain the fully connected feature map of key knowledge;
[0110] The classification layer (Softmax classifier) is used to make classification decisions and extract relations on the fully connected feature graph of key knowledge, obtain the relational features presented in the text information of key knowledge, and output them.
[0111] Assume the input is x, the specific expression is:
[0112]
[0113] Among them, x represents the input of the activation function, and π is the ratio of pi.
[0114] S4, using the multi-dimensional knowledge extraction model built by S3, extracts structured fill-in data from the experience feedback information obtained by S1, and maps the extracted fill-in data to the ontology model of S2, generates the corresponding experience feedback map, and stores it in the knowledge base.
[0115] The purpose of knowledge-based experience feedback is to enable the key information of experience feedback to be transmitted, retrieved and shared in actual nuclear power construction.
[0116] After the experience feedback information is extracted through the multi-dimensional cascade knowledge extraction model, the original large amount of unstructured quality event text is transformed into the key domain of experience feedback rich in structured semantic knowledge. On this basis, the Neo4j graph database is further used to construct the nodes and edges in the structure, complete the ontology model mapping and knowledge specification storage, and realize the visual integration of experience feedback knowledge. The resulting nuclear power construction experience feedback map is as follows: Figure 6 shown.
[0117] By centrally managing the organized nuclear power construction experience feedback key knowledge domains in the knowledge base, its independent and scattered nature can be changed to ensure that the experience feedback knowledge can be presented in a unified and clear manner. On this basis, we further establish the experience feedback knowledge reasoning rules, transform the experience feedback information into a knowledge structure with semantic association and logical reasoning capabilities, and realize the coupling interaction between knowledge, that is, trigger the corresponding reasoning process according to specific conditions, so that the associated knowledge can be obtained. The application framework is as follows: Figure 7 For example, if a certain failure mode occurs, what specific solutions or corrective actions are taken.
[0118] It is possible to combine machine learning and data mining techniques to automatically discover potential rules and patterns by analyzing large amounts of experience feedback data, assist in establishing more intelligent and efficient reasoning rules, and use professional knowledge graphs or expert system technologies to build a reasoning engine that converts reasoning rules into a form that can be understood and executed by computers, thereby improving the interpretability of knowledge and realizing the automatic reasoning and application of experience feedback knowledge from nuclear power construction.
[0119] S5. When a quality incident occurs during the nuclear power construction process, the reasoning engine is used to search the knowledge base to obtain the required knowledge to assist nuclear power professional and / or management personnel in analysis and processing.
[0120] This method breaks the traditional text recording method and uses the knowledge graph carrier to present the knowledge-based results of experience feedback. It builds a unified data experience feedback graph, uses natural language processing technology to intelligently extract experience knowledge under the guidance of the key knowledge domain of experience feedback, stores structured knowledge through a graph database, and establishes an inference engine to provide knowledge retrieval tools to realize the knowledge-based experience feedback.
[0121] This method can transform the experience feedback information of the quality event records of nuclear power construction into knowledge, and construct a unified experience feedback map through multi-dimensional cascade knowledge mining. Specifically, after obtaining the experience feedback information, the characteristics of the experience feedback information are analyzed, and multiple categories of high-value knowledge are hierarchically associated and coupled to define the organizational structure of the key knowledge domain of the experience feedback, so as to construct the ontology model (i.e., framework model) of the experience feedback map; then, a multi-dimensional cascade knowledge extraction model is constructed based on the model architecture of the ontology model, and each category of key knowledge is obtained from the experience feedback information, and the relationship characteristics of each key knowledge are extracted and mapped to the ontology model to obtain the corresponding experience feedback map, which is stored in the knowledge base. This method breaks the traditional text recording method and uses the knowledge map carrier to present the experience feedback knowledge result. The experience feedback map of unified data is constructed, and under the guidance of the key knowledge domain of experience feedback, natural language processing technology is used to intelligently extract experience knowledge, and structured knowledge is stored through the graph database to realize the knowledge of experience feedback. This method realizes the transformation of nuclear power construction experience feedback information from document-centric to knowledge-centric, and stores the experience feedback knowledge in the knowledge base in a unified and clear way. In this way, when a quality incident occurs during the nuclear power construction process, the required knowledge can be quickly and accurately obtained through retrieval in the knowledge base to assist nuclear power professional and / or management personnel in analysis and processing.
