254 results about "Neural network learning" patented technology

The invention relates to the technical field of mutual inductor tests, and provides a mutual inductor test abnormal data automatic filtering analysis method and system, and the method comprises the steps: obtaining magnetic field data, electric field data and thermal field data, respectively carrying out the feature extraction and weighted fusion, and obtaining a multi-physical field fusion feature vector; constructing an electromagnetic induction chain type propagation graph, calculating the weight of an edge in the graph, constructing a magnetic flux conservation constraint graph convolutional neural network, learning propagation and evolution modes of transformer test abnormity, and obtaining a coupling abnormity feature vector; the coupling anomaly feature vectors are classified, normal data, single physical field abnormal data and multi-physical field coupling abnormal data are recognized, and corresponding data are judged as abnormal data and filtered; and carrying out physical mechanism analysis on the filtered abnormal data to obtain an abnormal detection report. According to the invention, high-precision automatic identification, filtering analysis and physical mechanism traceability diagnosis of transformer test abnormal data are realized.

The invention relates to an unmanned aerial vehicle track prediction method based on multi-feature LSTM, and belongs to the technical field of data processing, and the method comprises the following steps: 1, constructing a six-degree-of-freedom global motion state matrix of a target unmanned aerial vehicle; 2, performing joint estimation to obtain a three-dimensional wind field vector acting on the target unmanned aerial vehicle, and extracting the real-time speed direction of the target unmanned aerial vehicle; 3, fusing the historical state sequence, the three-dimensional wind field vector and the speed direction unit vector, and constructing a multi-dimensional time sequence feature vector; 4, the LSTM neural network learns a nonlinear maneuvering mode of the target unmanned aerial vehicle under the influence of the wind field through a door control mechanism of a forgetting door, an input door and an output door; and step 5, outputting the position probability distribution of the target unmanned aerial vehicle at a plurality of time points in the future through the LSTM neural network. The method has the advantages that the space-time cone representing the future trajectory is formed, and trajectory prediction of the target unmanned aerial vehicle is completed.

The invention relates to the technical field of electric vehicle charging pile planning, in particular to an intelligent charging pile layout optimization method based on a multi-scale space-time diagram neural network. Comprising the following steps: S1, constructing a heterogeneous dynamic graph which represents a potential charging pile position node set, epsilon t represents a time-varying edge set, At represents a time-varying adjacent matrix, and Xt represents a node feature matrix; s2, calculating a self-adaptive adjacency matrix with a specific relationship, and calculating a multi-type dynamic relationship between capture nodes of the self-adaptive adjacency matrix based on the constructed heterogeneous dynamic graph model; s3, updating a structure bias matrix, and capturing dynamic change characteristics of the network; s4, calculating distance measurement between nodes, and fusing geographic information and semantic information; s5, through graph neural network learning, based on the constructed heterogeneous dynamic graph and the calculated adaptive adjacency matrix, executing a graph neural network learning process, and extracting node space-time representation; and S6, predicting a future charging demand based on node representation obtained by graph neural network learning, and generating based on the future charging demand.

The invention belongs to the field of track planning and intelligent control of a reusable vehicle, and relates to an intelligent track planning method for a cruise section of the reusable vehicle. The method comprises the following steps: constructing a three-degree-of-freedom centroid motion dynamics model of a cruise section vehicle, generating an offline optimal trajectory sample library covering wide working conditions based on a Gaussian pseudo-spectral method, designing a mapping relation between deep neural network learning task parameters and trajectory features, and outputting a high-quality trajectory planning initial value online. The distribution of collocation points is dynamically adjusted in combination with a self-adaptive collocation point method, a nonlinear programming problem is iteratively solved with a high-quality initial value as a starting point, and efficient and accurate optimization of the trajectory is achieved. Simulation results show that the method significantly reduces the sensitivity of online optimization to initial guess, still has fast response ability and high robustness under complex constraint and uncertainty working conditions, effectively improves the autonomous trajectory planning ability of the cruise section of the reusable vehicle, and ensures flight reliability and control precision.

