537 results about "Graph embedding" patented technology

The invention relates to a risk management and control method and system based on real-time behavior analysis, and the method comprises the steps: carrying out the structural processing of multi-source behavior data through lightweight protocol decoding and behavior label embedding, and constructing an original behavior data set of a user and an entity; extracting multi-dimensional behavior characteristics by using a sliding window analysis and sparse representation mechanism, and constructing a user behavior graph by combining graph embedding learning; constructing a time-sensitive behavior trend model through streaming modeling and an incremental learning strategy, identifying an abnormal evolution trajectory in real time, and introducing a dynamic risk threshold regulation and control mechanism; adopting a high-throughput flow data processing and fast similarity matching algorithm to construct a fusion discrimination model, giving risk levels to abnormal behaviors and classifying the abnormal behaviors; and finally, performing closed-loop optimization in combination with a historical treatment effect. The system has the advantages of high real-time performance, high calculation efficiency, adaptability to complex network environments and the like, and the network security protection capability can be effectively improved.

The invention discloses a knowledge question and answer processing method fusing a large model and a knowledge graph, and relates to the technical field of artificial intelligence and natural language processing. Aiming at the defects of a traditional retrieval enhancement generation technology in the aspects of complex semantic association, context consistency and dynamic knowledge updating, the scheme adopted by the invention comprises the following two stages: knowledge graph construction and mixed index generation: collecting and processing internal and external multi-source data of an enterprise, and performing cleaning preprocessing such as coding normalization and de-duplication to obtain a mixed index; entities and relations are extracted through a pre-training model to generate a triple, a knowledge graph is constructed and stored in Neo4j, then a mixed index is generated through text and graph embedding fusion, and two types of retrieval are supported; retrieval and answer generation: obtaining user query, preprocessing, vectorizing, obtaining a candidate list through low-level semantic retrieval and high-level reasoning retrieval, fusing multi-dimensional indexes, rearranging and screening top-M candidates through Cross-encoder, constructing a JSON evidence list, and generating traceable answers through small model draft, large model fine calibration and consistency verification.

The invention provides an intelligent risk identification and analysis method based on multi-modal heterogeneous data fusion, which relates to the technical field of risk management, and comprises the following steps of: acquiring multi-source data, constructing symbol-nerve double-space processing, generating semantic vectors, extracting features through orthogonal matrix decomposition and a bidirectional long-short-term memory network, and obtaining an intelligent risk identification and analysis result; according to the method, multi-modal information is fused by using an adaptive weight mechanism, and a risk propagation topology network is constructed by applying a graph embedding algorithm, so that accurate identification and early warning of risks are realized, and the risk prevention and control capability and prediction accuracy are effectively improved.

The invention discloses a cerebral apoplexy risk and prognosis prediction system and method, and relates to the field of intelligent medical treatment, and the system comprises a data processing and knowledge construction layer which is used for extracting, cleaning and constructing a space-time multi-modal knowledge graph and structured clinical features from multi-source heterogeneous medical data; the feature engineering and fusion layer is used for deeply fusing dynamic semantic information in the space-time multi-modal knowledge graph and the structured clinical features through a graph embedding and attention mechanism to generate a fusion feature vector for a cerebral apoplexy prediction task; and the prediction model and output layer is used for performing cerebral apoplexy risk and prognosis prediction based on the fusion feature vector to obtain a prediction result, and generating a decision result for assisting a doctor in understanding the model through an interpretable mechanism. The method provided by the invention can improve the accuracy of stroke recurrence, bleeding transformation or function prognosis prediction, and provides a new way for accurate stroke management.

The invention discloses a document analysis method based on an RAG model and a dynamic knowledge graph, and relates to the technical field of artificial intelligence. The method is combined with an RAG model and a dynamic knowledge graph technology, and is realized by the following steps of: performing entity relationship joint extraction on an input document, generating a structural triple, and constructing a dynamically updatable knowledge graph; based on the knowledge graph, mapping entities and relationships into low-dimensional vectors by adopting a graph embedding model, and constructing a local vector knowledge base with a topological structure; receiving user questions in real time, encoding the user questions into query vectors, executing approximate nearest neighbor search based on the vector knowledge base, and matching related map fragments; and combining the retrieved graph fragments with the large language model, and generating a structured answer through path constraint of the injection knowledge graph. The method is used for solving the problem that in the prior art, a model cannot capture document deep semantics and dynamic relations insufficiently.

