22 results about "Graph sequence" patented technology

The present application relates to the technical field of marine data processing and spatio-temporal prediction, and specifically discloses a sea surface core variable prediction method and system based on a wavelet-enhanced graph neural network.The method comprises the following steps: obtaining multivariate graph sequence data of a target sea surface; performing multilevel wavelet decomposition and gated fusion on each variable through a variable-level multiscale wavelet gated fusion module, and outputting enhanced multiscale time series features; inputting the enhanced multiscale time series features into a KAN-LSTM encoder module, performing recursive updating through gated fusion of a conventional convolution and a KAN convolution branch, and outputting encoder spatio-temporal features; inputting the encoder spatio-temporal features into a signed adaptive spatial graph convolution module, learning a signed sparse adaptive adjacency matrix and performing multi-order diffusion aggregation, and outputting a prediction result.The present application realizes long-term prediction of a target sea surface with high precision and high stability.

The application discloses an abnormal driving behavior detection method based on an improved residual double-layer graph attention network (Res-DBiGATv2), and belongs to the technical field of Internet of Vehicles. Firstly, the vehicle trajectory data is constructed into a space-time dynamic graph sequence according to time steps. Then, a ResBiGATv2 module is designed, and spatial feature aggregation is realized through double-layer graph attention, a multi-head mechanism and residual connection. Then, a ContraNorm contrast normalization layer is used to enhance feature uniformity and inhibit dimension collapse and oversmoothing. A graph external attention enhancement module GEA is introduced, and a learnable external memory unit is used to inject global information. Finally, a GRU is used to model time sequence features, and an abnormal node is identified through reconstruction error in an encoder-decoder framework. The application can accurately and timely detect various abnormal driving behaviors such as slow driving, overspeeding, following, and stagnation in complex traffic scenes, and significantly improves detection accuracy and robustness.

The application relates to a causal graph neural network prediction method and medium for eliminating AD false correlation. The method first constructs a graph structure of SNPs and brain images, extracts features using a time series graph network and maps the features to a disease quasi-time axis, and then divides stages and generates a graph sequence. The causal contribution degree between nodes is calculated through a time series prediction network to obtain a causal prior graph; direction correction is performed by using counterfactual flipping and graph convolution to obtain a refined causal graph. The features are subjected to causal enhancement, and double-mode information is fused through cross-attention. Finally, a classifier is input to realize AD staging prediction. The method overcomes the limitation that existing methods are difficult to distinguish between causality and false correlation in cross-sectional data, and through explicit modeling of disease time evolution and causal structure, the ability to capture early Alzheimer's disease and subtle pathological patterns is enhanced, thereby improving the accuracy, interpretability and cross-domain generalization ability of classification prediction.

This application provides a material vectorization method and system. The method includes: acquiring user profile data, material profile data, user behavior data, and material category structure data; constructing behavioral intent sequences and material click sequences based on user behavior data, and constructing a material graph structure based on the material category structure data; pre-training a material graph representation model by combining the material graph structure and material profile data; training a user behavior intent model by combining user profile data and behavioral intent sequences, and training a material click-through rate prediction model by combining the user behavior intent model, material click sequences, material graph representation model, and material profile data; and fine-tuning the parameters of the material graph representation model through backpropagation; after the material graph representation model gradually converges, using the fine-tuned material vectors as the vectors of the entire material system. This application offers more accurate and comprehensive vector representation, more comprehensive sequence representation, better model interpretability, and greater versatility.

This invention relates to the field of semiconductor equipment manufacturing technology, specifically to a method, system, device, and medium for semiconductor equipment fault prediction based on a time-series graph neural network. The method first collects time-series operational data of each component of the equipment to form a data matrix. The data matrix is ​​then preprocessed and segmented to obtain standardized segments. Next, a time-series graph is constructed for each segment: using components as nodes, edge weights are determined based on physical connections and Pearson coefficients to form a time-series graph sequence. This sequence is then divided into training and testing sets, input into a time-series graph neural network model for training, and optimized using a cross-entropy loss function. Finally, real-time data, processed through the above steps, is input into the trained model, outputting the fault type and probability, and displaying and pushing the results. By constructing a time-series graph data structure and designing a dedicated time-series graph neural network model, accurate early fault prediction is achieved.