[0122] In addition, when constructing the experience feedback graph, this method adopts a top-down and bottom-up hybrid construction method to sort out and reorganize a large amount of complex and lengthy quality information, and uses the key knowledge domain of experience feedback as the graph model layer; at the same time, according to the characteristics of nuclear power experience feedback text, natural language processing technology is used to design named entity recognition and relationship extraction methods, from which key knowledge groups in structured form are extracted as data layers and mapped to the ontology model to update the knowledge structure. In this way, the high efficiency and effectiveness of the obtained experience feedback graph can be guaranteed. This method can effectively solve the problems of weak timeliness of experience feedback in the nuclear power construction process, inconsistent cognition of experience knowledge among various participants, and low reuse rate of experience feedback knowledge.
[0123] Embodiment 2
[0124] In order to illustrate the effectiveness of the knowledge mining model and relationship extraction model in this method, the following description is made:
[0125] The quality event records in the information system of nuclear power enterprises are collected as original training data. After cleaning, the data length is standardized according to the characteristics of the information text. After preliminary regular collection, screening, pre-experimental verification, label annotation and relationship annotation are performed according to the annotation strategies commonly used in natural language processing tasks to further complete data preprocessing and provide a data basis for text mining. This data set consists of 350 quality event records, and the data is divided into training set, test set and validation set in a ratio of 7:1.5:1.5 to calculate the specific evaluation indicators of the model.
[0126] Model parameter settings
[0127] By consulting the literature and conducting preliminary experiments, the main hyperparameter settings in the knowledge mining model are shown in Table 2 below. Other parameters use the default values. The experience feedback datasets preprocessed above are put into the model for training.
[0128] Table 2 Model parameter settings
[0129]
[0130]
[0131] Knowledge extraction effect
[0132] In order to verify the extraction effect of knowledge extraction model and relation extraction model in key semantic features in experience feedback information, several traditional knowledge mining models were selected for experimental comparison with the BERT-BiLSTM-CRF model of this method under the same data set. The results are as follows: Figure 8 shown.
[0133] From the performance comparison results, it can be seen that the BERT-BiLSTM-CRF model with integrated attention mechanism used in this method has achieved the best performance in all indicators in the entity recognition task of experience feedback information. The precise, recall and F1 values (the composite indicator of precision and recall) reached 91.49%, 93.14% and 92.31%, respectively, and the model recognition effect is good.
[0134] Similarly, the improved PCNN model in this method is compared with the ordinary PCNN, and the results are shown in Table 3.
[0135] Table 3. Comparison results of relation extraction model performance
[0136]
[0137] According to the performance comparison experimental results of the relationship extraction task in Table 3, the precision, recall and F1 value of the improved PCNN relationship extraction model proposed in this method are higher than those of the ordinary PCNN, reaching 93.18%, 93.36% and 92.87% respectively. It can better identify the relationship characteristics of nuclear power quality texts and provide data support for the next step of building a complete experience feedback map.
[0138] Embodiment 3
[0139] Based on the experience feedback graph generated by the neo4j graph database, cypher language instructions can be used for further in-depth knowledge mining and application.
[0140] Query and reuse of quality problem handling measures
[0141] By searching a specific knowledge base through the experience feedback map, you can query the treatment measures for specific quality problems in a targeted manner. For example, if a quality incident of gear collision occurs during the construction of a nuclear power plant, you can use the command in the experience feedback map:
[0142] “MATCH(n)-[r]-()
[0143] WHERE n.gz CONTAINS 'collision'
[0144] RETURN n,r"
[0145] Retrieve past experience and knowledge to obtain relevant "collision" fault nodes, expand the nodes to further obtain the corresponding experience feedback graph and node information statistics, as follows: Fig. 9 shown.
[0146] Compared with the traditional way of recording quality events in text, the unified and standardized graph structure is more conducive to managers to organize and manage experience feedback knowledge, which means that all quality event records are organized and stored according to the same structure and specifications, ensuring that all kinds of complex information can be expressed and stored in a consistent way. Users do not need to read a lot of text documents. Even non-professionals can quickly understand the key content, obtain quality problem handling measures and other relevant knowledge by browsing knowledge nodes and connection relationships. At the same time, it can avoid cognitive bias between different people or teams on experience feedback information with different writing styles, improve the comprehensibility and reusability of experience feedback information, and thus improve the efficiency and quality of experience feedback.