The invention relates to the field of medical artificial intelligence, in particular to a laryngeal cancer early intelligent diagnosis system and method based on multi-modal deep learning, and the system comprises a multi-modal data collection module, a cross-modal feature extraction module, a cross-modal attention network module, an expert-level knowledge distillation network module, and a focus evolution prediction and diagnosis decision and visualization module. Endoscope images, acoustic features and clinical data are collected, a system extracts high-dimensional feature vectors, a cross-modal attention network is used for feature fusion, an expert-level knowledge distillation network is combined with pathology and expert experience, neural network learning is guided, a lesion evolution prediction module tracks lesion changes, risk prediction is generated, and finally, the lesion evolution prediction module is used for predicting the lesion change. And the diagnosis decision and visualization module generates a diagnosis result and explanation. The system improves the early detection rate of laryngeal cancer through multi-modal data fusion.

The invention provides a disease and pest epidemic prevention system based on image recognition, and the system comprises a data collection and preprocessing module which is used for collecting optical images, sound wave images and environment information of crops, and carrying out the preprocessing and feature extraction; the image enhancement module is used for processing the sound wave features, enhancing the optical features through a sound wave weight map obtained through processing, and marking the pest and disease region through a neural network to obtain an optical enhancement image; the image segmentation module is used for carrying out region division on the optical enhancement image and then carrying out pixel-level segmentation to obtain a to-be-identified region map; and the disease and pest recognition module is used for outputting a disease and pest recognition result graph through a disease and pest recognition model after learning a double mapping relation between the to-be-recognized area graph and the environment weight vector through a neural network. According to the invention, through comprehensive application of multi-modal information fusion and a deep learning technology, the efficiency and precision of crop disease and insect pest detection are significantly improved.

The invention discloses a single cell space transcriptome data analysis method, device and system, and a storage medium. The method comprises the following steps: acquiring single cell space transcriptome data; the spatial transcriptome data comprises a gene expression map and spatial site coordinates; constructing a graph structure considering gene expression similarity and spatial position continuity on the basis of the data, and performing data representation on a gene expression graph by using an adversarial auto-encoder; in combination with a graph neural network, robust potential characterization is learned, training loss is constructed through reconstruction of a gene expression map, and meanwhile, a clustering prediction result of spatial sites is obtained in combination with mcluster clustering. By adopting the technical scheme of the invention, the spatial clustering analysis with higher precision can be realized, and the key spatial functional region in the biological tissue can be identified.

The invention provides a three-dimensional reconstruction method based on a structured Gaussian and neural kernel field, and the method comprises the steps: taking a multi-view image, internal and external parameters of a camera, and a point cloud as input, firstly constructing a structured Gaussian representation, optimizing the structured Gaussian representation, then constructing a neural kernel function field of a scene, taking a sparse voxel grid and an anchor point as a basic space structure of a kernel function, and carrying out the optimization of the neural kernel function field; and related features of the kernel function are predicted through neural network learning. According to the method, a mutual guidance mechanism between the structured Gaussian representation and the neural kernel function field is introduced to collaboratively optimize the two representations. And finally, performing efficient view rendering by using the structured Gaussian representation, and outputting a high-precision three-dimensional surface model. According to the method, high-quality rendering and reconstruction output can be realized by utilizing compactness and efficiency of structured Gaussian representation in processing a large-scale scene and combining scalability and robustness of a neural kernel function field in implicit surface learning, and the method is suitable for understanding and application of complex and large-scale three-dimensional scenes.

The invention discloses an air purification and temperature and humidity coordination control method and system for a central air conditioner, and the method comprises the steps: learning past and future environment change trends based on a Bi-LSTM bidirectional neural network, adding a one-dimensional CNN convolution layer before Bi-LSTM, and extracting a local space mode of sensor data; introducing an attention mechanism to dynamically weight a pollutant concentration peak value and a temperature and humidity sudden drop point in the input sequence; inputting the multi-modal data into the hybrid prediction model for prediction, and outputting a prediction result; establishing a dynamic optimization model by taking minimization of energy consumption, maximization of air cleanliness and maintenance of temperature and humidity stability as targets through an NSGA-II multi-objective optimization algorithm; and inputting the prediction result, the multi-modal data and a threshold value set by a user into a dynamic optimization model, and outputting an air purification adjustment parameter and a temperature and humidity adjustment parameter. The efficiency and accuracy of temperature and humidity control and air purification control are improved.

