150 results about "Sparse constraint" patented technology

The invention discloses a road cavity volume calculation method and device based on a three-dimensional FK migration algorithm, equipment and a medium, and relates to the technical field of road cavity detection, and the method comprises the steps: obtaining three-dimensional ground penetrating radar data, and carrying out the preprocessing of the data, and obtaining target radar data; converting the target radar data to a frequency wavenumber domain to obtain frequency wavenumber domain radar data; establishing a road hierarchical velocity model, and performing hierarchical wave field extrapolation and double weight correction based on the frequency wave number domain radar data and the road hierarchical velocity model to obtain corrected wave field data; performing sparse constraint optimization imaging processing on the corrected wave field data to obtain a target imaging result; and carrying out boundary extraction processing on the target imaging result to obtain a three-dimensional cavity boundary point set, and carrying out voxelization processing based on the three-dimensional cavity boundary point set to obtain a road cavity volume. According to the invention, high-precision restoration of the shape of the road cavity and automatic extraction of the volume can be realized in three-dimensional GPR imaging.

The invention provides an anomaly detection method and device for image data and a storage medium, and relates to the technical field of image processing, and the method comprises the steps: carrying out the feature extraction of the image data, fusing the spectrum and spatial features to obtain a joint feature matrix, and inputting an anomaly detection model; the model adopts an alternating direction multiplier algorithm to solve a low-rank sparse decomposition problem, and an objective function comprises a data fidelity item, a regularization item and a waveband weight item; the regularization item comprises a low-rank constraint and a sparse constraint, and the wave band weight item acts on the low-rank constraint in a weighting form; updating a background low-rank tensor, an abnormal sparse tensor, a Lagrange multiplier, a sparse constraint weight, a wave band weight item and penalty parameters of an algorithm by adopting an iteration mode in a solving process; and repeating iteration until a preset termination condition is reached, calculating an abnormal score graph pixel by pixel based on the abnormal sparse tensor, and comparing to determine an abnormal target. The problem that an abnormal target is difficult to accurately recognize in a complex scene can be solved, and detection precision and efficiency are improved.

The invention discloses a power adaptive digital pre-distortion method, system and device based on sparse GRU and a medium. The method comprises the steps that input and output signals under different powers are collected, a composite feature vector is constructed through amplitude normalization and phase alignment, and a pre-distortion training target is generated in combination with indirect learning; constructing a sparse GRU neural network, introducing L1 regularization constraint, and obtaining a sparse pre-distortion model through joint optimization of mean square error and sparse constraint; partial weight updating is realized based on power grading, and weight fusion output is carried out during power interval switching so as to complete power adaptive optimization; pruning and compressing the sparse GRU model and then deploying the sparse GRU model in a pre-distortion module to compensate the nonlinearity of the power amplifier; according to the method, the model complexity is reduced through rarefaction, the dynamic working condition stability is improved through a power adaptive mechanism, pruning compression adapts to low-power-consumption hardware, high precision, low complexity and adaptive capacity are considered, and the method is suitable for the low-power-consumption and high-performance requirements of a broadband wireless communication system.

The invention discloses an automatic business travel itinerary optimization method, and relates to the technical field of intelligent itinerary planning, and the method comprises the steps: integrating the multi-source heterogeneous data of enterprise policies, personal preferences and real-time traffic through a federated learning framework, and achieving the cross-domain knowledge sharing; the method comprises the following steps: constructing a staged optimization engine by adopting an attention mechanism to dynamically balance cost, time, comfort and sustainability targets: in the first stage, modularly disassembling a travel through sparse constraint linear programming, and quickly generating a Pareto frontier candidate set; in the secondary stage, on the basis of a multi-agent reinforcement learning framework, complex interaction is simulated through a Markov decision process, and strategy iteration is driven through a special reward function for quantifying a comfort index; in order to cope with real-time disturbance, event-driven edge computing nodes are deployed, flight delay and traffic jam emergencies are responded in real time, an incremental topology updating algorithm is triggered, and only affected sub-modules are reconstructed to reduce computing complexity. According to the invention, the bottleneck of dynamic adjustment efficiency and multi-target balance capability is solved.

