Nuclear power construction quality formation characterization method

By building a collaborative mass flow chart and BFAC model, the redundancy and silo problems of quality data management in nuclear power construction are solved, the knowledge and clear traceability of quality data are realized, and the safety and information sharing of the construction process are improved.

CN118982273APending Publication Date: 2024-11-19CHONGQING UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410998540.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

During the construction of nuclear power, there are data redundancy, multi-source heterogeneity, and data island problems in quality data management, which makes it difficult to clearly and traceable for the construction quality formation process and cannot meet high safety requirements.

Method used

A method of characterizing nuclear power construction quality, collecting multi-source heterogeneous quality data, constructing a collaborative quality metadata model, using BiLSTM-CRF model and multi-head attention mechanism, constructing a BFAC model, and performing instance extraction of unstructured data and Knowledge mapping, combining Neo4j graph database to store and visualize quality knowledge, and form a collaborative quality flow diagram.

Benefits of technology

The knowledge of quality data is realized, the accuracy of learning key semantic features of quality instances is improved, multi-source heterogeneous data is integrated, information islands are broken, and a clear and traceable quality knowledge base is formed, supporting the shared use of collaborators.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982273A_ABST
    Figure CN118982273A_ABST
Patent Text Reader

Abstract

The invention discloses a nuclear power construction quality formation characterization method, which comprises the following steps: collecting and storing quality data of different morphological characteristics in multi-source heterogeneous nuclear power construction according to collaborative business activities in nuclear power construction quality formation; orderly organizing the quality data to form a collaborative quality metadata model; according to the method, a collaborative quality element is used as a data source, a business layer ontology model is constructed through a quality knowledge triple, a BFAC model is proposed to carry out quality instance extraction on quality data and unstructured data, knowledge mapping is carried out on structured data in a business platform based on an R2RML method, instantiated knowledge filling of a business layer ontology is completed, and the business layer ontology is obtained. And a Neo4j graph database is utilized to fuse storage quality knowledge, and a collaborative quality flow graph is constructed. According to the method, nuclear power construction quality formation is effectively represented and analyzed, so that integration and utilization of multi-source heterogeneous quality data resources are promoted, and the high requirement that nuclear power construction is clear and traceable is met.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear power intelligent construction quality control, and in particular to a method for characterizing nuclear power construction quality. Background Art

[0002] As a clean, low-carbon, high-energy and high-efficiency high-quality energy, nuclear power has attracted much attention from all over the world. Especially in today's severe situation of global warming and increasing resource shortages, the active and orderly development of nuclear power is of great significance for building a new energy system, ensuring energy supply security and helping to achieve environmental protection goals. However, nuclear power development also has certain risks. For example, nuclear power plant accidents may lead to leakage of radioactive materials and nuclear pollution, posing a profound and serious threat to the surrounding environment and human health. Therefore, nuclear power technology is a "double-edged sword". It can only benefit mankind under the premise of fully ensuring safety, and ensuring the safety of nuclear power needs to be based on the material foundation of good construction quality of nuclear power equipment.

[0003] Under the organization and implementation of the engineering general contracting, nuclear power construction is manifested as a value chain collaboration model with a leading enterprise, multiple follower enterprises and N participating enterprises. The enterprises have established an industrial cluster value chain strategic collaboration relationship under the control of the group. Under the leadership of the general contracting unit, the follower and participating enterprises undertake the construction outsourcing business, which presents the characteristics of small batch and discrete production in the macro sense. However, due to the high safety and joint responsibility of nuclear power, under the joint influence of national legal and policy orientation and industrial characteristics, its construction quality control cannot be completed independently by the outsourcing unit alone, but adopts a collaborative quality control model. Different from the general quality management that only focuses on whether the final "drawing requirements" are met and the accountability after the problem occurs, this model follows the principle of "defense in depth" and adopts multi-level supervision and collaborative quality witnessing to monitor quality in practice. Through cross-enterprise multi-party collaborative participation, the items, services and various activities that affect the quality of nuclear power construction are supervised and verified in the "whole process" according to the quality assurance outline, ensuring that each link is placed under the defense of overlapping measures, providing multi-level guarantees for the achievement of nuclear safety-level quality assurance requirements.

[0004] However, in collaborative quality control, on the one hand, with the development of cross-enterprise and multi-party collaborative quality business, the multi-dimensional interaction of quality elements such as "people, machines, materials, methods, environment, and measurement", the cascade coupling of quality activities between different subjects such as team level, department level, and enterprise level, the reciprocating iteration of multi-level quality control such as primary, rework, and maintenance, and the complexity of nuclear power as a giant system itself, make its construction-related coupling relationship more and more complex. In addition, as its construction and service process progresses, new collaborative quality business activities are constantly intertwined, and the above problems will be further aggravated, which will eventually lead to the obscurity of the quality of nuclear power construction and the difficulty in meeting the high safety requirements of clear and traceable entire construction process.

[0005] On the other hand, as my country's nuclear power is transforming and upgrading towards "intelligent construction", new generation technologies such as the Internet of Things and cloud computing are being rolled out throughout the nuclear power industry. Considering that data on quality business activities recorded using information technology under intelligent construction technology is an indispensable basic data for analyzing quality formation. However, the management of nuclear power construction quality data is still in the initial stage of emphasizing records, and there are problems such as data redundancy, multi-source heterogeneity, and data islands. This is another important reason why the process of nuclear power construction quality formation is difficult to be "clearly traceable".

[0006] Therefore, how to efficiently characterize and analyze quality formation in the above-mentioned complex business scenarios of nuclear power construction and in the context of digital transformation and intelligent improvement has become an important issue that needs to be urgently addressed in the collaborative quality control of nuclear power construction. Summary of the invention

[0007] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for characterizing the quality formation of nuclear power construction, which is used to effectively characterize and analyze the quality formation of nuclear power construction, thereby promoting the integrated utilization of multi-source heterogeneous quality data resources and ensuring the high requirements of clear and traceable nuclear power construction.

[0008] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0009] A method for characterizing the quality of nuclear power construction comprises the following steps:

[0010] S1. According to the collaborative business activities in the formation of nuclear power construction quality, collect and store quality data of different forms and characteristics in multi-source heterogeneous nuclear power construction, wherein the quality data of different forms and characteristics include unstructured quality data, semi-structured quality data and structured quality data;

[0011] S2, organizing the quality data in an orderly manner to form a collaborative quality metadata model;

[0012] S3. Based on the collaborative quality metadata model, summarize the quality elements of the collaborative quality metadata model and define the relationships and attributes involved in quality formation, and build a business layer ontology model;

[0013] S4. Based on the BiLSTM-CRF model, by integrating the multi-head attention mechanism and the fast gradient ascent algorithm, a BFAC model based on key semantic feature learning reinforcement and imbalanced sample adversarial enhancement is constructed;

[0014] S5. Based on the BFAC model, extract instances of the unstructured quality data and the semi-structured quality data to obtain quality instances that match the business layer ontology, and store them as structured data;

[0015] S6. Based on the R2RML method, the structured data after instance extraction is mapped to the collected structured quality data to obtain the business layer ontology instantiation knowledge;

[0016] S7. The business layer ontology and the business layer ontology instantiated knowledge are integrated with quality knowledge, and quality knowledge is stored and visualized based on the Neo4j graph database to obtain a visualized collaborative quality flow diagram for nuclear power construction, thereby completing the characterization of the quality formation of nuclear power construction.

