Knowledge graph fusion method and system based on multi-level alignment

Abstract

The present invention provides a knowledge graph fusion method and system based on multi-level alignment, which relates to the field of data processing. The method comprises: obtaining a design inversion fault knowledge graph, a debugging joint test fault knowledge graph and a cross-domain collaborative fault knowledge graph; aiming at the problem that the existing embedding-based knowledge graph fusion method cannot treat multiple features such as entities, relationships and attributes equally, a multi-level aligned knowledge graph fusion model is constructed, and the model comprises a knowledge graph input module, a knowledge graph embedding module and a knowledge graph fusion output module; the knowledge graph input module pre-processes the knowledge graph, focusing on solving the problems of inconsistent standards at the ontology layer and large data noise, so as to provide a reference for the knowledge graph fusion method. The subsequent embedding module and fusion output module provide a consistent foundation; the knowledge graph embedding module extracts the embedding features of entities, relationships and attributes in the design inversion fault knowledge graph, the debugging and joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph respectively through different encoders in an equal manner; the knowledge graph fusion output module fuses the design inversion fault knowledge graph, the debugging and joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph based on the embedding features of entities, relationships and attributes extracted by the knowledge graph embedding module to generate a fused fault knowledge graph; based on the fused fault knowledge graph, electronic equipment fault diagnosis is performed.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a knowledge graph fusion method and system based on multi-level alignment. Background Art

[0002] Intelligent manufacturing is an advanced production method based on the deep integration of new generation information technology and advanced manufacturing technology, which runs through all aspects of manufacturing activities such as design, production, management, and service, and aims to improve the quality, efficiency, and core competitiveness of the manufacturing industry. It can be seen that the development of an integrated design, manufacturing, and operation and maintenance platform for the entire manufacturing process is the core and key difficulty in realizing intelligent manufacturing.

[0003] The integrated design, manufacturing and operation and maintenance platform is mainly aimed at the new needs of multi-enterprise collaborative design, distributed manufacturing and multi-regional operation and maintenance in the manufacturing industry. Combined with the hierarchical and graded security management needs of business data, it solves the scientific problems of multi-system model integration and transmission, and breaks through key technologies such as the construction of complex electronic equipment fault knowledge graphs, data homology and automatic adaptation of polymorphic heterogeneous models with large differences, and distributed cross-network collaborative data security storage. It establishes a distributed manufacturing technology system that integrates design, manufacturing and operation and maintenance, and ultimately realizes dynamic sharing of data, mutual adjustment of equipment functions, mutual recognition of debugging results, and cross-domain collaboration of services for multi-enterprise collaborative design, distributed manufacturing, and multi-regional operation and maintenance.

[0004] The construction of the integrated design, manufacturing and operation and maintenance platform has gone through a bottom-up, self-built, demand-oriented, and step-by-step R&D process. The entire process lacks integrated, collaborative, and holistic planning, which makes the information systems of different departments and businesses relatively isolated, leading to inconsistent data standard definitions, collaboration, and sharing difficulties among the various subjects responsible for multi-enterprise collaborative design, distributed manufacturing, and multi-regional operation and maintenance. This is particularly evident in the fault diagnosis of complex electronic equipment.

[0005] The collaborative design department has built a design inversion fault knowledge graph for fault data in distributed manufacturing and multi-regional operation and maintenance processes, aiming to guide the design by analyzing fault data in the design and manufacturing processes; the distributed manufacturing department has built a debugging and joint testing fault knowledge graph for fault phenomena in module debugging and whole machine joint testing, and through the research of automatic fault analysis technology based on knowledge graph, it solves the problems of low fault coverage, time-consuming and labor-intensive, and high personnel quality requirements in traditional fault location methods such as product BIT, serial testing and manual experience; the multi-regional operation and maintenance department has built a cross-domain collaborative fault knowledge graph for fault phenomena in the operation and maintenance process, solving the problems of lack of automated tool support in the traditional fault diagnosis process, difficulty in fault reproduction and location, and lack of network interconnection between base-level manufacturers and related supporting enterprises and between regions. Ultimately, it realizes the accurate identification and location of fault modes and significantly shortens the average maintenance cycle of cross-domain faults.

[0006] Due to the lack of overall planning for the three knowledge graphs and the correlation between various businesses in the early stage, the contents of the three knowledge graphs overlap and complement each other, resulting in resource waste, inability to provide services for various businesses in a unified manner, and difficulty in subsequent maintenance. Therefore, based on the original knowledge graph, standardization of data representation between heterogeneous knowledge graphs and research on knowledge graph fusion technology are carried out to finally form a unified knowledge graph to provide users with more complete and high-quality fault location, fault diagnosis and other services, which has become an urgent need and important research direction for the current integrated platform. Summary of the invention