The invention relates to the technical field of satellite navigation signal processing, and discloses a signal noise suppression method in a Beidou navigation system, which comprises the following steps of: performing three-dimensional modeling on navigation signals, and extracting time-frequency characteristics to generate a signal model; constructing an interference heat prediction map, and dynamically updating and identifying a high-interference area; analyzing a signal phase change, calculating a phase rate, and identifying abrupt change interference; adopting adaptive filtering and L1 norm sparsification to remove multi-source interference; correcting deviation in signal modeling by using a redundant navigation module; and adjusting a signal processing strategy based on a real-time control and scheduling mechanism. According to the method, a signal modeling guide mechanism based on an interference heat prediction map is introduced, and the dynamic change of electromagnetic environment interference is combined with a signal modeling process, so that an interference sensitive area is recognized in advance, and structural sparse constraint is applied; and a blind area lacking sensing of interference at the initial modeling stage in the traditional technology is eliminated.

The invention relates to the technical field of big data processing, and discloses a data extraction method and system for geological mineral exploration, and the method comprises the steps: carrying out the standardization preprocessing of original data containing space coordinates, lithology texts, mineral components, logging curves and mineralization labels; spatial, semantic, concentration and time sequence deep representations are extracted in parallel through a multi-modal geologic feature encoder and fused into high-dimensional vectors; applying structured sparse constraint to the features by using a graph attention mechanism guided by an expert knowledge graph, and strengthening a mineralization association dimension; a dynamic incremental learning engine is combined with an elastic weight solidification mechanism to realize local fine tuning of model parameters and historical knowledge retention; and finally outputting a mineralization potential score, a mineralization factor sequence and an abnormal element combination. According to the method, through triple mechanisms of multi-modal fusion, knowledge embedding and incremental evolution, the accuracy, interpretability and timeliness of data extraction are improved, and the real-time analysis requirement of large-scale mineral exploration is met.

The invention discloses a million-frame-level industrial vision system and method based on event driving and compressed sensing, and the system is characterized in that an event camera imaging module in the system captures the brightness change of each pixel in a field of view of the event camera imaging module in an asynchronous manner, and generates an event containing a pixel coordinate, a timestamp and change polarity for each change; a compressed sensing coding module constructs sparse image vectors for events in a time window, and the sparse image vectors are projected to low-dimensional observation vectors through an observation matrix phi; the sparse image reconstruction module is used for optimizing an objective function through sparse constraint and total variation regularization; a dynamic ROI compression module controls a compression mask function according to the event density and a gradient threshold. The method can break through the limitation of the traditional frame rate, has the advantages of high precision, high efficiency, low power consumption, strong robustness and the like, and has a wide industrial application prospect.

The invention discloses a dual-channel label-guided multi-label feature selection method and system, and belongs to a feature engineering technology. The method mainly comprises the steps of obtaining a feature matrix and a positive label matrix of a multi-label data set, performing logic negation on the positive label matrix to generate a mirror image negative label matrix, and constructing a graph Laplacian matrix based on the feature matrix; constructing a multi-label model based on the preprocessed data, wherein an objective function of the multi-label model at least comprises a positive label regression loss item, a negative label regression loss item, a label alignment constraint item, a graph regularization item and a sparse constraint item; constructing an optimization function through relaxation processing constraint and in combination with a Lagrangian multiplier method, iteratively solving the objective function according to a KKT condition, and evaluating feature importance based on a projection matrix for associating features and positive tags after iterative convergence; according to the method, the requirement of multi-label learning for accurate and efficient feature screening is met, label information can be comprehensively utilized, the anti-interference capability is enhanced, and the efficiency is considered.

The invention discloses a multi-view clustering method based on tensor feature extraction, and the method comprises the steps: inputting a multi-view data matrix, constructing a similarity matrix of each view through a K-NN algorithm and a Gaussian kernel function, and carrying out the spectral clustering to obtain a sample embedding matrix; performing singular value decomposition on an original data matrix of each view, taking first c left singular vectors to construct a feature embedding matrix, applying 2, 1 norm group sparse constraint on the feature embedding matrix, and connecting a sample embedding matrix through a bigraph to extract features; and normalizing the sample embedded matrix, and reconstructing a block diagonal matrix into a third-order tensor. And integrating the sample embedding matrix, the feature embedding matrix and global tensor learning to construct a target function, and optimizing through an alternating direction multiplier method until convergence. And finally, the normalized samples are embedded into the matrix to form block diagonals to form a consistent similarity graph, and a clustering result is obtained by using an N-Cut or k-means algorithm.