[0017] As a preferred solution, in step S2, the collaborative quality metadata model includes static quality data resources, process quality data resources and result quality data resources;

[0018] The static quality data resources include basic information, inherent attributes and guiding quality documents;

[0019] The process quality data resources include dynamic information, dynamic attributes and process flow information;

[0020] The result quality data resource includes a result record quality file.

[0021] As a preferred solution, in step S3, the specific process of summarizing the quality elements of the collaborative quality metadata model and defining the relationships and attributes involved in quality formation is:

[0022] S301. According to the data resources in the collaborative quality metadata model, the quality elements are classified into static quality elements, process quality elements and result quality elements, and the hierarchical relationship of the quality elements is established by using the classes and subclasses in the network ontology language;

[0023] S302, using the data attributes in the network ontology language to set corresponding attributes for each quality factor;

[0024] S303: According to the quality factors and the corresponding attributes of each quality factor, using the object attributes in the network ontology language, construct the intra-process relationship and the extra-process relationship.

[0025] As a preferred solution, in step S303, the intra-process relationship is used to construct the relationship between the quality elements of the process, including personnel quality relationship, item quality relationship and document quality relationship, according to the logic of the output in the process and the on-site business logic;

[0026] The external process relationships are used to establish logical connections and synergies between different processes, including process flow relationships and structural relationships.

[0027] 4. The method for characterizing the quality of nuclear power construction according to claim 1 is characterized in that, in step S4, the specific processing process of constructing the BFAC model by integrating the multi-head attention mechanism and the fast gradient ascent algorithm is:

[0028] S401, the non-structural quality data is used as the input of the Embedding layer for vector conversion and used as the original sample.

[0029] S402, using the BiLSTM layer to perform preliminary learning on the original sample and extract the semantic feature matrix of the context;

[0030] S403, using the semantic feature matrix as the input of the multi-head attention mechanism, and concatenating the outputs of all attention heads to obtain a key semantic information matrix;

[0031] S404, inputting the key semantic information matrix into the CRF layer, performing sequence labeling, and obtaining labeled target instances;

[0032] S405. By using the fast gradient ascent algorithm to add perturbations to the Embedding layer vector of the original sample, an adversarial sample is generated. The original sample and the adversarial sample are used as inputs of the BFAC model to perform adversarial training on the BFAC model. The BFAC model parameters are updated through an iterative process to obtain the final trained BFAC model.

[0033] 5. The method for characterizing nuclear power construction quality according to claim 4, characterized in that in step S403, the calculation formula of each attention head is as follows:

[0034]

[0035] Head i =Attention(QW i Q ,KW i K ,VW i V );

[0036] Where Q, K, and V are the semantic vectors after the semantic feature matrix is ​​mapped, and W i Q , W i K , W i V are the weight matrices learned during mapping, d k is the dimension of the input vector; the subscript i represents the i-th attention head in the multi-head attention mechanism, and i = 1, 2, …, h, h represents the number of attention heads in the multi-head attention mechanism;

[0037] The calculation formula of the key semantic information matrix is ​​as follows:

[0038] MultiheadATT=Concat(Head0,...,Head h )W O ;

[0039] Where W O is the learnable weight matrix when concatenating attention heads.

[0040] 6. The method for characterizing the quality of nuclear power construction according to claim 4, characterized in that, in step S405, the adversarial training of the BFAC model and updating the BFAC model parameters through an iterative process to obtain the final trained BFAC model are specifically carried out as follows:

[0041] S4051, initializing the Embedding layer parameters;

[0042] S4052, perform normal forward propagation and loss calculation on the original samples of the current batch, and then perform back propagation to calculate the network back propagation parameter gradient of the tth step;

[0043] S4053, generating a sample perturbation of the tth step according to the calculated network back propagation parameter gradient of the tth step;

[0044] S4054, adding the sample perturbation of the tth step to the Embedding vector of the original sample to obtain an adversarial sample;

[0045] S4055, using the adversarial sample to perform forward propagation and loss calculation, and then performing back propagation to obtain the adversarial training gradient, and accumulating the adversarial training gradient to the network back propagation parameter gradient of the tth step;

[0046] S4056, restore the Embedding layer parameters, and use the accumulated gradients to update the network parameters;

[0047] S4057. Repeat steps S4051 to S4056. When the number of iterations is equal to or greater than the preset number of iterations, end the loop.

[0048] 7. The method for characterizing the quality of nuclear power construction according to claim 6, characterized in that, in step S4052, the expression for calculating the gradient of the network back propagation parameter in the t-th step is:

[0049] g0=▽ x L(θ,x,y);

[0050] Where g0 is the network back propagation parameter gradient at step t,x is the gradient of the original sample x, L is the loss function, θ is the network parameter, x is the original sample, and y is the true label.

[0051] 8. The method for characterizing nuclear power construction quality according to claim 6, characterized in that, in step S4053, the expression for generating the sample disturbance of step t is:

[0052] r adv =ε·g0 / ||g0||2;

[0053] In the formula, r adv is the sample disturbance at step t, and ε is the disturbance factor.

[0054] 9. The method for characterizing nuclear power construction quality according to claim 6, characterized in that, in step S4054, the expression of the adversarial sample is:

[0055] x adv =x+r adv

[0056] In the formula, x adv is an adversarial sample, and x is the original sample.

[0057] Compared with the prior art, the present invention has the following technical effects:

[0058] (1) The present invention organizes multi-source heterogeneous quality data in an orderly manner to form a collaborative quality metadata model. By taking collaborative quality metadata as the data source and using quality knowledge triples, a collaborative quality flow diagram is constructed to characterize the collaborative interaction relationship between personnel, items and documents, thereby realizing the knowledge-based quality data and characterizing the formation of nuclear power construction quality.

[0059] (2) The present invention integrates the multi-head attention mechanism with the fast gradient ascent algorithm to construct a BFAC model based on key semantic feature learning reinforcement and imbalanced sample adversarial enhancement. This solves the problem that in the construction of collaborative quality flow graphs, the lack of quality instance corpus extracted from quality data and the imbalance of labels limit the model learning ability. It improves the learning ability of key semantic features of quality instances and ensures the accuracy of model extraction.

