61 results about "Graph regularization" patented technology

The invention relates to the technical field of nondestructive testing, and discloses an aircraft defect identification method and system based on tensor decomposition and an attention mechanism, and the method comprises the following steps: obtaining multi-modal data of an aircraft; constructing the multi-modal data into a fourth-order space-time-modal tensor, and generating a dynamic graph structure based on the modal characteristics of the tensor; applying mixed constraint to obtain a core tensor and a factor matrix during four-order tensor decomposition; features are extracted through multi-scale pooling, weights are distributed in combination with topological persistent coherence and a gating attention mechanism, and feature fusion is achieved; and identifying defect types based on fusion features, and positioning defect regions by using factor matrix gradient amplitudes and dynamic thresholds. According to the method, through multi-modal data fusion, dynamic graph regularization constraint and mixed tensor decomposition technologies, the detection sensitivity and the positioning precision of the small defects on the surface of the aircraft are improved, and meanwhile, the physical interpretability of the characteristics and the robustness of the algorithm to complex working conditions are enhanced.

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 relates to the technical field of price monitoring, in particular to a building material multi-source price anomaly detection method, which comprises the following steps: firstly, uniformly metering and pricing calibers, learning a conversion coefficient, and constructing a replaceable relation graph; multi-source distribution is aligned through optimal transmission, residual errors and shadow prices are obtained based on structure invariants and variational inequality, and abnormal evidences are formed through hypergraph propagation and persistent coherence; generating a valence band reference in combination with a convex hull method and distribution robust optimization under the equilibrium clearing of graph regularization; a feasible set is defined by price bands and constraints, a weighted maximum satisfactory model is constructed to position a minimum default set, a minimum correction suggestion is generated by using vector optimal transmission and packet sparsity, and an executable closed loop is realized through satisfactory model theory verification.

The invention discloses an overlapping community detection method, system and device and a medium, and particularly relates to the technical field of community detection, and the technical key points are as follows: extracting node information data from a pre-constructed adjacent matrix of an overlapping community, and inputting the node information data into a point mutual information function to calculate and obtain point mutual information; obtaining a superpoint mutual information matrix by combining the point mutual information with the hypergraph structure; constructing a target function by using the super-point mutual information matrix and a pre-constructed three-factor illegal matrix decomposition optimization model based on graph regularization; decomposing the super-point mutual information matrix in the objective function to obtain an indication matrix, and solving the indication matrix by using an alternating iteration method to obtain an optimal solution of the indication matrix; and detecting the overlapping community based on the optimal solution of the indication matrix to obtain an overlapping community detection result.

The invention relates to a terahertz spectrum detection technology, in particular to a pericarpium citri reticulatae producing area identification method based on graph regularization sparse principal component analysis and a support vector machine algorithm, and belongs to the field of pericarpium citri reticulatae producing area quality detection. The method comprises the following steps: grinding and tabletting a dried orange peel sample to be detected, firstly detecting the dried orange peel sample in a nitrogen environment by adopting a terahertz time-domain spectroscopy system in a transmission mode to obtain a terahertz time-domain spectroscopy signal of the sample, and performing Fourier transform on the time-domain spectroscopy signal to obtain a frequency-domain spectrum of the sample; and obtaining a corresponding terahertz absorption spectrum according to the frequency domain spectrum. Savitzky-Golay smoothing preprocessing is carried out on the obtained terahertz absorption spectrum, the terahertz absorption spectrum is divided into a training set and a test set, and then feature extraction is carried out on data by utilizing sparse principal component analysis in combination with graph regularization. And taking the processed data as the input of a classification model. And finally, a combined parameter of an optimal regularization coefficient c and a kernel function parameter g of the support vector machine is obtained through a particle swarm optimization algorithm, so that an optimal pericarpium citri reticulatae producing area classification model is established. The method provided by the invention is convenient in sample preparation and simple in operation, can effectively realize rapid and accurate identification of dried orange peel from different producing areas, and provides a new method for identification of dried orange peel in high-value producing areas.

