14 results about "Centering matrix" patented technology

The invention discloses a high-dimensional data clustering method and device, equipment, a medium and a product, and relates to the field of clustering analysis in machine learning, and the method comprises the steps: updating a membership matrix according to a cluster center matrix, a projection matrix and a cluster pair similarity matrix; the membership matrix is constructed based on a local information factor of a high-dimensional data sample; updating the cluster center matrix; according to the original high-dimensional data matrix, the updated membership matrix, the updated cluster center matrix and the cluster pair similarity matrix, an auxiliary matrix is constructed, eigenvalue decomposition is solved, and a projection matrix is updated; constructing a convex optimization problem according to the updated membership matrix, the updated cluster center matrix and the updated projection matrix, and updating a cluster pair similarity matrix; judging whether a termination condition is met or not; if yes, dividing the high-dimensional data sample according to the updated membership matrix, and outputting a clustering result; and if not, returning to the membership degree matrix updating step. According to the method, the clustering result of the clustering method on the high-dimensional data can be improved.

The invention provides a matrix multiplication calculation method and device based on a three-dimensional accelerated calculation architecture, and the method comprises the steps: segmenting a quantized target matrix into bit slice matrixes along a bit dimension, and carrying out the clustering of the bit slice matrixes along a row vector dimension, and obtaining a clustering center matrix and a class residual matrix; combining repeated operation between different columns of the clustering center matrix to obtain an intermediate activation vector; multiplying the enumeration matrix by the intermediate activation vector to reconstruct a calculation result of the clustering center; multiplying the similar residual matrix by the input activation, and adding with the calculation result of the corresponding clustering center to obtain the calculation result of the similar residual matrix; and obtaining a multiplication operation result of the target matrix and the input activation based on calculation results of all the clustering centers and the class residual matrix. According to the method, through joint optimization of three dimensions of bit, row and column, the redundancy calculation is minimized, the number of necessary operations is effectively reduced, and the complexity and calculation overhead of large-scale matrix multiplication are reduced.

The application discloses an anchor point guided clustering based data classification method, device and equipment, relates to the technical field of digital data processing, and comprises the following steps: obtaining an original data set to be classified, converting each sample data into a numerical vector, and constructing a data matrix; initializing a clustering center matrix and an anchor point matrix, and setting a fuzzy coefficient; obtaining the anchor point matrix and the clustering center matrix after joint iterative optimization; calculating the fuzzy membership degree of each sample vector; the category with the maximum fuzzy membership degree is taken as the final category label of the sample vector, and the classification results of all sample vectors are output. The application solves the problems that the existing fuzzy clustering method is sensitive to initial conditions, is easy to fall into a suboptimal solution, leads to unstable and inaccurate classification results, and cannot be directly solved by using a gradient descent algorithm, and is difficult to be applied to large-scale data sets, and realizes the enhancement of complex data classification precision and the applicability in different data scale scenes.

The application provides a hyperspectral image band selection method, device and electronic equipment. The method comprises: sampling each band image of a hyperspectral image in a three-dimensional space to obtain a sampling two-dimensional matrix; calculating the information entropy of each sampling band in the sampling two-dimensional matrix, grouping the information entropy, selecting a sampling band corresponding to the maximum information entropy in each group to form an initial clustering center matrix; performing bias processing on the initial clustering center matrix to obtain a target clustering center matrix; based on the sampling two-dimensional matrix and the target clustering center matrix, calculating the membership degree of each sampling band in the sampling two-dimensional matrix to each clustering center in the target clustering center matrix to obtain an initial membership matrix; calculating a kernel matrix based on the sampling two-dimensional matrix; and determining a target band selected in the hyperspectral image based on the kernel matrix and the initial membership matrix. The method is efficient in calculation and improves the local optimal solution problem.