[0147] Query and push of quality issue related entities
[0148] In the actual construction process of nuclear power plants, experience feedback engineers are not archival professionals and have limited information retrieval capabilities. They often use keywords to search for resources, which results in low completeness and precision and coarse granularity, making it difficult to discover the implicit knowledge of experience feedback information and delaying its transmission. The associated knowledge retrieval based on the experience feedback graph breaks the traditional keyword retrieval method. Managers can quickly query various entities related to the push of specific quality issues based on conditions such as quality issue nodes and associated relationships.
[0149] For example, there is no direct relationship between the unit to which the equipment belongs and the stage at which the fault occurs. However, the knowledge graph multi-hop search method can be used to query the implicit path association between entities. Enter the command:
[0150] "MATCH q=(a)-[*1..3]-n
[0151] WHRER n.jd CONTAINS 'Construction Phase'
[0152] AND a.jz CONTAINS 'Unit 3'
[0153] RETURN q"
[0154] Perform a multi-level jump query to find the indirect relationship between "Unit 3" and "Construction Phase". The results are as follows Fig.10 shown.
[0155] As can be seen from the figure, a total of three devices in Unit 3 had different types of quality problems during the construction phase. Further selecting nodes to expand can obtain the corresponding fault causes, personnel and other information, which can assist management personnel in clarifying the relevant quality business interfaces and personnel responsibility implementation between enterprises, and thus provide feedback to the corresponding departments. In the subsequent debugging phase, relevant personnel need to focus on the quality status of these equipment to reduce the possibility of the same quality problem occurring again on the same equipment or the possibility of other quality problems arising from the same equipment. Pushing query results to report generation tools, data analysis platforms, etc. for further data analysis will help understand the scope of quality problems and the status of related entities, expand the trade-offs in the analysis of related factors in a graphical way, support decision-making and the implementation of improvement measures, and make collaborative quality control more accurate, comprehensive and efficient.
[0156] Experience feedback pedigree knowledge statistics and prevention
[0157] By clustering the same label class recorded in the experience feedback graph, we can obtain the experience feedback knowledge pedigree used to characterize the knowledge base in a specific field. Based on the experience feedback knowledge pedigree, various types of experience knowledge can be quickly accessed and retrieved, and experience feedback engineers do not need to spend a lot of time and energy to search and collect.
[0158] For example, enter the command: "MATCH(n:gz)RETURN n"
[0159] Get the fault knowledge spectrum as Fig.11 As shown in the figure, by counting the number of quality events in the project and analyzing the most frequent failures of the same type, it is helpful to identify potential risks and early warning signals, and take corresponding preventive measures to reduce the occurrence of quality problems.
[0160] Key knowledge management and statistics based on experience feedback spectrum provide users with powerful search and filtering functions, which help to reveal problems and trends in the nuclear power construction process and help managers identify common quality events, failure modes and risk factors, thereby providing key insights to assist decision makers and teams to better understand and respond to problems.
[0161] Case Studies
[0162] Integrating experience feedback graphs and deep knowledge mining functions into the experience feedback system of nuclear power companies provides original text records, graph display, association retrieval and data analysis functions to assist managers in making nuclear power construction decisions, transforming traditional decisions based on manual experience into knowledge-driven intelligent decision-making models, and effectively improving the interpretability of experience feedback information in the nuclear power construction process.
[0163] Take the reactor pressure vessel (RPV) as an example. RPV is one of the main equipment of nuclear power. It has many parts and its quality status is crucial to the safety of nuclear power plants. During the installation or commissioning process, when a component of the reactor pressure vessel has a quality problem, it can be recorded in the project experience feedback system by on-site personnel, and the text library can be intelligently parsed through multi-dimensional cascade knowledge extraction. A visual experience feedback map can be generated in real time in the "map display" function module. Based on a unified and standardized map, relevant managers can query instructions through the inference engine and conduct in-depth mining on the reactor pressure vessel in the "association retrieval" module to obtain the associated knowledge of the occurrence of RPV failures. Fig.12 As shown. Managers can quickly learn structured associated knowledge from the RPV experience feedback map, that is, in the quality events recorded in the history of this project, five different types of failures occurred in seven reactor pressure vessel-related equipment (including sub-equipment), including the insulation support body, workstation monitoring software, top cover assembly, and nozzle seat, belonging to Units 3 and 4, and the root cause of the failure and corresponding measures can be obtained. Further, through the "data analysis" function module, statistical information on the failure occurrence stage, the types of parts that often have quality problems, the response measures to past quality events, the personnel involved, etc. can be obtained, and statistical reports can be generated and fed back to the relevant enterprise departments involved, so as to achieve the purpose of accurate transmission of experience feedback information, thereby efficiently, scientifically, and comprehensively planning and formulating nuclear power equipment construction plans to ensure efficient and safe coordinated quality control of nuclear power equipment. For specific application processes, see Fig.13 .