[0007] The construction of different knowledge graphs (KG) involves different data sources due to specific requirements and differences in business functions, resulting in problems such as heterogeneity, redundancy and complementarity between knowledge graphs. In order to solve these problems, knowledge graph fusion (KF) technology came into being, which aims to align and merge heterogeneous, redundant and complementary information in the graph to form a globally unified knowledge identification and association. Entity alignment (EA) technology aims to automatically identify entities with equivalent relationships from different knowledge graphs and solve the naming differences of the same entity in different graphs. It is a key step to improve the integrity of knowledge graphs and realize the automatic fusion of multi-source heterogeneous knowledge graphs. Among the existing entity alignment methods, there are two main categories: traditional entity alignment methods and embedding-based entity alignment methods. Traditional entity alignment methods focus on relational reasoning and entity characters, and usually complete entity alignment with the help of entity character similarity comparison or relational reasoning. Entity alignment based on entity character similarity comparison mostly calculates similarity through traditional technologies such as term frequency-inverse document frequency (TF-IDF), machine learning discriminators, and synonym dictionaries. Entity alignment based on relational reasoning mostly relies on relations within KG. The core concept is to establish mechanisms such as relational similarity functions, probability functions, and relational comparability functions to evaluate the semantic equivalence between relations and achieve entity alignment under corresponding relations. Traditional entity alignment methods usually focus on relational reasoning and entity characters. The entity attributes involved in the alignment process are different and the domains are quite different, making it difficult to unify the similarity algorithms. At the same time, traditional entity alignment methods also ignore the semantic information between relational triple structures, fail to fully express the semantic association of entities, and have certain limitations. Embedding-based entity alignment methods mainly use knowledge graph embedding models to learn embedding representations that map entities to low-dimensional vector spaces, calculate the similarity of the embedding representations of entities, and finally obtain aligned entities. Some embedding-based entity alignment methods mainly use the relational features of knowledge graphs to align entities. For example, the TransE model regards relational embedding as a transformation embedding between the head entity embedding and the tail entity embedding. Specifically, given a triple (h, r, t), we expect the sum of the head entity embedding and the relation embedding to be close to the tail entity embedding, i.e., h+r≈t. MTransE encodes the relations and entities of each KG into different embedding spaces based on TransE, and provides transformation rules to calculate the embedding distance between entities to achieve entity alignment. IPTransE and BootEA are two self-training methods that embed two knowledge graphs into a unified embedding space and iteratively label new entity alignment information as a supervisory signal.GCN-Align learns the embedding representation of entities on the graph structure, and continuously optimizes the embedding of entity alignment in an iterative manner, so that similar entities in two knowledge graphs can be closer in the shared embedding space, thereby achieving the entity alignment task. The above methods mainly rely on the relational features in the knowledge graph for entity alignment, and lack the application of other features such as attributes and text descriptions. Therefore, some embedding-based entity alignment methods strengthen the relation-based embedding by integrating different types of features. JAPE combines the relations and attributes in the graph to achieve matching between entities. KDCoE jointly trains relational embedding and text description embedding, while AttrE adds character-level text embedding. However, these features are not treated equally, but are used to improve relation-based embedding and cannot integrate new features.

[0008] The present invention provides a knowledge graph fusion method based on multi-level alignment, including: obtaining a design inversion fault knowledge graph, a debugging joint test fault knowledge graph and a cross-domain collaborative fault knowledge graph; in view of the problem that the existing embedding-based knowledge graph fusion method cannot treat multiple features such as entities, relationships and attributes equally, a multi-level aligned knowledge graph fusion model is constructed, wherein the multi-level aligned knowledge graph fusion model includes a knowledge graph input module, a knowledge graph embedding module and a knowledge graph fusion output module; the knowledge graph input module pre-processes the design inversion fault knowledge graph, the debugging joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph, focusing on solving the problem of the present invention. The problems of inconsistent body layer standards and large data noise are solved; the knowledge graph embedding module extracts the embedded features of entities, relationships and attributes in the pre-processed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph respectively through different encoders in an equal manner; the knowledge graph fusion output module fuses the pre-processed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the embedded features of entities, relationships and attributes extracted by the knowledge graph embedding module to generate a fused fault knowledge graph; based on the fused fault knowledge graph, electronic equipment fault diagnosis is performed.

[0009] Furthermore, in order to eliminate the differences and inconsistencies between data sources and provide a consistent basis for subsequent fusion processing, the design inversion fault knowledge graph, the debugging and joint testing fault knowledge graph and the cross-domain collaborative fault knowledge graph are pre-processed, including: unifying the ontology layer standards of the design inversion fault knowledge graph, the debugging and joint testing fault knowledge graph and the cross-domain collaborative fault knowledge graph; denoising and text standardization of the design inversion fault knowledge graph, the debugging and joint testing fault knowledge graph and the cross-domain collaborative fault knowledge graph.

[0010] Furthermore, the knowledge graph embedding module extracts embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, including: for each entity in the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, using a discrete text embedding model to extract discrete text embeddings of tokens contained in the entity and a symbolic text embedding model to extract symbolic text embeddings of tokens contained in the entity, and generating embedding features of the entity.

[0011] Furthermore, the knowledge graph embedding module extracts embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, including: extracting embedding features of relationships in the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph through the TransE model.

[0012] Furthermore, the knowledge graph embedding module extracts embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, including: for each attribute in the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, the embedding of the attribute and the embedding of the value of the attribute are spliced ​​into a matrix, and the matrix is ​​input into a convolutional neural network to obtain the embedding feature of the attribute.

[0013] Furthermore, in order to measure the rationality of the obtained attribute embedding, the following loss function is established to train the convolutional neural network:

[0014]

[0015] SCattr(h (3) ,a,v)=-||h (3) -CNN(<a,v> )||

[0016] Among them, φ (3) represents the comprehensive entity embedding only at the attribute embedding level, h (3) is the embedding of the attribute, a is the attribute, v is the value of the attribute, CNN(·) is the convolution operation, and Z + =Z a ∪Z b ∪Z c Represents the set of real attribute triplets in the design inversion fault knowledge graph, debugging and joint testing fault knowledge graph, and cross-domain collaborative fault knowledge graph to be fused.

