Conjugate gradient modification method for identifying anti-geometric condition based on dual reciprocity boundary element method
By using power series expansion method and adaptive grid division in the integration process, combining precondition matrix and strong Wolfe conditions, the problems of low singular integral accuracy and local optimal solutions in the traditional method are solved, and the conjugate gradient is improved The solution efficiency and accuracy of the method.
Patent Information
- Application Number
- CN202411045523.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-11-26
AI Technical Summary
The traditional modified conjugate gradient method based on the double reciprocity boundary element method to identify inverse geometric conditions is poor in processing singular integrals, and it is difficult to break out of the local optimal solution during the iteration process, and the solution efficiency is low.
By applying the power series expansion method to process singular points in the integration process, adaptive grid segmentation and refinement, introducing precondition matrix and modifying search direction, using strong Wolfe conditions to determine the step size, ensuring that the value of the objective function is reduced and avoiding local optimality untie.
The calculation accuracy and convergence speed of iterative algorithms are improved, ensuring global optimization of effective step size solutions in steep areas.
Smart Images

Figure CN119026402A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of modified conjugate gradient methods, in particular to a modified conjugate gradient method for identifying inverse geometric conditions based on a double reciprocal boundary element method. Background Art
[0002] The dual reciprocal boundary element method is an improved form of the boundary element method, which is particularly suitable for dealing with problems involving inhomogeneous materials and complex geometric shapes, while the conjugate gradient method is an iterative method for solving linear equations and nonlinear optimization problems. It only uses first-order derivative information and avoids the complexity of storing and calculating the Hessian matrix. The modified conjugate gradient method based on the dual reciprocal boundary element method to identify inverse geometric conditions can handle problems with complex geometries and inhomogeneous materials. However, the traditional modified conjugate gradient method based on the dual reciprocal boundary element method to identify inverse geometric conditions still has shortcomings. First, in the traditional dual reciprocal boundary element method, when dealing with integrals with singularity or near-singularity, the accuracy of the calculation results may be poor. Second, during the iteration process, the conjugate gradient method can be modified according to the characteristics of the problem to improve the solution efficiency. Third, when the objective function value moves along the search direction, it may produce too small a step size in the steep canyon area, making it difficult to jump out of the local optimal solution. Therefore, it is necessary to design a modified conjugate gradient method based on the dual reciprocal boundary element method to identify inverse geometric conditions. Summary of the invention
[0003] The object of the present invention is to provide a modified conjugate gradient method for identifying inverse geometric conditions based on a dual reciprocal boundary element method to solve the problems raised in the above background technology.
[0004] To achieve the above object, the present invention provides the following technical solution: a modified conjugate gradient method for identifying inverse geometric conditions based on a dual reciprocal boundary element method, comprising step one, identification conversion; step two, iterative optimization; step three, introduction of preconditions; step four, modification of preconditions; step five, conjugate solution; step six, result output;
[0005] In the above step 1, the problem conditions are identified and converted into integral equations;
[0006] Wherein in the above step 2, iterative optimization is performed during the integration process of step 1;
[0007] Wherein in the above step 3, a suitable preconditioning matrix is selected or designed for import;
[0008] Wherein in the above step 4, the preconditioning matrix is modified;
[0009] In the above step 5, a conjugate solution is performed on the iteration point;
[0010] In the above step six, the result is iterated and output according to the convergence criterion.
[0011] As a further technical solution of the present invention, in the step one, the boundary conditions of the problem are first clarified, and according to the physical properties and boundary conditions of the problem, the basic principles of the double reciprocity boundary element method are used to derive the corresponding boundary integral equation; a series of control points are selected on the boundary, and a NURBS curve or surface describing the boundary is defined by connecting the control points and applying the NURBS basis function, and the continuous variables in the boundary integral equation are replaced by discrete variables on the NURBS curve or surface. By solving the discretized boundary integral equation, an approximate solution of the unknown quantity on the boundary can be obtained.
[0012] As a further technical solution of the present invention, in the step 2, during the integration process of step 1, the singular point in the boundary element method is determined, and the Taylor formula is used to expand near the singular point. A suitable truncation point is selected according to the limitations of calculation accuracy and calculation resources, and the truncated power series is used to replace the original integral kernel function, and then the integral is calculated.