[0060] (3) The present invention analyzes quality formation through the Neo4j graph database, integrates multi-source heterogeneous data, breaks down information silos, and forms a unified quality knowledge base. At the same time, it performs semantic query and visualization on the nodes and relationships in the collaborative quality flow graph to achieve shared use by all collaborative parties. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] 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:

[0062] Figure 1 A flow chart of a method for characterizing the quality of nuclear power construction is disclosed for the present invention;

[0063] Figure 2 This is a schematic diagram of the main quality business process of steel lining welding of the present invention and the data generated;

[0064] Figure 3 A schematic diagram of the characterization connotation and method of nuclear power construction quality based on quality data knowledge in the present invention;

[0065] Figure 4 A schematic diagram of a coordinated mass flow diagram for nuclear power construction of the present invention;

[0066] Figure 5 Constructing a flow chart for the collaborative mass flow diagram of the present invention;

[0067] Figure 6 It is a schematic diagram of quality factors related to the steel lining welding business of the present invention;

[0068] Figure 7 It is a graphical schematic diagram of the steel lining welding business layer ontology model of the present invention;

[0069] Figure 8 This is the BiLSTM-FGM-Attention-CRF model structure diagram of the present invention;

[0070] Fig. 9 This is the structure diagram of the multi-head attention mechanism of the present invention;

[0071] Fig.10 A schematic diagram of information transmission for cross-platform quality formation of the present invention;

[0072] Fig.11 A visual representation diagram is formed for the quality defects of the present invention;

[0073] Fig.12 This is a schematic diagram of data related to steel lining welding quality activities in an embodiment of the present invention;

[0074] Fig.13 This is a schematic diagram of the steel lining welding business layer ontology model in an embodiment of the present invention;

[0075] Fig.14 A comparison chart of the effects of various models is extracted for the NCR text quality example of steel lining welding in the embodiment of the present invention;

[0076] Fig.15 A schematic diagram of a quality knowledge triple set instantiated for the lining welding business layer ontology according to an embodiment of the present invention;

[0077] Fig.16 It is a partial diagram of the coordinated mass flow diagram of a weld of a steel lining according to an embodiment of the present invention;

[0078] Fig.17 This is a schematic diagram of cross-platform review of a steel lining welding inspection report according to an embodiment of the present invention;

[0079] Fig.18 The figure is a schematic diagram for visualizing the quality defects of the steel lining truncated cone weld (H1J***006) in an embodiment of the present invention. DETAILED DESCRIPTION

[0080] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but only represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0081] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0082] In the existing technology, quality data is only a series of identifiable symbols and their combinations obtained by observing and recording nuclear power construction quality activities using various tools and means. Most of the current research focuses on the processing of manufacturing data and its knowledge service application, but no research has yet proposed the connotation, characteristics and implementation methods of nuclear power construction quality formation representation based on quality data knowledge. Faced with the complex scenario of cross-enterprise, cross-platform and cross-domain collaborative control of nuclear power construction, the special requirements of clear and traceable nuclear safety regulations, and the unique characteristics of the field of quality data with a large span, mixed sources, long retention and strict confidentiality, there has been no research on the use of knowledge graphs to represent the formation of nuclear power construction quality, and there is a lack of effective forms of knowledge acquisition, organization and management in the field of nuclear power construction quality formation. In view of the above problems and deficiencies, the present invention discloses a method for characterizing the formation of nuclear power construction quality. Based on the analysis of the collaborative quality control business and data characteristics of nuclear power construction, the method proposes a method for characterizing the formation of nuclear power construction quality based on quality data knowledge, and constructs a collaborative quality element and a collaborative quality flow diagram; the BFAC model is proposed to extract quality instances from the unstructured data of quality text, and the structured data in the business platform is mapped based on the R2RML method to complete the instantiation of knowledge filling of the business layer ontology; a quality knowledge fusion storage and presentation method based on the Neo4j attribute graph mode is proposed to construct a collaborative quality flow diagram. Through the collaborative quality flow diagram, the quality formation of nuclear power construction can be analyzed, and the quality formation of the concerned point can be visualized, and the quality formation information can be transmitted across platforms.

[0083] Specifically, if Figure 1 As shown, a method for characterizing the quality of nuclear power construction proposed by the present invention comprises the following steps:

[0084] S1. According to the collaborative business activities in the formation of nuclear power construction quality, collect and store quality data of different forms and characteristics in multi-source heterogeneous nuclear power construction, wherein the quality data of different forms and characteristics include unstructured quality data, semi-structured quality data and structured quality data;

[0085] S2, organizing the quality data in an orderly manner to form a collaborative quality metadata model;

[0086] S3, summarizing the quality elements of the collaborative quality metadata model and defining the relationships and attributes involved in quality formation, and constructing a business layer ontology;

[0087] S4. Based on the BiLSTM-CRF model, by integrating the multi-head attention mechanism and the fast gradient ascent algorithm, a BFAC model based on key semantic feature learning reinforcement and imbalanced sample adversarial enhancement is constructed;

[0088] S5. Based on the BFAC model, extract instances of the unstructured quality data and the semi-structured quality data to obtain quality instances that match the business layer ontology, and store them as structured data;

[0089] S6. Based on the R2RML method, the structured data after instance extraction is mapped to the collected structured quality data to obtain the business layer ontology instantiation knowledge;

[0090] S7. The business layer ontology and the business layer ontology instantiated knowledge are integrated with quality knowledge, and quality knowledge is stored and visualized based on the Neo4j graph database to obtain a visualized collaborative quality flow diagram for nuclear power construction, thereby completing the characterization of the quality formation of nuclear power construction.

[0091] The method for characterizing the quality of nuclear power construction of the present invention is described in more detail below.

[0092] 1. Analysis of collaborative quality control business and data characteristics of nuclear power construction

[0093] The quality control of nuclear power construction has different levels for equipment with different nuclear safety levels and quality assurance levels. The higher the level, the stricter the control. In the present invention, the welding construction of the containment steel lining, which is the highest quality assurance level of nuclear power plant protective equipment, is taken as a specific embodiment. Welding is the main way to join steel linings and is also one of the frequently used processing techniques in nuclear power construction. The quality of the weld directly determines the overall quality of the containment, which is related to whether it can play a key role in sealing and shielding radioactive substances under extreme conditions. Figure 2 As shown, with the implementation of QC control work at all levels, the quality process is recorded synchronously in accordance with the requirements of HAF003 and the quality assurance outline, and a large amount of quality data is generated, thus forming a comprehensive and complete weld construction quality document history, which is stored in various business platforms to achieve the quality control requirements of having rules to rely on, traces to follow, and evidence to check.