The invention discloses a data classification method, device and equipment based on anchor guide clustering, and relates to the technical field of electrical digital data processing, the method comprises the following steps: obtaining a to-be-classified original data set, converting each sample data into a numerical vector, and constructing a data matrix; initializing a clustering center matrix and an anchor point matrix, and setting a fuzzy coefficient; obtaining an anchor point matrix and a clustering center matrix after joint iterative optimization; calculating the fuzzy membership degree of each sample vector; the category with the maximum fuzzy membership degree value serves as a final category label of the sample vectors, and classification results of all the sample vectors are output. The problems that an existing fuzzy clustering method is sensitive to initial conditions and is prone to falling into a suboptimal solution, so that a classification result is unstable and inaccurate, and a gradient descent algorithm cannot be directly used for solving and is difficult to adapt to a large-scale data set are solved; the method achieves the enhancement of the classification precision of complex data and the applicability in different data scale scenes.

The invention discloses a feature selection method and system based on ocean data. The method comprises the steps that original ocean data are acquired and integrated into a data set matrix, and a feature selection matrix, a clustering center matrix and an auxiliary matrix are initialized to be in a preset initial state; updating a fuzzy membership matrix based on the data set matrix, the feature selection matrix and the clustering center matrix; updating a weight vector according to the data set matrix, the feature selection matrix, the clustering center matrix and the fuzzy membership matrix; and updating the feature selection matrix through the data set matrix, the auxiliary matrix, the clustering center matrix, the fuzzy membership matrix and the weight vector. According to the method, the fuzzy membership degree is introduced in the unsupervised feature selection process, so that the data can be shared among multiple clusters, and the complex structure and uncertainty of the data are better reflected.

The application discloses a power distribution operation scene construction method and device based on load clustering, and the method comprises the following steps: acquiring multiple groups of historical load data; performing feature extraction on each group of historical load data respectively to obtain a load feature matrix; combining a preset initial cluster number to determine an initial cluster center set from the load feature matrix; performing clustering analysis based on each initial cluster center in the initial cluster center set to obtain a historical clustering center matrix; acquiring a synthetic clustering center matrix, fusing the historical clustering center matrix, and giving a mixed clustering center matrix; performing secondary clustering analysis based on the mixed clustering center matrix to obtain a target clustering center matrix, so as to complete the construction of the power distribution operation scene. Through clustering of historical load data of a power distribution network, load characteristic analysis is realized, a load feature matrix is obtained through feature extraction, and potential change rules are mined based on the load feature matrix, so that the target clustering center matrix is given, and the accuracy of the construction of the power distribution operation scene is improved.

The application provides a random subspace-based coastal zone ecological disaster mitigation assessment method, device, equipment and storage medium, a sample matrix is constructed, and a feature transformation matrix, a discrete coding matrix and a clustering center matrix are initialized, in an iterative updating process, a similarity matrix of each sample belonging to a feature transformation matrix array vector set is calculated, different types of coastal zone samples can be associated to different feature subspaces, and therefore the limitation of a single subspace assumption is broken through; finally, key risk assessment indexes are screened out according to L2 norm ordering of each row vector of the feature transformation matrix, and targeted ecological disaster mitigation risk assessment on a coastal zone region with spatial heterogeneity is realized.

The invention provides a spot day-ahead market decision sample matrix clustering preprocessing method and system, and belongs to the technical field of electricity market and data preprocessing, and the method comprises the steps: determining the time granularity and statistical period of a sample, and constructing a sample matrix; introducing a dimension weight matrix for each cluster, defining a cluster center matrix, updating a cluster center under a typical fuzzy clustering framework, and adaptively updating a dimension weight according to a weighted variance in the cluster; explicitly introducing a nuclear norm regular term into the clustering objective function; solving the minimum value of the clustering objective function by adopting an alternating iterative optimization strategy; and outputting the cluster to which each sample belongs, the membership of the cluster, the typical operation mode corresponding to each cluster and the abnormal sample set. According to the method, abnormal working condition samples such as extreme climate, sudden maintenance and data errors can be effectively identified, the robustness of a subsequent prediction and optimization model is improved, and the efficiency and precision of links such as day-ahead market quotation strategy optimization, flexibility evaluation and safety check are improved.

