JP2023183671A - Design variable optimization device, design variable optimization method, and program - Google Patents

Abstract

To provide a design variable optimization device, a design variable optimization method, and a program that can be applied to problems in which analysis results may not be obtained.SOLUTION: A design variable optimization device includes: a space limiting unit that limits a search space to be searched in a multidimensional space that includes a large number of design variables as elements; and an optimization unit that searches for an optimal solution, by using a combination of design variables that satisfies predetermined constraint conditions and optimizes a predetermined objective function as a solution, and repeatedly executing arithmetic processing to search for a new best solution by limiting the search space to the vicinity of the best solution using the search space as a starting point. The space limiting unit re-limits the search space including the solution searched by the optimization unit. The optimization unit excludes from the search space a region in the vicinity of a combination of design variables for which arithmetic processing has failed.SELECTED DRAWING: Figure 1

Description

The present disclosure relates to a design variable optimization device, a design variable optimization method, and a program.

Patent Document 1 proposes a method to efficiently search for an optimal solution by dividing a large-scale, complex optimization problem into smaller, simpler problems, and selecting and executing a method that is effective for each optimization problem. ing. However, Patent Document 1 assumes a separable problem, and is difficult to apply to problems that are difficult to separate.

Non-Patent Document 1 discloses that after dividing a large design space based on UCB values (Monte Carlo tree search), a response is performed within a confidence region (hypercube) centered on the best solution, starting from a space with a good UCB value. A method of evaluating a curved surface and searching for an optimal solution (Trust Region Bayesian Optimization (TuRBO)) has been proposed. This method can be applied to problems with about 50 dimensions of design parameters without dividing the problem. However, there is a problem that constraints on design variable values and response variable values cannot be considered. Note that the UCB (Upper Confidence Bound) value is an index that increases as the number of samples decreases and the evaluation function value increases. Further, the response surface is a mathematical model for calculating an evaluation function from design variables, and can be, for example, a mathematical model based on Gaussian process regression.

Non-Patent Document 2 proposes a method for searching for an optimal solution in TuRBO by evaluating a response surface and constraint values within a confidence region (hypercube) centered on the best solution. This makes it possible to solve large-scale optimization problems that include constraints on design variable values and response variable values.

Japanese Patent Application Publication No. 2002-183707

Linnan Wang, Rodrigo Fonseca, Yuandong Tian, “Learning Search Space Partition for Black-box Optimization using Monte Carlo Tree Search,” Neurips(Advances in Neural Information Processing Systems 33), 2020 David Eriksson, Matthias Poloczek, “Scalable Constrained Bayesian Optimization,” AISTATS, 2021

In the methods described in Patent Document 1 and Non-Patent Documents 1 and 2, in order to update the response surface model etc. based on the analysis results, analysis solver calculations are performed depending on the combination of design parameters, such as in axial flow compressor analysis. There is a problem in that it cannot be applied to problems where analysis results may not be obtained due to failure. Note that the analysis solver is software that solves equations by numerical calculation, such as fluid analysis software and structural analysis software, for example.

The present disclosure has been made to solve the above problems, and provides a design variable optimization device, a design variable optimization method, and a program that can be applied to problems in which analysis results may not be obtained. The purpose is to provide.

In order to solve the above problems, a design variable optimization device according to the present disclosure includes a space limiter that limits a search space to be searched in a multidimensional space including many design variables, and a predetermined constraint condition. Using the combination of design variables that satisfies and optimizes a predetermined objective function as a solution, repeating arithmetic processing to search for a new best solution by limiting the search space to the vicinity of the best solution using the search space as a starting point. and an optimization unit that searches for an optimal solution; the space limitation unit relimits the search space including the solution searched by the optimization unit; A region in the vicinity of the combination of design variables determined is excluded from the search space.

Further, the design variable optimization device according to the present disclosure includes a space limiting unit that limits a search space to be searched in a multidimensional space that includes many design variables as elements, and a space limiting unit that limits a search space that is a search target and that satisfies a predetermined constraint condition and performs a predetermined objective function. Using the combination of the design variables to be optimized as a solution, the optimal solution is searched for by repeatedly executing calculation processing to search for a new best solution by limiting the search space to the vicinity of the best solution using the search space as a starting point. and an optimization unit that relimits the search space including the solution searched by the optimization unit, and also defines a region near the combination of design variables that does not satisfy the constraint. excluded from the search space.

Further, the design variable optimization method according to the present disclosure includes the steps of limiting a search space to be searched in a multidimensional space including many design variables, and optimizing a predetermined objective function by satisfying predetermined constraints. a step of searching for an optimal solution by repeatedly executing arithmetic processing to search for a new best solution by using the combination of design variables as a solution and limiting a search space to the vicinity of the best solution using the search space as a starting point; and relimiting the search space including the solution searched for by the calculation process, and excluding from the search space a region in the vicinity of the combination of design variables for which the calculation process failed.

Further, the program according to the present disclosure includes the steps of: limiting a search space to be searched in a multidimensional space including many design variables; The step of searching for the optimal solution by repeatedly executing arithmetic processing to search for a new best solution by limiting the search space to the vicinity of the best solution using the search space as a starting point, using the combination of A program to be executed, which relimits the search space including the solution searched for by the calculation process, and excludes from the search space a region in the vicinity of the combination of design variables for which the calculation process has failed.

