31 results about "LU decomposition" patented technology

The invention relates to a hierarchical iterative parallel computing method and device for a load flow sparse equation of multiple leading edge surfaces, and the method comprises the steps: distributing different numbers of computing resources for each branch and each node of a Jacobian matrix parallel hierarchical elimination tree according to the structural characteristics of the Jacobian matrix parallel hierarchical elimination tree; according to the allocated computing resources, taking a branch subordinate to a root node of the Jacobian matrix parallel hierarchical elimination tree as a unit, and carrying out sub-matrix computing on nodes in parallel from a lower-layer node to an upper-layer node; storing a Jacobian sub-matrix obtained by sub-matrix calculation, and storing an LU decomposition and contribution sub-matrix of a node dense sub-matrix obtained by LU decomposition of the sub-matrix; and superposing the contribution sub-matrix calculated by the lower-layer node to the corresponding node of the upper layer, and synchronously updating the data in the Jacobian sub-matrix. The power flow equation matrix decomposition efficiency can be improved, and the power flow calculation effectiveness is improved.

The invention discloses an aircraft simulation method, device and equipment based on parallel block sparse matrix incomplete LU decomposition and a medium, and relates to the technical field of simulation, and the method comprises the steps: building a block sparse matrix comprising a plurality of zero sub-blocks and non-zero sub-blocks based on the topological relation of each grid point in an aircraft grid and simulation physical parameters, determining dependency pairs respectively corresponding to a plurality of non-zero sub-blocks from the non-zero sub-blocks, and storing the dependency pairs by adopting a preset data structure; wherein the dependent pair comprises a pair of matched non-zero sub-blocks; respectively distributing the plurality of non-zero sub-blocks to a plurality of GPU thread bundles for parallel processing so as to update the corresponding non-zero sub-blocks based on the updated dependency pairs after the GPU thread bundles complete updating on the dependency pairs corresponding to the corresponding non-zero sub-blocks; and after the plurality of non-zero sub-blocks are updated, updating the block sparse matrix based on the plurality of updated non-zero sub-blocks, and determining an incomplete LU decomposition result according to the updated block sparse matrix so as to perform aircraft simulation.

The application belongs to the technical field of power systems, and discloses an EIGD time-delay power system stability analysis method and device, wherein the method comprises the following steps: a linear model of the time-delay power system is established, and the differential equation in the model is converted into an abstract Cauchy problem by using an infinitesimal generator; a group of discrete points is selected for each time-delay interval of the time-delay power system, a discrete function space is established according to the discrete points, and the time-delay variable is discretized to generate a low-order partial infinitesimal generator discretization matrix; displacement and inverse transformation are performed on the discretization matrix to obtain an inverse matrix, and the characteristic value of the system is obtained according to the inverse matrix; the characteristic value is checked by using the Newton method to obtain the accurate characteristic value and the characteristic vector, and the stability of the time-delay power system is analyzed and judged. The method solves the problem of large matrix LU decomposition calculation amount in the existing EIGD time-delay power system characteristic value sparse calculation method based on the DDAE model.

This invention relates to a hierarchical iterative parallel computation method and apparatus for sparse power flow equations with multiple front edges. The method includes: allocating different amounts of computational resources to each branch and node of the parallel hierarchical elimination tree of the Jacobian matrix based on its structural characteristics; performing submatrix computation on nodes in parallel from lower-level nodes to upper-level nodes, taking the branches under the root node of the parallel hierarchical elimination tree as units; storing the Jacobian submatrices obtained from the submatrix computation, and storing the LU decomposition of the node dense submatrices and contribution submatrices obtained from the submatrix LU decomposition; superimposing the contribution submatrices calculated by the lower-level nodes onto the corresponding nodes in the upper level, and synchronously updating the data in the Jacobian submatrices. This invention can accelerate the efficiency of power flow equation matrix decomposition and improve the effectiveness of power flow computation.

The invention discloses a multi-industrial equipment-oriented ELM fault identification method fusing feature attention and manual LU decomposition, and belongs to the technical field of industrial equipment fault diagnosis. The core of the method is that a feature attention mechanism and a manual LU decomposition module are fused: firstly, the feature contribution degree is quantified through a channel attention mechanism, key fault features are automatically focused, weak correlation feature interference is inhibited, and the quality of a hidden layer output matrix is improved; secondly, the manual LU decomposition module realizes autonomous controllability and numerical stability of matrix inverse operation through row principal component selection and regularization optimization, and is separated from the dependence of a third-party math library. According to the method, the accuracy of equipment fault identification is remarkably improved, the omission ratio and the false alarm rate are effectively reduced, memory limitation of edge equipment is overcome, and an efficient and reliable real-time identification scheme is provided for bearing abrasion and gearbox abnormity of a wind power variable pitch system and faults of other industrial equipment such as a machine tool and an industrial motor.

