69 results about "Matrix form" patented technology

The invention provides a high-speed NTT method and system based on GPU matrix-thread collaborative optimization. The method comprises the steps that an input twiddle factor is converted into a matrix form; transmitting the updated twiddle factor matrix and the input array to a graphics processor; according to the input array and the size of the twiddle factor matrix, the number of blocks in the grid and the number of threads included in each block are set, so that each block is responsible for the operation of multiplying one row by one column, and each thread in the block completes the operation of multiplying one pair of data in the corresponding row-by-column process to obtain a group of intermediate results; and obtaining output data corresponding to the block according to an intermediate result in the block, and sending the output data to a corresponding position of an output matrix to serve as a multiplication result. According to the method, NTT calculation is reconstructed into parallel matrix-vector operation, the maximum utilization of GPU calculation resources is realized through the combination of an efficient thread allocation strategy and an optimized memory access mode, and the calculation efficiency is remarkably improved.

The invention discloses a non-uniform sparse sampling single-section near-field test system and method, and belongs to the technical field of test. According to the method, the single-section radiation field is expressed in a matrix form, the unknown transmission coefficient is solved from the system of linear equations through an inverse matrix method, and the defect that a traditional single-section near-field test method based on fast Fourier transform depends on the distribution uniformity of near-field samples is overcome; under the assumption that antenna transmission coefficients have sparsity, a compressed sensing technology is introduced, L1 norm minimization constraint is applied to a transmission coefficient vector q to ensure the sparsity, and under the condition that the number M of single-section near-field samples is smaller than the number J of unknown coefficients, the transmission coefficients can be solved from an underdetermined linear equation set; according to the invention, a non-uniform sampling-based single-section near-field test method is realized, and the flexibility of a sampling mode is improved; the number of sampling points and the scanning time are reduced, and the testing efficiency is improved; and the far-field reconstruction precision is higher.

The invention discloses a multi-period non-stationary hydrological variable valuation method, and relates to the technical field of hydrological prediction. Comprising the steps of obtaining hydrological variables of a plurality of observation stations in a drainage basin at different time points as hydrological time sequences of the observation stations; constructing an experimental variation function according to the hydrological time sequence of each observation station; selecting a drift model through a trend surface analysis method and model inspection; constructing a multi-period non-stationary variable equation set comprising a Lagrange multiplier and a hydrological variable weight based on the experimental variation function and the drift model; expressing the multi-period non-stationary variable equation set as a matrix form, and calculating a Lagrange multiplier and a hydrological variable weight through a matrix inversion method; and constructing an estimation formula according to the calculation result of the Lagrange multiplier and the hydrological variable weight, and performing multi-period spatial estimation on the hydrological variable of the to-be-measured drainage basin through the estimation formula. Dependence on the number of observation stations can be reduced, and accurate estimation can be achieved only through a small amount of observation station data.

The invention discloses a renewable energy source data clustering method and system based on an optical quantum computer, and the method comprises the steps: building a renewable energy source data clustering optimization problem based on a quadratic unconstrained binary optimization (QUBO) model based on a discrete optimization objective function, carrying out the matrix of the objective function, carrying out the clustering optimization of the objective function through a kernel method, and carrying out the clustering optimization of the objective function. The method comprises the following steps of: establishing a target function item, converting the target function item into a matrix form, constructing a matrix of a new clustering model, finally solving a new QUBO optimization problem by using CIM, and converting the new QUBO optimization problem into an optimal clustering result. According to the method, data similarity measurement is improved by constructing the QUBO model and combining a kernel method, and a renewable energy source data clustering problem is converted into a form suitable for CIM solution, so that the calculation bottleneck of a classical calculation method in high-dimensional data processing is avoided. According to the method, the nonlinear characteristic of the data can be kept, efficient clustering optimization calculation can be realized, and the calculation efficiency and the clustering precision are remarkably improved.

The invention discloses a fixed time inclusion control method and system for a second-order nonlinear multi-agent system, and the method comprises the steps: defining a state tracking error according to a control target, augmenting the state tracking error into a matrix form, and designing an integral sliding mode surface for suppressing disturbance; designing a distributed fixed time inclusion consistency protocol and a dynamic event triggering mechanism based on position information and speed information according to the integral sliding mode surface; and controlling the multi-agent system model to reach an integral sliding mode surface according to a distributed fixed time inclusion consistency protocol and a dynamic event triggering mechanism, then sliding on the integral sliding mode surface, and controlling the state of the multi-agent system model to realize inclusion consistency within fixed time. According to the invention, the energy consumption of the system and the updating frequency of the controller can be reduced, the service life of the system is prolonged, and under a new distributed fixed time inclusion consistency protocol, the system firstly reaches the sliding mode surface and then slides on the sliding mode surface, so that the inclusion consistency of the system state is finally realized within fixed time.