[0164] This method focuses on the needs of collaborative quality control of nuclear power equipment. Aiming at the problems of weak timeliness of experience feedback between enterprises, inconsistent knowledge cognition of experience feedback among all parties, and low reuse rate of experience feedback information, an experience feedback map with a unified data model is constructed based on the text data of quality events. The characteristics of experience feedback information in the nuclear power construction process and the difficulties faced under the "people-oriented" approach are analyzed, and it is proposed to turn experience feedback into knowledge. The construction scheme of key knowledge domains of experience feedback is discussed in detail. For the data mining task in the experience feedback map, a multidimensional information perception model is designed, which makes full use of context and key semantic features, improves the mining effect of named entity recognition and relationship extraction in the nuclear power field, and is verified by data on the actual experience feedback information of a nuclear power construction enterprise, forming a visual map and conducting deep knowledge mining. Finally, a case study is conducted on the reactor pressure vessel to obtain its associated knowledge and statistical information, providing intelligent and refined quality knowledge management tools for relevant quality assurance personnel to assist management decision-making, which is helpful for nuclear power enterprises to deeply reuse quality data resources and break the barriers between quality information of distributed enterprises in different locations, thereby improving the quality and efficiency of nuclear power construction.
[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the technical solution. Those skilled in the art should understand that those modifications or equivalent substitutions of the technical solution of the present invention that do not depart from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.
Claims
1. A nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction, characterized in that: The following steps are involved: S1. Obtaining experience feedback information on quality incident records generated internally during the entire life cycle of nuclear power construction; S2. Analyze the characteristics of experience feedback information to obtain key knowledge categories, hierarchically associate and couple various key knowledge categories, define the organizational structure of the experience feedback key knowledge domain, and obtain the ontology model of the experience feedback map; S3, based on the model architecture of the ontology model of S2, construct a multi-dimensional cascade knowledge extraction model to obtain the filling data of the ontology model from the experience feedback information; the multi-dimensional cascade knowledge extraction model includes a knowledge mining model and a relationship extraction model; the knowledge mining model is used to perceive and capture each category of key knowledge in the experience feedback information; the relationship extraction model is used to extract the relationship characteristics of each key knowledge in the experience feedback information; S4, using the multi-dimensional knowledge extraction model built by S3, extracts structured filling data from the experience feedback information obtained by S1, and maps the extracted filling data to the ontology model of S2, generates the corresponding experience feedback map, and stores it in the knowledge base; S5. When a quality incident occurs during the nuclear power construction process, the knowledge base is searched to obtain the required knowledge to assist nuclear power professional and / or management personnel in analysis and processing.
2. The nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction according to claim 1, characterized in that: S2 includes: Analyze the characteristics of experience feedback information to obtain key knowledge categories related to fault manifestations; Each knowledge category is formed into a knowledge layer of the key knowledge domain of experience feedback, and the domain knowledge spectrum is obtained; at the same time, the association coupling is realized through the interactive relationship between each knowledge layer, and the spatial knowledge structure with semantic association and logical reasoning ability is obtained as the organizational structure of the key knowledge domain of experience feedback; Afterwards, the class and object attributes are constructed for the different levels of knowledge in the key knowledge domain of experience feedback, as well as the associations between the layers, so as to establish the semantic relationships between node types and concepts or instances, obtain the knowledge form with the <entity-relationship-entity> structure, and explicitly represent the key knowledge domain of experience feedback to form an ontology model of the experience feedback map.