The invention discloses a node zero-trust trusted access method and device for a computing power network, and relates to the technical field of network security. And based on a zero-trust principle, credibility evaluation is carried out in a registration stage and a task execution stage of the computing power node. In the registration stage, a multi-source trust modeling mechanism fusing identity attributes, capability observation and organization reputation is constructed, and a Bayesian reasoning and graph regularization method is introduced to realize initial trust evaluation; in the operation stage, a multi-modal log anomaly detection method is designed, high-dimensional behavior modeling is carried out in combination with semantics, time, parameters and quantity characteristics, and the anomaly recognition capability is enhanced by utilizing long sequence dependence; and a dynamic trust adjustment strategy is provided by further combining time decay and a user feedback mechanism, so that continuous evolution and refined access control of node trust are realized. According to the method and the system, trusted access of the computing power nodes is realized by the scheduling system in the environment of the computing power network, and the safety and the reliability of the computing power network system are effectively improved.

The invention provides a ground penetrating radar clutter suppression method and system and a medium, the method highlights a texture change area through first-order difference, can accurately capture change information of echo intensity and suppress background fluctuation and noise, meanwhile, introduces graph regularization to ensure similarity of adjacent pixels, suppresses isolated noise points, and improves noise suppression efficiency. And the stability of the texture structure is enhanced. By means of the combination, clutters can be effectively removed when the method is used for processing complex backgrounds, and meanwhile key structure information is reserved. Complex echo texture data are projected to a singular value feature space through singular value decomposition, singular values with a large change rate are screened out for reconstruction, redundant interference can be effectively removed, and stable echo textures are reserved. According to the mechanism, when the method is used for processing multi-target, dispersed and irregular echo signals, target signals and clutters can be effectively separated, and the detection precision is influenced.

Embodiments described herein provide an approach (referred to as “Co-training” mechanism throughout this disclosure) that jointly learns two representations of the training data, their class probabilities and low-dimensional embeddings. Specifically, two representations of each image sample are generated: a class probability produced by the classification head and a low-dimensional embedding produced by the projection head. The classification head is trained using memory-smoothed pseudo-labels, where pseudo-labels are smoothed by aggregating information from nearby samples in the embedding space. The projection head is trained using contrastive learning on a pseudo-label graph, where samples with similar pseudo-labels are encouraged to have similar embeddings.

The invention provides a multi-view subspace clustering method (TNMVSC) based on a non-convex tensor nuclear norm. According to the method, the low-rank representation theory (LRR), the graph regularization technology and the tensor decomposition theory are combined, and the global structure and the local attribute of each view can be captured at the same time. Besides, the TNMVSC decomposes each low-rank representation sparse matrix into three factor matrixes through a matrix three-factor decomposition theory so as to realize alignment of the representation matrixes, thereby ensuring that the generated core matrix can effectively retain key information. Meanwhile, a non-convex low-rank tensor nuclear norm is adopted to capture high-order correlation among a plurality of core matrixes. In the aspect of algorithm optimization, on the basis of the established optimization model, an alternating direction multiplier method (ADMM) is adopted to carry out optimization solution on the representation matrix of each view. And then, performing angle correction on the fused representation matrix to obtain a similarity matrix among the samples, and clustering the similarity matrix by using a spectral clustering method to realize clustering of the samples. In order to verify the effectiveness of the TNMVSC, tests are performed on reference data sets in multiple fields, including computer vision, bioinformatics, social media analysis and the like. A large number of simulation experiment results show that the TNMVSC is excellent in clustering precision and robustness, and the performance of multi-view clustering can be effectively improved.

The invention discloses a spatial domain identification method and system based on spatial multi-omics data. The method comprises the following steps: obtaining an expression profile and a spatial position coordinate through spatial multi-omics sequencing, and preprocessing the expression profile and the spatial position coordinate; decomposing and extracting a shared low-dimensional incidence matrix of the multi-omics data by using an anchor concept; constructing an intercellular similarity matrix based on the spatial coordinates, and generating a spatial location map by combining the k-nearest neighbor map and the high-order adjacency matrix; integrating multi-base clustering results by adopting an element-by-element weighting strategy to form an integrated graph; solving an optimal low-dimensional feature through a double-graph regularization objective function (fusing Laplacian matrix constraints of a spatial position graph and an integrated graph); and finally realizing high-precision spatial domain clustering. The method breaks through the limitation that a traditional method ignores spatial dependency and integrated interference, effectively fuses the high-order relation and spatial information of multi-omics data through anchor concept decomposition and a double-graph regularization framework, and remarkably improves cross-scene robustness; the method is suitable for biomedical scenes such as tumor microenvironment analysis and cell heterogeneity analysis.