[0094] However, in a cross-enterprise, cross-platform, and cross-domain collaborative construction environment, nuclear power generates massive amounts of quality data, presenting some complex characteristics:

[0095] (1) Multi-source heterogeneity. Data comes from complex collaborative businesses. Due to the differences in recording standards and usage methods of various collaborative entities, there are structured data such as inspection reports and personnel information records, semi-structured data such as quality control sheets and repair records, and unstructured data such as non-conformance documents recorded in natural language.

[0096] (2) Data redundancy. The data is large in scale, the degree of standardization needs to be improved, and the redundancy of records is high. Figure 1There are many overlapping data items between the weld repair record sheet and the repair control sheet, which not only greatly wastes storage resources but also causes data conflicts.

[0097] (3) Discrete storage. Most existing data is stored by system or department. Information is distributed in a scattered manner and has weak relevance. Data islands exist, which greatly hinders the integration of quality information and its efficient use.

[0098] The present invention divides quality data into three types according to its morphological characteristics: unstructured quality data, semi-structured quality data and structured quality data;

[0099] According to the role of quality data, it can be divided into three types:

[0100] Static quality data refers to quality data, information, documents, etc. that do not change with quality activities over a period of time. Such as basic information such as the name, number, and qualifications of personnel, inherent attributes such as the model and number of materials and equipment, and guiding quality documents such as applicable documents.

[0101] Process quality data refers to dynamic quality data, information, and documents generated as quality activities are carried out, such as dynamic information such as personnel operation time and plan changes, and dynamic attributes such as item usage time and consumables status.

[0102] Result quality data refers to the result-oriented quality data, information, and documents generated in quality activities, such as test results, rework notices, quality non-conformity reports, completion reports, etc.

[0103] 2. Knowledge-based nuclear power construction quality data

[0104] At present, the massive quality data from multiple sources and heterogeneity are only piled up in various data warehouses and platforms. The patterns are rigid and the information is redundant, which cannot directly show the process and details of the formation of nuclear power construction quality. Therefore, it is proposed to make quality data knowledgeable in order to effectively characterize and analyze the formation of quality and promote the value of quality data resources.

[0105] like Figure 3 As shown, in the present invention, the multi-source heterogeneous quality data collected and stored in the quality activities are orderly organized and processed, effective information that can reflect the facts of quality formation is extracted, and organized in a certain form into quality knowledge that can characterize the quality formation pattern, rules, and laws.

[0106] Considering that the process is the basic unit of quality formation, it integrates quality elements such as "people, machines, materials, methods, environment, and measurement", and is the source of output of various quality data. To this end, the present invention proposes a "collaborative quality element" to describe the process quality data model and a "collaborative quality flow diagram" for the flow of quality data between processes, in order to extract knowledge from quality data, visualize the effective information therein, and characterize the quality formation process of nuclear power construction. The following will introduce them in detail.

[0107] 2.1. Constructing a collaborative quality metadata model

[0108] Collaborative quality element refers to the set D of quality elements and data such as personnel, items, and documents that interact collaboratively with the process as the basic unit in collaborative quality control. The collaborative quality element D for the i-th process is i Describe the following expression:

[0109] D i ={s i ,d i ,r i |Obj}, i=1,2,...,n;

[0110]

[0111] r i =F i r ;

[0112] Where: Obj is the quality control object, such as specific process entities such as welds or equipment entities at each level; s i It is a collection of static quality data and its related elements; F i s It is a collection of guiding quality documents; It is a set of inherent attributes of personnel; is the set of inherent attributes of the item; d i It is a collection of process quality data and its related elements; i Gathering for personnel; is a set of dynamic attributes of personnel; i is a collection of items; is the dynamic attribute set of the item; γ i is the process flow information set; i is the result quality data and its related elements; F i r A collection of quality documents to record the results.

[0113] According to the above collaborative quality element D i Expression,classifies the collaborative quality elements into static quality data resources,,process quality data resources and result quality data resources;

[0114] The static quality data resources include basic information, inherent attributes and guiding quality documents;

[0115] The process quality data resources include dynamic information, dynamic attributes and process flow information;

[0116] The result quality data resource includes a result record quality file.

[0117] 2.2. Define collaborative quality flow diagram

[0118] Collaborative quality flow diagram refers to a form of nuclear power quality formation knowledge organization and management that takes collaborative quality element as the basic unit, extracts ordered information that can characterize quality formation from the quality elements and data sets it encapsulates, and associates them using quality knowledge triples. It depicts the collaborative interaction between personnel, items, and documents, as well as quality formation facts such as process flow, in order to reveal a clear and traceable collaborative quality control chain.

[0119] The knowledge organization method under the collaborative quality flow diagram framework is the quality knowledge triple. The data source is the collaborative quality element. The quality data knowledge needs to be completed through the construction of the business layer and its instantiation. The final presentation form is a directed semantic graph, such as Figure 4 shown.

[0120] The quality knowledge triples are endowed with directed semantic symbols that represent the activities, patterns, facts, etc. that characterize the formation of quality. They are a formal expression of quality knowledge that transforms quality data into quality knowledge through quality business logic. The expression of the quality knowledge triples is:

[0121] K = (dom, r, ran);

[0122] In the formula: dom is the semantic starting point, r is the relationship, and ran is the semantic end point.

[0123] In the collaborative quality flow diagram, four types of quality knowledge are set up: feature class, instance class, instance-feature class, and attribute class, as shown in Table 1.

[0124] Table 1 Quality knowledge types

[0125]

[0126] The quality business logic is an abstract distillation of the quality elements involved in the process and the relationships between them. In the collaborative quality flow diagram, the business layer is used for knowledge description. On the basis of the collaborative quality element, combined with the characteristics of the control object, the business logic of collaborative quality witnessing, quality inspection, rework and rectification, process flow, etc. is described through the quality knowledge triple, providing rules and framework for the subsequent characterization of quality facts.

[0127] Quality activity facts are the concrete manifestation of quality instances such as quality objects, processes, personnel, items, documents and their relationships under business logic. In the collaborative quality flow diagram, knowledge is described through instantiation at the business layer. Under the predefined rules at the business layer, instances are extracted from the quality data organized by the collaborative quality element and organized into a set of triples to obtain a series of quality knowledge that represents the objective facts of quality formation.

[0128] 3. Build a collaborative quality flow diagram

[0129] The construction of collaborative quality flow graphs requires the extraction of knowledge from multi-source heterogeneous quality data stored across platforms. The knowledge graph integrates a series of knowledge modeling and extraction technologies such as ontology models and data mining, which can better integrate multi-source heterogeneous quality data, represent objective facts through graph structure semantic networks, and better realize the knowledge construction and organizational management of quality data.

[0130] like Figure 5 As shown, the construction of the collaborative quality flow diagram includes: business layer ontology construction, business layer ontology instantiation knowledge filling and quality knowledge fusion after visualization map presentation;

[0131] The business layer ontology construction is a bridge for transforming quality data into knowledge. It is used to summarize quality elements and define the relationships and attributes between them using the ontology model, and to construct a business layer that can characterize the collaborative quality activity model and logic.