This invention provides a reconstruction-based graph-text alignment method for large language models, belonging to the field of large language model technology. It includes: acquiring a text attribute graph and text instructions, serializing the text attribute graph according to a neighborhood detail template to obtain a graph label sequence, aligning it in the text space using a projector to obtain a prefix graph label sequence, and inputting this sequence and the text instructions into a large language model to output graph label hidden representations and prediction outputs; a reconstructed label representation is obtained by aggregating the mean of the graph label hidden representations; a graph reconstruction loss is calculated based on the reconstructed label representation, and a text autoregressive loss is calculated using the prediction output, and a joint loss is obtained; the projector and the large language model are optimized by minimizing the joint loss to complete the graph-text alignment. This invention solves the problems of existing graph labeling large models relying solely on text instructions for supervision, resulting in insufficient graph-text alignment, inadequate utilization of graph structure information, and low performance.

The application discloses a speech separation method and related device, electronic equipment and storage medium, comprising: acquiring audio stream and video stream synchronously collected in a speech dialogue scene; detecting and tracking based on the video stream to obtain a first graph sequence of the face of each of a plurality of objects; based on a specified instruction for the plurality of objects, selecting a specified object as a target object; based on a second graph sequence of the lips of the target object, extracting features to obtain a time sequence lip shape feature, and based on the first graph sequence of the target object, determining whether the target object is a registered object; based on whether the target object is a registered object, selecting a target voiceprint feature of the target object obtained by any one of a registered voiceprint library, a temporary voiceprint library, and a temporarily generated voiceprint; and based on the time-frequency speech feature of the audio stream, the time sequence lip shape feature, and the target voiceprint, performing prediction to obtain audio data of the target object in the audio stream. The above scheme can improve the accuracy and robustness of speech separation.

The application relates to the technical field of traffic network topology optimization and graph sequence technology, and particularly relates to a traffic network optimization method based on a graph sequence problem of a weighted graph, which comprises the following steps: S1: acquiring a network topology structure and a degree sequence of a target traffic network and a preset weighted graph type; S2: selecting a corresponding weighted graph solving algorithm based on the preset weighted graph type; S3: judging whether the degree sequence of the target traffic network is a graph sequence based on the weighted graph solving algorithm; if yes, executing step S4; S4: generating a weighted graph of a corresponding type based on the selected weighted graph solving algorithm and the degree sequence of the target traffic network; and S5: optimizing the network topology structure of the target traffic network based on the weighted graph of the target traffic network, and generating an optimized traffic network. The application can improve the accuracy and reliability of traffic network topology optimization.

This invention discloses a multi-task spatiotemporal prediction method based on snapshot cues. It combines dynamic graph pre-training with multi-task prediction of future snapshots and consists of three stages: In the pre-training stage, the pre-training task is redefined using dynamic subgraph sequences, and a multi-granularity evolutionary graph convolution based on self-supervised training is proposed to extract local and global features from the dynamic graph. In the cues stage, a novel subgraph cues method is introduced, including trainable cues for nodes and edges in the context graph, which can better capture the evolutionary information of future snapshots. In the fine-tuning stage, meta-learning is used to update the subgraph cue parameters, enabling the cues to effectively adapt to diverse downstream tasks. Extensive experiments on real-world datasets demonstrate that this method achieves state-of-the-art performance.

The application relates to the technical field of unmanned aerial vehicle trajectory prediction, and discloses a multi-source data driven unmanned aerial vehicle flight trajectory prediction method and system, multi-source data of an unmanned aerial vehicle flight state and an external environment are collected and preprocessed, abnormal data is detected, interpolation completion is carried out based on adjacent time points, a spatiotemporal graph sequence is formed based on a spatiotemporal decoupling graph construction method, an input sequence is intercepted through a trend smoothing adaptive window selection algorithm, an improved Graph Transformer model is input, double-branch coding is carried out to extract and fuse features, and then sparse self-attention decoding is carried out to predict a trajectory. The scheme effectively mines a spatiotemporal coupling relationship, reduces a calculation amount, and realizes high-precision and high-real-time prediction of an unmanned aerial vehicle trajectory in a complex environment.

