59 results about "Partition matrix" patented technology

The invention provides an inventory optimization method and system based on a deep learning model, and relates to the technical field of deep learning, and the method comprises the steps: obtaining the multi-dimensional data of a commodity, constructing a dynamic convolution kernel through employing a graph neural network, constructing a soft clustering distribution matrix in combination with a multi-layer perceptron, and obtaining a long-term dependence feature fusing a space dimension and a time dimension. And fusing the short-term and long-term features, inputting the fused short-term and long-term features into a bidirectional long-short-term memory network to generate a sales prediction sequence, and calculating a prediction confidence interval to obtain a final sales prediction result. And calculating a safe inventory level based on a sales prediction result, generating a replenishment decision by using a dynamic planning model based on deep reinforcement learning in combination with an inventory cost function, and finally sending the decision to a supply chain execution system. The sales volume is accurately predicted through the deep learning model, the inventory management is optimized, the storage cost, the stockout cost and the overstock cost are reduced, and the supply chain efficiency is effectively improved.

The invention relates to the technical field of wireless communication and edge computing fusion, in particular to a wireless network resource allocation method based on edge intelligence, which comprises the following steps: S1, collecting a multi-dimensional parameter set of network edge nodes in real time; s2, converting node position coordinates into a network topological structure map, and mapping service type labels and bandwidth requirements into service requirement feature vectors; s3, generating a domain division matrix based on weight distribution of the network topological structure atlas; s4, outputting a resource block configuration tensor according to the time-frequency feature of the business demand feature vector; s5, generating a resource allocation mapping relation table; and S6, generating and executing a node-level resource scheduling instruction set. According to the invention, by constructing a resource scheduling mechanism fusing a topological structure and service features, precise resource allocation and dynamic load balancing of a cooperative domain level and a node level are realized, and the efficiency and the resource utilization rate of multi-task concurrent processing in an edge computing environment are improved.

The invention relates to a heterogeneous GPU cluster scheduling optimization method and system based on resource affinity, and belongs to the field of computer resource scheduling. Comprising the steps of 1, according to an affinity historical library of historical jobs, matching the historical jobs with relatively high similarity through job characteristics, and estimating resource requirements; 2, for a new job, receiving the job, performing probe testing on each type of GPU, and obtaining and recording job performance data; 3, establishing an affinity evaluation model for each job by using historical information or test data, evaluating affinity values of the jobs under different resource configurations, and providing data support for resource allocation decisions; and 4, entering a scheduling optimization period cycle, constructing an allocation matrix according to a target function of an optimization problem, and allocating corresponding resources to each job. According to the method, accurate resource matching and data support are achieved, the dynamic adaptive capacity is guaranteed, and the system expandability is enhanced.

The invention relates to the technical field of bridge safety monitoring, and comprises a bridge safety monitoring analysis system based on big data, and the system comprises a data collection processing module, a manifold feature dimension reduction module, an entropy change partition evaluation module, a health state dynamic analysis module, and a bridge safety early warning module. According to the method, a bridge measuring point topological relation matrix is established, measuring point spatial distribution characteristics are analyzed, spatial relevance of a bridge structure is reflected, measuring point local curvatures are calculated, a local curvature matrix is established, a neighborhood similarity matrix is combined, low-dimensional projection transformation is executed, local topological consistency of data is kept in the dimensionality reduction process, an entropy change partition matrix is established, and the spatial relevance of the bridge structure is reflected. The health states of different structural parts of the bridge are subjected to differential analysis based on mechanical characteristics, the entropy change trend and damage probability of a measuring point are calculated, and the overall safety state level of the bridge is judged through a damage risk threshold, so that the local damage severity of the bridge can be comprehensively considered in safety assessment, and the stability and adaptability of safety state judgment are improved.

The invention provides an active power distribution network reactive voltage optimization method and related equipment. The method comprises the following steps: constructing a centralized optimization model; decomposing the boundary constraint conditions according to sub-partitions, and determining a sub-partition optimization problem; solving each sub-partition optimization model based on the sub-partition optimization problem, converting the sub-partition optimization model into a sequential quadratic programming sub-partition optimization model based on the solved first power flow solution, and determining each sub-partition matrix; extracting coupling information in adjacent sub-partition matrixes to obtain an optimization sub-problem about a Lagrange multiplier coefficient; determining the optimal search direction of each sub-partition; and correcting the first search direction by utilizing the optimal search direction, carrying out iterative updating until a preset condition is met, and ending optimization processing on the reactive voltage of the active power distribution network. According to the embodiment of the invention, in the calculation process, the partitions are mutually independent, only the coupling information of the adjacent nodes of each partition is exchanged, and the information privacy among the sub-partitions is protected.