[0132] The business layer ontology instantiation knowledge filling is used to extract unstructured data quality instances based on the BFAC model and store them in a relational database to achieve structured governance. The business layer-driven R2RML mapping method is used to transform the quality instances extracted in the previous step and the structured data stored in the business platform into quality knowledge that can characterize the facts of quality formation.

[0133] The quality knowledge is fused and presented in a visualized graph, which is used to fuse the business layer ontology and its instantiated quality knowledge, resolve the semantic conflicts therein, and complete the storage and visualization of quality knowledge based on the Neo4j graph database under the attribute graph data mode, and finally obtain a visualized presentation form of the nuclear power construction collaborative quality flow diagram.

[0134] Among them, in the instantiation knowledge filling of the business layer ontology, the mapping of structured data to quality knowledge based on the R2RML method and the visualization map presentation after the quality knowledge fusion, the knowledge fusion and visualization map presentation based on Neo4j can all be processed using standard knowledge map methods. The following is a detailed introduction to the construction of the business layer ontology and the extraction of unstructured data quality instances based on the BFAC model.

[0135] 3.1. Business layer ontology construction

[0136] In this embodiment, the formation of steel lining welding quality is taken as the research object, and various types of quality elements such as static, process, and result organized in the collaborative quality metadata model are used as data sources. The quality elements involved in the formation of steel lining welding quality, the relationship between quality elements, and the attributes of quality elements are analyzed and defined. The various parts of the business layer ontology model are preliminarily determined, and the business layer ontology model is constructed using the improved skeleton method.

[0137] (1) Analysis and definition of business layer ontology

[0138] The specific process of analyzing and defining the quality factors involved in the formation of steel lining welding quality, the relationship between the quality factors, and the attributes of the quality factors is as follows:

[0139] S301. Based on the data resources in the collaborative quality metadata model, the quality elements are classified into static quality elements, process quality elements and result quality elements, and the hierarchical relationship of the quality elements is established using the classes and subclasses in the network ontology language.

[0140] In this embodiment, under the collaborative quality metadata model, the quality elements that can characterize the welding business logic and rules are summarized according to the quality control objects, personnel, items, static files, and result record files, and the hierarchical relationship of the quality elements is established using the class and subclass in the network ontology language. Figure 6 Shown are the sorted quality factors related to steel lining welding business.

[0141] S302: Using the data attributes in the network ontology language, the quality factors are assigned corresponding attributes.

[0142] In this embodiment, in order to accurately describe the details of the quality of nuclear power construction, it is necessary to add attributes to the quality factors. For example, in the actual nuclear power construction, the complete description of a weld needs to indicate the characteristics of the weld name, weld number, manufacturer, unit, project, technical route, QA level, etc. Accordingly, the data attributes in the network ontology language are used to set them as the attributes of the weld quality factors, and the data type of each attribute is specifically specified.

[0143] S303: According to the quality factors and the corresponding attributes of each quality factor, using the object attributes in the network ontology language, construct the intra-process relationship and the extra-process relationship.

[0144] The intra-process relationship is used to construct the relationship between the quality elements of the process according to the logic of the output in the process and the on-site business logic, including personnel quality relationship, item quality relationship and document quality relationship; the extra-process relationship is used to establish logical connections and synergy between different processes, including process flow relationship and structural relationship.

[0145] In this embodiment, as shown in Table 2, the relationship within the process is constructed. According to the logic that "personnel", "items", and "static files" flow into the "process", and the "result record file" is output from the "process", as well as the business logic of on-site supervision of QC at all levels in collaborative quality witnessing, the relationship between the quality elements within the process is established, mainly including personnel quality relationship, item quality relationship, document quality relationship, etc.

[0146] Establishment of process flow relationship. Establish process flow relationship according to the business logic of process flow after passing quality inspection and issuing non-conforming items for repair and rectification after failing.

[0147] Construction of structural relationships. In order to ensure the coherence and integrity of knowledge expression in the collaborative quality flow diagram, it is also necessary to establish structural relationships, such as the quality control "start" and "end" relationship between "weld" and "process".

[0148] Table 2 Relationship of steel lining welding quality factors

[0149]

[0150]

[0151] (2) Business-level ontology evaluation and verification

[0152] In this embodiment, by checking the clarity, completeness, consistency, scalability, physical errors and logical errors of the established business layer ontology, the quality factors, quality relationships and quality attributes are based on the actual nuclear power welding and support the description of its business logic and rules; the quality factors are divided reasonably, there is no intersection between them, and there is no logical conflict; the quality factors and the relationships between them are set with the process as the knowledge interface, which has good scalability and is conducive to the subsequent management, maintenance and updating of quality knowledge.

[0153] (3) Graphicalization of business layer ontology model

[0154] like Figure 7 (a) shows the graphical representation of the triples of quality factor classes and quality factor attribute classes composed of quality factors and relationships; the steel lining welding business layer ontology model composed of the quality knowledge triple set is shown in Figure 7 (b) Figure 7 (c) Combined with process examples, several situations of process flow relationships are specifically demonstrated: when no quality non-conformity event occurs, the processes flow through the "qualified_flowTo" or "NA_flowTo" relationship; when a quality event occurs, the NCR is opened and flows to the first rework process through the "disqualified_flowTo" relationship, and returns to the original process sequence after the rework is completed; the flow between rework processes is the same as above.

[0155] 3.2. Extraction of quality instances of unstructured data based on BFAC model

[0156] There is a large amount of unstructured data in the nuclear power quality text recorded in natural language, which contains several details of quality formation. The quality instances that match the predefined ontology model can be extracted through "named entity recognition". The main process is to build a natural language processing model (such as BiLSTM-CRF model), collect corpus and annotate target instances for model training, and save the model with better training performance for quality instance extraction tasks. However, there is no mature large-scale corpus for nuclear power construction quality text to provide model training, so this paper needs to build its own corpus. The more sentences available for model training, the better the effect, but at the same time it increases the target instance annotation cost. Therefore, when the number of sentences and annotation scale in the self-built corpus is small, it is a problem that needs to be solved to build a model with better performance to ensure the accuracy of extraction.

[0157] In the traditional model training process, after collecting quality text, after instance annotation and vectorization, the text vector x is put into the BiLSTM layer to mine contextual semantic information, capture the long-distance dependencies between quality instance characters, and obtain the preliminary semantic feature vector H of the quality text. Given the limited amount of corpus available for model training, if H is put into the CRF layer for instance prediction at this time, it is difficult to achieve the expected effect.