The invention belongs to the technical field of computers, and particularly relates to a personalized recommendation method and system based on a dynamic heterogeneous graph and reinforcement learning. The method comprises the following steps: firstly, constructing a global heterogeneous information graph of multiple types of nodes offline, and learning static embedding of the nodes by using a graph neural network; secondly, dynamically constructing a session history into a session graph in a real-time interaction process of the user, and aggregating by adopting a graph convolutional network to generate a dynamic state vector of the user; inputting the dynamic state vector into an actor and commentator reinforcement learning framework; and finally, using a dominant function calculated by the commentator network as a stable learning signal, and performing end-to-end joint training on the whole model to optimize long-term cumulative return. According to the method, by introducing the session graph volume accumulation device, the accuracy of dynamic state representation is remarkably improved; and an actor commentator framework is adopted, so that the problem of high variance of a traditional strategy gradient method is effectively solved, and the training stability and efficiency are improved.

The invention belongs to the technical field of six-axis robots, and discloses a method for predicting robot joint friction based on an improved PINN principle and a LuGre model, and the method specifically comprises the following steps: 1, generating a robot excitation track, and collecting the positions, speeds and currents of six joints of a robot; 2, current estimation torque of each joint of the robot is calculated according to the current, and a friction force-speed-displacement mapping of the joints of the robot is obtained; 3, constructing a friction model of the robot based on the improved PINN principle and the LuGre model; and 4, designing a double-neural network learning strategy of the robot friction model. According to the method, the improved physical information neural network PINN and the LuGre model are fused, the multi-joint coupling effect of the six-axis robot and the friction characteristic in the full-speed range can be considered at the same time, accurate prediction of the joint friction force of the six-axis robot is achieved, the prediction precision is improved, and the prediction efficiency is improved. And the problem of physical information loss caused by model simplification is also avoided.

The invention discloses a vehicle refitting scheme generation method and system based on a large model and a knowledge graph, and relates to the technical field of electric vehicle networking, and the method comprises the steps: inputting an obtained user refitting demand text into a cloud large model, analyzing the refitting demand text, extracting key information, converting the key information into a query vector, and sending the query vector to a cloud server; the method comprises the following steps: establishing a vehicle refitting knowledge graph, performing matching retrieval with entity embedding in the commercial vehicle refitting knowledge graph, constructing cue words according to the retrieved knowledge, generating a vehicle refitting scheme through cloud big model thinking chain reasoning, calling a 3D component to render a scheme effect graph, and feeding back the scheme effect graph to the local; vehicle data, accessory data, scene data, law and regulation data and historical modification case data are obtained in advance, after data preprocessing, entities and relationships are obtained through named entity recognition and relationship extraction, a commercial vehicle modification knowledge graph is constructed, and entity embedding in the knowledge graph is obtained through graph neural network learning. According to the invention, a refitting scheme which better meets user requirements can be obtained.

The invention relates to the technical field of energy conservation and carbon reduction in the high-energy-consumption industry, and discloses a system air leakage and carbon emission intelligent monitoring method in the high-energy-consumption industry, which comprises a data preprocessing link, a gas detection link, a carbon emission accounting link, a data fusion weight adjustment link and a carbon emission abnormity check link. The system innovatively develops a data deep processing mechanism, and adopts a multi-dimensional data calibration technology to carry out total factor correction on factors such as temperature and pressure fluctuation and humidity interference; the method comprises the following steps: measuring and calculating air leakage volume flow, calculating air leakage oxygen molar flow by combining with an oxygen proportion characteristic in air, carrying out cross validation on carbon emission through an oxygen balance and material balance double algorithm, learning historical working condition data by using an LSTM neural network, dynamically adjusting an algorithm weight built-in constant verification mechanism, and starting a retest and calibration program when the data is abnormal. According to the invention, through full-process intelligent cooperation, accurate monitoring of carbon emission and dynamic verification of abnormity are integrated, and a systematic solution is provided for low-carbon transformation in the high-energy-consumption industry.

The invention provides a three-dimensional flow field acquisition method and system based on two-dimensional through-flow and a neural network, belongs to the field of fluid mechanics calculation, and can at least partially solve the problem that in the prior art, calculation efficiency is low, and two-dimensional through-flow cannot reflect three-dimensional features. Geometric parameters and working condition parameters of fluid machinery are input, and flow parameters on a two-dimensional flow surface are rapidly output; constructing and training a neural network model, wherein the neural network learns a mapping relation between a two-dimensional flow surface calculation result and a three-dimensional flow field; for the fluid machinery under the target working condition, a result is obtained through two-dimensional through-flow calculation, the result is preprocessed and then input into the trained neural network, the low-dimensional representation of the three-dimensional flow field is output, and complete three-dimensional flow field parameters are obtained through reconstruction. The calculation period is remarkably shortened, and rapid optimization of multiple schemes is supported.