The invention discloses a multi-probe spherical near-field sampling system based on compressed sensing and a test method, and belongs to the technical field of measurement. According to the method, the antenna transmission equation is expressed in a matrix form, the unknown transmission coefficient is solved from the linear equation set through the matrix method, and the defect that a traditional spherical near-field test method based on FFT depends on the distribution uniformity of near-field samples is overcome; according to the method, under the assumption that antenna transmission coefficients have sparsity, a compressed sensing technology is introduced, L1 norm minimization constraint is applied to a transmission coefficient vector q to ensure the sparsity of the transmission coefficient vector q, and under the condition that the number M of spherical near-field samples is smaller than the number J of unknown coefficients, the transmission coefficients can be solved from an underdetermined linear equation set; according to the invention, on the basis of sampling coordinates obtained based on prior information of an antenna, in combination with an adaptive positioning algorithm, each probe is controlled to autonomously select an optimal path and move to a specified position along an annular frame to carry out interference-free sampling, so that data acquisition of a pitching section at any angle is realized.

The invention relates to the technical field of artificial intelligence and knowledge engineering, and discloses a knowledge graph completion method, system and device based on entity relation attention and a medium, and the method comprises the steps: distributing an embedded vector with a fixed length for each entity and relation in a knowledge graph, and employing a first initialization method to initialize the embedded vector; respectively mapping and remodeling the embedded vectors of the head entity and the relationship into a matrix form; capturing a deep relationship between the head entity and the relationship through an attention mechanism, and generating a joint feature vector; calculating the joint feature vector and the embedded vector of the tail entity to obtain a scoring result of the triple; and sorting and selecting all the candidate triads according to the scoring result, and adding the first selection result as a completion result into the knowledge graph. According to the method, the deep semantic association between the entity and the relationship can be effectively captured, and key information is dynamically focused through an attention mechanism, so that knowledge graph completion with better performance is realized.

THE WAY IN WHICH THE ESSENTIAL ASPECT OF THE STRATEGE "FROM NOTHING - SOMETHING" IS IMPLEMENTED IN INFORMATION TECHNOLOGIES. The initial essence "Nothing" (0, Zero) of the method is the abstract essence. The mathematical model of the Projected Object Entity (PEO) is embodied in material, software, and mixed entity objects from various disciplines—the "something" of the method. The methodology of comparative cause-and-effect analysis is applied in the creation of the PEO, and the methodology of essential synthesis is used in its design: subject-specific and essential interdisciplinary properties of similarity, universal mathematical models in the form of orgraphs and their matrices, graphs with algorithms of nodes and their matrices, as well as essential identities, equations, and differential equations that reflect the balance of actions and counteractions.And since the professional and interdisciplinary properties of similarity for individual objects of different disciplines continue to apply, the transfer of the parental properties, qualities and the collective image of "nothing" and its "something" takes place "according to the image and likeness" of the essence.

The invention discloses a carbon price prediction method and system based on a graph G and a model prediction algorithm, and the method comprises the steps: firstly building a carbon price network, representing the network in the form of the graph G, then introducing a structural equation model to capture key variables of the network, and enabling the structural equation model to be expressed in the form of a matrix, adding additional items into the structural equation model to dynamically and accurately capture the carbon price fluctuation in the complex environment, and finally, constructing a carbon price prediction model by taking the minimum weighted prediction error and policy deviation cost as objective functions, and solving the model by adopting an artificial hummingbird algorithm and adding a chaotic mapping Bernoulli initialized population method. And obtaining the predicted carbon price. According to the carbon price prediction method, the carbon emission price with obvious fluctuation characteristics can be accurately predicted, and the effectiveness of the carbon emission trading market can be realized.