The invention relates to the field of concrete 3D printing, and discloses a path optimization method for concrete 3D printing, which comprises the following steps: firstly, slicing a three-dimensional model of a to-be-printed concrete member, converting the sliced model into a geometric model, and initializing position information of geometric points in the geometric model; according to a graph theory principle, dividing the geometric model into a plurality of areas which can be printed at one time; the method comprises the following steps: constructing slices of a printing model into geometric figures which can be identified by a computer, extracting and numbering vertexes of the geometric figures, respectively recording coordinates of the vertexes, setting an adjacent matrix to describe a connection condition between the vertexes, then storing the vertexes into a partition matrix according to a rule, the partitions can be traversed at a time; finding out a shortest printing path in the plurality of divided areas by using an ant colony algorithm; and the printing platform is controlled to move according to the planned optimal printing path.

The invention provides a global structure sensing vector quantization method based on optimal transmission, which comprises the following steps of: mapping input data to a continuous hidden variable space through an encoder to generate continuous representation; constructing a distance matrix between the continuous representation and the codebook, and solving a distribution matrix by using an optimal transmission problem; carrying out normalization processing on the distance matrix, and generating an initial value of a distribution matrix by adopting a special initialization method; performing row and column normalization iteration on the distribution matrix through a Sinkhorn-Knopp algorithm until the distribution matrix is converged; and after iteration is finished, quantitative characteristics are determined based on the maximum value of the distribution matrix, the quantitative characteristics are mapped back to an original data space through a decoder, and a reconstructed data result is obtained. Based on the scheme provided by the invention, a codebook utilization rate close to 100% and an excellent data reconstruction result can be ensured, and a powerful tool is provided for discrete compression scene reconstruction of continuous data such as images and videos.

The invention relates to a high-order analysis power flow and carbon flow tracking method based on full-pure embedding, and the method comprises the specific steps: introducing a complex embedding parameter s, constructing a node voltage vector, an injection power vector and a tracking matrix in a power grid power flow equation into an analysis power series and an analysis power series of the complex embedding parameter s based on the analysis continuation characteristic of full-pure embedding, solving a node voltage initial value when the power grid is in a basic operation state by adopting a Newton-Raphson method; calculating a zero-order line active power vector based on the node voltage initial value, and constructing a zero-order distribution matrix representing a power distribution relation in a basic operation state of the power grid; solving a zero-order power flow tracking matrix based on the zero-order line active power vector and the zero-order power distribution matrix; calculating node indirect total carbon emission based on the zero-order power flow tracking matrix; and replacing the power distribution matrix, the node outflow power and the power flow direction matrix in the zero-order power distribution matrix with power series expansion of corresponding variables.

The invention relates to the technical field of data processing, and discloses a high-rise building elevator partition intelligent distribution method and system and a storage medium. The method comprises the steps that floor people flow data and elevator calling frequency are collected, and a floor weight incidence matrix is constructed in combination with functional attributes; optimizing the partition boundary through a partition boundary game algorithm; load distribution is monitored, the unbalanced state is recognized, and an elevator-partition distribution matrix is constructed; the unbalanced partition pairs are screened to execute boundary floor transfer operation; and scheme verification is carried out based on the reconstruction cost and the energy-saving income constraint, and an elevator partition intelligent distribution scheme is generated. The technical problems that in the prior art, a static partition strategy cannot adapt to dynamic people flow changes, partition loads are unbalanced, and multi-target comprehensive optimization is lacked are solved. And the energy consumption control level and the load balancing performance of the elevator system are improved.

The invention belongs to the field of matrix operation, and particularly discloses a matrix preprocessing method, system and equipment, and the method comprises the steps: converting a finite element grid corresponding to a sparse matrix into a graph structure; simplifying the graph structure by adopting a repeated edge matching algorithm to obtain a coarsened graph structure; dividing the coarsened graph structure into a plurality of sub-graphs through a direct division algorithm; the divided sub-graphs are refined and amplified through a vertex splitting and partition refining algorithm, and a plurality of refined graphs are obtained; performing corresponding mapping on the sparse matrix by referring to a plurality of refined images to obtain a plurality of partitioned matrixes; performing reduced filling sorting on each partition matrix through a multi-layer nested decomposition normal form algorithm to obtain a rearranged matrix of a plurality of partitions; the rearrangement matrix of the multiple partitions is used for parallel decomposition solving to achieve solving of the sparse matrix. Through the method, the limitation that an existing preprocessing method is large in calculation amount and difficult to parallel is overcome, and support is provided for efficiently solving a large-scale matrix.