This invention discloses a method and system for extracting S-parameters of microwave passive devices, comprising: geometrically modeling the microwave passive device and then meshing it to obtain a processed device model; discretizing the processed device model according to basis functions, and performing mode function expansion on the ports of the processed device model to establish a transfinite element matrix equation; performing partial LU decomposition of the system matrix in the transfinite element matrix equation using the multi-wavefront method to obtain the Schur complement matrix of the system matrix and the decomposed system matrix; obtaining the dense S-parameter matrix equation for the microwave passive device based on the Schur complement matrix of the system matrix and the decomposed system matrix; and solving the dense S-parameter matrix equation to obtain the S-parameters of the microwave passive device. This invention effectively reduces the memory usage of matrix calculations and reduces computation time.

The present invention provides a processor for matrix inversion and matrix multiplication, comprising a processing unit (PU) array, a divider array, a memory, and a finite state machine; the processing unit (PU) array is used to perform matrix multiplication and matrix inversion operations; the processing unit (PU) array includes a systolic array. The systolic array in the present invention can implement both matrix multiplication and LU decomposition of matrices and inversion of upper and lower triangular matrices, thereby implementing arbitrary matrix inversion, rather than using two different hardware resources to implement matrix multiplication and matrix inversion separately, thereby significantly improving hardware utilization.

The present invention combines the advantages of fast running speed of spatial domain digital watermarking algorithm and high robustness of frequency domain digital watermarking algorithm, and discloses a spatial domain color digital image blind watermarking method integrating LU decomposition. According to the correlation of the elements of the lower triangular matrix obtained after LU decomposition, the present invention obtains the high correlation elements of the lower triangular matrix obtained after LU decomposition of the image block in the spatial domain, and uses these elements to complete the embedding and blind extraction of the digital watermark in the spatial domain without the need for real LU decomposition. The invention can embed the color image digital watermark into the color host image, and not only has good watermark concealment and strong robustness, but also has good real-time performance, which solves the problem of slow running speed of large-capacity color image digital watermark, and is suitable for occasions of fast and efficient digital media copyright protection.

The invention discloses a linear pre-correction method, device and equipment, a medium and a program product, and the method comprises the steps: obtaining a cross-correlation matrix of a primary function matrix based on the primary function matrix of a current input signal, and judging whether the primary function matrix meets a preset orthogonalization processing condition or not; when the primary function matrix meets the orthogonalization processing condition, performing LU decomposition on the primary function matrix based on a coarse-grained reconfigurable computing architecture to obtain a decomposition matrix; performing orthogonalization processing on the primary function matrix by using the decomposition matrix to obtain a primary function orthogonal matrix; and on the basis of the primary function orthogonal matrix, a correction parameter used for linearization pre-correction is obtained by using a minimum root mean square algorithm. According to the method, the processing efficiency of orthogonalization of the primary function matrix can be improved by utilizing the coarse-grained reconfigurable computing architecture, so that the real-time performance of linear pre-correction in a large-bandwidth or large-scale MIMO scene can be ensured, and the scene limitation is relatively small.

This application discloses a sparse LU decomposition solution method and related apparatus based on matrix computation unit optimization, belonging to the field of numerical algorithm optimization technology. The method includes: reordering the sparse matrix using an improved PRCM algorithm to increase data local density, obtaining a bandwidth-compressed sparse matrix; then converting it into a regular, dense tile block structure and storing it in pre-allocated contiguous memory space; and finally, employing a right-look-based LU decomposition algorithm to numerically decompose the bandwidth-compressed sparse matrix. During the numerical decomposition process, a matrix computation unit is used for acceleration to obtain the LU matrix, which can then be used to efficiently solve the linear equation. This application's solution fully utilizes the spatial locality of the sparse matrix while leveraging the powerful computational capabilities of the matrix computation unit, reducing computational load while improving computational efficiency, significantly reducing the time spent on symbolic and numerical decomposition stages of sparse LU decomposition.