The invention discloses a drug recommendation system based on causal co-occurrence enhancement, which fuses a co-occurrence relationship and a causal relationship between medical information to realize high-precision and high-safety personalized drug recommendation. Comprises: a medical information representation generation module based on a medical information co-occurrence relationship, used for constructing a medical entity co-occurrence graph and obtaining embedded representation through graph attention neural network learning; the patient treatment characterization generation module based on the medical information causal relationship constructs a treatment-level medical causal graph by using a causal discovery algorithm, and performs weighted fusion according to the causal roles (such as cause nodes and intermediate nodes) of entities to form patient treatment characterization; according to the drug recommendation module based on the co-occurrence causal relationship, co-occurrence enhancement factors and a causal effect matrix calibration mechanism are introduced on the basis of preliminary prediction, and it is ensured that a recommendation result has statistical rationality and causal interpretation at the same time. The accuracy, interpretability and clinical applicability of drug recommendation are remarkably improved.

The invention relates to the technical field of leather products, and particularly discloses a leather product feature recognition and authentication method and system based on deep neural network learning. According to the scheme, visible light, near-infrared reflection and polarized light interference images of leather are obtained through a multi-mode imaging technology, surface texture and color, internal fiber structure and section fiber arrangement features are extracted respectively and converted into standardized feature vectors, the three types of features are fused to construct a comprehensive leather feature vector, and the comprehensive leather feature vector is obtained. A deep neural network is input for authenticity identification, a confidence score is output, objective quantification of full-dimensional physical characteristics from the surface to the interior is realized by outputting the confidence score, subjectivity and one-sidedness of traditional artificial identification are overcome, and the method has efficient and large-scale accurate identification capability.

This invention discloses a method for predicting and protecting the overheating risk of asynchronous motors, specifically relating to the field of motor control and protection technology. Based on an intelligent fusion model, it estimates temperature and thermal stress field, calculates the rate of change of thermal stress, non-uniformity, and hotspot trends, and calculates a dynamic health index using historical data. These parameters are then input into a multi-objective reinforcement learning controller to optimize long-term health and short-term performance, generating a thermal shaping control vector to regulate the motor. This invention combines physical mechanisms with data-driven approaches through an intelligent fusion model, utilizing a graph neural network to learn the structure of the heat conduction graph and verify physical laws, thereby improving the accuracy of thermal state estimation. It achieves a multi-dimensional risk characterization combining transient impact and cumulative effects through the rate of change of thermal stress, non-uniformity, hotspot trends, and dynamic health index. By optimizing long-term health and short-term performance losses, a thermal shaping control vector is generated to achieve regulation from passive protection to active prevention, extending the motor's service life.

The invention relates to a commodity sales volume prediction method and device based on multi-modal feature fusion and a storage medium, and is applied to the technical field of store commodity sales volume prediction, and the method comprises the steps: obtaining static and dynamic multi-modal features affecting the store commodity sales volume, and carrying out the unique coding representation of a store and a commodity; the feature processing layer converts unique coding representation into dynamic embedded features and then performs weighted fusion on the dynamic embedded features and static features through feature gating, the feature gating has the advantage of dynamic weight distribution, and adaptive adjustment cannot be performed according to input sample features due to the fact that static weight distribution is adopted in a traditional method; then, the dynamic features continue to be fused through a cross attention mechanism, and soft alignment and weighted fusion of cross-modal features are achieved through the attention mechanism; then, the incidence relation between fusion features is learned through a multi-layer feedforward neural network of a calculation layer; and inputting the final fusion feature into a task layer to obtain a sales prediction result.

The embodiment of the invention provides an obstacle avoidance track generation method based on a deep network, and the method comprises the steps: 1, generating expert data, generating an initial path based on a given starting point and a given terminal point in a randomly generated scene containing an obstacle, carrying out the optimization of the initial path, and carrying out the generation of the expert data; generating a data set which can be divided into a training set and a test set; 2, in an imitation learning and online reasoning stage, constructing a deep neural network, and training the data set to enable the deep neural network to learn a mapping relation from environment information and a target position to a planned trajectory; and 3, online application: inputting environment perception data acquired in real time and a current target point into the trained deep neural network, and outputting a smooth obstacle avoidance trajectory through the deep neural network. According to the method, high-quality expert data is generated by using a traditional planning algorithm, a deep neural network is trained to simulate planning behaviors of experts, and then rapid online reasoning is performed by using the trained network.