3. The nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction as claimed in claim 2, characterized in that: The knowledge categories include the equipment information, causes, countermeasures, time of occurrence, stage, related equipment, unit, main responsible personnel, and enterprise departments of the fault; the association relationships include cause-and-effect relationships, dependency relationships, hierarchical relationships, and time relationships.
4. The nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction according to claim 1, characterized in that: In S3, the knowledge mining model is a multi-layer neural network model that integrates the attention mechanism and is built using a cascade strategy; the knowledge mining model includes a word embedding layer, a knowledge perception layer, and a knowledge prediction layer that are set in sequence.
5. The method for nuclear power construction experience feedback based on multi-dimensional cascade knowledge extraction as claimed in claim 4, characterized in that: The word embedding layer is designed based on Transformer's bidirectional encoder representation technology BERT; the working process of the word embedding layer includes: the input has three different embedding layers, which respectively represent the word information, sentence paragraph information and sentence position information of the input quality event text, and the three-layer information vector is concatenated as the input of BERT, and the output vector containing semantic features of multiple dimensions is obtained through multi-layer Transformer.
6. The method for nuclear power construction experience feedback based on multi-dimensional cascade knowledge extraction according to claim 4, characterized in that: The knowledge perception layer is constructed based on the bidirectional long short-term memory network BiLSTM fused with the attention mechanism Attention; the bidirectional long short-term memory network BiLSTM is used to bidirectionally mine the historical information and future information in the quality event text segment, and comprehensively connect the long-term features; The attention mechanism is used to calculate the degree of correlation between word features in the experience feedback information, so that the knowledge perception layer can focus on areas related to nuclear power when processing information to obtain effective semantic information.
7. The nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction as claimed in claim 4, characterized in that: The knowledge prediction layer is built based on conditional random fields and is used to optimize the output results of the knowledge perception layer. During prediction, the emission matrix and the transfer matrix are combined to calculate the highest-scoring annotation sequence corresponding to the sentence label through the state score and transfer score output by the upper layer as the corresponding key knowledge.
8. The method for nuclear power construction experience feedback based on multi-dimensional cascade knowledge extraction according to claim 1, characterized in that: The relation extraction model is optimized by the segmented convolutional neural network PCNN based on multi-instance learning, including: A vector expression layer, used to convert the input key knowledge into an embedding vector expression matrix; the embedding vector expression matrix includes a word embedding vector and a position embedding vector, the word embedding vector is used to indicate the text words in the key knowledge, and the position embedding vector is used to indicate the position information of each text word in the key knowledge; The convolution layer is used to extract local features of the embedded vector expression matrix by region to obtain local feature maps of different regions; The segmented maximum pooling layer is used to perform segmented maximum pooling operations on the local feature maps of different regions to obtain the pooled maximum value feature maps corresponding to different regions; The fully connected layer is used to concatenate the pooled maximum feature maps corresponding to different regions, and then use the Gaussian error linear unit GELU as the activation function to obtain the fully connected feature map of key knowledge; The classification layer is used to make classification decisions and extract relationships on the fully connected feature graph of key knowledge, obtain the relationship features presented in the text information in the key knowledge, and output them.
9. The method for nuclear power construction experience feedback based on multi-dimensional cascade knowledge extraction as claimed in claim 8, characterized in that: The activation function expression of the Gaussian error linear unit GELU is: Among them, x represents the input of the activation function, and π is the ratio of pi.
10. The nuclear power construction experience feedback method based on multi-dimensional cascade knowledge extraction according to claim 1, characterized in that: In S5, the knowledge base is searched through the reasoning engine to obtain the required knowledge.
Citation Information
Patent Citations
Transformer operation and maintenance processing system based on knowledge graph
CN114943396A
Knowledge graph construction method and device based on document relationship extraction
CN115269857A
Emergency plan text extraction method and system based on natural language processing
CN116108170A
Nuclear power plant maintenance experience feedback knowledge base construction method based on knowledge graph
CN116109289A
Nuclear power plant operation event report knowledge graph construction and application method
CN116127021A
Cited By
Nuclear power DCS signal distribution method, device, equipment, medium and product
CN119322501A
Knowledge graph construction system and method for water conservancy and hydropower engineering
CN120671788A
Metro operation accident knowledge graph construction method and system, terminal and storage medium
CN121189452A