The application belongs to the field of tunnel engineering intelligent construction, and discloses a tunnel intelligent supporting method and system based on a graph structure, which comprises a time-varying graph structure model; in response to the occurrence of an engineering event, the time-varying graph structure model is dynamically updated to form a current subgraph matching the current engineering state; based on the subgraph sequence of the time-varying graph structure model at the current and historical time slices, a predicted value of a supporting parameter in the next stage is output; when the risk of the current supporting effect exceeds a threshold value, a risk factor chain is extracted to update the correlation relationship weight in the time-varying graph structure model; based on the updated weight, the predicted value of the supporting parameter is iteratively optimized to output a final supporting scheme, and the model is updated by using the actual engineering data after the implementation of the scheme and enters the next decision cycle. The application solves the problems of mismatching of supporting schemes, high safety risks and poor economy caused by insufficient data correlation expression, static design and lagging risk feedback in the existing tunnel supporting technology.

The application provides a network violation external connection detection method and system, and belongs to the technical field of network security. The method constructs a network topology graph sequence containing historical and current states according to time; a structure prediction model is used to process the historical sequence to generate a current time prediction graph level representation, and a structure prediction error of the current actual graph level representation is calculated; when it is determined that there is an anomaly, an intervention graph is constructed for a target node and an error after intervention is determined; a difference between the structure prediction error and the error after intervention is taken as an error improvement amount, and a node with an improvement amount greater than a threshold is determined as a high-risk node. The application introduces a reverse intervention mechanism based on the structure prediction error, quantifies the contribution degree of a specific node to the deviation of the overall structure evolution, realizes accurate positioning of the violation external connection node, and significantly improves the accuracy and interpretability of detection.

This application relates to a machine learning-based method and system for quality monitoring of self-piercing riveted joints. The method includes: acquiring dynamic physical process diagrams of acoustic emission events during the self-piercing riveting process, and summarizing these events chronologically to obtain a sequence of dynamic physical process diagrams; constructing and training a multimodal spatiotemporal graph convolutional network model based on a physical constraint loss function; inputting the sequence of dynamic physical process diagrams into the trained graph convolutional network model, outputting a three-dimensional residual stress field, and mapping it onto the geometric model of the riveting area to obtain a three-dimensional stress cloud map and perform quality assessment to obtain the joint quality grade; combining the spatial coordinates of the three-dimensional stress cloud map, the joint quality grade, and the maximum residual tensile stress value to obtain the joint quality monitoring result. This method, by embedding a network model with physical constraints, can still output a physically consistent stress field distribution even in the absence of a large amount of labeled data, significantly improving the depth and reliability of monitoring.

This application provides a distributed data consistency verification method and system based on graph neural networks, belonging to the field of data verification technology. The method includes: deploying data probes at all nodes in the power grid meter reading data flow chain to synchronously collect and acquire power data status distribution sequences within historical time periods, and constructing a global data status graph sequence; identifying the temporal fluctuation patterns of the global data status graph sequence and obtaining the graph fluctuation deviation sequence; based on the graph fluctuation deviation sequence, using a data anomaly diagnostic tool built on graph neural networks to perform anomaly detection and root cause tracing on the global data status graph sequence, outputting the distribution of inconsistent nodes and the probability distribution of anomaly diagnosis, and generating a consistency repair strategy to perform closed-loop consistency governance of power grid meter reading data. This solves the technical problems of low anomaly diagnosis accuracy, inaccurate root cause tracing, and lack of targeted repair strategies in existing distributed data consistency verification technologies.