The invention provides a long text intelligent review and prediction method fusing a dynamic knowledge evolution mechanism, and relates to the technical field of text review, and the method comprises the steps: carrying out the structural analysis of a long text, constructing an initial knowledge graph, and generating an evolution knowledge graph through combining a time sequence change mode of an entity relationship in a historical text; calculating semantic similarity between word vectors and graph embedding to realize information interaction; deep semantic features are extracted to calculate the mahalanobis distance between the deep semantic features and an abnormal category prototype to determine an abnormal mode; and combining historical evolution trajectory modeling time sequence characterization to predict an abnormal development trend. And the accuracy and prediction capability of long text review can be effectively improved.

The invention relates to the technical field of intelligent operation and maintenance, in particular to an artificial intelligence operation and maintenance decision support method and system for multi-source information fusion. The method comprises the following steps: acquiring multi-source operation and maintenance data to perform multi-dimensional feature extraction to obtain multi-dimensional operation and maintenance feature data; performing continuous spatial cross-modal embedding according to the multi-dimensional operation and maintenance feature data to obtain cross-modal embedded data; performing heterogeneous feature coupling graph generation on the cross-modal embedded data to obtain coupling graph data; performing heterogeneous space fusion coding according to the coupling graph data to obtain fusion coding data; performing expert knowledge driving graph embedding on the fusion coding data to obtain operation and maintenance fusion graph data; performing root cause positioning reasoning according to the operation and maintenance fusion graph data to obtain root cause positioning data; and performing operation and maintenance decision generation according to the root cause positioning data to obtain operation and maintenance decision data. Through multi-source information fusion and intelligent reasoning, the root cause positioning accuracy and intelligent operation and maintenance decision efficiency of the system can be effectively improved.

The invention provides a tunnel risk reasoning method fusing a knowledge graph and a large language model, which comprises the following steps of: obtaining structured monitoring data and unstructured text data, adopting methods such as field standardization for the structured monitoring data to realize a unified format, adopting methods such as sentence segmentation and word segmentation for the unstructured text data to realize the unified format, and obtaining the structured monitoring data and the unstructured text data; the method comprises the following steps of: extracting entities from data by utilizing a model, extracting a relationship between the entities based on the entities, forming basic triads, forming a sub-graph by the basic triads, integrating to form a knowledge graph, generating a natural language, extracting the sub-graph related to the natural language from the knowledge graph, and converting the sub-graph into a sub-graph in a vector form by utilizing a graph embedding algorithm. The entities and the relation paths of the entities serve as explicit reasoning clues, the natural language, the sub-maps in the vector form and the explicit reasoning clues are input into a large language model, natural language output is generated, multi-source data information is integrated, and high-precision and interpretable tunnel risk early warning is output through the large language model.

The invention provides a cold rolling mill roller micro displacement real-time monitoring and dynamic compensation control early warning method, which relates to the technical field of monitoring control, and comprises the following steps: monitoring roller temperature field distribution through infrared thermal imaging, establishing a mapping relation between displacement and temperature gradient, calculating thermal deformation and compensating displacement drift; constructing a rolling mill equipment domain knowledge graph and a diagnosis rule base, and performing fault diagnosis by utilizing graph embedding learning; and generating a compensation optimization strategy and establishing fuzzy correlation mapping to calculate an optimal parameter. According to the method, accurate monitoring and compensation of the micro displacement of the roller are achieved, and the equipment fault early warning capacity and the rolling precision are improved.

The invention belongs to the field of bioinformatics, and relates to an improved integrated deep learning cell communication ligand-receptor interaction prediction method. The method comprises the following steps: firstly, carrying out extraction and dimensionality reduction on biological sequence features of a ligand and a receptor, and constructing multi-modal feature input; secondly, constructing an improved deep neural network branch, introducing a batch normalization layer and a Leaky ReLU activation function, solving the problems of gradient disappearance and neuronal necrosis, and improving regularization strength to prevent overfitting; meanwhile, an enhanced heterogeneous graph auto-encoder branch is constructed, the graph embedding dimension is remarkably expanded to improve the feature capacity, and full convergence of the model is ensured by increasing training rounds; thirdly, fusing the improved deep network with the prediction probability of a heterogeneous graph auto-encoder by adopting a weighted integration strategy; and finally, outputting a potential interaction relationship based on the fusion probability. By optimizing the architecture and the strategy, the prediction accuracy and robustness are remarkably improved, and a reliable tool is provided for analyzing a complex cell communication network.