The invention relates to the technical field of sensor data analysis, in particular to a multivariable time series anomaly detection method and system based on a multi-scale graph neural network, and the method comprises the following steps: data preprocessing, multi-scale feature extraction, graph structure learning, graph neural network learning, and prediction and anomaly score calculation. The method has the beneficial effects that by learning the similarity, changing along with time, between the sensors, the complex relation between the sensors can be modeled more accurately, and therefore abnormity caused by interaction between the sensors can be detected more effectively; by learning the similarity between the sensors, the method can provide information about which relationships between the sensors are most important for anomaly detection, thereby enhancing the interpretability of the model. Therefore, the user can understand the cause of the abnormity and take corresponding measures.

The present disclosure relates to the technical field of automatic intelligent auxiliary detection of electrocardiogram, and proposes a self-supervised neural network learning heartbeat classification method and system based on pre-labeling, which comprises the following steps: preprocessing the acquired electrocardiogram data to obtain heartbeat data; performing time-frequency analysis on the acquired one-dimensional heartbeat data to convert it into a two-dimensional time-frequency graph; inputting the converted two-dimensional time-frequency graph into a trained self-supervised learning network model to classify the heartbeat data and obtain a classification result; the self-supervised learning network model is trained in combination with the SimCLR method and the clustering method, so that the loss function term changes from one item to two items, the learning effect is strengthened, and the classification accuracy of the heartbeat data can be improved.

This application discloses a method for identifying groove areas in aircraft parts, relating to the field of machining, and aims to solve the technical problem of low efficiency in existing methods for identifying groove areas in aircraft parts. The method includes the following steps: performing a first identification on a three-dimensional model image of the aircraft part to obtain a first identification result, wherein the first identification involves identifying groove areas in the three-dimensional model image of the aircraft part. The first identification includes a first coarse identification based on mathematical morphological pattern difference evaluation, a second coarse identification based on the feature patch attributes of the groove region's neighborhood, and a third coarse identification based on deep neural network learning; obtaining groove area identification result data based on the first identification result; and interactively judging the groove area identification result data with the three-dimensional model image of the aircraft part to obtain the groove areas in the three-dimensional model image of the aircraft part. This application can achieve automated identification of groove areas in aircraft parts, improving the efficiency of groove area identification.

A neural network is trained for estimating patient hemodynamic data using a plurality of extravascular imaging data sets and a plurality of intravascular imaging data sets that are each co-registered to a corresponding extravascular imaging data set. A plurality of hemodynamic data sets are provided, each hemodynamic data set co-registered with the corresponding extravascular imaging data set. The neural network learns what hemodynamic data to expect for a given intravascular imaging data set. An intravascular imaging event is subsequently performed in which an intravascular imaging element is translated within a blood vessel of the patient to produce one or more intravascular images. The neural network uses its training to predict hemodynamic values corresponding to the one or more intravascular images from the intravascular imaging event, and the one or more intravascular images are outputted in combination with the predicted hemodynamic values.

A memory device is configured to perform neural network learning, the memory device including a first dedicated memory corresponding to a first link included in a neural network and configured to store a first forward propagation weight and at least one first candidate weight for the first link, and a first processing element (PE) configured to perform a multiplication operation between an input and the first forward propagation weight stored in the first dedicated memory, for the first link, in which the first forward propagation weight stored in the first dedicated memory is configured to be updated with one of the at least one first candidate weight after the multiplication operation for the first link corresponding to the first dedicated memory.

The invention relates to a regional ocean sound propagation field millisecond-level prediction method based on a deep neural network, and the method comprises the steps: constructing two factors which have the greatest influence on a sound propagation field for a target region: a sound velocity profile and a terrain; an environment information data set of an area sound propagation field is constructed by constructing an annual sound velocity profile of a target area node and building a large number of two-dimensional terrain models by extending around the node by 360 degrees, and the sound propagation fields of the area under different environment conditions are calculated by using a traditional sound field calculation model to serve as a training set; feature simplification parameters of the sound velocity profile and the terrain in each environment sample are extracted to serve as guide information of a corresponding sound propagation field, and a deep neural network is built to learn distribution rules of the sound propagation fields under different environment conditions in the area; and finally, the trained model can quickly predict the corresponding sound propagation field in the region according to different environmental condition characteristic batches.