[0017] Furthermore, in order to complete entity alignment, the knowledge graph fusion output module fuses the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the embedding features of entities, relationships and attributes extracted by the knowledge graph embedding module to generate a fused fault knowledge graph, including: fusing the embedding features of entities, relationships and attributes of the design inversion fault knowledge graph extracted by the knowledge graph embedding module to obtain a comprehensive entity embedding of the design inversion fault knowledge graph; fusing the embedding features of the debugging joint test fault knowledge graph extracted by the knowledge graph embedding module to obtain a comprehensive entity embedding of the design inversion fault knowledge graph; The embedding features of entities, relationships and attributes of the debugging and joint test fault knowledge graph are merged to obtain the comprehensive entity embedding of the debugging and joint test fault knowledge graph; the embedding features of entities, relationships and attributes of the cross-domain collaborative fault knowledge graph extracted by the knowledge graph embedding module are merged to obtain the comprehensive entity embedding of the cross-domain collaborative fault knowledge graph; based on the comprehensive entity embedding of the design inversion fault knowledge graph, the comprehensive entity embedding of the debugging and joint test fault knowledge graph and the comprehensive entity embedding of the cross-domain collaborative fault knowledge graph, the entity similarity between different knowledge graphs is calculated, the entity alignment is completed, and the fused fault knowledge graph is generated.

[0018] Furthermore, in order to integrate the contributions of each embedding level and more accurately capture the semantic features of the text, the embedding features of the entities, relationships and attributes of the design inversion fault knowledge graph extracted by the knowledge graph embedding module are fused to obtain a comprehensive entity embedding of the design inversion fault knowledge graph, including: based on weights and or shared space, the embedding features of the entities, relationships and attributes of the design inversion fault knowledge graph extracted by the knowledge graph embedding module are fused to obtain a comprehensive entity embedding of the design inversion fault knowledge graph.

[0019] Furthermore, the entity similarity between different knowledge graphs is calculated based on the following formula:

[0020]

[0021] Among them, DIS(e m ,e n ) is the entity e in the design inversion fault knowledge graph m and debugging joint test fault knowledge graph entity e n The similarity of To design the entity e in the fault knowledge graph m Comprehensive entity embedding of To debug the entity e in the joint test fault knowledge graph n The comprehensive entity embedding of , ||·||1 is the L1 norm.

[0022] The present invention provides a knowledge graph fusion system based on multi-level alignment, including: a graph acquisition module, used to acquire a design inversion fault knowledge graph, a debugging joint test fault knowledge graph and a cross-domain collaborative fault knowledge graph; a model building module, used to construct a multi-level aligned knowledge graph fusion model, to solve the problem that the existing embedding-based knowledge graph fusion method cannot treat multiple features such as entities, relationships and attributes equally, wherein the multi-level aligned knowledge graph fusion model includes a knowledge graph input module, a knowledge graph embedding module and a knowledge graph fusion output module; a graph fusion module, used to fusion the design inversion fault knowledge graph, the debugging joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph based on the knowledge graph input module. The knowledge graph is preprocessed; the graph fusion module is also used to extract the embedded features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the knowledge graph embedding module; the graph fusion module is also used to fuse the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the embedded features of entities, relationships and attributes extracted by the knowledge graph embedding module by the knowledge graph fusion output module to generate a fused fault knowledge graph; a fault diagnosis module is used to perform electronic equipment fault diagnosis based on the fused fault knowledge graph.

[0023] Compared with the prior art, the knowledge graph fusion method and system based on multi-level alignment provided by the present invention have at least the following beneficial effects:

[0024] 1. In view of the characteristics of knowledge graphs in the integrated design, manufacturing and operation and maintenance platform, such as inconsistent ontology layer and large data noise, data preprocessing is performed on each knowledge graph in the platform to provide a consistent basis for subsequent integration.

[0025] 2. Aiming at the problem that the existing embedding-based knowledge graph fusion methods cannot treat multiple features such as entities, relationships and attributes equally, a knowledge graph fusion method based on multi-level alignment is proposed. This method first extracts the embedding features of entities, relationships and attributes respectively through different encoders in an equal manner, then fuses multiple embedding features to obtain comprehensive entity embedding, and finally calculates the similarity of comprehensive entity embeddings between different knowledge graphs through L1 distance to find aligned entities. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0027] Figure 1It is a flowchart of a knowledge graph fusion method based on multi-level alignment according to some embodiments of this specification;

[0028] Figure 2 It is a schematic diagram of a debugging joint test fault knowledge graph according to some embodiments of this specification;

[0029] Figure 3 is a schematic diagram of a cross-domain collaborative fault knowledge graph according to some embodiments of this specification;

[0030] Figure 4 is a schematic diagram of a fused fault knowledge graph according to some embodiments of this specification;

[0031] Figure 5 is a schematic diagram of a multi-level aligned knowledge graph fusion model according to some embodiments of this specification;

[0032] Figure 6 is a schematic diagram of a triplet according to some embodiments of the present specification;

[0033] Figure 7 It is a module schematic diagram of a knowledge graph fusion system based on multi-level alignment according to some embodiments of this specification;

[0034] Figure 8a is a schematic diagram of original data of a knowledge graph according to some embodiments of this specification;

[0035] Figure 8b It is a schematic diagram of data after cleaning, denoising and standardization of the knowledge graph shown in some embodiments of this specification. DETAILED DESCRIPTION

[0036] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of this specification. For ordinary technicians in this field, this specification can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.

[0037] Figure 1 is a flow chart of a knowledge graph fusion method based on multi-level alignment according to some embodiments of this specification, such as Figure 1 As shown, the knowledge graph fusion method based on multi-level alignment can include the following steps.

[0038] Step 110, obtaining the design inversion fault knowledge graph, the debugging joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph.

[0039] The collaborative design department has built a design inversion fault knowledge graph for fault data in distributed manufacturing and multi-regional operation and maintenance, aiming to guide design by analyzing fault data in the design and manufacturing process. As shown in Table 1-1, the design inversion fault knowledge graph contains seven entities, including complete machine, module, printed circuit board, etc., and different attributes are designed for each entity.