This invention relates to the field of code summarization technology, specifically to a code summarization method based on attention mechanism and graph sequence association, comprising: 1. Processing the code: segmenting the source code into words to obtain a code sequence; parsing the code to obtain an abstract syntax tree; and obtaining a code graph by adding a sequence stream and a data stream; 2. Encoding the code sequence and code graph using a sequence encoder and a graph encoder respectively, to obtain a sequence encoding vector and a graph encoding vector; 3. At the decoder end, inputting the starting word of the code summarization. <bos>In the decoder, attention is calculated simultaneously based on both the sequence encoding vector and the graph encoding vector. After fusion, the decoder output, i.e., the predicted word, is obtained through a fully connected mapping. Fourth, the output word of the previous time step is used as the input of the decoder at the current time step, and step three is repeated until the decoder output encounters the terminating word. <eos>End. This invention provides a superior code summary.< / eos> < / bos>

A method, a device and a recording medium for extracting a qualitative characteristic of a video may include segmenting a video into at least one video segment, and based on an edge representing a relationship between objects of the video segment and the objects, generating video graph sequences in a graph form of the video segment, and generation of the video graph sequences may be performed based on an object feature vector expressing a qualitative characteristic of the objects as a vector and an edge feature vector expressing a qualitative characteristic of the edge as a vector.

The present application belongs to the technical field of power system intelligent operation and maintenance, and discloses a power distribution room early warning method based on power semantic constraints and a space-time graph neural network; a power distribution room video stream and an electrical state signal are acquired, a virtual synchronous layer mapping is constructed to generate a synchronous multi-modal event; an entity object is extracted as a graph node, an electrical potential energy decay attribute is injected into a device node to calculate a non-static safety boundary, and a dynamic semantic graph is constructed; a dynamic semantic graph sequence is input into a space-time graph neural network, a physical law constraint layer is embedded to perform physical consistency correction on personnel trajectory and arc spread prediction; a cross-modal micro-causal edge is constructed based on the micro-time difference of operation and response as an attention constraint of time coding; and finally, a risk evolution probability is output and a graded early warning instruction is generated. The present application realizes deep fusion of multi-modal data, and significantly improves the accuracy and rationality of risk prediction.

The application provides a knowledge integration-based Internet of Things network situation deduction method. The method comprises the following steps: constructing a network time sequence knowledge graph according to entity concepts, resource mapping and data attributes of the Internet of Things; constructing a historical evolution space-time graph sequence based on the network time sequence knowledge graph, and obtaining a space-time evolution factor of a node in the Internet of Things through a space-time graph encoder and a historical information embedder according to the historical evolution space-time graph sequence; designing a historical evolution encoder according to the space-time evolution factor of the node, and solving a network space-time evolution factor through the historical evolution encoder; and perceiving a network situation of the Internet of Things at the next moment through a situation prediction decoder according to the network space-time evolution factor. The method constructs an Internet of Things time sequence knowledge graph, effectively integrates network information of the Internet of Things, further constructs a network space-time graph sequence, and cooperatively designs a historical evolution encoder and a situation prediction decoder to perceive network situation information at a future moment.

This invention discloses a traffic flow prediction method based on graph serialization and bidirectional state space. First, a spatiotemporal graph of traffic flow is constructed. Topological features of nodes are extracted using a graph attention network, and a double-random permutation matrix reflecting the optimal traversal path is generated through a differentiable relaxation sorting operator, adaptively mapping the two-dimensional graph structure data to one-dimensional serialized features. Then, a bidirectional gated state space model is constructed, and the serialized features are recursively scanned from both the forward and reverse directions using discretized state space parameters, and a hybrid hidden state is obtained through fusion using gated units. Finally, the transpose of the permutation matrix is ​​used to perform an inverse transformation on the hybrid features to restore the spatial structure, and the future traffic flow is output through a prediction head. This invention effectively solves the problem that traditional state space models struggle to handle graph structure data, achieving end-to-end joint optimization of the graph traversal strategy and the prediction model, significantly improving the accuracy and efficiency of long-sequence traffic prediction.