A load balancing scheduling method, system and device relate to the technical field of resource scheduling, and the method comprises the following steps: setting a node label for each working node in a cluster to label a hardware resource type, and obtaining a load index of each working node; receiving a task request, obtaining a task stage and a task type, and forwarding the request to the Service corresponding to the task stage and the task type; task resource requirements, Pod topological constraints and resource utilization rates of scheduling nodes are obtained to construct a QUBO model, a task-to-node allocation matrix is obtained through a quantum annealing algorithm, and target nodes are determined; and starting the pod at the target node through the pre-deployed KubeRay, and when a preset elastic expansion and contraction condition is triggered, performing elastic expansion and contraction. According to the method, the hardware resource type is labeled through the node label, and the task demand and the node resource are dynamically matched in combination with the QUBO model and the quantum annealing algorithm, so that the resource utilization rate is effectively improved.

The application relates to the technical field of digital site selection planning, and particularly provides a reservoir resettlement area site selection method and system. The method comprises the following steps: constructing a reservoir resettlement area planning model under multi-target optimization, generating an initial site selection planning scheme; performing topological representation on the initial site selection planning scheme to obtain a current resident gathering point-resettlement area distribution matrix, and constructing a solution space; estimating an initial resettlement area quantity; realizing global search by using a simulated annealing algorithm, avoiding local optimization by probabilistically accepting a poor solution, combining the current resident gathering point-resettlement area distribution matrix and an adaptive multi-step neighborhood generation algorithm meeting the resettlement area quantity constraint, iteratively optimizing and adaptively adjusting a search strategy under the feasibility constraint, and converging to obtain an optimal resettlement area site selection scheme. The application organically combines a policy gradient method in reinforcement learning and a simulated annealing framework, and solves the problem of insufficient intelligence of reservoir resettlement area site selection.

A neural processing device and method for pruning thereof are provided. The neural processing device includes a processing unit configured to perform calculations, an L0 memory configured to store input and output data of the processing unit, wherein the input and output data include a two-dimensional weight matrix and a weight manipulator configured to receive the two-dimensional weight matrix and partition it into preset sizes to thereby generate partitioned matrices, to generate a pruning matrix by pruning the partitioned matrix, and to transmit the pruning matrix to the processing unit.

The invention discloses a parallel online similarity analysis method for trajectory flows, and relates to the technical field of trajectory analysis, and the method comprises the following steps: S1, collecting a first trajectory position flow and a second trajectory position flow of a moving object, and carrying out load partitioning to obtain a plurality of partitioning matrixes; s2, carrying out load balancing processing on each partition matrix; s3, based on the partition matrix subjected to load balancing processing, similarity increment updating is carried out; and S4, on the basis of a similarity increment updating result, calculating trajectory similarity by using an attenuation factor when new data is added. In order to improve the efficiency, a multi-stage pruning technology is implemented in two dimensions of space and time, and similarity calculation is accelerated.

The application discloses a multi-view clustering method and system based on self-step learning and view weighting, and belongs to the technical field of multi-view data processing. The method comprises the following steps: normalizing a multi-view data set, splicing the multi-view data, initializing each view clustering kernel and distribution matrix by using a kmeans algorithm, and calculating each view weight; each view sample weight matrix, each view clustering kernel and distribution matrix are iteratively updated in sequence through a target function; and when an iteration end condition is met, a final clustering kernel and distribution matrix are output. The clustering system comprises an acquisition module, a preprocessing module, a construction module, an optimization module and a clustering output module. The self-step learning model is used to sequentially learn clustering data and finally obtain a clustering result. Through view weighting, the model can selectively learn information of different views, thereby effectively improving clustering accuracy. The application can be applied to retrieval of an image database, a text database and the like.