[0040] Table 1-1

[0041]

[0042]

[0043] In addition, the design inversion fault knowledge graph also contains four types of relationships, such as fault and suggestion, as shown in Table 1-2. These relationships provide semantic association information between ontologies, which helps to understand the connection and meaning between various ontologies in the knowledge graph. For example, the "suggestion" relationship exists between the fault ontology and the suggestion ontology, indicating that a certain suggestion can solve a certain fault.

[0044] Table 1-2

[0045]

[0046] Figure 2 is a schematic diagram of a design inversion fault knowledge graph according to some embodiments of this specification, such as Figure 2 As shown in Table 1-3, the knowledge graph of debugging and joint testing faults is constructed for the fault phenomena in the process of module debugging and whole machine joint testing. By studying the automatic fault analysis technology based on the knowledge graph, the traditional fault location methods such as product BIT, serial testing and manual experience are solved, such as low fault coverage, time-consuming and labor-intensive, and high personnel quality requirements. As shown in Table 1-3, the knowledge graph of debugging and joint testing faults contains five ontologies such as components, faults, and phenomena, and different attributes are designed for each ontology.

[0047] Table 1-3

[0048]

[0049] In addition, the debugging and joint test fault knowledge graph also contains four types of relationships, such as belongs to and causes, as shown in Table 1-4. These relationships provide semantic association information between ontologies, which helps to understand the connection and meaning between various ontologies in the knowledge graph. For example, the "cause" relationship exists between the fault ontology and the phenomenon ontology, indicating that a certain fault will cause a certain phenomenon to occur.

[0050] Table 1-4

[0051]

[0052] Figure 3 is a schematic diagram of a debugging joint test fault knowledge graph according to some embodiments of this specification, such as Figure 3 As shown in Table 1-5, the multi-regional operation and maintenance departments have built a cross-domain collaborative fault knowledge graph for the fault phenomena in the operation and maintenance process, solving the problems of the lack of automated tool support in the traditional fault diagnosis process, the difficulty in fault reproduction and location, and the lack of network interconnection between base-level manufacturers and related supporting enterprises and regions. Ultimately, the fault mode is accurately identified and located, and the average maintenance cycle of cross-domain faults is greatly shortened. As shown in Table 1-5, the cross-domain collaborative fault knowledge graph contains six entities such as the whole part, component, and fault, and different attributes are designed for each entity.

[0053] Table 1-5

[0054]

[0055] In addition, the cross-domain collaborative fault knowledge graph also contains four types of relationships, such as fault_fault phenomenon, fault_fault cause, etc., as shown in Table 1-6. These relationships provide semantic association information between ontologies, which helps to understand the connection and meaning between various ontologies in the knowledge graph. For example, the "fault_fault cause" relationship exists between the fault ontology and the fault cause ontology, indicating that a certain fault is caused by a certain cause.

[0056] Table 1-6

[0057]

[0058] Step 120, construct a multi-level aligned knowledge graph fusion model.

[0059] Figure 5 is a schematic diagram of a multi-level aligned knowledge graph fusion model according to some embodiments of this specification, such as Figure 5 As shown, the multi-level aligned knowledge graph fusion model includes a knowledge graph input module, a knowledge graph embedding module and a knowledge graph fusion output module.

[0060] Step 130, the knowledge graph input module preprocesses the design inversion fault knowledge graph, the debugging joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph.

[0061] In some embodiments, the design inversion fault knowledge graph, the debugging joint test fault knowledge graph, and the cross-domain collaborative fault knowledge graph are preprocessed, including:

[0062] Unify the ontology-level standards for the design inversion fault knowledge graph, the debugging and joint testing fault knowledge graph, and the cross-domain collaborative fault knowledge graph;

[0063] The design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph are denoised and text standardized.

[0064] Specifically, based on the design inversion fault knowledge graph ontology layer, the debugging joint test fault knowledge graph ontology layer, and the cross-domain collaborative fault knowledge graph ontology layer, the ontology after knowledge graph fusion, the relationship between ontologies, and the attributes of the ontology are sorted out with reference to relevant literature, industry standards, and other materials, and the unification of the ontology layer standards is completed. As shown in Table 1-7, the fused knowledge graph ontology layer contains 8 ontologies such as components, modules, and devices, and different attributes are designed for each ontology.

[0065] Table 1-7

[0066]

[0067]

[0068] In addition, the fused knowledge graph ontology layer also contains five types of relationships, such as belongs to, causes, and is associated with, as shown in Table 1-8. These relationships provide semantic association information between ontologies, which helps to understand the connection and meaning between various ontologies in the knowledge graph. For example, the "association" relationship exists between the phenomenon ontology and the condition and opportunity ontology, indicating that a certain fault phenomenon has an association relationship with a certain condition and opportunity.

[0069] Table 1-8

[0070]

[0071] The knowledge graph data obtained from the design inversion fault knowledge graph, debugging joint test fault knowledge graph, and cross-domain collaborative fault knowledge graph contain a lot of noise, such as garbled characters, repeated content, etc. These noise information not only reduces the data quality, but also increases the complexity of model calculation, which needs to be processed.

[0072] Noise can be removed by any noise processing method (e.g., regularization, statistical model, confidence learning, etc.). As an example only, given the nature of noise, this article chooses regularization to clean up the noise in the text. Regular expression matching and replacement can effectively remove noise and improve the quality and clarity of the text.

[0073] Compared with ordinary text, the knowledge graph data of the design inversion fault knowledge graph, the debugging joint test fault knowledge graph, and the cross-domain collaborative fault knowledge graph are different in that their formats are not fixed and the data length varies greatly due to different diagnostic subjects. Long texts may contain a lot of repeated and redundant information, and directly inputting the model will affect the knowledge graph fusion effect. Therefore, it is necessary to standardize long texts so that their length meets the model input standard.