The invention discloses a data processing method and device and related equipment, which can perform LU decomposition of R columns of sub-matrixes on a to-be-decomposed matrix needing to be decomposed in each iteration process in a process of performing LU decomposition calculation on applied to-be-processed data, namely a dense matrix, and calculate a matrix multiplication taking a product of R and sub-matrix dimensions as a dimension. The dimension of matrix multiplication calculation is large, and efficient calculation of matrix multiplication can be achieved. In this way, the dense matrix can be divided into the sub-matrixes with the small granularity, it is guaranteed that the calculation efficiency of matrix multiplication is high on the premise that process loads are balanced, and then the overall efficiency of matrix decomposition calculation is improved.

The application particularly relates to a linear equation set solving method for semiconductor device simulation, which comprises the following steps: determining a target linear equation set, and performing nested partitioning and reordering on a coefficient matrix of the target linear equation set to obtain a transformed coefficient matrix and a corresponding right end vector; based on a preset decomposition precision, the transformed coefficient matrix is decomposed through a selected main element LU decomposition, a forward substitution algorithm and a back substitution algorithm to obtain a current solution vector, and based on a preset working precision, when an initial residual vector does not satisfy a preset solving condition, a modified vector is obtained through a preconditioned GMRES algorithm; based on the preset working precision, the current solution vector and the modified vector, an updated solution vector is obtained as a current solution vector, and a step of calculating a residual vector is re-executed until a latest residual vector after iteration satisfies the preset solving condition, and a final solution is obtained based on the latest current solution vector. Thus, the problems of low solving precision and slow speed in semiconductor device simulation are solved.

The application discloses a ground penetrating radar time domain finite element fast simulation method, system and equipment. The method comprises the following steps: setting geometric parameters and electrical parameters of a two-dimensional complex geoelectric model; constructing a weak form of a two-dimensional electromagnetic wave equation; forming a time domain finite element equation; establishing a linear equation group after central difference time discretization; giving a time step, judging whether the time step meets a CFL stability condition, if yes, directly solving the linear equation group, if not, performing LU decomposition on a coefficient matrix of the linear equation group, and disturbing unstable eigenvalues in the coefficient matrix until absolute values of all eigenvalues of the decomposed coefficient matrix are less than or equal to 1, and generating a reconstruction coefficient matrix; and solving the reconstructed linear equation group, so that the time step breaks through the CFL stability condition. The application can improve the simulation precision of the ground penetrating radar.

The present invention provides a high-quality imaging method for full matrix data based on frequency-domain reverse time migration. The excitation signal, sound velocity distribution, and full matrix data are respectively preprocessed to obtain an excitation sound source matrix, an impedance matrix, and a received sound source matrix. Using the above preprocessing results as inputs, LU decomposition processing is performed on the impedance matrix to solve for the transmitted wave field and the received wave field. Finally, an imaging result is obtained using an imaging condition. This imaging method utilizes the characteristic that the impedance matrix is fixed at the same frequency. Through LU decomposition, the sound pressure fields generated by multiple excitations at the same frequency can be obtained at one time, greatly saving calculation time and improving calculation efficiency. Based on the full wave equation for wave field forward modeling, there is little simplification of wave field propagation, which is suitable for measuring defects in any complex sound velocity distribution and can ensure high quality of the obtained imaging result.

The application discloses a kind of AC / DC hybrid distribution network power flow calculation method and device considering incomplete LU decomposition preconditioning, and relates to the field of smart grid.The method comprises the following steps: establishing active / reactive control equation based on VSC steady-state equation;Based on Kirchhoff's current law, establish AC / DC node power imbalance equation;Taylor expansion is carried out on the above equation, and high-order terms are ignored, to obtain linearized unified power flow equation set and matrix form Jacobian matrix;Incomplete LU decomposition preconditioning is carried out on Jacobian matrix, and stable double-conjugate gradient method is used to solve, to obtain power flow solution.The application establishes unified linearized power flow model, combines incomplete LU decomposition preconditioning and stable double-conjugate gradient method, effectively reduces the sensitivity to initial value, improves the robustness and convergence speed of complex AC / DC hybrid system power flow calculation.