The present invention discloses a method, a system and a device for extracting vibration signal characteristics of a hydropower unit, relating to the field of feature extraction, including: obtaining the vibration signal of the hydropower unit; hierarchically decomposing the time series of the vibration signal by using moving average and moving difference methods based on conditional bubble entropy to obtain the matrix form of the corresponding operator for each layer; constructing a vector of non-negative integers e and obtaining the matrix form of the corresponding operator for the vector at each layer, and obtaining the hierarchical component at each layer node e by multiplying the matrix form of the corresponding operator for each layer by the time series; calculating the CBE value of each layer, and defining the set of CBE values of all components as the improved hierarchical conditional bubble entropy value HCBE; finding the optimal parameter of HCBE, and when the non-negative integer e is less than the threshold, outputting IHCBE as the feature characterizing the fault. The present invention proposes an improved hierarchical conditional bubble entropy, which makes the obtained information more detailed and improves the accuracy of hierarchical bubble entropy calculation.

The invention relates to a matrix type differential passivity index determination method and device, equipment and a storage medium. The method comprises the following steps: acquiring a linear system model corresponding to target equipment in a power system, wherein the power system is a multiple-input multiple-output system; determining a corresponding differential passivity equivalent system model according to the linear system model; based on the linear matrix inequality, a matrix type passivity index of the differential passivity equivalent system model is solved, the matrix type passivity index is used as a matrix type differential passivity index of the linear system model, the matrix type differential passivity index is in a matrix form, and the matrix type differential passivity index is used as a differential passivity index of the linear system model; matrix elements in the ith row and the jth column in the matrix represent coupling passivity characteristics between an input channel i and an output channel j in the power system. By adopting the method, the passivity characteristics in the MIMO dynamic system can be represented more accurately and comprehensively.

The present invention provides a dense modal response reconstruction method based on a state transfer matrix, belonging to the field of structural response reconstruction, and comprising the following steps: S1, decomposing each order modal response, separating the initial phase from the initial amplitude in the response, and expressing each order modal response in matrix form; S2, constructing the state matrices of the measured point and the point to be reconstructed, respectively, and obtaining the modal mode ratio matrix of the measured point and the point to be reconstructed; S3, constructing the state transfer matrix between the point to be reconstructed and the measured point; S4, reconstructing the response of the point to be reconstructed according to the state transfer matrix. The dense modal response reconstruction method based on a state transfer matrix proposed by the present invention, based on the modal superposition method of linear systems, separates the initial phase from the initial matrix in the modal response, and reconstructs the response according to the state transfer matrix between the point to be reconstructed and the measured point. This method can handle the response reconstruction of structures with dense modal characteristics and ensure the accuracy of the response reconstruction.

This invention presents a method for establishing large deformation viscoelastic constitutive relations based on Prony technology. It addresses the problems of existing viscoelastic constitutive models lacking universality, and the complexity and programming difficulties associated with their mathematical relationships. The invention includes: decomposing the stress tensor into the sum of the deviatoric stress tensor and the mean stress; decomposing the strain tensor into the sum of the deviatoric strain tensor and the mean strain; and providing a convolutional expression for the deviatoric stress tensor and the mean stress. The increment of the deviatoric stress tensor is expressed as a function of the deviatoric strain increment, and the increment of the mean stress is expressed as a function of the mean strain increment. The tensor form of the incremental constitutive relation is written in matrix form. This method has been programmed, and the numerical results obtained from the program agree well with the experimental results in specimen-level experiments on solid rocket motor propellant grains.

The present invention discloses a method, system, medium and product for error identification of a coordinate measuring machine. The present invention includes obtaining the measurement coordinates during the movement of the space diagonal of a standard sphere in a preset body space and establishing an error equation for each measurement position, which includes the influence caused by the probe error and the error value composed of the polynomial coefficients of the geometric error and the perpendicularity error; after expanding the error equation, neglecting the second-order small quantities to construct a linear equation system in matrix form for multiple measurement positions; subtracting the linear equation systems in matrix form of adjacent measurement positions to obtain a linear equation system in matrix form of the differences between adjacent measurement positions, thereby decoupling the influence caused by the probe error and obtaining the error value composed of the polynomial coefficients of the geometric error and the perpendicularity error, as well as the geometric error and the perpendicularity error. The purpose of the present invention is to simply and reliably achieve the accurate identification of the geometric error of the coordinate measuring machine in order to achieve the precise compensation of the geometric error of the coordinate measuring machine.