[0074] Text standardization can be performed by any method (for example, truncation, keyword extraction, and text summarization). The truncation method makes the text length meet the standard by truncation of the head, tail, or head and tail, but it may cause the loss of text sequence information. Keyword extraction extracts and combines keywords in the text through a keyword dictionary. Among them, building a keyword dictionary requires a lot of human resources and the dictionary has poor portability. Text summarization uses open source tools or neural network models to extract text summaries, including extractive and generative types. The former extracts key information from the original text to form a summary, and the latter summarizes the original text to generate new phrases and words. Compared with truncation and keyword extraction, text summarization controls the text length within the standard range while retaining key information. For example, text standardization can be performed using the TextRank4ZH model. The text standardization process is: first check whether the text length meets the standard. If it does, use it directly; if it exceeds the standard, use the TextRank4ZH model for summary extraction. As an example only, Figure 8a is a schematic diagram of original data of the knowledge graph according to some embodiments of this specification, Figure 8b It is a schematic diagram of data after cleaning, denoising and standardization of the knowledge graph shown in some embodiments of this specification.

[0075] It can be understood that through the processing of the knowledge graph input module, the differences and inconsistencies between data sources can be eliminated, providing a consistent basis for subsequent fusion processing.

[0076] Step 140, the knowledge graph embedding module extracts the embedding features of entities, relationships and attributes in the pre-processed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph.

[0077] Entity embedding describes the basic and key features of an entity, while the relationships and attributes defined in the knowledge graph describe the rich external and internal information of the entity. Based on the above hierarchical division, the knowledge graph is formalized into a 7-tuple G = {E, R, A, V, N, Y, Z}, where E and R represent the set of entities and relationships respectively, A represents attributes, and V represents entity values ​​or attribute values. represents the entity embedding level, represents the relation embedding level, It represents the attribute embedding hierarchy.

[0078] Just as an example, given two knowledge graphs G a =(E a ,R a ,A a ,V a ,N a ,Y a ,Z a) and G b =(E b ,R b ,A b ,V b ,N b ,Y b ,Z b ), entity fusion aims to find a set of identical entity pairs C = {(e i ,e j )∈E a ×E b |e i ≡e j}, where "≡" indicates an equivalence relation. First, we learn the d-dimensional embedding, and then find equivalent entity pairs based on the embedding to complete the fusion of the knowledge graph. For the notation, the bracketed superscript of the knowledge graph embedding indicates which level the embedding comes from, and entities, relations, and attributes are marked with (1), (2), and (3), respectively.

[0079] In some embodiments, the knowledge graph embedding module extracts embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, including: for each entity in the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, using a discrete text embedding model to extract discrete text embeddings of tokens contained in the entity and a symbolic text embedding model to extract symbolic text embeddings of tokens contained in the entity, to generate embedding features of the entity.

[0080] As the basis of multi-level embedding, text embedding represents discrete words and symbolic words through a d-dimensional embedding vector. Without loss of generality, let t = (s1, s2, ..., s n ) represents a text with n tokens. TP(·) is defined as a lookup function that maps an input text to an embedding. To account for the different representations of text and capture its inherent semantics, consider the following case:

[0081]

[0082] Among them, word_embed(s i ) returns the input token s i The corresponding discrete text embedding, while char_embed(s i ) returns the input token s i The corresponding symbolic literal embedding of .

[0083] As an example only, the pre-trained language model BERT (Bidirectional Encoder Representation from Transformers) is used as a discrete text embedding model. The BERT model is based on a bidirectional Transformer encoder and combines text features from left to right and from right to left to capture the semantic information of the text more comprehensively. Compared with the ELMO (Embeddings from Language Models) language model based on bidirectional LSTM (Long Short-Term Memory), the Transformer encoder uses a multi-head self-attention mechanism to more effectively extract the key information of the text, thereby better capturing the local features of the text. Taking into account the global and local features of the text, the BERT model can deeply explore the semantic features of the text and improve the effect of text processing.

[0084] The pre-trained language model Skip-Gram is used as the symbolic text embedding model. The Skip-Gram model uses a flexible context prediction method and pays more attention to the prediction of contextual text. Therefore, it is more flexible and adaptable when processing symbolic text data, and can better learn rare features.

[0085] Let θ(t) represent the text embedding of the input text, which is calculated as follows:

[0086] θ(t)=encode(TP(s1); TP(s2);...; TP(s n ))

[0087] Where encode(·) returns the compressed representation of the input embedding vector, and ";" indicates a concatenation operation.

[0088] Due to the superiority of pre-trained language models in capturing text semantic information and mining text semantic features, this paper directly uses the above text embedding as entity embedding. Given an entity h, its entity embedding is defined as follows:

[0089] h (1) =θ(name(h))

[0090] Among them, name(·) is used to extract the input entity name information.

[0091] In some embodiments, the knowledge graph embedding module extracts the embedded features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, the debugging and joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph, including: extracting the embedded features of the relationships in the preprocessed design inversion fault knowledge graph, the debugging and joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph through the TransE model.

[0092] Entities in the knowledge graph are connected to each other through relations, so relation embedding describes the structure of the knowledge graph. In order to preserve such relational structure, the TransE model is used to implement relation embedding. TransE is a commonly used knowledge graph embedding model used to learn low-dimensional vector representations of entities and relations. Its core idea is to represent a relation as a translation vector from the head entity to the tail entity. Specifically, Figure 6 is a schematic diagram of a triplet according to some embodiments of this specification, such as Figure 6 As shown in the figure, for a given triple (h, r, t), where h represents the head entity, r represents the relation, and t represents the tail entity, TransE assumes that the head entity h is translated through the relation r to obtain the tail entity t.