[0158] Based on the BiLSTM-CRF model, the present invention constructs a BFAC model based on key semantic feature learning reinforcement and imbalanced sample adversarial enhancement by integrating the multi-head attention mechanism (Multi-Head Attention) and the fast gradient ascent algorithm (Fast Gradient Method, FGM), so as to efficiently complete the extraction of quality instances. Figure 8 As shown, the specific process is:

[0159] S401, the non-structural quality data is used as the input of the Embedding layer for vector conversion and used as the original sample.

[0160] S402, using the BiLSTM layer to perform preliminary learning on the original sample and extract the semantic feature matrix of the context;

[0161] S403, using the semantic feature matrix as the input of the multi-head attention mechanism, and concatenating the outputs of all attention heads to obtain a key semantic information matrix;

[0162] S404, inputting the key semantic information matrix into the CRF layer, performing sequence labeling, and obtaining labeled target instances;

[0163] S405. By using the fast gradient ascent algorithm to add perturbations to the Embedding layer vector of the original sample, an adversarial sample is generated. The original sample and the adversarial sample are used as inputs of the BFAC model to perform adversarial training on the BFAC model. The BFAC model parameters are updated through an iterative process to obtain the final trained BFAC model.

[0164] The key semantic feature reinforcement learning based on the multi-head attention mechanism and the adversarial enhancement training of imbalanced samples based on FGM are introduced in detail below.

[0165] (1) Reinforcement learning of key semantic features based on multi-head attention mechanism

[0166] like Fig. 9 As shown in the figure, by splicing a multi-head attention mechanism after the BiLSTM layer, multiple groups of attention weights are assigned to the output semantic feature matrix H. Each group of weights learns different semantic information, captures richer feature information, and splices and intersects to obtain the key semantic information vector, so that the model pays more attention to the quality instances to be extracted and ignores the secondary information.

[0167] The calculation formula of each attention head is as follows:

[0168]

[0169] Head i =Attention(QW i Q ,KW i K ,VW i V );

[0170] Where Q, K, and V are the semantic vectors after the semantic feature matrix is ​​mapped, and W i Q , W i K , W i V are the weight matrices learned during mapping, d k is the dimension of the input vector; the subscript i represents the i-th attention head in the multi-head attention mechanism, and i = 1, 2, …, h, h represents the number of attention heads in the multi-head attention mechanism;

[0171] The calculation formula of the key semantic information matrix is ​​as follows:

[0172] MultiheadATT=Concat(Head0,...,Head h )W O ;

[0173] Where W O is the learnable weight matrix when concatenating attention heads.

[0174] (2) Adversarial Enhancement Training for Imbalanced Samples Based on FGM

[0175] Considering that the number of "O" labels that are not within the extraction target range is greater than the target labels corresponding to the instances in the quality text, the label imbalance limits the model's learning ability. For this reason, the FGM algorithm is used to apply perturbation r to the Embedding layer vector x adv Conduct adversarial enhancement training. The basic idea is to find a set of adversarial samples that maximize the loss in the sample space to update the model parameters so that the expected loss of the model on the adversarial sample set is minimized. The specific process is:

[0176] S4051, initializing the Embedding layer parameters;

[0177] S4052, perform normal forward propagation and loss calculation on the original samples of the current batch, and then perform back propagation to calculate the network back propagation parameter gradient of the tth step;

[0178] g0=▽ x L(θ,x,y);

[0179] Where g0 is the network back propagation parameter gradient at step t, x is the gradient of the original sample x, L is the loss function, θ is the network parameter, x is the original sample, and y is the true label.

[0180] S4053, generating a sample perturbation of the tth step according to the calculated network back propagation parameter gradient of the tth step;

[0181] r adv =ε·g0 / ||g0||2;

[0182] In the formula, r adv is the sample disturbance at step t, and ε is the disturbance factor.

[0183] S4054, adding the sample perturbation of the tth step to the Embedding vector of the original sample to obtain an adversarial sample;

[0184] x adv =x+r adv

[0185] In the formula, x adv is an adversarial sample, and x is the original sample.

[0186] S4055, using the adversarial sample to perform forward propagation and loss calculation, and then performing back propagation to obtain the adversarial training gradient, and accumulating the adversarial training gradient to the network back propagation parameter gradient of the tth step;

[0187] S4056, restore the Embedding layer parameters, and use the accumulated gradients to update the network parameters;

[0188] S4057. Repeat steps S4051 to S4056. When the number of iterations is equal to or greater than the preset number of iterations, end the loop.

[0189] The key semantic feature vector MultiheadATT learned through adversarial optimization is input into the CRF layer for prediction that fully considers the transfer characteristics and sequentiality of the labels, which can achieve better model training performance.

[0190] 4. Analysis of nuclear power construction quality formation based on collaborative quality flow diagram

[0191] In practical applications, the collaborative quality flow diagram can be used to integrate and organize the multi-source heterogeneous quality data that were originally separated by different systems and departments, break the data silos, and coordinate the construction of shared and efficient information management methods. Fig.10 As shown, each unit processes, interprets and uses it efficiently according to unified standards, promoting the transmission of quality information across platforms and entities throughout the value chain. Secondly, based on the collaborative quality flow diagram, the present invention can quickly visualize and characterize the formation process of quality defect focus points, such as Fig.11 As shown, it is to quickly retrieve the process chain of quality defects, describe the interaction process between each quality element and its instances, reproducibly restore business activities, and assist engineers to quickly establish macro-cognition such as traceability direction; check the NCR file node, check the specific quality defect description, and locate the specific "trap process"; expand the node associated with the process, check the specific information of the relevant "executor", "quality control person", "item", and "applicable document"; combine specific analysis tools to mine potential useful information. For example, analyze indicators such as the "trap rate" of all operators, and take corresponding quality control measures such as suspension of work and retraining, revocation of nuclear component welding qualification certificates, and welding examinations for workers with higher defects. Monitor the locations prone to defects, strengthen quality control to prevent risks, and accumulate and summarize quality experience for subsequent construction.

[0192] 5. Embodiment

[0193] This example takes the characterization and analysis of the welding quality of the containment steel lining of a domestic nuclear power plant as an example, and sorts out the data and quality elements related to the steel lining welding quality activities from the aspects of quality objects, personnel, items, documents, etc. based on the collaborative quality element. The data used is collected from a nuclear engineering company in China, such as Fig.12As shown in the figure, the structured data is mainly extracted from quality business platforms such as the construction management system, test report platform, and non-conformity system, and stored in the MySQL relational database for subsequent use. The unstructured data mainly comes from the defect description statements in the welding NCR, with a total of more than 3,900 items collected.

[0194] 5.1. Construction of collaborative mass flow diagram for steel lining welding

[0195] 5.1.1. Construction of ontology model of steel lining welding business layer

[0196] This embodiment uses the ontology modeling software protégé5.5.0 to construct the steel lining welding business layer ontology. After the welding quality elements, attributes and relationships are constructed, the resulting steel lining welding business layer ontology graphical model is as follows: Fig.13 shown.