The invention discloses a method for calculating the inherent frequency of a porous two-dimensional functionally graded material framework, and belongs to the technical field of structural design, the framework is simplified into a beam model with two clamped ends, a motion control equation of a porous two-dimensional functionally graded material beam in a humid and hot environment is deduced through a Hamilton principle based on a Timoshenko beam theory, and the inherent frequency of the porous two-dimensional functionally graded material beam is calculated. Simplifying the equation by using a variable separation method; the inherent frequency of the beam is solved by adopting a differential quadrature method; a motion control equation is discretized by utilizing a differential quadrature method discretization principle, then the equation is converted into a matrix form, the problem of solving the motion equation is converted into the problem of solving a matrix eigenvalue by adopting a matrix transformation method, and finally solving is carried out by writing a corresponding program. The corresponding solving program is clear in logic and high in uniformity, the calculation efficiency and precision are greatly improved, and a theoretical reference can be provided for the design of the functional gradient skeleton in the later period.

The invention discloses a BP16 matrix multiplication and addition operation circuit based on an RISC-V framework, and relates to the technical field of processor microstructures and artificial intelligence calculation acceleration. The circuit comprises an input module, a matrix multiplication and addition operation module, an accumulation operation module and an output module, wherein the input module is used for inputting a first operand in a BP16 format, a second operand in the BP16 format and a third operand in a preset format; the matrix multiplication and addition operation module periodically performs row and column multiplication operation on a plurality of first floating point elements of a preset row number in the first operand and a second floating point element in the second operand in a matrix form to generate a fourth operand; the accumulation operation module performs accumulation calculation on the fourth operand and the third operand to generate a fifth operand; and the output module outputs the target operand based on the fifth operand. The throughput rate and the energy efficiency ratio of AI acceleration calculation under the RISC-V architecture can be remarkably improved.

The application discloses a kind of based on time delay recursive discrete grey model's weld seam tracking error real-time prediction method, the method includes the following steps: step S1: obtaining real-time weld seam tracking error data;Step S2: according to new information priority principle, using new information priority accumulation operation to time series data smoothing weld seam tracking error data;Step S3: using time delay grey differential equation and discrete equation establishes the matrix form of TRDGM;Step S4: using recursive least square method to derive the recursive equation of TRDGM structure parameter;Step S5: construct the nonlinear optimization objective function of TRDGM;Step S6: using heuristic optimization algorithm to optimize the hyperparameter of TRDGM;Step S7: select prediction accuracy index and computational efficiency index, evaluate the prediction performance of TRDGM.The TRDGM proposed in the above can quickly and accurately predict weld seam tracking error under small sample condition, help to predict in advance the risk of weld seam tracking failure.

The application discloses a kind of meta-heuristic test design optimization methods based on matrix transformation, belong to computer-aided test design and statistical modeling technical field, this method uses a two-stage optimization framework: first, in global exploration stage, with minimizing global uniformity index as target to carry out unconstrained optimization, to obtain well space coverage sample set;Subsequently in local development stage, with minimizing adjacent distance variance as target, optimization is carried out under the constraint of maintaining the good global coverage obtained, to fine adjustment sample point local spacing distribution, improve uniformity;The application encodes test design scheme into matrix form individual, is updated iteratively by meta-heuristic algorithm, and adopts global normalization strategy to ensure that sample is always located in design space;The application can systematically consider the global coverage of sample and local uniformity, significantly improve the sample distribution quality in high-dimensional space, generate comprehensive performance excellent test design scheme.

The embodiment of the application provides a kind of based on agent model's oil reservoir remaining oil prediction method and system, the method includes: setting oil reservoir simulation data, combining oil reservoir simulation module, generate sample data set;Based on the data type of sample data set, corresponding data preprocessing is carried out, and data is converted into two-dimensional matrix form;Residual oil prediction model is constructed, and residual oil prediction model is the encoder-decoder network of fusion residual structure, and coordinate attention mechanism is introduced;Two-dimensional matrix form sample data set is input data, and residual oil prediction model is iteratively optimized, and iteration is until convergence, to obtain the trained residual oil prediction model;The complete input parameter of target oil reservoir is acquired, and the trained residual oil prediction model of input complete parameter is output target oil reservoir remaining oil distribution result.

The present invention relates to the field of grain temperature technology and discloses a thermodynamically based grain pile temperature calibration method and system. This method aims to address the high cost and poor accuracy of existing grain pile temperature detection methods. The solution primarily includes: installing a sensor matrix in the grain pile and obtaining the temperature of each sensor-detected point in the sensor matrix; discretizing the space where the grain pile resides into grain pile grids, determining the temperature of each grain pile grid based on the temperature detected by each sensor, and generating a temperature distribution field representing the temperature distribution; constructing a thermodynamic equation for each grain pile grid, discretizing the thermodynamic equation to generate a discrete equation, and converting the discrete equation into a linear equation system in matrix form; solving the linear equation system using the temperature of each sensor-detected point as a boundary condition, and calibrating the temperature based on the solution results. This method reduces cost and improves accuracy, making it particularly suitable for granaries.

