14 results about "Krylov subspace" patented technology

The application relates to the technical field of power electronics, and discloses a press-contact type IGBT device packaging structure optimization method and system. The method comprises the following steps: constructing an equivalent model of a press-contact type IGBT device; constructing a differential algebraic equation for uniformly describing the equivalent model of the press-contact type IGBT device and circuit port variables; converting the differential algebraic equation into a nonlinear algebraic equation group through time domain discretization, and constructing an iteration matrix with a global state vector as a derivation object according to a solving mode of the nonlinear algebraic equation group; performing low-dimensional subspace projection on the iteration matrix based on a Krylov subspace, and combining residual minimization to solve a linear subproblem of the iteration matrix, so that current uniformity characteristic results of the press-contact type IGBT device are obtained; and the packaging structure of the press-contact type IGBT device is optimized based on current density distribution reflected by the current uniformity characteristic results. The application can improve the current uniformity and operation reliability of the system.

The present application relates to the field of electromagnetic numerical calculation, and discloses a kind of method for solving electromagnetic scattering based on adaptive block Krylov subspace base function, method includes the following steps: step one: target region is carried out adaptive region decomposition, and target subdomain is obtained;Step two: target subdomain is expanded, and extended subdomain is formed;Step three: using optimized extended subdomain, the solving calculation of electromagnetic scattering is carried out.The main advantage of the present application is that the efficiency of constructing Krylov subspace base function is higher by using adaptive block technology, which significantly reduces the generation time of Krylov subspace base function based on block, optimizes the block expansion of subdomain, improves the solving calculation time, uses clustering algorithm to carry out region decomposition on the calculation target, and uses the average distance of clustering data points to cluster center to expand each subdomain to ensure the continuity of current, which not only ensures the calculation accuracy, but also improves the construction efficiency of base function, significantly reduces the generation time of Krylov subspace base function based on block.

The application discloses a quantum linear solving method and device based on quantum adiabatic linear algorithm, and a medium, and relates to the technical field of quantum computing. The method comprises the following steps: obtaining a linear system to be processed, and constructing a Krylov subspace matched with the linear system; determining a subspace equation set of the linear system based on the Krylov subspace according to the linear system and the Krylov subspace; constructing a quantum circuit corresponding to the quantum adiabatic linear algorithm according to the subspace equation set, and solving an approximate solution of the linear system in the Krylov subspace according to the quantum circuit. The application can combine the subspace method with the quantum adiabatic linear algorithm, perform dimension reduction processing on the original high-dimensional linear equation set in the quantum adiabatic linear algorithm through the subspace method, so that the dimension of the linear equation set actually solved in the iteration process is smaller than the dimension of the original linear equation set. Therefore, the demand of the quantum discrete adiabatic linear algorithm for computing resources is reduced, and the quantum discrete adiabatic linear algorithm can be implemented in an NISQ chip and a classical computer.

The application discloses a material deformation displacement calculation method and a calculation system. The method comprises the following steps: in response to obtained design parameters, a corresponding structure and a geometric model are constructed, and a stiffness matrix K and a load matrix F are generated; the stiffness matrix K is stored in a sparse form by row compression, the load matrix F is input, and a Krylov subspace dimension m, an initial value U0 and an allowable error ε are selected; a set of standard orthogonal bases of the Krylov subspace is constructed; the best approximation U of the displacement in the Krylov subspace is calculated m ; the residual ||F-KU m || is calculated, and if the residual is less than the allowable error ε, the best approximation U m is output as the material deformation displacement; if the residual is greater than the allowable error ε, the best approximation U m is taken as a new initial value, the Krylov subspace is updated, and the material deformation displacement is recalculated. The application provides a stiffness calculation method which can be flexibly deployed and efficiently parallel, and realizes acceleration of finite element analysis.

This invention discloses a method and apparatus for intelligent task navigation and multi-dimensional collaborative management, applicable to the optimization and execution control of complex workflows, such as hospital information systems and intelligent systems. The method achieves personalized task generation, multi-role collaborative task allocation, and dynamic monitoring and optimization in complex scenarios by constructing a multi-level knowledge graph, applying the fully discrete finite element method, the nonlinear GCR-type Krylov subspace method, and the unfitted Green's function method.

This invention discloses a deep learning-based engine condition monitoring method, comprising the following steps: collecting multi-source sensor data and corresponding speed data during engine operation; preprocessing the original multi-source data to obtain preprocessed multi-source data; performing condition-locked resampling to obtain a set of window sequences; constructing a Krylov subspace and performing orthogonalization to obtain an orthogonal basis for the subspace; performing Krylov subspace change point and state transition detection to obtain the current monitoring state; generating a subspace fingerprint and selecting a deep learning monitoring branch based on the subspace fingerprint routing structure to obtain a health index and anomaly score; writing the health index and anomaly score into a candidate alarm queue, switching the current monitoring state, and outputting alarm results based on the candidate alarm queue. This invention improves the accuracy of engine condition monitoring by combining condition-locked resampling with Krylov subspace change point detection.

This invention discloses a method, apparatus, and medium for solving variable quantum linear problems in subspaces. The method includes: determining the system of linear equations to be solved, constructing a variable quantum circuit and a Krylov subspace, and using the generalized minimum residual method and the variable quantum circuit to calculate an approximate solution of the system of linear equations to be solved in the subspace. This addresses the shortcomings of existing technologies and can reduce the time complexity and computational load of solving linear problems.