The invention relates to the technical field of environment informatization, in particular to a new pollutant environment fate simulation and risk assessment system, which comprises a data prediction module used for receiving molecular structures of target new pollutants, space-time emission source item data and basin environment medium parameters, and complementing missing physicochemical property parameters by using a model; the effect analysis module is used for acquiring new pollutant toxicity data and deducing and predicting non-effect concentration and sensitive point risk weight in combination with a mixed toxicity model; the sample generation module is used for simulating various discrete emission scenes based on a mechanism driving model and generating a space-time environment concentration response data set; and the agent model module is used for training a machine learning agent model according to the emission scene and the space-time environment concentration response data set. According to the method, the input and output relation of a deep neural network learning mechanism driving model is utilized through the agent model module, and the problem that a traditional mechanism model cannot support rapid iterative optimization of mass emission scenes due to slow operation is solved.

The application belongs to the technical field of software engineering and cloud computing, and particularly relates to a distributed system call chain anomaly detection method based on a graph neural network and PU learning. The application is based on call chain data generated during distributed system runtime, constructs a call chain causal graph according to attributes and parent-child relationships on the call chain, and learns normal and abnormal patterns of the call chain causal graph generated by a microservice system by using a graph neural network. When used online, a newly generated call chain causal relationship graph is constructed in real time, and abnormal call chains are identified. Specifically, the application includes call chain span vectorization, call chain causal graph construction, graph neural network training, and online anomaly detection. The application can help operation and maintenance personnel and developers quickly and accurately discover system anomalies at a very low data labeling cost, generate corresponding alarm information, speed up distributed system fault positioning and online problem solving, reduce labor costs, and improve system reliability.

Techniques for performing secure computing of softmax functions at high speed and with high accuracy are provided. A secure softmax function calculation system that calculates a share ([[softmax (u1)]], . . . , [[softmax (uJ)]]) from a share ([[u1]], . . . , [[uJ]]) includes a subtraction means for calculating a share ([[u1−u1]], [[u2−u1]], . . . , [[uJ−uJ]]), a first secure batch mapping calculation means for calculating, [[exp (u1−u1)]], [[exp (u2−u1)]], . . . , [[exp (uJ−uJ)]], an addition means for calculating a share([[∑ j=1J⁢exp⁡(uj-u1)]],… ,[[∑ j=1J⁢exp⁡(uj-uJ)]],and a second secure batch mapping calculation means for calculating a share ([[softmax (u1)], . . . , [[softmax (uJ)]]).

This application relates to the field of high-speed imaging technology, specifically to a high-speed imaging method, apparatus, and electronic device, which can solve the problem that the computation time of the existing DeSCI algorithm is too long and cannot meet the requirements of real-time imaging. The high-speed imaging method includes: determining initial information based on known information, wherein the known information includes a pre-set mask matrix and acquired compressed signals, the initial information being obtained based on the relationship between the mask matrix and the acquired compressed signals; inputting the initial information into an end-to-end neural network including a feature extraction module and a feature fusion module to obtain an inverse mapping for signal reconstruction, wherein the inverse mapping is learned by the end-to-end neural network under supervised training with a large amount of known information; and outputting reconstructed information based on the inverse mapping, the reconstructed information being obtained by the inverse mapping according to the initial information.

The invention relates to a time-space dynamic evolution-oriented man-machine cooperation assembly cognitive inference method, which comprises the following steps of: extracting visual features in an assembly scene, generating a scene graph, constructing a time hyperedge, a space hyperedge and a task hyperedge, and fusing the three types of hyperedges to form a hyperedge set; time-varying non-pairwise relationships among human operators, robots, assembly operations and various assembly component nodes are represented, a hyperedge incidence matrix is defined, a man-machine cooperation assembly knowledge space-time hypergraph is constructed, and hyperedge representation among assembly components is realized; designing a stacked graph neural network with a self-excitation characteristic based on a multi-event hokes process, learning high-order task association among assembly nodes, and updating the change of the high-order task association along with time to realize space-time hypergraph representation; and modeling a self-excitation process among different sub-tasks, and capturing individual features and collective association to realize man-machine cooperation assembly. The problem that man-machine cooperation cognition in a time-varying task is difficult to infer is solved, and man-machine cooperation assembly efficiency and initiative are improved.