[0197] 5.1.2. Steel lining welding quality example extraction based on BFAC model

[0198] Taking the quality instance extraction of NCR defect description text as an example, the BFAC model input is the defect description sentence recorded in natural language, and the output is the quality instance attribute values ​​corresponding to the two types of quality factor attributes "welding defect (Def)" and "defective weld (Num)" in the business layer ontology. The extraction targets are shown in Table 3.

[0199] Table 3 Data input and example output

[0200]

[0201] The example extraction comprises the following steps:

[0202] (1) Corpus division and annotation. The collected more than 3,900 sentences describing steel lining welding quality defects were randomly divided into training set, validation set, and test set according to the ratio of 8:1:1. With reference to GB6417-86 "Classification and Description of Metal Fusion Welding Defects", the training set and validation set were annotated using the BIO system.

[0203] (2) Sentence trimming and alignment. The defect description sentences recorded in steel lining welding vary in length and need to be trimmed to a uniform length for matrix operations in the model. According to the overall statistical analysis of the corpus based on the character length (tokens length), when the sentence length is "140", it is sufficient to cover about 95% of the samples. This is used as the standard length for trimming and alignment. The specific method is to fill "0" on the right side to align to the standard length when the sentence length is less than 140; when the sample sentence length is greater than 140, it is truncated from the right side and trimmed to the standard length.

[0204] (3) Model performance evaluation indicators. The precision rate P, recall rate R, and F1 value are used to evaluate the performance of model entity recognition. The F1 value is the harmonic mean of the precision rate and the recall rate. The value range of the three is [0, 1]. The closer to 1, the better the performance of the model.

[0205] (4) Model settings and training prediction results. The BiLSTM-CRF model, BiLSTM-FGM-CRF model, BiLSTM-Attention-CRF model and BFAC model were set up for comparative experiments. All models were written in Python 3.7 under the tensorflow 1.15.0 and keras 2.3.1 deep learning frameworks. The computer environment and model parameter settings used are shown in Table 4.

[0206] Table 4 Environment and model parameter settings

[0207]

[0208] Finally, the effects of each model on the same test set are as follows Fig.14 As shown in the figure. The experimental results show that when the FGM algorithm is added to the BiLSTM-CRF model alone, the F1 value is increased by 7.78%; when the multi-head Attention mechanism is added alone, the F1 value is increased by 5.08%; the BFAC model that combines the two achieves the best results, with precision, recall, and F1 value increased by 9.84%, 7.05%, and 8.45% respectively. The F1 value of the final BFAC model is above 90%, which can effectively realize the extraction of steel lining welding quality instances.

[0209] 5.1.3 Knowledge mapping of steel lining welding structured data quality based on R2RML method

[0210] The business layer-driven R2RML method is used to transform the data of various business platforms related to welding quality activities into instantiated quality knowledge. The input is the MySQL data table related to the quality elements of the business layer such as "process", "executor", "quality control person", and "applicable documents", and the output is a set of quality knowledge triples.

[0211] exist Fig.13 Driven by the established steel lining welding business layer ontology model, the knowledge mapping of quality instances and relationships is carried out on the data tables related to quality elements such as "process", "executor", "quality control person", and "applicable documents" in turn. The quality knowledge triple set instantiated by the steel lining welding business layer ontology is finally obtained. Fig.15 shown.

[0212] 5.1.4 Visualization of collaborative mass flow diagram for steel lining welding based on Neo4j

[0213] right Fig.13 The steel lining welding business layer ontology model constructed Fig.15 The quality knowledge triples instantiated in the steel lining welding business layer ontology in the fusion of quality knowledge are uploaded to the Neo4j5.8.0-AuraDB cloud knowledge service platform and integrated into the nuclear power equipment full life cycle value chain collaboration platform developed by a Chinese nuclear engineering company to support the construction of nuclear power construction knowledge space for sharing by all collaborators. Finally, the collaborative quality flow diagram of steel lining welding is obtained as follows: Fig.16 shown.

[0214] 5.2 Analysis of steel lining welding quality based on collaborative mass flow diagram

[0215] 5.2.1. Information on steel lining welding quality is transmitted across platforms

[0216] Information transmission is spread throughout the entire process of steel lining welding quality control. Taking the review of the "inspection report" by the QC department of the general contractor as an example, various inspection reports generated in the procurement, manufacturing, construction, commissioning and other stages involving quality inspection of items such as welding machines and welding rods, non-destructive testing of welds, and commissioning inspection are scattered in different systems and departments, such as Fig.17 (a) shows. To review it, it is necessary to obtain the authority to log in to different quality business platforms and systems, and the scope of the review is limited to a certain extent by manual experience. The above work can be integrated by using the collaborative quality flow diagram. Fig.17 As shown in (b), all inspection report nodes involved in the welding process of Units 1 and 2 of a nuclear power plant are searched in the search box. The content details can be viewed by further expanding the attribute list, which greatly improves the review efficiency and promotes the information transmission of steel lining welding quality.

[0217] 5.2.2 Visual representation of steel lining welding quality

[0218] Take the visualization of the defects of "lack of fusion" and "dense holes" in the steel lining truncated cone weld (H1J***006) as an example. The graphical operation process is as follows: Fig.18As shown in (a), first, starting from the "NCR" node, the associated nodes are reversely expanded through the "process flow relationship", and the process path of defect formation is traced back to grasp the overall traceability scope and direction; secondly, combined with the relevant information in the "preliminary cause analysis" attribute of the "NCR" node, the "welding" process, which is the main defect-causing process, is quickly located; then the associated nodes of "welding" are expanded, and the specific information such as the executor, witness, items used, applicable documents, etc. involved are visually queried in detail to quickly grasp the key information; finally, this traceability operation can be saved to the query template library. At any time during the construction and service of nuclear power plants, managers can perform visual queries again to quickly restore the details of its quality formation and assist in nuclear power quality management. The visualization characterization results are shown in Figure 1. Fig.18 (b) as shown.

[0219] 6. Overview

[0220] In summary, the method of the present invention has the following technical advantages:

[0221] (1) The present invention organizes multi-source heterogeneous quality data in an orderly manner to form a collaborative quality metadata model. By taking collaborative quality metadata as the data source and using quality knowledge triples, a collaborative quality flow diagram is constructed to characterize the collaborative interaction relationship between personnel, items and documents, thereby realizing the knowledge-based quality data and characterizing the formation of nuclear power construction quality.

[0222] (2) The present invention integrates the multi-head attention mechanism with the fast gradient ascent algorithm to construct a BFAC model based on key semantic feature learning reinforcement and imbalanced sample adversarial enhancement. This solves the problem that in the construction of collaborative quality flow graphs, the lack of quality instance corpus extracted from quality data and the imbalance of labels limit the model learning ability. It improves the learning ability of key semantic features of quality instances and ensures the accuracy of model extraction.