The invention relates to a distributed source-load collaborative optimization method based on high-order topology and multi-scale attention, and the method comprises the steps: firstly providing a high-order graph construction method driven by structural interaction, and achieving the structural embedded expression of a physical interaction relation between multi-source equipment through a hyperedge-line graph mapping mechanism and functional attribute coding; secondly, a graph feature extraction method based on a multi-scale joint attention mechanism is designed, topology and state information are fused, and the inter-node adjustment collaboration recognition capability is improved; further constructing a source-load collaborative optimization scheduling model, introducing a particle swarm optimization algorithm to obtain an initial feasible strategy, and establishing a state-action mapping relation based on a deep reinforcement learning framework driven by graph embedding to realize autonomous learning and rolling optimization of a distributed control strategy; and finally, constructing an operation feedback closed loop mechanism, and introducing a graph structure migration and strategy adaptive updating method to enhance the response capability of the system to topological change and dynamic disturbance.

The invention discloses a power grid state characterization method and system based on multi-modal fusion. The method comprises the following steps: characterizing power grid topological structure data through a graph embedding algorithm to generate topological feature vectors; encoding the time sequence operation data through a long short-term memory network to generate a time sequence feature vector; extracting an equipment state feature vector through a multi-layer perceptron; projecting the three types of feature vectors to a shared semantic space, and realizing cross-modal feature alignment by using cosine similarity loss; and adopting a gating multi-mode unit GMU to carry out adaptive weighted fusion, and generating a unified power grid state representation vector. The system comprises a multi-modal data acquisition and preprocessing module, a topological structure characterization module, a time sequence dynamic characterization module, an equipment state characterization module and a cross-modal fusion and decision module. According to the method, the problems of incomplete single-mode characterization, characteristic isomerism and insufficient dynamic characteristic capture are solved, the accuracy and robustness of power grid state sensing are improved, and the method is suitable for real-time monitoring and decision support of an intelligent power grid.

The invention discloses a knowledge graph-based old building reconstruction scheme automatic recommendation method. The method comprises the following steps of S1, obtaining and preprocessing old building house data; s2, extracting a semantic entity and attribute relationship from a transformation case library, a building specification library and a construction scheme library, and constructing a knowledge graph; s3, mapping the house data to knowledge graph entity nodes, and executing graph embedding to generate building semantic representation; s4, constructing a graph neural network, calculating node semantic relevancy, and generating a building state vector; s5, inputting the building state vector into the reinforcement learning decision network, and optimizing the strategy to obtain an optimal transformation action; s6, screening reconstruction measures from the knowledge graph according to the optimal reconstruction action, and performing scoring to form candidate schemes; and S7, sorting the candidate schemes, selecting the scheme with the highest score, and recommending and updating the knowledge graph. According to the method, intelligent generation and self-optimization recommendation of the old building reconstruction scheme are realized, and the reconstruction efficiency of the scheme is remarkably improved.

The invention relates to the field of data processing, and provides an intelligent prediction recommendation method and system based on full-link data of a supply chain. The method comprises the steps of collecting multi-source heterogeneous supply chain data and performing standardization processing to obtain a standardized data set; according to the standardized data set, performing quantitative modeling on the supply chain state features to obtain a supply chain state feature vector; on the basis of the supply chain state feature vector, performing supply chain entity relationship modeling through a heterogeneous graph neural network to obtain a graph embedding vector; according to the graph embedding vector, performing dynamic generation on a candidate set through a multi-objective optimization algorithm to obtain a dynamic candidate set; and performing personalized matching of user requirements based on the dynamic candidate set to obtain a recommendation result. According to the invention, intelligent recommendation of the supply chain is realized, the decision-making efficiency is improved, and the operation cost is reduced.

The invention discloses an Internet of Vehicles information age optimization method and system based on graph reinforcement learning, and the method comprises the steps: firstly constructing a batch modeling and limited buffering queue structure of a vehicle state, carrying out the modeling of perception data into a multi-data-packet batch, and carrying out the queue management; modeling V2V link topology by using a graph neural network, and extracting large-scale channel embedding representation reflecting a topological structure; a multi-agent reinforcement learning system based on a centralized training distributed execution framework is constructed, each agent makes a decision according to a local state containing graph embedding features, a mixed action space is output, and an AoI opposite number of a receiving end is used as a reward; a graph embedding supervision mechanism based on a dominant function is introduced, so that topological features are aligned with a long-term optimization target; and network parameters are updated through multiple rounds of training, and finally, autonomous optimization control of each agent on packet loss and power is realized. According to the invention, data packet queue management and wireless resource allocation can be effectively coordinated, and efficient and low-overhead AoI minimization is realized in a complex dynamic topology environment.