The invention relates to an interpretable CNN (Convolutional Neural Network) training method and device based on Bernoulli sampling and a medium, and the method comprises the following steps: inputting a picture into a CNN, and obtaining a response feature map of a filter; performing Bernoulli sampling on the response feature map to obtain a binary distribution matrix; calculating a filter average weight matrix of each picture category according to the binarization distribution matrix, and calculating the sum of pairwise differences; calculating the Hadamard product of the distribution vector of the binary distribution matrix and the response feature map to obtain a mask feature map; respectively inputting the response feature map and the mask feature map into a CNN full connection layer to respectively obtain classification prediction probability vectors, and respectively calculating cross entropy loss with a real label; and according to the sum of the pairwise differences and the cross entropy loss, using a stochastic gradient descent method to realize network training, and obtaining an interpretable CNN for image classification. Compared with the prior art, the method has the advantages of high adaptability, high interpretability and the like.

The invention provides a project scale and progress prediction method, electronic equipment and a storage medium, and relates to the technical field of data processing, and the method comprises the steps: obtaining target enterprise project information, carrying out the multi-layer nested analysis of a project according to a plurality of project modules, and constructing a multi-layer nested project structure; on the basis of a multi-layer nested project structure, probability transformation is carried out on the membership relation between the nodes and the clusters, and a soft distribution matrix set is obtained; obtaining a plurality of project module structure entropy values; respectively analyzing the plurality of real-time acquisition data stream sets and the soft distribution matrix sets based on a dual-channel sequential network to obtain a plurality of prediction results of the plurality of optimization project modules; and obtaining a target project scale and progress prediction result. The technical problems that in the prior art, project scale and progress prediction is analyzed from a single dimension, the prediction result deviates from reality, and reliability is low are solved. The technical effects of considering hierarchical dependence between the nodes and improving the prediction accuracy are achieved.

The invention relates to a single-cell RNA (Ribonucleic Acid) sequencing data clustering method, device and equipment based on graph structure information fusion and a medium, and the method comprises the following steps: respectively calculating soft distribution matrixes corresponding to cell attribute representation, cell structure representation and consensus embedding representation, and deriving target distribution based on the soft distribution matrixes of the cell consensus embedding representation, all parameters of a zero-expansion negative binomial autoencoder, a graph autoencoder and a cross-modal fusion module are jointly optimized by minimizing the KL divergence between target distribution and each soft distribution matrix, iterative training is performed until the model reaches a preset convergence condition, and optimized cell consensus embedding representation is determined; and carrying out clustering analysis on the optimized cell consensus embedded representation by adopting a preset K-means clustering algorithm to obtain a cell type tag corresponding to the gene expression data of single cell RNA sequencing. According to the method, the problems of zero expansion, sparsity and over-dispersion characteristics of single-cell RNA sequencing data can be solved.

The invention discloses a source network load storage economic dispatching method considering multi-partition risk in extreme weather, which comprises the following steps: representing a plurality of substations in a power grid system as nodes to obtain a plurality of nodes; generating a typical section according to historical data, and generating a node load shedding risk distribution matrix Msr and a node wind and light abandoning risk distribution matrix Mcr based on the typical section; constructing a target optimization function and a constraint condition of Msr and Mcr based on a multivariable conditional value-at-risk formula, and solving to obtain a load shedding expected value-at-risk and a confidence coefficient of the plurality of nodes and a wind and light abandoning expected value-at-risk and a confidence coefficient of the plurality of nodes; and constructing an objective function of the source network load storage economic dispatching model, and optimizing reserve capacity constraints in constraint conditions of the objective function according to a solving result to obtain the source network load storage economic dispatching model. The invention relates to the technical field of power systems, and solves the technical problems that the risk assessment of the power system is inaccurate and the economy is difficult to guarantee.

The application discloses a kind of dynamic blockchain sharding method and system based on composite clustering algorithm, the method includes: based on blockchain sharding system, determine transaction frequency matrix and sharding allocation matrix, construct the target optimization transaction model of blockchain sharding system;Based on the target optimization transaction model of blockchain sharding system, preliminary global division processing is carried out through agglomerative hierarchical clustering algorithm;The preliminary blockchain sharding result is divided and processed by DBSCAN clustering algorithm, and the blockchain sharding result after secondary division is obtained;Based on the blockchain sharding result after secondary division, carry out noise account division processing, obtain the final blockchain sharding result.The application can preferentially aggregate high-frequency associated accounts and classify low-frequency accounts, minimizing the proportion of cross-shard transactions and achieving load balancing.The application is a kind of dynamic blockchain sharding method and system based on composite clustering algorithm, which can be widely applied in the field of blockchain transaction technology.