The invention discloses a three-dimensional Gaussian splashing method and system for sparse view curve reconstruction, and the method reconstructs a three-dimensional curve scene through a sparse image: firstly, obtaining a dense point cloud and a camera pose, extracting a two-dimensional line segment graph, employing a line segment guide Gaussian initialization strategy, and obtaining a three-dimensional curve scene; generating an initial Gaussian set on the basis of the dense point cloud and the two-dimensional line segment graph; structure perception Gaussian pruning is executed in the optimization process of the initial Gaussian set, and spatial outliers and visibility redundant gauss are removed; and finally, calculating the total loss containing sparse regular terms through the difference between the rendered image and the sparse image, and iteratively optimizing Gaussian parameters until the total loss converges, thereby obtaining a final three-dimensional curve reconstruction result. According to the method, the problem of geometric prior deficiency under a sparse view is solved through line segment guide initialization, overfitting and artifacts are effectively inhibited by using structure pruning and sparse constraint, and high-quality three-dimensional curve reconstruction is realized.

The invention relates to the technical field of machine learning feature engineering, in particular to an image data robust discrimination feature selection method and system based on an improved trace difference criterion, and the method comprises the steps: randomly selecting sample data with a label type from an image source sample data set as training sample data, and enabling the remaining sample data to serve as to-be-processed data; constructing a target function by adopting a trace difference criterion and introducing L2, p norm sparse constraints, training a feature selection model based on the target function by utilizing training sample data, and obtaining model optimal hyper-parameters and a row sparse projection matrix through cross validation and grid search; the trace difference criterion is constructed by an intra-class divergence matrix, an inter-class divergence matrix and a projection matrix of sample data; and performing feature screening on the to-be-processed data based on the optimal hyper-parameter and the row coefficient projection matrix, and outputting an optimal feature subset. The method can solve the problem that the feature selection effect on a small sample or a poor-quality data set is not ideal, and has a good application prospect in the field of image classification and the like.

The invention discloses a screening method of a protein marker combination, which is used for risk prediction of abdominal aortic aneurysm or ruptured abdominal aortic aneurysm, and optimizes the stability of a protein expression profile by constructing an initial protein expression profile; screening differential expression proteins by adopting statistical analysis, and preliminarily locking candidate proteins remarkably associated with a disease endpoint based on survival analysis or a risk regression model; performing cross validation by combining a sparse constraint algorithm and a nonlinear feature selection algorithm, and extracting a robust core marker; the independent contribution degree of each marker is evaluated through the standardized weight, and finally the optimal protein combination for disease prediction or diagnosis is determined. According to the method, six core proteins obtained through multi-algorithm cross screening can be detected in serum, multivariable logistic regression and external verification set performance both show good robustness, and an aorta protein fingerprint feature set which can be popularized and explained is formed.

The invention discloses a high-resolution video generation method and system based on multi-modal semantic guidance, and belongs to the technical field of video generation. According to the method, firstly, a low-resolution video is generated according to a text prompt, an edge map, a depth map and optical flow features of the low-resolution video are extracted through a space-time perception module, the features are combined with the text prompt, and a saliency region mask is generated after saliency weighting. And then the edge map, the depth map, the optical flow features and the text prompt are input into a DiT model. In each layer of Transform of DiT, after input is converted into corresponding embedded representation, structural feature embedding and text embedding are aligned in a space-time dimension, global semantic association is established through a cross attention mechanism, text embedding is processed by MLP to form conditional constraints, and sparse constraints and mask guidance are applied to an attention matrix. And finally, after multi-layer Transform processing, the model outputs a high-resolution video. In the whole process, the generation from the text to the high-resolution video is realized, and the details of the saliency region are emphatically protected.

The invention provides a photoelectric volume pulse wave identity recognition method and system, and relates to the technical field of identity recognition, and the method comprises the steps: obtaining a to-be-recognized PPG signal; and inputting the PPG signals into a trained identification model, firstly extracting linear features and nonlinear features, then projecting the features fused by the two features into a feature space by using a learned discriminant projection matrix to obtain multi-view features, and finally classifying the PPG signals by using the multi-view features to obtain an identity identification result. According to the method, manifold regularization and inter-class-error double sparse constraints are combined, the problems of intra-class discretization and inter-class overlapping of a linear model are solved, a graph structure learning method for adaptive local density adjustment is provided, and the manifold modeling precision of non-stationary PPG signals is improved.