The invention belongs to the technical field of power system fault prediction, and particularly relates to an anticipated fault analysis chart parallel computing method and system.The method includes the steps that when parallel computing is conducted on a power system anticipated fault analysis chart, a power system topological structure chart is established firstly; establishing a power flow model of the power system under the expected fault according to the power topological graph; converting the power flow model into a matrix form to form a matrix form equation set; and finally, carrying out parallel calculation on the matrix form equation set. According to the scheme, rapid analysis under the anticipated fault of the power system can be realized, and the practicability of static analysis of the power grid is improved.

The invention provides a multi-omics graph representation learning method and device based on a multi-head self-attention mechanism, and the method comprises the steps: carrying out the preprocessing of multi-omics original data, converting the data into a matrix form, and constructing or obtaining a multi-source priori knowledge network according to the existing priori knowledge; encoding the priori knowledge network by using a graph neural network to obtain a feature-level embedding representation; inputting the multi-omics feature matrix into a multi-head self-attention mechanism, capturing potential association inside each group of omics data and among different omics data, and performing dimension reduction processing on original high-dimensional data to obtain low-dimensional sample-level embedding representation; splicing and fusing feature-level embedding and sample-level embedding to obtain unified representation of comprehensive data features and prior knowledge; and finally, inputting the fused representation into a task specific module, performing supervised optimization by combining a cross entropy loss function of a real label, and learning model parameters. In the application stage, a classification result can be output only by inputting the to-be-analyzed multi-omics data into the trained model.

This invention relates to the field of traffic safety technology, specifically to an efficient simulation method for quantum systems based on parallel order reduction. The method includes the following steps: receiving the physical parameters of the quantum system and simulation requirements, whereby the physical parameters include electron mass, potential well size, and initial wave packet parameters; and the simulation requirements include simulation physical time and accuracy requirements. Based on the physical parameters, a time-dependent Schrödinger equation describing the dynamic behavior of the quantum system is established, and the Schrödinger equation is rearranged into matrix form using the Kronecker product. This invention utilizes Arnoldi to reduce the order of the matrix-form Schrödinger equation. By reducing the order, the large matrix in the original space is projected into a relatively small subspace, improving the efficiency of the simulation solution. The reduced-order Schrödinger equation is solved using a time-parallel algorithm. This approach overcomes the time step limitation imposed by the CFL condition, ensuring that a stable solution can be obtained with fewer time steps.

The application provides a small spherical crown characteristic parameter extraction method for precision instrument lenses, comprising: obtaining small spherical crown three-dimensional space data through laser wave number scanning and shift axis Fizeau interference measurement, considering the distortion of actual data, linearizing the spherical crown mathematical model and mapping it into a matrix form through the characteristic mathematical formula of a space sphere, finally using multiple linear regression to fit the small spherical crown curvature and the spherical crown center, and checking the goodness of fit. Since uncontrollable deformation is generated when the actual small spherical crown is collected through a special camera system, the axial distortion error model is considered, and a scalar least square formula is established. In order to facilitate subsequent fitting and calculation, the scalar least square formula needs to be separated through known quantities and unknown quantities, and the matrix form of the specific linear equation is constructed. Multiple linear regression is performed for fitting, and the accurate spherical center position, distortion coefficient and spherical crown curvature of the small spherical crown are obtained.

The invention relates to the technical field of intelligent power grids, in particular to a power transmission and transformation grid monitoring method and system based on the Internet of Things, and the method comprises the steps: representing the noise complexity of an initial data matrix through the reconstruction difference evaluation of the initial data matrix formed by monitoring data vectors, thereby determining the number of denoising iterations, and improving the denoising precision. According to the method and the system, the target data matrix is subjected to fault identification, iteration denoising is carried out through the denoising model, the expected denoising effect can be achieved with the number of denoising iterations as small as possible, and therefore on the premise that the reliability of the target data matrix is guaranteed, that is, on the premise that the accuracy of power grid monitoring is guaranteed, denoising time consumption is reduced, and in addition, fault identification is carried out in a matrix mode. The power grid early warning information is determined according to the obtained fault identification probability vector, and compared with a time sequence prediction model iterative prediction mode in the prior art, time consumed by iterative prediction does not need to be introduced, and the timeliness of power grid monitoring is effectively improved.