[0093] To train the TransE model, a scoring function is defined, usually in the form of margin-based, which aims to increase the gap between the correct triple score and the incorrect triple score. By minimizing the scoring function, the model can learn a vector representation that makes the correct triple score higher than the incorrect triple score. The scoring function is defined as follows:

[0094] SC rel (h (2) ,r,t (2) )=-||h (2) +rt (2) ||

[0095] Here, ||·|| represents the L2 vector norm.

[0096] The probability that (h,r,t) is a true relationship is calculated based on the following formula:

[0097] P rel (ξ (h,r,t) =1|φ (2) )=Sigmoid(SC rel (h (2) ,r,t (2) ))

[0098] Among them, φ (2) represents the comprehensive entity embedding only at the relation embedding level, ξ (h,r,t) is the label of the triple (h, r, t), and the label value is "1" or "-1", where "1" means (h, r, t) is a true triple and "-1" means (h, r, t) is not a true triple. Finally, φ is parameterized by minimizing the following logistic loss function (2) :

[0099]

[0100] Among them, Y + = Y a ∪Y b ∪Y c represents the set of true attribute triples in the design inversion fault knowledge graph, commissioning and trial operation fault knowledge graph, and cross-domain collaborative fault knowledge graph to be fused. Y - represents the set of wrong relation triples, that is, replacing the head entity or tail entity of the true relation triple with other entities.

[0101] In some embodiments, the knowledge graph embedding module extracts the embedding features of entities, relations, and attributes in the preprocessed design inversion fault knowledge graph, commissioning and trial operation fault knowledge graph, and cross-domain collaborative fault knowledge graph, including: for each attribute in the preprocessed design inversion fault knowledge graph, commissioning and trial operation fault knowledge graph, and cross-domain collaborative fault knowledge graph, concatenate the embedding of the attribute and the embedding of the value of the attribute into a matrix, and input the matrix into a Convolutional Neural Network (CNN) to obtain the embedding feature of the attribute.

[0102] Specifically, the CNN vectorizes the text through components such as convolutional layers, pooling layers, and fully connected layers. The convolutional layer extracts local features through a sliding window, the pooling layer compresses the size of the feature map and retains key information, and finally, the fully connected layer outputs the vector representation of the text. Its advantage is that it can effectively capture the local features and semantic information of the text, and at the same time reduces the number of model parameters through parameter sharing and local connection, improving the computational efficiency and generalization ability.

[0103] Concatenate the embedding of an attribute a and the embedding of its value v into a matrix, denoted by and input it into the CNN to obtain the attribute embedding representation:

[0104] CNN(<a,v>) = ρ(vec(ρ(<a,v>*Ω))W)

[0105] where CNN(·) represents the convolution operation, slides a convolution kernel Ω of size 2×c (c < d) on the input embedding matrix to extract high-level features. Then, the feature map tensor is reshaped into a vector through vec(·), and finally, it is projected into the knowledge graph embedding space using a densely connected layer parameterized by the weight W. ρ(·) is an activation function.

[0106] In some embodiments, the following loss function is established to train the convolutional neural network:

[0107]

[0108] SC attr (h(3) ,a,v)=-||h (3) -CNN(<a,v> )||

[0109] Among them, φ (3) represents the comprehensive entity embedding only at the attribute embedding level, h (3) is the embedding of the attribute, a is the attribute, v is the value of the attribute, CNN(·) is the convolution operation, and Z + =Z a ∪Z b ∪Z c Represents the set of true attribute triplets in the design inversion fault knowledge graph, debugging joint test fault knowledge graph, and cross-domain collaborative fault knowledge graph to be fused. SC attr (h (3) ,a,v)=-||h (3) -CNN(<a,v> )|| is used to measure the rationality of the attribute embedding obtained by the convolution operation CNN(·). The closer the head entity is to its attributes and values, the higher the score, which can minimize the loss Γ(φ (3) ) to achieve it.

[0110] Step 150, the knowledge graph fusion output module fuses the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the embedded features of entities, relationships and attributes extracted by the knowledge graph embedding module to generate a fused fault knowledge graph.

[0111] Specifically include:

[0112] The embedding features of entities, relationships and attributes of the design inversion fault knowledge graph extracted by the knowledge graph embedding module are integrated to obtain the comprehensive entity embedding of the design inversion fault knowledge graph;

[0113] The embedding features of entities, relationships and attributes of the debugging and joint test fault knowledge graph extracted by the knowledge graph embedding module are integrated to obtain the comprehensive entity embedding of the debugging and joint test fault knowledge graph;

[0114] The embedding features of entities, relationships and attributes of the cross-domain collaborative fault knowledge graph extracted by the knowledge graph embedding module are integrated to obtain the comprehensive entity embedding of the cross-domain collaborative fault knowledge graph;

[0115] Based on the comprehensive entity embedding of the design inversion fault knowledge graph, the comprehensive entity embedding of the debugging joint test fault knowledge graph, and the comprehensive entity embedding of the cross-domain collaborative fault knowledge graph, the entity similarity between different knowledge graphs is calculated, the entity alignment is completed, and the fused fault knowledge graph is generated. For example, the entity pair with the highest similarity is selected to complete the entity alignment. As an example only, the fused fault knowledge graph can be as follows Figure 4shown.

[0116] In some embodiments, the embedding features of entities, relationships and attributes of the design inversion fault knowledge graph extracted by the knowledge graph embedding module are fused to obtain a comprehensive entity embedding of the design inversion fault knowledge graph, including: based on weights and or shared space, the embedding features of entities, relationships and attributes of the design inversion fault knowledge graph extracted by the knowledge graph embedding module are fused to obtain a comprehensive entity embedding of the design inversion fault knowledge graph.