The invention relates to the technical field of intelligent power distribution operation and maintenance, in particular to an Internet of Things power distribution box monitoring method, which comprises the following steps of implementing multi-domain active excitation in a safety interlocking time window, acquiring multi-domain response, and generating an excitation-response observation packet; carrying out feature distribution alignment on the observation package, obtaining a degradation state field and an uncertainty field on a node-edge graph model through joint inversion containing group sparse constraint and physical prior, and generating a sensitivity field through a risk density function; establishing model predictive control under action mutual exclusion and resource constraint, and jointly arranging a source channel, an environment channel and a local channel to output an optimal action trajectory; and single-channel amplitude difference comparison is implemented in adjacent micro-regions, and an acquisition time slot selection scheme and a channel priority of a next period are generated in combination with an evidence ratio, an action Shapley attribution and a second-order optimal transmission distance write-back weight, so that in-service interpretable monitoring and self-calibration control are realized.

The application is suitable for the technical field of radar signal processing, and provides a deep unfolding ISAR super-resolution imaging method based on sparse-neighborhood joint constraint, comprising: firstly, constructing an ISAR image degradation model, constructing an ISAR super-resolution imaging problem based on the ISAR degradation model and the compressed sensing theory, using ADMM to solve the ISAR super-resolution imaging problem, unfolding the process of solving the ISAR super-resolution imaging problem by ADMM into a multi-level neural network, and finally inputting low-resolution ISAR echoes into the trained neural network to obtain an ISAR super-resolution imaging result. The application combines sparse constraint and neighborhood amplitude constraint, deduces a corresponding signal model and an ADMM solving algorithm, effectively improves the reconstruction ability of the algorithm to complex target details and the representation ability of the algorithm to structural information, and can realize effective super-resolution imaging based on narrowband short-aperture echoes.

The application discloses a kind of double-channel label guide multi-label feature selection method and system, belong to feature engineering technique.Method mainly includes: obtaining the feature matrix and positive label matrix of multi-label data set, by performing logical negation to positive label matrix, generate mirror negative label matrix, and construct graph Laplacian matrix based on feature matrix;Based on the data after pre-processing, a multi-label model is constructed, and the objective function of the multi-label model includes at least positive label regression loss term, negative label regression loss term, label alignment constraint term, graph regularization term and sparse constraint term;The constraint is processed by relaxation, and the optimization function is constructed by combining the Lagrange multiplier method, and then the objective function is iteratively solved according to the KKT condition, and after iterative convergence, the feature importance is evaluated based on the projection matrix used to associate features and positive labels;The application meets the demand of multi-label learning for accurate and efficient feature selection, can fully utilize label information, enhance anti-interference ability and consider efficiency.

The present disclosure provides a neural network training method and a computing device, which relates to the technical field of artificial intelligence. The scheme includes: splitting a weight matrix into В weight blocks each having the same size and satisfying N:M sparsity constraint; and then by using В sparse mask sets defined for the В weight blocks respectively and В random number indices generated for the В weight blocks respectively, generating a mask matrix for implementing sparse matrix multiplication operation. There is no need to pre-train a dense weight matrix of huge-scale in advance nor to fine-tune a trained model again for the N:M sparsity constraint, which saves a lot of training time and improves training efficiency.

The application discloses a hyperspectral sparse unmixing method of a learnable implicit variable iterative unfolding network, and comprises the following steps: constructing an unmixing model by sparse constraint optimization; constructing an alternating direction iteration by variable splitting and an augmented Lagrangian method; modeling an alternating direction iteration step as an implicit unfolding network module, including a learnable layer of abundance variable and multiplier variable; constructing a learnable smooth convolution layer to improve the smoothness of abundance patches; realizing spectral unmixing and reconstruction by a coding-decoding structure; and realizing model training by adopting an unsupervised loss function. The application uses the optimization mechanism of a classical sparse unmixing model to design a learnable network, the network layer is designed based on algorithm iteration steps, the sparsity and patch smoothness of hyperspectral abundance are fully met, and the explainability and transparency are enhanced; an unsupervised training mechanism is introduced, the network availability is enhanced; the model parameter size and overfitting phenomenon are reduced through model driving and network parameter sharing, and the model is lightened.