The invention discloses an injection molding equipment anomaly detection method based on a graph neural network, and the method comprises the following steps: collecting multi-dimensional monitoring data in the operation process of injection molding equipment, and carrying out the normalization, denoising and missing value filling, and obtaining structured monitoring sequence data; dividing the injection molding production process into a plurality of process stages based on timestamps and process stage labels in the structured monitoring sequence data, and constructing a corresponding stage sub-graph set; based on the stage sub-graph set, the energy transfer path, the material flow path and the historical abnormal propagation path, constructing a nested graph structure; inputting the nested graph structure into an improved dynamic graph convolutional network for feature extraction to obtain a time-aware graph embedding representation; and performing anomaly detection on the time perception type graph embedded representation to generate an anomaly detection result of the injection molding equipment. According to the method, nested graph structure modeling and the dynamic graph convolutional network are adopted, and high-precision detection and rapid positioning of the abnormal state of the injection molding equipment are achieved.

The invention discloses a small sample remote sensing image classification method based on hierarchical spatial structure learning. The method comprises the following steps: firstly, extracting multi-scale features of a remote sensing image by using a ViT (Visual Transform) model, and capturing rich semantic information and spatial structure relationships in the image; secondly, constructing a graph structure based on spatial adjacency and attention weight to model a structured relationship between samples, and encoding graph node features through a graph convolutional network (GCN) so as to enhance the discrimination ability of the features in a structural semantic space; thirdly, a residual enhancement mechanism is introduced to fuse global semantic information, and the discrimination capability of graph embedding is improved; then, based on the structural similarity between the support set and the query set, performing classification decision, and realizing accurate classification under a small sample condition; and finally, carrying out joint optimization on the whole model by adopting a training strategy of a small sample meta learning task and a supervision loss function.

The embodiment of the invention provides a terminal threat event detection method and device, computer equipment and a storage medium. Comprising the steps of obtaining interaction data of a target terminal to construct a heterogeneous network behavior graph, and determining a target node needing to be detected and an associated path element graph; for each path element graph, according to the node similarity between the neighbor node and the target node, adjusting the neighbor node vector to obtain a first node embedding vector, and obtaining a graph embedding representation based on the plurality of first node embedding vectors; for each neighbor node, fusing the neighbor node vector and the graph embedding representation according to the calculated attention score to obtain a second node embedding vector; aggregating the plurality of second node embedding vectors and the target node to obtain a comprehensive embedding representation of the path element graph; and fusing the plurality of comprehensive embedding representations to obtain a target node, embedding and inputting the target node into the target classification model to obtain a threat event detection result. Therefore, the efficiency and accuracy of threat event detection can be improved.

The invention discloses an enterprise and policy service automatic matching method based on machine learning, and the method comprises the steps: carrying out the processing of enterprise multi-source data, constructing an enterprise portrait graph, and generating an enterprise graph embedding and structuring feature set; performing policy text analysis and condition recognition, constructing a policy condition graph and generating condition graph embedding representation; constructing a condition constraint field based on policy conditions, and generating a cross-graph alignment relationship and fusion features; integrating a multi-source feature input improved model to carry out joint modeling, and outputting a basic matching score; constructing a condition boundary manifold, and generating an anti-fact feature sample and a matching elasticity score; and based on the basic and elastic scores, generating a comprehensive score and outputting matching result information. According to the invention, by introducing graph structure perception alignment, conditional constraint guide interaction and a multi-channel scoring aggregation mechanism, high-precision, high-interpretability and intelligent reachability automatic matching between enterprises and policy services is realized.

The invention relates to an intelligent procurement collaboration method and system, and the method comprises the following steps: S1, building a supplier knowledge graph based on multi-source heterogeneous data fusion through employing a graph embedding algorithm, achieving the associated storage of industrial and commercial information, performance records and quality reports through a Neo4j graph database, calculating the weight of a supplier node through employing a PageRank improved algorithm, and carrying out the calculation of the weight of the supplier node; s2, based on the three-dimensional supplier portrait matrix, using an improved collaborative filtering algorithm to carry out demand matching, analyzing a purchase demand document through an ElasticSearch semantic analysis engine, combining TF-IDF weighted cosine similarity calculation to realize intelligent recommendation, and generating a purchase demand scheme with a weight score. The supplier knowledge graph is constructed through the multi-source heterogeneous data fusion and graph embedding algorithm, the scattered industrial and commercial information, performance records and quality reports are stored in an associated mode, traditional data island limitation is broken through, and effective integration of supplier multi-dimensional features is achieved.