The invention discloses a feature selection method for multi-label prediction of diabetic complications, and relates to the technical field of machine learning and medical data processing. The method comprises the following steps: carrying out label layered independent weight calculation on original characteristics of diabetic complications and complication label data to obtain a label weight matrix of a plurality of complications; performing graph regularization optimization based on layered independent weights on the label weight matrix of the multiple complications, training a multi-label classification model, reducing irrelevant and redundant features in the multi-label classification model, and obtaining feature subsets with importance perception, high discrimination and robustness; and screening the original data of the diabetic complications by using the optimized multi-label classification model, calculating a feature contribution matrix, and for positive prediction labels, outputting the first three features of contribution values for constructing the features of diabetic complication prediction. According to the method, key features can be screened out, and the accuracy of model prediction is improved.

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.

This disclosure relates to a method, apparatus, and product for full-waveform reflection inversion based on dynamic graphic warping, applicable to the field of seismic exploration technology. In this disclosure, seismic data of the target area is acquired; wavefield simulation is performed based on the seismic data to determine the background wavefield and scattered wavefield. Based on dynamic graphic warping, full-waveform reflection inversion is performed on the background wavefield and scattered wavefield to determine the inversion target model. The inversion target model is optimized to determine the target image of the target area. Determining the inversion target model through the dynamic graphic warping algorithm can improve the resolution of the inversion target model and reduce the model's ambiguity and computational load. Furthermore, iterative optimization of the inversion target model can improve imaging quality. Simultaneously, using the dynamic graphic warping algorithm can reduce the accuracy requirements of high wavenumber models, enabling more accurate imaging of complex geological features. Therefore, the accuracy of imaging is improved.

The present invention discloses a smoothed group information-guided independent component analysis method for brain functional network analysis, which belongs to the technical field of independent component analysis of brain images. The present invention includes performing independent component analysis to obtain components at the group level as reference signals; calculating voxel features to construct a graph regularization term; using voxel features and reference signals as guidance, using multi-objective functions to perform iterative solutions to estimate the independent components of individual subjects; and calculating the time series corresponding to each component in the individual subject based on the extracted components. The present invention overcomes the limitation of the group information-guided independent component analysis method currently widely used in the field of brain functional network extraction that does not optimize the smoothness of the extracted components, and obtains a more accurate brain functional network. Voxel features are introduced as a guide in the process of constructing the objective function, which enhances the spatial smoothness and functional correlation of the results, and can help the new method learn a network that is more in line with the actual working mechanism of the brain.

The application provides a double relaxation image classification method based on a width learning system, and steps are as follows: firstly, a feature data set and a corresponding class label matrix are acquired, and the feature data set generates width conversion features through a standard width learning network; secondly, a double relaxation technique and a graph regularization technique are introduced, and a double relaxation image classification optimization objective function based on the width conversion features is constructed; finally, the double relaxation image classification optimization objective function is solved by using iterative optimization, a classification result is obtained, and the classification result is evaluated. The manifold regularization technique is applied to the width learning network, and a double relaxation method is used to obtain greater freedom, so that the data geometric structure is mined, and the learning of the intra-class similarity realizes the relaxed regression of the target. The application has the characteristics of higher classification precision, relatively less training time, higher model flexibility and the like, and the introduction of the double relaxation method makes the model have stronger discrimination ability.

The invention provides a semi-supervised newborn movement type prediction method and system based on graph regularization loss, and the method comprises the steps: collecting an original video of the limb movement of a newborn, and carrying out the manual segmentation of the original video of the limb movement, and obtaining effective video segments of the limb movement; converting the limb action effective video segment into a limb joint point sequence; inputting the limb joint point sequence into a neural network to obtain a predicted action category of the newborn; obtaining a graph regularization loss based on the predicted action category and the real category, calculating a cross entropy loss and a ternary loss of the neural network, and carrying out weighted summation on the graph regularization loss, the cross entropy loss and the ternary loss to obtain a total loss function of the neural network; and training the neural network based on the total loss function, wherein the trained neural network is used for predicting the neonatal motion type. The image regularization is added into the loss function, the feature extraction effect is improved, and the classification accuracy is effectively improved.