This invention discloses a sparse array optimization method and related equipment based on amplitude mapping. The method includes: acquiring array information and preset sparse constraints of an antenna array; initializing an amplitude weight matrix based on the array information and sparse constraints; mapping the amplitude weight matrix to a sparse binary layout; quantizing the sidelobe level based on the sparse binary layout; if the sidelobe level meets preset requirements, the corresponding sparse binary layout is used as a sparse array; otherwise, the amplitude weight matrix is ​​updated through an optimization algorithm, and the step of mapping the amplitude weight matrix to a sparse binary layout is returned to be executed until the sidelobe level meets the preset requirements. This invention transforms the discrete, massive cell position combination search problem into a continuous, relatively low-dimensional amplitude weight optimization problem, which can greatly compress the solution space and thus efficiently complete the optimization of large-scale arrays. It can be widely applied in the field of data processing technology.

The application discloses a gearbox fault diagnosis method based on adaptive decomposition and transfer learning, comprising: collecting multi-condition vibration signals as original input; adopting an exponential coupling resonance search-Fourier adaptive modal decomposition method to adaptively decompose the signals, accurately decoupling multi-component signals and suppressing modal aliasing; extracting multi-domain features and utilizing weighted multi-objective probability principal component analysis dimension reduction to obtain a sensitive low-dimensional feature set; constructing a sparse excitation residual network with a sparse constraint channel attention mechanism as a feature extractor and pre-training; adopting a confidence-aware incremental open set transfer learning strategy to fine-tune the high layer of the network, realizing known fault diagnosis and open set identification of unknown fault samples. The method effectively improves the fault feature decoupling capability and the generalization performance of the diagnosis model, and is suitable for intelligent fault diagnosis and state monitoring of the gearbox under multiple conditions.

The invention relates to the field of computer vision, artificial intelligence and medical image analysis, in particular to a disease prediction method for supervising multi-omics tensor fusion, and the method comprises the steps: obtaining a tensor covariance among multi-omics data according to the processed multi-omics data, and defining a target function of tensor canonical correlation analysis; according to the tensor covariance and the target function, introducing structured sparse constraint and disease supervision information, and constructing a multi-modal image correlation analysis model based on tensor; solving the multi-modal image correlation analysis model by using an alternating iteration method to obtain a typical weight of each omics data; and predicting the disease according to the typical weight of each piece of omics data. According to the supervised multi-omics data fusion method based on the tensor, high-order related information can be effectively mined, the classification accuracy of chronic diseases is improved, and powerful technical support is provided for clinical diagnosis.

This invention relates to the field of seismic data processing technology, specifically disclosing a method for removing strong reflection shielding in hidden river channels based on adaptive hybrid L0-L1 norm. The method includes: first, establishing an optimization objective function based on Bayes' theorem using hybrid L0-L1 norm; second, dynamically adjusting the L0-L1 norm weights using an adaptive weight function driven by the seismic signal, combined with reflection coefficient amplitude and residual information, enhancing sparsity constraints in strong reflection zones and reducing constraint strength in weak reflection zones; then, constructing a convex upper bound for the objective function using a minimization framework, and solving it iteratively in stages using an accelerated rapid iterative threshold shrinkage algorithm, while incorporating prior knowledge of seismic wave propagation laws and river channel deposition patterns to ensure the geological rationality of the solution. This invention solves the technical problems of traditional sparse processing methods, such as fixed parameters, lack of geological constraints, and inability to simultaneously address strong reflection suppression and weak signal protection, significantly improving the separation accuracy of strong reflections and the recovery rate of weak signals, effectively overcoming the strong reflection shielding effect.

The invention relates to an incomplete multimedia recommendation method based on a unified heterogeneous hypergraph, and belongs to the field of multimedia data recommendation. According to the method, a unified heterogeneous hypergraph construction framework HIRE for incomplete multimedia recommendation and a sparse constraint enhanced version HIRE S of the unified heterogeneous hypergraph construction framework HIRE are provided, and the recommendation problem of modal information missing in a multimedia sharing platform is solved. The HIRE initializes a hypergraph structure through a K-means algorithm, and captures a cross-modal high-order relationship in combination with a heterogeneous hypergraph convolution mechanism to complement missing multi-modal features. Meanwhile, a self-supervised contrast learning mechanism of text alignment is adopted, and a hypergraph structure is jointly optimized to improve multimedia recommendation performance. A sparse optimization strategy is introduced into HIRES, a hypergraph structure is refined through optimal transmission and # imgabs0 # norm constraint, noise interference is reduced, and recommendation accuracy is improved. The method is obviously superior to a baseline method, and an efficient and explainable solution is provided for multimedia recommendation in an incomplete scene.