[0117] Specifically, the weight distribution method can more finely consider the contribution of each embedding level to the overall embedding representation, thereby more accurately capturing the semantic features of text or data. In this way, the embedding information at different levels can be better balanced, improving the effect and performance of embedding fusion. represents the comprehensive entity embedding, which is calculated as follows:

[0118]

[0119] Among them, w j Yes (j) The weight is calculated as follows:

[0120]

[0121] in, It indicates the mean of embeddings at different levels. It can be seen that if the embedding of a level is far away from the average embedding, its weight will be lower.

[0122] Based on the assumption that potential shared hierarchical embeddings can generate embeddings at each level, shared space-based embedding fusion aims to find an orthogonal mapping matrix from each level embedding space to the shared hierarchical embedding space. is the comprehensive entity embedding matrix, where each row corresponds to an entity, H (j) For the entity embedding matrix at the jth level, minimize the mapping loss:

[0123]

[0124] in, Z (j) Represents the mapping matrix from the j-th level embedding space to the shared level embedding space. Represents the Frobenius norm. The first Frobenius norm in the formula is used as a soft constraint on the orthogonalized mapping matrix, where I is the identity matrix. This orthogonality helps to keep the distance between embeddings in a specific hierarchical space unchanged during the transformation process, thereby enabling the shared space to learn alignment information from multiple embedding spaces.

[0125] In some embodiments, the entity similarity between different knowledge graphs is calculated based on the following formula:

[0126]

[0127] Among them, DIS(e m ,e n ) is the entity e in the design inversion fault knowledge graph m and debugging joint test fault knowledge graph entity e n The similarity of To design the entity e in the fault knowledge graph m Comprehensive entity embedding of To debug the entity e in the joint test fault knowledge graph n The comprehensive entity embedding of , ||·||1 is the L1 norm.

[0128] Step 160 , performing electronic equipment fault diagnosis based on the fused fault knowledge graph.

[0129] It can be understood that electronic equipment fault diagnosis based on the fused fault knowledge graph can effectively improve the accuracy and efficiency of electronic equipment fault diagnosis.

[0130] Figure 8 is a module diagram of a knowledge graph fusion system based on multi-level alignment according to some embodiments of this specification. As shown in Figure 8, the knowledge graph fusion system based on multi-level alignment may include a graph acquisition module, a model building module, a graph fusion module and a fault diagnosis module.

[0131] The graph acquisition module can be used to obtain the design inversion fault knowledge graph, the debugging and joint testing fault knowledge graph and the cross-domain collaborative fault knowledge graph.

[0132] The model building module can be used to construct a multi-level aligned knowledge graph fusion model, wherein the multi-level aligned knowledge graph fusion model includes a knowledge graph input module, a knowledge graph embedding module and a knowledge graph fusion output module.

[0133] The graph fusion module can be used to preprocess the design inversion fault knowledge graph, the debugging and joint testing fault knowledge graph and the cross-domain collaborative fault knowledge graph based on the knowledge graph input module to obtain the preprocessed design inversion fault knowledge graph, the debugging and joint testing fault knowledge graph and the cross-domain collaborative fault knowledge graph.

[0134] The graph fusion module can also be used to extract the embedded features of entities, relationships and attributes in the pre-processed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the knowledge graph embedding module.

[0135] The graph fusion module can also be used to fuse the preprocessed design inversion fault knowledge graph, debugging and joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the embedded features of entities, relationships and attributes extracted by the knowledge graph embedding module by the knowledge graph fusion output module to generate a fused fault knowledge graph.

[0136] The fault diagnosis module can be used to diagnose electronic equipment faults based on the fused fault knowledge graph.

[0137] The knowledge graph fusion system based on multi-level alignment can be used to execute the knowledge graph fusion method based on multi-level alignment. For more descriptions of the knowledge graph fusion system based on multi-level alignment, please refer to the relevant descriptions of the knowledge graph fusion method based on multi-level alignment, which will not be repeated here.

[0138] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. A knowledge graph fusion method based on multi-level alignment, characterized in that: include: Obtain the design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph; Constructing a multi-level aligned knowledge graph fusion model, wherein the multi-level aligned knowledge graph fusion model includes a knowledge graph input module, a knowledge graph embedding module and a knowledge graph fusion output module; The knowledge graph input module preprocesses the design inversion fault knowledge graph, the debugging joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph; The knowledge graph embedding module extracts the embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph; The knowledge graph fusion output module fuses the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the embedded features of entities, relationships and attributes extracted by the knowledge graph embedding module to generate a fused fault knowledge graph; Based on the fused fault knowledge graph, electronic equipment fault diagnosis is performed; The knowledge graph embedding module extracts the embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, including: For each entity in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph, and cross-domain collaborative fault knowledge graph, a discrete text embedding model is used to extract discrete text embeddings of tokens contained in the entity and a symbolic text embedding model is used to extract symbolic text embeddings of tokens contained in the entity to generate embedding features of the entity; The knowledge graph embedding module extracts the embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, including: Extracting embedded features of the relationships in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph, and cross-domain collaborative fault knowledge graph through the TransE model; The knowledge graph embedding module extracts the embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, including: For each attribute in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph, and cross-domain collaborative fault knowledge graph, concatenate the embedding of the attribute and the embedding of the value of the attribute into a matrix, input the matrix into a convolutional neural network, and obtain the embedding feature of the attribute; The knowledge graph fusion output module fuses the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the embedded features of entities, relationships and attributes extracted by the knowledge graph embedding module to generate a fused fault knowledge graph, including: The embedding features of entities, relationships and attributes of the design inversion fault knowledge graph extracted by the knowledge graph embedding module are integrated to obtain a comprehensive entity embedding of the design inversion fault knowledge graph; The embedding features of entities, relationships and attributes of the debugging joint test fault knowledge graph extracted by the knowledge graph embedding module are integrated to obtain a comprehensive entity embedding of the debugging joint test fault knowledge graph; The embedding features of entities, relationships and attributes of the cross-domain collaborative fault knowledge graph extracted by the knowledge graph embedding module are integrated to obtain a comprehensive entity embedding of the cross-domain collaborative fault knowledge graph; Based on the comprehensive entity embedding of the design inversion fault knowledge graph, the comprehensive entity embedding of the debugging and joint testing fault knowledge graph, and the comprehensive entity embedding of the cross-domain collaborative fault knowledge graph, the entity similarity between different knowledge graphs is calculated, the entity alignment is completed, and the fused fault knowledge graph is generated.