The present application belongs to the technical field of sparse matrix decomposition, and relates to a block method of performance-aware block sparse LU decomposition for a GPU platform, comprising the following steps: S1, constructing a performance function of a GPU computing kernel based on statistical data; S2, based on the performance function, designing a decision method for deciding whether to block a sparse matrix at a specified position; and S3, decomposing the overall block problem of a large-scale sparse matrix into multiple sub-problems decided in sequence, solving each sub-problem one by one based on a greedy heuristic strategy and using the decision method, and obtaining a final specific block scheme. Compared with existing methods, the method of the present application is more excellent in performance in the numerical decomposition stage, and is not only suitable for a heterogeneous platform of multi-core CPU and GPU, but also suitable for a pure CPU architecture.

The application provides a method for solving dynamic voltage drop of a chip-package based on LU decomposition combined with an SMW formula. The chip-package power supply network is regarded as a composite system connected by an interconnection node Pad between a package part and a plurality of chip sub-networks. A chip-package partition solving strategy is introduced. According to the chip and package characteristics, a large-scale sparse matrix of the composite system is disassembled into a structural model with a plurality of diagonal sub-blocks, so as to convert the chip-package power supply network equation into a block structure, and avoid the inefficient redundancy of full matrix repeated decomposition by focusing on necessary coupling calculation. Meanwhile, the SMW formula is used for low-rank incremental updating of the package network influence, so as to quickly update the influence introduced by the package and quickly correct the solution vector. The calculation task is split into a plurality of sub-problems for synchronous execution, so that the plurality of sub-blocks are solved synchronously. The application can provide a high-precision and high-efficiency solution for dynamic voltage drop analysis of a large-scale chip package power supply network.

The present invention provides a method and a related device for calculating the solution time of a node voltage equation in an electromagnetic transient simulation. The method determines the number of non-zero elements and the number of zero elements in each column of an equivalent conductance matrix of the node voltage equation in the electromagnetic transient simulation; solves the element substitution time of the equivalent conductance matrix based on the number of non-zero elements in each column of the equivalent conductance matrix; solves the element solution time of the equivalent conductance matrix and the first matrix solution time of the equivalent conductance matrix after substitution; solves the second matrix solution time based on LU decomposition of the equivalent conductance matrix of the node voltage equation in the electromagnetic transient simulation; solves the order threshold of the equivalent conductance matrix based on the first matrix solution time and the second matrix solution time, selects a solution method with a shorter matrix solution time based on the order of the equivalent conductance matrix, and obtains the corresponding solution time. The present invention provides the matrix solution time of the equivalent conductivity matrix of the node voltage equation in electromagnetic transient simulation and the solution time of the equivalent conductivity matrix based on LU decomposition, which can quantify the computational performance of the node voltage equation and theoretically quantify the performance advantages and disadvantages of different solution methods.

The invention discloses a reconfigurable matrix operation circuit and system, and belongs to the field of high-performance computing hardware accelerators, and the circuit comprises a first operation input end which is used for receiving matrix data; the first operation output end is used for outputting a matrix operation result; the system comprises M cascaded processing units, an independent operation part, a result collection unit and a module top layer control logic, the first processing unit in the M processing units is connected with a first operation input end, the Mth processing unit is connected with a first operation output end, and each processing unit comprises an internal state machine and an operation core; the module top layer control logic is used for setting an internal state machine in the processing unit to realize Cholesky decomposition, LU decomposition, QR decomposition, lower triangular matrix inversion and matrix transposition multiplication operation of a matrix; the result collection unit is used for collecting operation results; and the independent operation component is used for realizing rooting operation and division operation. By means of the reconfigurable matrix operation circuit, hardware multiplexing is achieved, and resources are saved.

A calculation method for the solving time of a node voltage equation in electromagnetic transient simulation, and a related apparatus. The method comprises: for an equivalent conductance matrix of a node voltage equation in electromagnetic transient simulation, determining the number of non-zero elements and the number of zero elements in each column of the equivalent conductance matrix; on the basis of the number of non-zero elements in each column of the equivalent conductance matrix, calculating the element permutation time of the equivalent conductance matrix; for the equivalent conductance matrix which has been subjected to permutation, calculating the element solving time of the equivalent conductance matrix and a first matrix solving time; for the equivalent conductance matrix of the node voltage equation in electromagnetic transient simulation, calculating a second matrix solving time based on LU decomposition; and on the basis of the first matrix solving time and the second matrix solving time, calculating an order threshold value of the equivalent conductance matrix, and on the basis of the order of the equivalent conductance matrix, selecting a solving method involving less matrix solving time, and obtaining a corresponding solving time. The present invention provides the matrix solving time of the equivalent conductance matrix of the node voltage equation in electromagnetic transient simulation, and an equivalent conductance matrix solving time based on LU decomposition, which can quantify the computational performance of the node voltage equation and theoretically quantify the performance advantages and disadvantages of different solving methods.