The invention discloses a robust speech enhancement method based on adaptive beam forming and sparse spectrum constraint. The robust speech enhancement method comprises the following steps: receiving a multichannel observation signal through a microphone array and constructing a signal model of a generalized sidelobe canceller structure; based on the signal model, a beam forming optimization model is constructed in combination with the least square criterion and the speech spectrum sparse constraint; performing iterative solution on the beam forming optimization model by adopting a Lagrange multiplier alternating direction method to obtain an adaptive filter weight vector in a generalized sidelobe canceller; and calculating and outputting an enhanced voice signal based on the weight vector. According to the method, the voice quality and intelligibility can be effectively improved in a strong interference and reverberation environment, interference and reverberation are remarkably suppressed, and the robustness of an algorithm is improved.

The invention discloses a two-dimensional wave field reconstruction method based on compressed sampling, and belongs to the technical field of signal processing and sensors. According to the method, original time domain signals of a two-dimensional wave field are expressed as a three-dimensional matrix, random sampling is carried out through a Bernoulli distribution random sampling matrix, after the three-dimensional matrix is reduced into a two-dimensional matrix, a group sparse perception convolutional neural network is constructed to serve as a non-convex optimization solver, a real part and an imaginary part of a complex signal are processed, group sparse constraint is applied, and a non-convex optimization solver is constructed. And reconstructing the basis coefficient matrix to obtain an original signal reconstruction result, and finally raising the dimension to restore the original signal into a three-dimensional wave field. According to the method, the sampling rate is reduced, the reconstruction precision is improved, large-scale data set training is not needed, each training is an independent reconstruction process, and the method is suitable for the fields of acoustic imaging, seismic exploration and the like.

The application is suitable for the technical field of device management, and provides a behavior recognition method based on space-time relationship and an electronic device. The method comprises the following steps: receiving target video data to be recognized; introducing the target video data into a preset inter-frame action extraction network to obtain inter-frame action feature data; introducing the inter-frame action feature data into a feature extraction network to output sparse feature data corresponding to the target video data; the feature extraction network is generated by performing sparse constraint processing on each convolution kernel in a pooling fusion network through selected weights; introducing the target video data into a context attention network to determine gait behavior data of a target object in the target video data; and obtaining a behavior category of the target object according to the gait behavior data and the sparse feature data. The above method can greatly reduce the calculation cost of video data in the behavior recognition process, thereby improving the operation efficiency.

This invention relates to a method and system for EEG emotion recognition based on an adaptive multi-view graph neural network, belonging to the field of brain-computer interface and emotion computing technology. The method includes: dividing multi-channel EEG signals into continuous time windows, and using four adjacent time windows as temporal input samples; extracting multi-band differential entropy features of each time window as initial node features; fusing prior knowledge of electrode spatial proximity and brain biological symmetry to construct a basic matrix, and modulating and applying sparse constraints through a learnable attention mechanism to generate an individualized brain functional connectivity topology; designing a parallel bi-branch deep network, where a graph convolutional branch extracts global spatiotemporal features from the graph structure sequences corresponding to the four time windows, and a one-dimensional convolutional branch extracts and fuses local frequency-spatial features; and during training, comprehensively applying node-level domain adversarial and graph structure collaborative regularization to output the emotion category. This invention is beneficial for improving cross-subject recognition performance.

The application provides an industrial network security situation prediction method and system based on a generative large model, relates to the technical field of industrial network security, and comprises the following steps: acquiring time series network behavior data of a plurality of monitoring nodes in an industrial network, and extracting multi-dimensional features to obtain a security feature vector; the feature vector is decomposed into a periodic baseline component and a transient disturbance component through frequency domain transformation, and a decomposition situation feature is obtained through sparse constraint screening; semantic space mapping is performed by using a generative large model to obtain a semantic enhanced situation representation, the correlation between security anomalies among nodes is calculated based on the decomposition situation feature, and a dynamic correlation matrix is constructed; a propagation operator is constructed based on the dynamic correlation matrix, the semantic enhanced situation representation is subjected to multi-step iterative propagation, a decay factor is introduced to simulate abnormal influence diffusion, and a security situation prediction result in a future time window is obtained.