The application discloses an image processing model training method, an image processing method and an electronic device, relates to the technical field of image processing model training, and comprises the following steps: obtaining a current batch of samples in a training image set, removing the category labels of the samples to obtain a first category set, and extracting a first category center vector set from a target category center matrix; a second category set is formed by the remaining category labels in the training image set which do not belong to the first category set, a category is sampled from the second category set according to an importance weight, a second category center vector set is extracted, and the importance weight is determined according to the correlation degree between a category center vector and a sample feature vector and the gradient norm of the category center vector in a preset number of iterations; the first and second category center vector sets are merged, and the training parameters are updated according to the loss value calculated by the sample feature vector and the merged set, thereby solving the problems of large memory consumption and insufficient training of few sample categories during image processing model training, and improving the training reliability.

The embodiment of the specification provides a method and device for multi-party joint dimension reduction processing for privacy protection. The method comprises the following steps: reading local features of M samples in a current batch according to a predetermined sample sequence to construct a local original matrix; locally calculating a first matrix obtained by multiplying a transpose matrix of the local original matrix with itself, and multiplying the transpose matrix of the local original matrix with original matrices of other parties by using multi-party secure calculation to obtain a second matrix and multiplying transpose matrices of the original matrices of the other parties with the local original matrix to obtain a third matrix; splicing the first matrix, the second matrix, the third matrix and first matrices of the other parties to form a merged matrix; performing mean value processing on the merged matrix to obtain a center matrix; accumulating the center matrices of each batch to obtain a covariance matrix; and performing multi-party joint dimension reduction processing on sample feature data of each batch by using the covariance matrix. The processing speed can be improved.

The invention relates to a large-scale federated multi-view clustering method, system and device and a medium. The method comprises the following steps: a client generates local data representation based on a linear kernel matrix of view data; the server integrates local data representations of the clients and performs clustering to obtain consistent labels and clustering centers, the consistent labels serve as pseudo labels of the clients, orthogonalization is performed on the clustering centers to obtain consistent center matrixes, then segmentation is performed, and segmentation results serve as center matrixes of the clients; in order to maximize the consistency between local data representation of each client and a linear kernel matrix, the consistency between local data representation and a pseudo tag and a central matrix, and the consistency between server integrated data representation and a consistent central matrix and a consistent tag, local data representation orthogonal constraints and central matrix orthogonal constraints are used as constraint conditions; constructing a target function; and solving the objective function according to a loop optimization strategy to obtain a clustering result. By adopting the method, the efficiency and precision of federal multi-view clustering can be improved.

The invention discloses a method for realizing nondestructive testing of a fuse through an equivalence class division method, and relates to the technical field of power electronics. The method for realizing the nondestructive testing of the fuses through the equivalence class division method comprises the following steps: 1) testing all to-be-detected samples of the fuses by using small current, and collecting data; 2) initializing the collected data, giving a clustering category number, setting an iteration stop threshold, initializing a clustering center matrix, and setting an iteration counter b to be zero; 3) calculating or updating a membership matrix U of the data; (4) updating the clustering center matrix V of the data, and (5) stopping and outputting the membership matrix U and the clustering center matrix V if the variation dV of the clustering center matrix is smaller than a threshold value, otherwise, adding one to the counter, and repeating the step (4). The fuse detection method can avoid damage to the sample to be detected and obtain reliable data of sample quality.

The present application relates to a large-scale federated multi-view clustering method, system, device, and medium. The method includes: the client generates local data representations based on the linear kernel matrix of view data; the server integrates the local data representations of each client and performs clustering to obtain consistent labels and clustering centers, uses the consistent labels as the pseudo-labels of each client, orthogonalizes the clustering centers to obtain a consistent center matrix and then performs segmentation, and uses the segmentation result as the center matrix of each client; with the goal of maximizing the consistency between the local data representations of each client and the linear kernel matrix, the consistency between the local data representations and the pseudo-labels and center matrices, and the consistency between the server-integrated data representations and the consistent center matrix and consistent labels, and with the local data representation orthogonal constraint and the center matrix orthogonal constraint as the constraint conditions, a target function is constructed; the target function is solved according to the cyclic optimization strategy to obtain the clustering result. Using this method can improve the efficiency and accuracy of federated multi-view clustering.