The invention discloses a remote sensing image sub-region rapid classification method based on dual-scale multi-graph regularization non-negative matrix factorization, and provides a dual-scale basis selection strategy and a multi-graph regularization non-negative matrix factorization model for overcoming the defects of complex scene adaptation, small sample learning and classification speed and accuracy balance in the prior art. Unsupervised subregion classification is realized through the steps of data preprocessing, double-scale seed set division, basis matrix construction, double information graph regularization item construction, model optimization solution and the like. According to the method, the classification number does not need to be predefined, the basis matrix representativeness is enhanced through dual-scale basis selection, and the classification precision and robustness are improved by combining a dual-graph regularization item and utilizing a manifold structure and discrimination information of data at the same time; the efficient optimization solution of the non-negative matrix factorization realizes the rapid classification of the sub-regions, and improves the accuracy and efficiency of the classification of the sub-regions of the remote sensing image.

The application relates to the technical field of nondestructive testing, and discloses an aircraft defect identification method and system based on tensor decomposition and an attention mechanism. The aircraft defect identification method based on tensor decomposition and the attention mechanism comprises the following steps: acquiring multi-modal data of an aircraft; constructing the multi-modal data into a four-order space-time-modal tensor, and generating a dynamic graph structure based on the modal features of the tensor; applying a mixed constraint when decomposing the four-order tensor to obtain a core tensor and a factor matrix; extracting features through multi-scale pooling, combining topological persistent homology and a gated attention mechanism to distribute weights, and realizing feature fusion; identifying a defect type based on the fused features, and locating a defect area by using a factor matrix gradient amplitude and a dynamic threshold. Through multi-modal data fusion, dynamic graph regularization constraint and mixed tensor decomposition technology, the application improves the detection sensitivity and positioning accuracy of small defects on the surface of the aircraft, and enhances the physical interpretability of features and the robustness of the algorithm to complex working conditions.

The application relates to the technical field of intelligent report forms, in particular to a financial report form intelligent generation method and system fusing a variational Gaussian process and reinforcement learning. The method establishes an initial report form template by collecting original business data, completes business feature clustering and semantic mapping to identify a business type by using hierarchical fuzzy clustering, constructs a financial relationship graph based on sparse representation graph embedding and performs abnormality identification to generate a financial data set, fuses variational Gaussian process regression and reinforcement learning to complete budget prediction, finally fills a template with budget results and performs structure reconstruction by a graph regularization autoencoder to generate a final available financial report form, and realizes intelligent report form automatic generation under digital finance. The application significantly improves the automation degree, prediction accuracy and report form generation efficiency of financial data processing.

This invention proposes a structurally enhanced discriminative width learning system for image classification. First, it introduces graph regularization based on class relationships to ensure that similar samples remain closely distributed in the projection space. Second, it adds intra-class divergence regularization to further enhance the aggregation of samples within the same class. Furthermore, it introduces inter-class divergence regularization to ensure that the centers of different classes are far apart in the feature space. Finally, in the output layer regression stage, it replaces the Frobenius norm regularization in ridge regression with sparse norm regularization to effectively suppress noise and redundant information. This image classification scheme, while maintaining the simplicity and efficiency of the optimization process, organically integrates three different types of structured information—intra-class structure, inter-class structure, and local manifold information—into a unified model, enabling the learned feature representations to possess stronger structure preservation and discriminative capabilities.