[0013] As a further technical solution of the present invention, in the step 2, the entire computational domain is initially meshed, which may be non-uniform to adapt to the characteristics of the problem. The integral is calculated on each mesh unit, and the error of the integral result is estimated. Based on the result of the error estimation, those mesh units with large errors are marked, which are usually located in areas where the function changes dramatically or there are singularities. The marked mesh units are refined, the number of nodes in these units is increased, or the size of the units is reduced.
[0014] As a further technical solution of the present invention, in step 2, the power series expansion method is applied on the initial grid to identify and process those integral points with obvious singularities. In the process of adaptive unit subdivision, the power series expansion method is used to more accurately process the singular points in the units with large errors. After the grid is refined, the power series expansion method is applied again to verify the integral accuracy after refinement to ensure that the requirements are met.
[0015] As a further technical solution of the present invention, in step 3, the original linear equation system is assumed to be:
[0016] Ax=b
[0017] Where A is the coefficient matrix, x is the vector to be solved, and b is the known vector. After introducing the preconditioning matrix M, the new linear equation system is:
[0018] M -1 Ax=M -1 b
[0019] Among them, M -1 is the inverse matrix of the preconditioning matrix M.
[0020] As a further technical solution of the present invention, in step 4, during the iteration process of the conjugate gradient method, the search direction d needs to be calculated. k and step size α k , after the introduction of the preconditioning matrix, the calculation method of the search direction is modified to:
[0021]
[0022] in, is at the current iteration point x k The gradient at .
[0023] As a further technical solution of the present invention, in step 5, the step length α is determined using the strong Wolfe condition. k , ensuring that the objective function value is along the search direction d k Mobile α k After the distance is sufficient, use the calculated search direction d k and step size α k , update iteration point: x (k+1) =x k +α k *d k ; Calculate the new gradient
[0024] As a further technical solution of the present invention, in step 5, the strong Wolfe condition is expressed as: given a constant c1 between 0 and 1, given a constant c2 between 0 and c1, find a step length α k , so that:
[0025]
[0026] Among them, x k is the current iteration point, d k is the search direction, α k is the step length, is in x k The gradient at .
[0027] As a further technical solution of the present invention, in step 6, according to the preset convergence criterion, when the gradient norm is less than a certain threshold or the change of the objective function value is less than a certain threshold, it is determined whether the convergence condition is met. If not, the iteration process is returned to the next iteration. If the convergence condition is met, the iteration is stopped and the final solution x is output. k .
[0028] Compared with the prior art, the invention has the following beneficial effects: the modified conjugate gradient method for identifying inverse geometric conditions based on the dual reciprocal boundary element method, in the process of integrating the problem conditions by the dual reciprocal boundary element method, applies the power series expansion method on the initial grid to identify and process those integral points with obvious singularities; in the process of adaptive unit subdivision, uses the power series expansion method to more accurately process the singular points in the unit with large errors; after refining the grid, the power series expansion method is applied again to verify the refined integral accuracy to ensure that the requirements are met; by continuously and iteratively applying the adaptive unit subdivision technology and the power series expansion method, the calculation grid and the integral accuracy can be gradually optimized until the required calculation accuracy is met; by introducing and modifying the preconditioning matrix in the iterative process of the conjugate gradient method, the linear equations and the search direction are changed, and the new coefficient matrix M -1 A can have a better condition number, thereby accelerating the convergence of the iterative algorithm and improving the solution efficiency; by using the strong Wolfe condition to determine the step size α k , which ensures that the objective function value decreases sufficiently after moving a certain distance along the search direction, and that the gradient in the search direction also decreases sufficiently, which helps to avoid too small a step size in steep canyon areas and to escape from the local optimal solution. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0031] See also Figure 1 An embodiment of the present invention provides a modified conjugate gradient method for identifying inverse geometric conditions based on a dual reciprocal boundary element method, comprising step one, identifying transformation; step two, iterative optimization; step three, introducing preconditions; step four, modifying preconditions; step five, conjugate solution; step six, outputting results;
[0032] In the above step 1, the problem conditions are identified and converted into integral equations; first, the boundary conditions of the problem are clarified, and the corresponding boundary integral equations are derived based on the physical properties and boundary conditions of the problem and the basic principles of the double reciprocity boundary element method; a series of control points are selected on the boundary, and a NURBS curve or surface describing the boundary is defined by connecting the control points and applying the NURBS basis function, and the continuous variables in the boundary integral equation are replaced by discrete variables on the NURBS curve or surface. By solving the discretized boundary integral equation, an approximate solution of the unknown quantity on the boundary can be obtained;