The application provides a kind of casting billet online thermal state tracking method, comprising the following steps: according to the rule of block division, the asymmetric profiled blank is divided into area block, and the casting billet center area block and the casting billet surface area block are obtained respectively;The structured coarse grid division is carried out to the casting billet center area block, and the structured fine grid division is carried out to the casting billet surface area block, to obtain a plurality of structured grids, and the grid nodes of the structured grid are indexed and described, and the grid data of each structured block and the index are stored correspondingly;The heat transfer equation coefficient matrix and the solidification latent heat release source term of the grid unit of each structured block are calculated;The temperature field of the grid unit in the structured block is calculated by coupling the incomplete matrix LU decomposition and SIP algorithm in turn. By using the application, the problems of the prior art, such as unsuitable for asymmetric profiled blank and slab dual-purpose casting machine, reduced system reliability and time step limitation, can be solved.

The invention discloses a GPU acceleration LU algorithm of a CSC format sparse matrix in a power system. The GPU acceleration LU algorithm comprises the following steps: compressing and storing an input sparse matrix into a CSC format according to a column priority; based on the stored CSC format, realizing an LU decomposition numerical algorithm suitable for a sparse matrix; parallel parts in the LU decomposition numerical algorithm are realized on a GPU, and parallel design is carried out; and timing the solution. According to the method, by changing the matrix format, in the actual high-dimensional power grid matrix, the parallel LU decomposition efficiency of the power sparse matrix is improved, and the problem that time consumption is large in time domain simulation of power grid nodes in a power system is solved.

The invention discloses a hybrid multi-layer preprocessing method and system for an ill-conditioned power flow system, and mainly relates to the technical field of data calculation of a power system. Comprising the following steps: obtaining an original Jacobian matrix J based on a power flow calculation method of a power system, and executing l1 norm scaling on the original Jacobian matrix to obtain a scaled matrix; performing approximate minimum sorting on the scaling matrix to obtain a rearranged matrix; performing incomplete LU decomposition on the rearrangement matrix to obtain a decomposition matrix; and performing hybrid precondition device integration according to the scaling matrix, the rearrangement matrix and the decomposition matrix. The method has the beneficial effects that the risk of convergence failure is reduced, the decomposition efficiency is improved, and meanwhile, the precision and the calculation efficiency are balanced.

To accelerate a calculation speed in the structure analysis of a building by parallel processing by a plurality of information processors.SOLUTION: A parallel processing system including a plurality of information processing devices 210 sets a load vector = an entire rigidity matrix * a displacement vector (an analysis target expression) for a building to be analyzed by a finite element method. The load vector and the total stiffness matrix are known, and the displacement vector is unknown. The whole rigidity matrix is divided into N pieces, and an individual information processor is allocated to processing related to each matrix. The LU decomposition process is applied to each of the N matrices in parallel. Next, the forward substitution process is sequentially performed on the divided N matrices from the top to the bottom. When the processing of the lowermost matrix is completed, the backward substitution processing is sequentially performed on the N divided matrices from the bottom to the top. When the results of all the backward substitution processes are integrated, the value of the displacement vector which is initially unknown is calculated.SELECTED DRAWING: Figure 1

The present application provides a data processing method and apparatus for parallel LU direct solution oriented to SIMD. The method includes: obtaining data to be processed, including 2m pattern matrices, each pattern matrix having the same non-zero element distribution pattern; performing a real-number processing on each pattern matrix to obtain a corresponding real-number matrix for each pattern matrix; performing a permutation on the real-number matrix corresponding to each pattern matrix to obtain a corresponding transformation matrix and a corresponding permutation matrix, where the transformation matrix is a matrix obtained after the real-number matrix is transformed, and the permutation matrix is a matrix reflecting the transformation process of the real-number matrix; performing a vectorization process based on the transformation matrices and permutation matrices corresponding to the 2m pattern matrices to obtain a one-dimensional array for storage; using the SIMD extension component to read the vectorized one-dimensional array, performing parallel LU decomposition and direct solution, and outputting the direct solution result. Thus, parallel linear solution is performed through the SIMD extension component to achieve accelerated solution.