The invention relates to a hexagonal and rectangular grid data unified storage and calculation method and system, specifically, hexagonal grid data and rectangular grid data are uniformly stored in a matrix form, if the hexagonal grid data are stored in the matrix, a first row pixel indentation identifier of the matrix is specified, and if the matrix is not stored in the matrix, a second row pixel indentation identifier of the matrix is specified; recording the storage format of the hexagonal grid data matrix as an indentation format; reading image data stored in a matrix form, if the image data is hexagonal grid data, for each step in the workflow, if the image data is a geometric processing task and the output of the previous step is in an indented format, converting the indented format into a double optimization coordinate format, if the image data is a statistical task and the output of the previous step is in the double optimization coordinate format, and if the image data is not hexagonal grid data, converting the indented format into a double optimization coordinate format; and if so, converting the double optimization coordinate format into an indentation format.

The embodiment of the invention discloses an FP8 matrix multiplication and addition operation circuit based on an RISC-V framework, and relates to the technical field of processor microstructures and artificial intelligence calculation acceleration. The circuit comprises an input module used for receiving a first operand in an FP8 format, a second operand and a third operand in a preset format; the matrix multiplication and addition operation module is used for periodically carrying out row and column multiplication operation on a plurality of first floating point elements of a preset row number in the first operand and a second floating point element in the second operand in a matrix form; the accumulation operation module is used for performing accumulation calculation on the intermediate result and a third operand; the output module is used for outputting a target operand, the problem that in the prior art, FP8 matrix multiplication and addition lacks special hardware support is solved, the computing power density and the energy efficiency ratio in an AI scene are remarkably improved, high fusion with an RISC-V vector architecture can be achieved, and the method is suitable for a CPU, an AI accelerator and a heterogeneous computing platform.

The application provides a multi-omics graph representation learning method and device based on a multi-head self-attention mechanism, pre-processes multi-omics original data, converts the multi-omics original data into a matrix form, constructs or acquires a multi-source priori knowledge network according to existing priori knowledge, encodes the priori knowledge network by using a graph neural network to obtain a feature-level embedding representation, inputs the multi-omics feature matrix into the multi-head self-attention mechanism to capture potential correlations between different multi-omics data and between different multi-omics data, and performs dimension reduction processing on the original high-dimensional data to obtain a low-dimensional sample-level embedding representation, splices and fuses the feature-level embedding and the sample-level embedding to obtain a unified representation of comprehensive data features and priori knowledge, and finally inputs the fused representation into a task-specific module, performs supervised optimization by combining a cross-entropy loss function of a real label, and learns model parameters. In the application stage, only the multi-omics data to be analyzed needs to be input into the trained model, and a classification result can be output.

The application belongs to the field of model-based system engineering application and practice, and relates to a method for quantitatively analyzing complexity of an architecture model of an engineering complex system. A plurality of alternative system architecture models are designed in advance, and then an architecture model to be evaluated is selected. The system architecture model is abstracted according to graph theory to obtain a graph representation of the architecture model, and the complexity of each vertex in the graph is determined (1). Then, the graph representation of the architecture model is converted into a matrix form of the architecture model, and the complexity of each connection relationship in the matrix is determined (2). Then, the topological complexity of the matrix is determined (3). Then, the complexity of the architecture model is calculated according to (1), (2) and (3), and the complexity values of all the alternative architecture models are calculated until the complexity values of different architecture models are compared to evaluate, compare and optimize the architecture model of the system or product. The selected architecture model has the lowest complexity and the lowest implementation difficulty, and the calculation method is convenient for engineering implementation.

The embodiment of the invention discloses a matrix multiplication and addition operation circuit based on an RISC-V framework, and belongs to the field of processor microstructures and artificial intelligence accelerated calculation. The circuit comprises an input module, a matrix multiplication and addition operation module, an accumulation operation module and an output module, wherein the input module is used for inputting operands and carrying out precision identification on floating point elements; the matrix multiplication and addition operation module is used for carrying out row and column multiplication operation on a first floating point element of a preset row number in the first operand and a second floating point element in the second operand in a matrix form on the basis of the identification information to obtain a fourth operand; the accumulation operation module accumulates the fourth operand and the third operand to obtain a fifth operand; and the output module outputs a target operand based on the fifth operand, thereby realizing compatibility of FP8 and BP16 matrix multiply-add operation in a single circuit, improving hardware reuse rate, reducing control complexity and power consumption, and being adaptive to an RISC-V vector expansion instruction set.