The present invention belongs to the fields of deep learning and brain network structure, and particularly relates to a joint learning method for brain network structure and similarity based on graph attention network, including: performing cortical segmentation processing and morphological feature extraction on the obtained brain image data, and modeling the subject's brain network as a graph; estimating the initial brain network structure through Pearson correlation calculation; obtaining the similarity between brain network structures through a siamese graph attention learning network; calculating the graph regularization loss function and the siamese network loss function to constrain the characteristics of the initial brain network structure; updating the adjacency matrix of the brain network according to the embedding features of the brain network, and obtaining the updated brain network structure and calculating the similarity of the brain network structure. The present invention effectively estimates the morphological brain network by jointly optimizing the two tasks of brain network structure estimation and similarity learning, and provides valuable information for subsequent tasks such as individual recognition and disease auxiliary diagnosis.

The invention discloses a wireless network time synchronization method and system, and a storage medium, and the method comprises the steps: collecting network node time data, forming an original sampling sequence, and generating a whole network clock observation matrix; based on the connection strength between each pair of network nodes, constructing a Laplacian matrix; on the basis of an objective function of a graph regularization robust low-rank decomposition model, carrying out joint optimization solution on the whole-network clock observation matrix, and alternately updating a low-rank matrix and a sparse matrix; after convergence conditions are met, drift and offset correction values are calculated, and correction parameters are generated; and performing wireless network time synchronization according to the correction parameter. The problem that an existing synchronization algorithm is unstable in precision under the conditions of strong interference, packet loss and dynamic topology is solved, and unification of high precision and high robustness of clocks of the whole network is achieved.

A low-quality multi-view news data anchor graph regular division method based on diffusion completion belongs to the field of data division in low-quality multi-view news data, and comprises the following steps: firstly, inputting low-quality news data of each view and a corresponding similarity relation matrix into a heterogeneous relation convolutional network; to obtain a low-dimensional embedded representation of each view. Then, the low-dimensional embedded representation uses forward noise adding and reverse noise reduction processes of a conditional diffusion model to obtain predicted noise, and missing samples are complemented through the predicted noise; and then, an anchor point diagram is constructed by using the similarity between the complemented embedded sample and the anchor points, soft clustering distribution is obtained for the constructed anchor point diagram through an orthogonal normalization layer, and discriminative feature representation is obtained through anchor graph regularization constraint. And finally, soft clustering distribution is constrained by using a tensor Schatten p-norm so as to fully mine complementarity information and a sparse structure between the views.

The invention discloses a structure constraint low-rank sparse representation arrhythmia classification method based on tag embedded graph regularization, and belongs to the field of electrocardiosignal processing. The method comprises the following steps: constructing a training data set and a label matrix; a graph Laplacian matrix is constructed based on the label matrix, local structure information and label information of the electrocardiosignals are fully utilized to construct a low-rank sparse representation model fusing a local constraint item and a graph regularization item, and a discriminative dictionary is obtained through dictionary learning; and finally, performing low-rank sparse representation on a test sample by using the learned dictionary, and completing classification according to a minimum reconstruction error principle. According to the method, by introducing a tag embedded graph regularization item, the model keeps a local manifold structure of data and enhances intra-class compactness and inter-class separability in a sparse representation learning process, so that the classification accuracy of electrocardiosignals, particularly arrhythmia classes, is remarkably improved.

The invention relates to the technical field of remote sensing image processing, and relates to a hyperspectral image dimension reduction method, device and equipment and a storage medium, and the method comprises the steps: dividing hyperspectral image data into N three-dimensional tensor blocks; taking the complete spectral vector of each pixel of the hyperspectral image data as a clustering input feature in a super-pixel segmentation algorithm, and generating a super-pixel semantic pseudo tag; on the basis of a maximum frequency principle or a voting strategy, according to the pixel semantic pseudo-tag of each pixel in the tensor block, determining a super-pixel semantic pseudo-tag of each tensor block; constructing a graph by taking the tensor blocks as nodes and taking the similarity between the tensor blocks as edges; and constructing a graph regularization dimensionality reduction model according to a locality preserving projection rule, solving the model to obtain an optimal projection tensor, and performing dimensionality reduction on the hyperspectral image by using the projection tensor. According to the method, the spatial-spectral information of the hyperspectral image can be considered at the same time, the discrimination and stability of intra-class structure expression and dimension reduction results are improved, and the method has good practical value and popularization potential.