The invention discloses a copious cooling rectification real-time monitoring and correction system based on digital twinning, and the system comprises a data processing module which is used for collecting and preprocessing data of a sensor; the graph structure construction module is used for constructing a twin graph structure and generating a topological adjacency matrix; the feature extraction module is used for executing graph embedding operation and extracting virtual twin features; the feature decoupling module is used for carrying out dynamic decoupling on the virtual twin tensor and constructing a feature sequence; the twinning correction module is used for inputting the state mapping vector into a twinning correction network and correcting a virtual twinning tensor; the variational inference module is used for inferring an output confidence interval and generating a reliability mask; and the parameter optimization module is used for adjusting the reflux ratio, the tower top temperature and the reboiling heat load in the credible area to realize real-time monitoring and correction. According to the invention, the intelligent monitoring capability and the self-adaptive control level of the cryogenic distillation process are obviously improved, and the system robustness, the energy-saving property and the industrial applicability are higher.

The invention belongs to the technical field of character pattern recognition, and particularly provides an oracle structure recognition method of a font pattern isomorphic network, which comprises three stages of font pattern structure feature extraction, font pattern isomorphic network model and oracle font structure recognition. Through oracle font skeleton extraction, oracle font skeleton singular point detection, skeleton burr and bifurcation optimization, and key point extraction as a node communication path as an edge, an oracle font graph structure is established, and graph structure nodes, edges and graph level features are calculated. The font graph isomorphic network integrates the advantages of a graph isomorphic network and dynamic edge and graph embedding, and learns vector representation of multi-level features such as nodes, edges and graphs of the oracle font graph structure so as to train and identify the font structure of the oracle. The method can be used for structure matching recognition of oracle character pattern images, has the advantage of being high in recognition accuracy, and is suitable for recognition of character pattern structures such as gold texts and seal scripts and recognition of social networks, biological information and molecular structures.

The invention discloses an energy storage system fault database indexing method, and particularly relates to the technical field of fault prediction. The method comprises the following steps: firstly, constructing an energy disturbance vector sequence matrix in a unified time window, extracting features based on a continuous variation rate and energy residual distribution, and generating an energy disturbance feature spectrogram; establishing an energy propagation path atlas in combination with a system module topological relation, and introducing a time sequence consistency identifier; a fault evolution fingerprint is generated through graph embedding coding, similarity index matching is carried out in combination with a standard fault trajectory, and a fault type and a positioning weight are output; real-time energy indexes are further fused, a micro-fault position prediction map is generated, and dynamic early warning is achieved; according to the method, accurate identification and visual positioning of the micro-fault of the energy storage system in a complex scene can be realized, and the operation safety and the intelligent operation and maintenance capability of the system are improved.

The invention discloses a multi-level network threat dynamic identification method based on a graph neural network, and the method comprises the following steps: collecting multi-source heterogeneous network security data, and carrying out the preprocessing; constructing a multi-level network threat graph; inputting the multi-level network threat graph into an improved GraphSAGE network to carry out graph embedding modeling; identifying an attack propagation path, and extracting a risk sub-graph region serving as a candidate attack chain; performing threat level evaluation on the risk sub-graph region, and calculating an overall threat score of the risk sub-graph region; and comparing the overall threat score with a preset threshold value, if the overall threat score exceeds the threshold value, determining that the threat is a high-risk threat, and outputting an early warning result. The multi-level threat graph is modeled through the graph neural network, attack chain recognition and threat evaluation are achieved, and the method has the advantages of being high in expressive power, accurate in recognition and fast in response.

The invention relates to a real-time interaction violation detection method, system and device and a medium. The method comprises the following steps: acquiring an online interaction session original information flow containing a user text sequence, a voice signal, an image file and an interaction behavior timestamp; extracting text semantic vectors, voice acoustic features and image visual content description, and integrating to generate an initial feature vector set; based on the interaction behavior timestamps, constructing an interaction time sequence diagram by taking the initial vectors as nodes, calculating multi-modal association weights among the nodes and updating connection edges to obtain a multi-modal fusion diagram; and inputting the fused graph into a graph neural network, outputting a global graph embedded vector through message passing and node aggregation, matching a preset violation mode vector library to calculate a similarity score, determining a violation type, and generating a risk assessment conclusion containing the violation type and confidence. According to the method, cross-statement and cross-modal context violation association is effectively captured, violation judgment accuracy is improved, and an intervention basis is provided for a platform.