The application relates to a subarray partitioning method based on rotational symmetry. The method comprises the following steps: according to a subarray type and an antenna array surface, a dictionary matrix is constructed, each array element on the antenna array surface is labeled according to an arrangement order to obtain an initial array element order, each array element on the antenna array surface is rotated for multiple times according to the symmetry, and the array element order after each rotation is recorded; then a row of candidate subarrays is randomly selected from the dictionary matrix, corresponding multiple rows of candidate subarrays are selected from the dictionary matrix according to the relationship between the initial array element order and the array element order after multiple rotations, the multiple rows of candidate subarrays are added to a subarray partitioning matrix, the multiple rows of candidate subarrays are deleted from the dictionary matrix, the rows related to the multiple rows of candidate subarrays are deleted from the dictionary matrix to form a new dictionary matrix, the deleting operation is continuously performed on the new dictionary matrix, and finally the subarray partitioning of the antenna array surface is realized. The method can improve the speed of subarray partitioning.

The application relates to the technical field of building three-dimensional modeling, and discloses a three-dimensional modeling information matching method in building operation based on AI, which comprises the following steps: extracting a native level bounding box tree for accelerated rendering in a three-dimensional modeling engine, mapping the hierarchical structure of the tree into a spatial matching potential field; driving a dynamic semantic node to complete gradient descent flow along the hierarchical bounding box tree, and determining a distribution matrix of the dynamic node to an initial grid model; calculating a Laplace differential coordinate array of a grid vertex to lock native high-frequency curvature features, combining position anchor point constraints to construct and solve a sparse linear equation set, and reconstructing an updated three-dimensional model; the application locks the curvature details of a model surface by using differential geometric features, makes the three-dimensional model maintain original geometric textures when responding to displacement anomalies, solves the problem of topological tearing caused in a dynamic updating process, and enhances the fidelity of spatial mapping.

The invention discloses a block chain phishing account detection method and device based on a dynamic weight transaction subgraph, and relates to the technical field of block chain security monitoring. The method comprises the following steps: constructing a dynamic weight transaction sub-graph network based on transaction information; obtaining a block chain phishing account detection model, wherein the block chain phishing account detection model comprises a first-stage graph neural network, a first-stage differentiable graph pooling layer, a second-stage graph neural network, a second-stage differentiable graph pooling layer, a third-stage graph neural network and a classifier module; inputting the dynamic weight transaction sub-graph network into a first-stage graph neural network, inputting an obtained first distribution matrix into a first-stage differentiable graph pooling layer, obtaining a first compression feature, inputting the first compression feature into a second-stage graph neural network, obtaining a second distribution matrix, inputting the second distribution matrix into a second-stage differentiable graph pooling layer, and inputting an obtained second compression feature into a third-stage graph neural network; and inputting the obtained global features of the graph into a classifier module to obtain a prediction result of the phishing account. By adopting the method and the device, the accuracy of block chain phishing account detection can be improved.

The application relates to a communication-efficient decentralized federated learning method based on subgraph division, comprising the following steps: constructing a physical network topology graph of a decentralized federated learning environment; dividing the physical network topology graph into non-conflict subgraphs, and setting a mixing matrix for each non-conflict subgraph; constructing a problem constraint model with the minimum total iteration time as the target; optimizing a bandwidth allocation matrix with the delay value of the link with the maximum communication delay in the current non-conflict subgraph as the optimization target; optimizing the mixing matrix with the minimum expected communication delay as the optimization target; optimizing the non-conflict subgraph sampling probability based on the optimized mixing matrix with the minimum expected communication delay as the optimization target; and performing global updating of the decentralized federated learning model based on the optimized non-conflict subgraph sampling probability. The application can significantly reduce the global model convergence delay.

The present application relates to the technical field of bridge safety monitoring, comprising a bridge safety monitoring and analysis system based on big data, which comprises a data acquisition and processing module, a manifold feature dimension reduction module, an entropy change partition evaluation module, a health state dynamic analysis module and a bridge safety early warning module.In the present application, a bridge measurement point topology relationship matrix is established, the spatial distribution characteristics of the measurement points are analyzed, the spatial correlation of the bridge structure is reflected, the local curvature of the measurement points is calculated and a local curvature matrix is established, the neighborhood similarity matrix is combined, and low-dimensional projection transformation is performed, so that the data maintains local topological consistency during dimension reduction, an entropy change partition matrix is established, the health states of different structural parts of the bridge are differentiated based on mechanical properties, the entropy change trend and damage probability of the measurement points are calculated, and the overall safety state grade of the bridge is judged through a damage risk threshold, so that the safety evaluation can comprehensively consider the severity of local damage of the bridge, and the stability and adaptability of safety state determination are improved.