[0033] Wherein, in the above step 2, during the integration process of step 1, the singular points in the boundary element method are determined, and the Taylor formula is used to expand near the singular points. A suitable truncation point is selected according to the limitation of calculation accuracy and calculation resources, and the truncated power series is used to replace the original integral kernel function, and then the integral is calculated; the entire calculation domain is initially meshed, which can be non-uniform to adapt to the characteristics of the problem, and the integral is calculated on each mesh unit, and the error of the integral result is estimated. According to the result of the error estimation, those mesh units with large errors are marked, which are usually located in the area where the function changes drastically or there are singularities. The marked mesh units are refined, and the number of nodes in these units is increased or the size of the units is reduced; the power series expansion method is applied to the initial mesh to identify and process those integral points with obvious singularities. In the process of adaptive unit subdivision, the power series expansion method is used to more accurately process the singular points in the units with large errors. After the mesh is refined, the power series expansion method is applied again to verify the precision of the refined integral to ensure that the requirements are met;
[0034] In the above step 3, a suitable preconditioning matrix is selected or designed for import; the original linear equation system is assumed to be: Ax = b, where A is the coefficient matrix, x is the vector to be solved, and b is the known vector; after the preconditioning matrix M is introduced, the new linear equation system is: M -1 Ax=M -1 b, where M -1 is the inverse matrix of the preconditioning matrix M;
[0035] In the above step 4, during the iteration of the conjugate gradient method, the search direction d needs to be calculated. k and step size α k , after the introduction of the preconditioning matrix, the calculation method of the search direction is modified to: in, is at the current iteration point x k The gradient at
[0036] In the above step 5, the iterative point is conjugate solved; the step length α is determined using the strong Wolfe condition k, ensuring that the objective function value is along the search direction d k Mobile α k After the distance is sufficient, use the calculated search direction d k and step size α k , update iteration point: x (k+1) =x k +α k *d k ; Calculate the new gradient The strong Wolfe condition is expressed as: given a constant c1 between 0 and 1, and a constant c2 between 0 and c1, find a step size α k , so that:
[0037] Among them, x k is the current iteration point, d k is the search direction, α k is the step length, is in x k The gradient at
[0038] In the above step 6, the result is iterated and output according to the convergence criterion; according to the preset convergence criterion, when the gradient norm is less than a certain threshold or the change of the objective function value is less than a certain threshold, it is judged whether the convergence condition is met. If not, the iteration process is returned to continue the next iteration. If the convergence condition is met, the iteration is stopped and the final solution x is output. k .
[0039] Working principle: When using, the problem is converted into a boundary integral equation using the double reciprocal boundary element method, and discretized using the NURBS basis function. Then, the modified conjugate gradient method is used to solve the boundary integral equation. By iteratively updating the search direction and step size, the solution of the problem is gradually approached. In the process of integrating the problem conditions using the double reciprocal boundary element method, the power series expansion method is applied to the initial grid to identify and process those integral points with obvious singularities. In the process of adaptive unit subdivision, the power series expansion method is used to more accurately process the singular points in the unit with large errors. After refining the grid, the power series expansion method is applied again to verify the refined integral accuracy to ensure that the requirements are met. By continuously and iteratively applying the adaptive unit subdivision technology and the power series expansion method, the computational grid and integral accuracy can be gradually optimized until the required computational accuracy is met. By introducing and modifying the preconditioning matrix in the iterative process of the conjugate gradient method, the linear equations and the search direction are changed, and the new coefficient matrix M -1A can have a better condition number, thereby accelerating the convergence of the iterative algorithm and improving the solution efficiency; by using the strong Wolfe condition to determine the step size α k , which ensures that the objective function value decreases sufficiently after moving a certain distance along the search direction, and that the gradient in the search direction also decreases sufficiently, which helps to avoid too small a step size in steep canyon areas and to escape from the local optimal solution.
[0040] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.
Claims
1. A modified conjugate gradient method for identifying inverse geometric conditions based on a dual reciprocal boundary element method, comprising step 1, identification conversion; step 2, iterative optimization; step 3, introduction of preconditions; step 4, modification of preconditions; step 5, conjugate solution; step 6, result output, characterized in that: In the above step 1, the problem conditions are identified and converted into integral equations; Wherein in the above step 2, iterative optimization is performed during the integration process of step 1; Wherein in the above step 3, a suitable preconditioning matrix is selected or designed for import; Wherein in the above step 4, the preconditioning matrix is modified; In the above step 5, a conjugate solution is performed on the iteration point; In the above step six, the result is iterated and output according to the convergence criterion.