The invention discloses a fusion graph recognition and symbolic expression-based resin-based composite material automatic native modeling and reverse design method and system, belongs to the field of composite material modeling and design, and particularly relates to a graph neural network and symbolic regression fusion-based composite material native modeling and structure reverse optimization method. The method comprises six steps of microcosmic image structure extraction, topological graph construction and graph embedding, constitutive relation symbol modeling, graph structure homogenization, performance-oriented reverse design and multi-modal performance prediction, and can realize an automatic process from microcosmic image to macroscopic performance prediction to structure optimization design. The problems that an existing method is low in efficiency, poor in interpretability and difficult in reverse design are solved, and the modeling efficiency and the design intelligence level of the composite material under multiple scales and multiple targets are improved.

Approaches in accordance with various embodiments provide unified training and inference frameworks useful for co-training, and performing inferencing using, models such as language and graphing models. The frameworks can be hierarchical, in that an inner layer can include a language model for generating semantic embeddings from input text content, and these semantic embeddings can be passed as input to an outer layer that can generate graph labels based, at least in part, upon a knowledge graph having these semantic embeddings substituted for textual content. Nodes of a knowledge graph that are determined to be related to the input content can then be used for various purposes, such as to generate tags for the content or determine how to route that content.

The invention discloses a communication network fault rapid positioning and recovery method based on self-supervised learning. The method comprises the following steps: S1, collecting and preprocessing communication network operation data; s2, constructing a communication network topological graph, and mapping the preprocessed data into node and edge attributes; s3, executing feature mask and structure disturbance, and constructing positive and negative comparison samples; s4, constructing a graph neural network model and pre-training by using positive and negative comparison samples; s5, calculating reconstruction errors of the nodes and the edges, and marking elements higher than a threshold value as abnormal candidate areas; s6, performing graph segmentation to extract continuous abnormal sub-graphs, and calculating sub-graph embedding and historical feature center similarity; and S7, matching an instruction sequence according to a positioning result, executing link scheduling, node reconfiguration and topology updating, and iteratively training the model. According to the method, high-precision rapid positioning and automatic recovery of communication network faults are realized under the condition of lack of a large number of labeled samples, and the method has a continuous optimization capability.

The invention belongs to the technical field of electricity price prediction, and particularly discloses an electricity price prediction method and system based on dynamic subgraph learning, a terminal and a medium. Comprising the steps of collecting multi-source electricity market data such as load, weather and market transaction, performing normalization and missing value filling, and constructing a dynamic electricity price information graph; dynamic sub-graph division is executed based on the edge weight calculated in real time among the nodes, and a density peak value clustering method is adopted to determine the center of the sub-graph and periodically update the center of the sub-graph; extracting spatial features in the sub-graph through a graph convolutional network, generating a sub-graph embedded vector, and inputting the sub-graph embedded vector into a bidirectional recurrent neural network to obtain time sequence features; utilizing a multi-head attention mechanism to realize interactive fusion among different sub-graphs to obtain global feature representation; and predicting the future electricity price in combination with the global features and the historical electricity price sequence. And in the face of new energy output fluctuation, load sudden change or market mechanism adjustment and the like, the prediction flexibility and accuracy are improved.

The invention discloses a zero-sample composite fault diagnosis method based on semantic graph embedding and multi-stage fusion, and the method comprises the following steps: firstly, generating an initial semantic descriptor through manual definition or statistical features according to known fault types, and constructing a fault relation structure diagram to represent the association between types; combining reconstruction, semantic comparison and propagation loss by using a graph convolutional network, and fusing the initial semantics and the relation graph to generate enhanced semantic features; meanwhile, a multi-modal model is constructed to extract vibration, temperature, acoustics and other signal features, and after multi-stage fusion of input-stage cross-modal attention, feature-stage Transform and output-stage semantic alignment, a semantic feature supervision training network is jointly enhanced; and finally, extracting unseen fault features in a zero sample scene and carrying out classified diagnosis. According to the method, through a multi-modal fusion and semantic enhancement strategy, the precision and generalization ability of zero-sample composite fault diagnosis are remarkably improved.