This disclosure provides for devices, methods, and systems for immobilizing partitions of a partition matrix, where partitions are rapidly generated from a sample, and for immobilizing a set of targets within a matrix, where examples of the matrix are optically clear. Embodiments of methods involve partition generation within a collecting container at a high rate (e.g., of at least 1 million partitions per minute, etc.), each of the partitions including an aqueous mixture. Embodiments of methods also involve stabilizing positions of targets within a matrix within a collecting container and performing reactions within the matrix at a temperature below the melting temperature of the matrix. Upon immobilization, readout of signals from targets within such partitions and / or the matrix can still be achieved in accordance with various assays.

The present application belongs to the technical field of nondestructive testing of concrete structures, and particularly relates to a method for identifying the topology of internal defects in concrete based on a graph neural network, comprising: obtaining three-dimensional detection data of a concrete test piece for voxel reconstruction and extracting a three-dimensional defect candidate; constructing an initial graph based on spatial adjacency relationships and extracting node features and edge features; recalibrating the adjacency relationships and edge weights based on the local graph structure topology index of the initial graph to obtain a topology-constrained graph, and performing message passing on the topology-constrained graph while updating the node embedding in combination with adaptive information aggregation weight; constructing a fitness function, optimizing the allocation matrix of nodes to clusters to determine graph coarsening mapping, and generating a multi-layer defect skeleton subgraph; extracting a cross-layer topology description vector from the multi-layer defect skeleton subgraph and outputting the classification and identification result of the defect topology structure. The present application can accurately classify and identify the topology of complex internal defects in concrete.

The application provides a pairing method and device for non-fixed additional pseudo codes, wherein the method comprises the following steps: setting pairing indexes, the first index being the aligned cross-correlation value absolute value, and the second index being the absolute value of the difference between the leading and lagging correlation values; determining a target function and a cost matrix based on the first index and the second index; establishing an additional pseudo code allocation model based on the cost matrix and an allocation matrix; dividing the additional pseudo codes into two parts, determining an initial population based on the division result, and solving the model by using a JVC algorithm to determine the initial pairing scheme corresponding to the individuals in the initial population; performing iterative updating on the population based on a meta-heuristic algorithm or an improved meta-heuristic algorithm, and performing iterative updating on the pairing scheme corresponding to the individuals based on the JVC algorithm until a preset maximum number of iterations is reached, and determining the optimal pairing scheme obtained in the last iteration process as the optimal pairing scheme of the non-fixed additional pseudo codes; wherein the optimal pairing scheme obtained in the iteration process is determined based on the target function.

The present invention provides an inventory optimization method and system based on a deep learning model, which relates to the technical field of deep learning. The method includes obtaining multi-dimensional data of commodities, constructing a dynamic convolution kernel using a graph neural network, and combining a multi-layer perceptron to construct a soft clustering assignment matrix to obtain long-term dependence features that integrate spatial and temporal dimensions. After fusing short-term and long-term features, the fused features are input into a bidirectional long short-term memory network to generate a sales volume prediction sequence, and the prediction confidence interval is calculated to obtain the final sales volume prediction result. Based on the sales volume prediction result, the safety inventory level is calculated, and combined with an inventory cost function, a dynamic programming model based on deep reinforcement learning is used to generate a replenishment decision, and finally the decision is sent to the supply chain execution system. The present invention accurately predicts sales volume through a deep learning model and optimizes inventory management, reduces warehousing costs, out-of-stock costs, and overstock costs, and effectively improves the efficiency of the supply chain.

The invention relates to a clustering federated learning method based on model decomposition and electronic equipment, and the method is realized based on model decomposition and low-rank similarity, and comprises the following steps: model decomposition and initialization: splitting a client model into a shallow embedding model and a deep decision model, initializing global embedding parameters and clustering center parameters of the server; based on a clustering algorithm of low-rank model similarity, performing low-rank decomposition on the client model splicing matrix through a dimension reduction technology to obtain a low-rank matrix, and calculating the similarity of the low-rank matrix; a knowledge integration and parameter updating mechanism: optimizing and updating client model parameters in combination with global embedded knowledge and clustering center knowledge in the step 1; and dynamically updating the parameters of the clustering distribution and clustering center model: dynamically adjusting the clustering distribution matrix based on the low-rank representation of the low-rank matrix, and updating the parameters of the global embedding and clustering center. Compared with the prior art, the method has the advantages of improving the performance and efficiency of federal learning in a heterogeneous environment and the like.