2. The modified conjugate gradient method for identifying inverse geometric conditions based on the dual reciprocal boundary element method according to claim 1, characterized in that: In the step 1, the boundary conditions of the problem are first clarified, and the corresponding boundary integral equation is derived according to the physical properties and boundary conditions of the problem by using the basic principle of the double reciprocity boundary element method; a series of control points are selected on the boundary, and a NURBS curve or surface describing the boundary is defined by connecting the control points and applying the NURBS basis function, and the continuous variables in the boundary integral equation are replaced by discrete variables on the NURBS curve or surface. By solving the discretized boundary integral equation, an approximate solution of the unknown quantity on the boundary can be obtained.
3. The modified conjugate gradient method for identifying inverse geometric conditions based on the dual reciprocal boundary element method according to claim 1, characterized in that: In the step 2, during the integration process of step 1, the singular point in the boundary element method is determined, and the Taylor formula is used to expand near the singular point. A suitable truncation point is selected according to the limitations of calculation accuracy and calculation resources, and the truncated power series is used to replace the original integral kernel function, and then the integral is calculated.
4. The modified conjugate gradient method for identifying inverse geometric conditions based on the dual reciprocal boundary element method according to claim 1, characterized in that: In the step 2, the entire computational domain is initially meshed, which may be non-uniform to suit the characteristics of the problem. The integral is calculated on each mesh unit, and the error of the integral result is estimated. Based on the error estimation result, those mesh units with large errors are marked, which are usually located in areas where the function changes dramatically or there are singularities. The marked mesh units are refined by increasing the number of nodes in these units or reducing the size of the units.
5. The modified conjugate gradient method for identifying inverse geometric conditions based on the dual reciprocal boundary element method according to claim 1, characterized in that: In the step 2, the power series expansion method is applied on the initial grid to identify and process those integral points with obvious singularities. In the process of adaptive unit subdivision, the power series expansion method is used to more accurately process the singular points in the unit with large errors. After the grid is refined, the power series expansion method is applied again to verify the integral accuracy after refinement to ensure that the requirements are met.
6. The modified conjugate gradient method for identifying inverse geometric conditions based on the dual reciprocal boundary element method according to claim 1, characterized in that: In step 3, the original linear equation system is assumed to be: Ax=b Where A is the coefficient matrix, x is the vector to be solved, and b is the known vector. After introducing the preconditioning matrix M, the new linear equation system is: M -1 Ax=M -1 b Among them, M -1 is the inverse matrix of the preconditioning matrix M.
7. The modified conjugate gradient method for identifying inverse geometric conditions based on dual reciprocity boundary element method according to claim 1, characterized in that: In step 4, during the iteration of the conjugate gradient method, the search direction d needs to be calculated. k and step size α k , after the introduction of the preconditioning matrix, the calculation method of the search direction is modified to: in, is at the current iteration point x k The gradient at .
8. The modified conjugate gradient method for identifying inverse geometric conditions based on the dual reciprocal boundary element method according to claim 1, characterized in that: In step 5, the step length α is determined using the strong Wolfe condition. k , ensuring that the objective function value is along the search direction d k Mobile α k After the distance is sufficient, use the calculated search direction d k and step size α k , update iteration point: x (k+1) =x k +α k *d k ; Calculate the new gradient 9. The modified conjugate gradient method for identifying inverse geometric conditions based on the dual reciprocal boundary element method according to claim 1, characterized in that: In step 5, the strong Wolfe condition is expressed as: given a constant c1 between 0 and 1, given a constant c2 between 0 and c1, find a step size α k , so that: Among them, x k is the current iteration point, d k is the search direction, α k is the step length, is in x k The gradient at .
10. The modified conjugate gradient method for identifying inverse geometric conditions based on dual reciprocity boundary element method according to claim 1, characterized in that: In step 6, according to the preset convergence criteria, when the gradient norm is less than a certain threshold or the change of the objective function value is less than a certain threshold, it is determined whether the convergence condition is met. If not, the iteration process is returned to the next iteration. If the convergence condition is met, the iteration is stopped and the final solution x is output. k .
Citation Information
Cited By
Industrial system identification method and system based on SWITCH mechanism and conjugate gradient
CN122449967A
An industrial system identification method and system based on switch mechanism and conjugate gradient
CN122449967B