The application discloses an aircraft simulation method and device based on parallel block sparse matrix incomplete LU decomposition, equipment and medium, relates to simulation technical field, including: based on each grid point in aircraft grid topology relation and simulation physical parameter constructs and contains several zero sub-block and non-zero sub-block block sparse matrix, and determines several non-zero sub-block corresponding dependent pair and stores by using preset data structure from it;Wherein the dependent pair includes a pair of matched non-zero sub-block;Several non-zero sub-block is respectively distributed to several GPU thread bundles and is handled in parallel, to utilize GPU thread bundle in corresponding non-zero sub-block dependent pair after completing update, based on the dependent pair of completing update, the corresponding non-zero sub-block is updated;After several non-zero sub-block all complete update, based on the several non-zero sub-block of completing update, the block sparse matrix is updated, and the incomplete LU decomposition result is determined according to the updated block sparse matrix to carry out aircraft simulation.

The invention provides a method for rapidly solving intensity directivity of a low-frequency sound target. The method comprises the following steps: (1) constructing a geometric model of an underwater structure and carrying out grid division on the geometric model; (2) establishing a linear equation set of the structure surface by using a finite element method, and solving the equation set by using an LU decomposition method to obtain a sound pressure level of the structure surface; (3) defining and initializing a matrix used for storing field point information according to structure surface nodes, discretizing a Helmholtz integral equation by using numerical integration, and constructing and storing as a field point matrix; (4) calculating the field point sound pressure based on the results of the structure surface sound pressure level and the field point matrix, and obtaining the full-space sound target intensity of the underwater structure; and (5) ensuring that the geometric model can realize sound target intensity directivity real-time calculation of the structure at different incident angles and excitation source positions under different incident angles and excitation source positions.

The present invention belongs to the technical field of electromagnetic exploration and numerical simulation. A regional decomposition magnetotelluric three-dimensional forward modeling method and system are provided, which introduces overlapping DDM to decompose the large-scale MT problem into multiple smaller subdomains, thereby reducing the computational complexity; a direct solver is used to solve each small linear system in the subdomain, which not only avoids the convergence problem of the iterative solver, but also reuses the LU decomposition results in the Schwarz iteration process to speed up the calculation speed; a multi-grid scheme is integrated to accelerate convergence, and good initial conditions are provided through coarse grid solutions, thereby reducing the number of iterations and further improving the computational efficiency; the use of unstructured grid technology enhances the adaptability of complex geological structures, thereby improving the simulation accuracy; the present invention develops an airborne magnetotelluric three-dimensional forward modeling technology with fast computing speed and low memory requirements, which provides a basis for theoretical research on airborne magnetotelluric and three-dimensional data inversion.

The invention specifically discloses a cutting temperature prediction model calculation method based on a polar coordinate grid, and relates to the technical field of metal cutting machining. The method comprises the following steps: firstly, obtaining cutting force, back force and cutter chip contact length through a cutting experiment; secondly, boundary processing is carried out, and the heat flux and the convective heat transfer coefficient of the rake face position are calculated; secondly, using a polar coordinate grid to divide a region to obtain a control body, establishing a heat balance equation of the control body and constructing an equation set, and solving the equation set through an LU decomposition and chasing method; and finally, carrying out iterative calculation until a set time threshold value is reached, and outputting a final temperature field value. The method adapts to the complex geometric structure and cooling condition of the tool nose, can be deployed on edge computing equipment, provides real-time data support for cutting process parameter optimization, and effectively prolongs the service life of the tool and improves the machining quality of the tool.

The invention provides a banded matrix LU decomposition fusion optimization method, which comprises the following steps of: firstly, generating a block division strategy based on a Right-looking structure, strictly limiting the calculation range of each round of iteration in a banded area, and processing data outside bandwidth through an auxiliary buffer area; then SCAL and GER operations in Panel decomposition are fused to a single kernel function, after SCAL is completed by a first thread block, a plurality of thread blocks execute GER in parallel, and an intermediate result is multiplexed through a register to reduce global memory access; and finally, based on the strategy and the kernel assembly execution process, dynamically determining the size of the blocks, sequentially calling the fused kernel and the optimized matrix to update the kernel in each round of iteration, and advancing along a main diagonal until decomposition is completed. The invention also provides a banded matrix LU decomposition fusion optimization system. In this way, the calculation parallelism, the data locality and the memory access efficiency are effectively improved, and strip matrix LU decomposition is remarkably accelerated.