2. The knowledge graph fusion method based on multi-level alignment according to claim 1 is characterized in that: Preprocessing the design inversion fault knowledge graph, the debugging joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph includes: Unifying the ontology layer standards of the design inversion fault knowledge graph, the debugging joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph; The design inversion fault knowledge graph, the debugging joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph are denoised and text standardized.

3. The knowledge graph fusion method based on multi-level alignment according to claim 1 is characterized in that: The following loss function is established to train the convolutional neural network: S Cattr (h (3) , a, v) = -||h (3) - CNN(<a, v>)||SC attr (h (3) , a, v) = -||h (3) - CNN(<a, v>)||where φ (3) represents the comprehensive entity embedding only at the attribute embedding level, h (3) is the embedding of the attribute, a is the attribute, v is the value of the attribute, CNN(·) is the convolution operation, and Z + =Z a ∪Z b ∪Z c Represents the set of real attribute triplets in the design inversion fault knowledge graph, debugging and joint testing fault knowledge graph, and cross-domain collaborative fault knowledge graph to be fused.

4. The knowledge graph fusion method based on multi-level alignment according to claim 1 is characterized in that: The embedding features of entities, relationships and attributes of the design inversion fault knowledge graph extracted by the knowledge graph embedding module are integrated to obtain a comprehensive entity embedding of the design inversion fault knowledge graph, including: Based on the weight and / or shared space, the embedding features of entities, relationships and attributes of the design inversion fault knowledge graph extracted by the knowledge graph embedding module are fused to obtain a comprehensive entity embedding of the design inversion fault knowledge graph.

5. The knowledge graph fusion method based on multi-level alignment according to claim 4 is characterized in that: The entity similarity between different knowledge graphs is calculated based on the following formula: Among them, DIS(e m ,e n ) is the entity e in the design inversion fault knowledge graph m and debugging joint test fault knowledge graph entity e n The similarity of To design the entity e in the fault knowledge graph m Comprehensive entity embedding of To debug the entity e in the joint test fault knowledge graph n The comprehensive entity embedding of , ||·||1 is the L1 norm.

6. Knowledge graph fusion system based on multi-level alignment, characterized by: include: A graph acquisition module is used to acquire the design inversion fault knowledge graph, the debugging joint test fault knowledge graph, and the cross-domain collaborative fault knowledge graph; A model building module, used to construct a multi-level aligned knowledge graph fusion model, wherein the multi-level aligned knowledge graph fusion model includes a knowledge graph input module, a knowledge graph embedding module and a knowledge graph fusion output module; A graph fusion module, used for preprocessing the design inversion fault knowledge graph, the debugging joint test fault knowledge graph and the cross-domain collaborative fault knowledge graph based on the knowledge graph input module; The graph fusion module is also used to extract the embedded features of entities, relationships and attributes in the pre-processed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the knowledge graph embedding module; The graph fusion module is also used to fuse the pre-processed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the embedding features of entities, relationships and attributes extracted by the knowledge graph fusion output module based on the knowledge graph embedding module, so as to generate a fused fault knowledge graph; A fault diagnosis module, used for performing electronic equipment fault diagnosis based on the fused fault knowledge graph; The knowledge graph embedding module extracts the embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, including: For each entity in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph, and cross-domain collaborative fault knowledge graph, a discrete text embedding model is used to extract discrete text embeddings of tokens contained in the entity and a symbolic text embedding model is used to extract symbolic text embeddings of tokens contained in the entity to generate embedding features of the entity; The knowledge graph embedding module extracts the embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, including: Extracting embedded features of the relationships in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph, and cross-domain collaborative fault knowledge graph through the TransE model; The knowledge graph embedding module extracts the embedding features of entities, relationships and attributes in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph, including: For each attribute in the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph, and cross-domain collaborative fault knowledge graph, concatenate the embedding of the attribute and the embedding of the value of the attribute into a matrix, input the matrix into a convolutional neural network, and obtain the embedding feature of the attribute; The knowledge graph fusion output module fuses the preprocessed design inversion fault knowledge graph, debugging joint test fault knowledge graph and cross-domain collaborative fault knowledge graph based on the embedded features of entities, relationships and attributes extracted by the knowledge graph embedding module to generate a fused fault knowledge graph, including: The embedding features of entities, relationships and attributes of the design inversion fault knowledge graph extracted by the knowledge graph embedding module are integrated to obtain a comprehensive entity embedding of the design inversion fault knowledge graph; The embedding features of entities, relationships and attributes of the debugging joint test fault knowledge graph extracted by the knowledge graph embedding module are integrated to obtain a comprehensive entity embedding of the debugging joint test fault knowledge graph; The embedding features of entities, relationships and attributes of the cross-domain collaborative fault knowledge graph extracted by the knowledge graph embedding module are integrated to obtain a comprehensive entity embedding of the cross-domain collaborative fault knowledge graph; Based on the comprehensive entity embedding of the design inversion fault knowledge graph, the comprehensive entity embedding of the debugging and joint testing fault knowledge graph, and the comprehensive entity embedding of the cross-domain collaborative fault knowledge graph, the entity similarity between different knowledge graphs is calculated, the entity alignment is completed, and the fused fault knowledge graph is generated.

Images