The invention discloses a feature representation method for realizing multi-view image alignment based on a dynamic graph. The feature representation method comprises the following steps: 1) initializing a multi-view structure; 2) common dynamic graph learning; 3) constructing a joint optimization model based on dynamic graph regularization; 4) solving a common source representation S and a projection matrix by minimizing projection features of each view; 5) initializing the dynamic similar matrix during the first iteration; 6) expanding a matrix dimension and constructing a coefficient matrix; 7) expanding and converting the expanded target function into a matrix trace form; 8) deducing a gradient expression of a dynamic graph learning objective function; and 9) setting an iteration stop criterion of the gradient descent method, and finally outputting an optimization result. Through combined utilization of the structure information of each feature view, adverse effects caused by insufficient samples or noise interference are reduced, and the image alignment precision in severe construction environments such as dark light is effectively improved.

This invention relates to the field of multi-sensor measurement and information fusion technology, and particularly to an intelligent spatial dynamic monitoring method based on multi-sensor collaboration, comprising the following steps: S1 Measurement state modeling and baseline generation; S2 Reference clock alignment and acquisition cycle tuning; S3 Anomaly and uncertainty-driven sampling and weighted linkage; S4 Constraint-aware multi-source fusion and uncertainty assessment; S5 Multi-evidence fault determination and graph regularization completion; S6 Measurement result indication and recording. This invention establishes a reference clock and maps each channel to a unified time axis, combines time quality and scene dynamic adaptive tuning sampling (including frequency boosting at key nodes), and outputs results with uncertainty through a two-stage fusion method of quality-aware weighted + gated network, supplemented by multi-evidence fault determination, graph regularization completion, and full-process recording, comprehensively achieving time-consistent, continuous, and traceable spatial dynamic monitoring.

The present invention discloses a prediction method for drug-disease association relationships based on graph-regularized matrix factorization. The method of the present invention extracts the semantic similarity between diseases according to the directed acyclic graph of each disease, and then combines the association relationships between each disease and drugs in the existing database. Using the graph convolution method, it extracts disease features to construct a disease feature matrix, determines the cosine similarity between diseases and fuses it with the semantic similarity. After obtaining the disease association relationship based on the directed acyclic graph, it constructs a drug feature matrix according to the drug features in the existing database, combines it with the disease feature matrix, establishes an association matrix between drug molecules and diseases, and performs eigen-decomposition on the association matrix based on the matrix factorization algorithm of graph regularization and kernel method. By constructing an objective function, it optimizes the neighbor relationships of nodes in the drug similarity network and the disease similarity network, fully utilizes disease features and drug features, and realizes the accurate prediction of the association relationship between drugs and diseases.

The present application discloses a method, device, electronic device and storage medium for recognizing chronic disease comorbidity patterns based on relaxed constrained symmetric low-rank representation. The method acquires medical data resources and uses the medical data resources to construct a chronic disease comorbidity network. The chronic disease comorbidity network is used to describe the correlation and influence between diseases. Low-rank representation learning is performed on the chronic disease comorbidity network to mine and output the community structure therein. The community partitioning mechanism is used to identify the community structure in the comorbidity network and explore the comorbidity pattern. The introduction of relaxed symmetric constraints in the learning objectives can well perceive the inherent symmetric structural characteristics of the network, and the introduction of graph regularization technology that captures local topological features effectively maintains the inherent geometric structural characteristics of the network.

The invention discloses a daily peak load prediction method and system based on graph regularization regression. The method comprises the following steps: acquiring multi-source historical data for daily peak load prediction; constructing multi-source features based on the multi-source historical data; establishing a feature association graph based on the multi-source features, and constructing a graph Laplacian matrix based on the feature association graph; constructing a graph regularization regression model based on the graph Laplacian matrix; solving the graph regularization regression model, determining an optimal parameter of the model, and determining an optimal model based on the optimal parameter; and predicting a future daily peak load based on the optimal model. According to the method, the daily peak load prediction model based on graph structure regularization regression is constructed, efficient and stable model solving is realized through the association relationship between feature graph modeling features and in combination with an alternating direction multiplier method ADMM, the future daily maximum load is predicted with high precision based on the model, the load peak can be accurately predicted in advance, and the prediction efficiency is improved. And important decision support is provided for power dispatching operation.