The invention discloses a micro-grid linearization model order reduction method, system, equipment and medium, and belongs to the field of micro-grid stability analysis and control. Models of all elements in a micro-grid are obtained, a nonlinear differential algebraic equation set is established, a stable equilibrium point of the nonlinear differential algebraic equation set is solved, linearization processing is carried out, and a linearized differential algebraic equation set is obtained; solving a rightmost characteristic value by adopting a characteristic value algorithm, if the system is not stable, performing leftward translation transformation on the system to stabilize the system, generating a Lyapunov equation according to the translated system, solving a controllable and observable Gramb matrix by adopting a Krylov subspace algorithm, and performing order reduction on the translated system by adopting a balanced truncation method to obtain a stable system; according to the method, an existing balance truncation method is improved, the precision, the stability, the controllability and the observability of a system after order reduction are guaranteed, meanwhile, the Grubrum matrix can be efficiently solved, and the method is made to be suitable for an unstable system.

This invention belongs to the field of nuclear reactor physics computational technology, and specifically discloses a coarse-grid nodal global efficient solution method and system for multi-group steady-state neutron transport. It effectively integrates and leverages the advantages of the coarse-grid nodal method and the Newton-Krylov method, constructing a general coarse-grid nodal transport computation model applicable to various polyhedral nodals. It achieves high-precision and high-efficiency transport scan computation under sparse nodals, significantly reducing the number of variables and compressing the computational scale of a single transport scan. A Newton-Krylov subspace global solution framework is established. Global solution variables are selected based on the variable coupling relationship of the coarse-grid nodal method, and the Krylov subspace method is used to solve the linear equations corresponding to the Newton step size. The obtained Newton step size is used to update the global solution variables to achieve fast and efficient convergence. This invention significantly improves computational efficiency and reduces computational scale while ensuring computational accuracy, making it particularly suitable for large-scale, high-dominance neutron transport problems.

The invention provides a large language model parameter optimization method and system based on random GMRES, and is applied to the technical field of large language models, and the method comprises the steps: determining at least one to-be-optimized problem in a large language model, and converting the to-be-optimized problem into a least square problem; inputting the least square problem into the constructed randomized iteration solver for processing, and outputting a solution vector; the randomization iteration solver comprises a random AB-GMRES solver and a random BA-GMRES solver; the random iteration solver comprises a random AB-GMRES solver and a random A random AB-GMRES solver and a random BA-GMRES solver project a high-dimensional orthogonalization process in a Krylov subspace into a low-dimensional space constructed by a random subspace embedding matrix by introducing the random subspace embedding matrix to carry out approximate calculation; parameters in the large language model are updated or generation of a next reasoning token is propelled based on the solution vector. According to the invention, on the premise of ensuring the model performance, the training speed is greatly improved and the reasoning delay is reduced.

This invention relates to a radio frequency (RF) simulation method, apparatus, computer device, storage medium, and program product. The RF simulation method includes: constructing a first Krylov subspace based on a first scanning frequency; generating a second Krylov subspace corresponding to a second scanning frequency by multiplexing vectors in the first Krylov subspace; and obtaining the circuit response when the scanning frequency is the second scanning frequency based on the second Krylov subspace. This invention can improve the calculation speed and simulation efficiency of RF simulation.

The invention relates to the technical field of electromagnetic simulation, in particular to a rapid electromagnetic simulation method based on a model reduced-order time domain fine integration method, which comprises the following steps of: establishing a Maxwell rotation equation containing magnetic conductivity, dielectric coefficient, conductivity, magnetic permeability and time change source items, and defining a magnetic field vector function and an electric field vector function; according to the method, a model order reduction technology is deeply embedded into a core equation of a time domain fine integration method, a correction Gram-Schmidt orthogonalization Arnoldi process of a Krylov subspace is combined, an original high-dimensional coefficient matrix is reduced into a low-order upper Hessenberg matrix, a compact projection matrix is generated, and the purpose of greatly reducing the coefficient matrix and matrix index dimensions is achieved; the matrix index is calculated by adopting the dimension expansion PI technology, direct solving of an inverse matrix of an original high-dimensional matrix is avoided, and the effect of reducing the calculated amount of the non-homogeneous ordinary differential equation set is achieved.

Calculation control for hybrid computing of Hamiltonian eigensolutions may be provided by preparing a plurality of states in a Krylov subspace of a full Hamiltonian that represents a chemical system; sampling a plurality of basis states from each state of the plurality of states in the Krylov subspace to produce a plurality of sampled basis states; processing the plurality of sampled basis states to yield an analysis selection; constructing a subspace Hamiltonian from the analysis selection; computing, via an eigensolver, an eigensolution for the chemical system from the subspace Hamiltonian; and outputting the eigensolution for the chemical system.

The invention provides an electromagnetic scattering calculation method and system based on a layered Krylov subspace, and relates to the technical field of electromagnetic numerical calculation, and the method comprises the steps: carrying out the blocking processing of a target electromagnetic scattering region, and obtaining a plurality of blocks; constructing an initial Krylov subspace primary function of each block to obtain an initial primary function set; performing projection, compression and orthogonalization operation on the initial primary function set to obtain a final global primary function matrix; and constructing an induced current sparse model, and based on the induced current sparse model, combining with the final global basis function matrix to obtain reconstructed induced current. According to the method, an algebraic multi-scale decomposition framework is adopted, a compact primary function set with good orthogonality is recursively constructed, and the calculation efficiency and the memory utilization rate are remarkably improved while the precision is guaranteed.