[0223] (3) The present invention analyzes quality formation through the Neo4j graph database, integrates multi-source heterogeneous data, breaks down information silos, and forms a unified quality knowledge base. At the same time, it performs semantic query and visualization on the nodes and relationships in the collaborative quality flow graph to achieve shared use by all collaborative parties.

[0224] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described with reference to the preferred embodiments of the present invention, it should be understood by those skilled in the art that various changes may be made in form and details without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A method for characterizing the quality of nuclear power construction, characterized in that: The steps include: S1. According to the collaborative business activities in the formation of nuclear power construction quality, collect and store quality data of different forms and characteristics in multi-source heterogeneous nuclear power construction, wherein the quality data of different forms and characteristics include unstructured quality data, semi-structured quality data and structured quality data; S2. Organizing the quality data in an orderly manner to form a collaborative quality metadata model; S3. Based on the collaborative quality metadata model, summarize the quality elements of the collaborative quality metadata model and define the relationships and attributes involved in quality formation, and build a business layer ontology model; S4. Based on the BiLSTM-CRF model, by integrating the multi-head attention mechanism and the fast gradient ascent algorithm, a BFAC model based on key semantic feature learning reinforcement and imbalanced sample adversarial enhancement is constructed; S5. Based on the BFAC model, extract instances of the unstructured quality data and the semi-structured quality data to obtain quality instances that match the business layer ontology, and store them as structured data; S6. Based on the R2RML method, the structured data after instance extraction is mapped to the collected structured quality data to obtain the business layer ontology instantiation knowledge; S7. The business layer ontology and the business layer ontology instantiated knowledge are integrated with quality knowledge, and quality knowledge is stored and visualized based on the Neo4j graph database to obtain a collaborative quality flow diagram for nuclear power construction visualization, thereby completing the characterization of nuclear power construction quality formation.

2. The method for characterizing nuclear power construction quality according to claim 1, characterized in that: In step S2, the collaborative quality metadata model includes static quality data resources, process quality data resources and result quality data resources; The static quality data resources include basic information, inherent attributes and guiding quality documents; The process quality data resources include dynamic information, dynamic attributes and process flow information; The result quality data resource includes a result record quality file.

3. The method for characterizing nuclear power construction quality according to claim 1, characterized in that: In step S3, the specific process of summarizing the quality elements of the collaborative quality metadata model and defining the relationships and attributes involved in quality formation is: S301. According to the data resources in the collaborative quality metadata model, the quality elements are classified into static quality elements, process quality elements and result quality elements, and the hierarchical relationship of the quality elements is established by using the classes and subclasses in the network ontology language; S302, using the data attributes in the network ontology language to set corresponding attributes for each quality factor; S303: According to the quality factors and the corresponding attributes of each quality factor, the object attributes in the network ontology language are used to construct the intra-process relationship and the extra-process relationship.

4. The method for characterizing nuclear power construction quality according to claim 3, characterized in that: In step S303, the intra-process relationship is used to construct the relationship between the quality elements of the process, including personnel quality relationship, item quality relationship and document quality relationship, according to the logic of the output in the process and the on-site business logic; The external process relationships are used to establish logical connections and synergies between different processes, including process flow relationships and structural relationships.

5. The method for characterizing nuclear power construction quality according to claim 1, characterized in that: In step S4, the specific process of constructing the BFAC model by integrating the multi-head attention mechanism and the fast gradient ascent algorithm is as follows: S401, the non-structural quality data is used as the input of the Embedding layer for vector conversion and used as the original sample. S402, using the BiLSTM layer to perform preliminary learning on the original sample and extract the semantic feature matrix of the context; S403, using the semantic feature matrix as the input of the multi-head attention mechanism, and concatenating the outputs of all attention heads to obtain a key semantic information matrix; S404, inputting the key semantic information matrix into the CRF layer, performing sequence labeling, and obtaining labeled target instances; S405. By using the fast gradient ascent algorithm to add perturbations to the Embedding layer vector of the original sample, an adversarial sample is generated. The original sample and the adversarial sample are used as inputs of the BFAC model to perform adversarial training on the BFAC model. The BFAC model parameters are updated through an iterative process to obtain the final trained BFAC model.

6. The method for characterizing nuclear power construction quality according to claim 5, characterized in that: In step S403, the calculation formula of each attention head is as follows: Head i =Attention(QW i Q ,KW i K ,VW i V ); Where Q, K, and V are the semantic vectors after the semantic feature matrix is ​​mapped, and W i Q , W i K , W i V are the weight matrices learned during mapping, d k is the dimension of the input vector; the subscript i represents the i-th attention head in the multi-head attention mechanism, and i = 1, 2, …, h, h represents the number of attention heads in the multi-head attention mechanism; The calculation formula of the key semantic information matrix is ​​as follows: MultiheadATT=Concat(Head0,...,Head h )W O ; Where W O is the learnable weight matrix when concatenating attention heads.

7. The method for characterizing nuclear power construction quality according to claim 5, characterized in that: In step S405, the adversarial training of the BFAC model is performed, and the BFAC model parameters are updated through an iterative process to obtain the final trained BFAC model in the following specific manner: S4051, initializing the Embedding layer parameters; S4052, perform normal forward propagation and loss calculation on the original samples of the current batch, and then perform back propagation to calculate the network back propagation parameter gradient of the tth step; S4053, generating a sample perturbation of the tth step according to the calculated network back propagation parameter gradient of the tth step; S4054, adding the sample perturbation of the tth step to the Embedding vector of the original sample to obtain an adversarial sample; S4055, using the adversarial sample to perform forward propagation and loss calculation, and then performing back propagation to obtain the adversarial training gradient, and accumulating the adversarial training gradient to the network back propagation parameter gradient of the tth step; S4056, restore the Embedding layer parameters, and use the accumulated gradients to update the network parameters; S4057. Repeat steps S4051 to S4056. When the number of iterations is equal to or greater than the preset number of iterations, end the loop.

8. The method for characterizing nuclear power construction quality according to claim 7, characterized in that: In step S4052, the expression for calculating the gradient of the network back propagation parameter in step t is: Where g0 is the network back propagation parameter gradient at step t, is the gradient of the original sample x, L is the loss function, θ is the network parameter, x is the original sample, and y is the true label.

9. The method for characterizing nuclear power construction quality according to claim 7, characterized in that: In step S4053, the expression for generating the sample disturbance of step t is: r adv =ε·g0 / ||g0||2; In the formula, r adv is the sample disturbance at step t, and ε is the disturbance factor.

10. The method for characterizing nuclear power construction quality according to claim 7, characterized in that: In step S4054, the expression of the adversarial sample is: x adv =x+r adv In the formula, x adv is an adversarial sample, and x is the original sample.