Further, the design variable optimization method according to the present disclosure includes the steps of limiting a search space to be searched in a multidimensional space including many design variables, and optimizing a predetermined objective function by satisfying predetermined constraints. a step of searching for an optimal solution by repeatedly executing arithmetic processing to search for a new best solution by using the combination of the design variables as a solution and limiting the search space to the vicinity of the best solution using the search space as a starting point; and relimiting the search space including the solution searched by the arithmetic processing, and excluding from the search space a region near the combination of design variables that does not satisfy the constraint.

Further, the program according to the present disclosure includes the steps of: limiting a search space to be searched in a multidimensional space including many design variables; The step of searching for the optimal solution by repeatedly executing arithmetic processing to search for a new best solution by limiting the search space to the vicinity of the best solution using the search space as a starting point, using the combination of A program to be executed, which relimits the search space including the solution searched by the calculation process, and excludes from the search space a region near the combination of design variables that does not satisfy the constraint.

According to the design variable optimization device, design variable optimization method, and program of the present disclosure, it can be applied to problems in which analysis results may not be obtained.

FIG. 1 is a block diagram illustrating a configuration example of a design variable optimization device according to a first embodiment of the present disclosure. 1 is a flowchart illustrating an example of the operation of the design variable optimization device according to the first embodiment of the present disclosure. 1 is a flowchart illustrating an example of the operation of the design variable optimization device according to the first embodiment of the present disclosure. FIG. 2 is a schematic diagram for explaining an example of the operation of the design variable optimization device according to the first embodiment of the present disclosure. FIG. 2 is a schematic diagram for explaining an example of the operation of the design variable optimization device according to the first embodiment of the present disclosure. FIG. 2 is a block diagram illustrating a configuration example of a design variable optimization device according to a second embodiment of the present disclosure. It is a flow chart which shows an example of operation of a design variable optimization device concerning a 2nd embodiment of this indication. FIG. 7 is a schematic diagram for explaining an operation example of a design variable optimization device according to a second embodiment of the present disclosure. FIG. 7 is a schematic diagram for explaining an operation example of a design variable optimization device according to a second embodiment of the present disclosure. It is a flow chart which shows an example of operation of a design variable optimization device concerning a 3rd embodiment of this indication. FIG. 1 is a schematic block diagram showing the configuration of a computer according to at least one embodiment.

Hereinafter, a design variable optimization device, a design variable optimization method, and a program according to an embodiment of the present disclosure will be described with reference to each figure. In addition, in each figure, the same reference numerals are used for the same or corresponding components, and the description thereof will be omitted as appropriate.

<First embodiment>
(Configuration of design variable optimization device)
FIG. 1 is a block diagram illustrating a configuration example of a design variable optimization device according to a first embodiment of the present disclosure. The design variable optimization device 1 can be configured using, for example, one or more computers such as a server or a personal computer, peripheral devices, and the like. Note that some of the plurality of computers may be configured, for example, on a cloud or the like via a network. The design variable optimization device 1 has a functional configuration consisting of a combination of one or more computers, hardware such as peripheral devices, and software such as programs executed by the computer, and includes a control unit 11, an input/output unit, etc. 12, a space limiting section 13 and an optimization section 14.

A design variable optimization device 1 shown in FIG. 1 is a device that finds an optimal combination of design variables that satisfies predetermined constraints and minimizes (or maximizes) one or more predetermined objective functions. In this embodiment, minimizing or maximizing the objective function is referred to as optimizing the objective function. Further, the design variables are also referred to as design parameters. The objective function is also called an evaluation function, response function (variable), etc. Furthermore, the combination of design variables that minimizes (or maximizes) the objective function is called an optimal solution. Furthermore, when an optimal solution is obtained by repeating the search process multiple times, the combination of design variables that makes the value of the objective function the best in each search process is called the best solution.

The design variables include, for example, variables representing the flow path (shape) of the turbine in axial flow compressor design. However, it is not limited to this example. In addition, constraints include, for example, constraints on the range, upper limit, and lower limit of each design variable or the relationship between design variables, and constraints determined by selected design variables that are not included in the design variables. Contains constraints on ranges, upper limits, and lower limits for elements.

The control unit 11 controls processing in the input/output unit 12, the space limiting unit 13, and the optimization unit 14. For example, the control unit 11 determines conditions for ending the processing of the space limitation unit 13 and optimization unit 14, and sets operating conditions for the space limitation unit 13 and optimization unit 14.

The input/output unit 12 accepts operations from an operator using, for example, a keyboard, a mouse, a display, a communication terminal, etc., inputs predetermined data to the space limiting unit 13 and the optimization unit 14, and Predetermined data is output from the limiting unit 13 and the optimizing unit 14.

The space limiting unit 13 limits a part of the multidimensional space having a plurality of design variables as elements of each dimension to a search space to be searched by the optimization unit 14. It is assumed that the multidimensional space includes a plurality of sample points that have already been determined as initial values. Each sample point (or each data) is defined by a combination of each value of the design variable and each value of the objective function. In addition, the space before limitation is called a whole space, and the space after limitation is called a partial space.

In this embodiment, the space limiter 13 limits the search space using the known Monte Carlo tree search method, which is one of the spatial search methods. The Monte Carlo tree search method is a tree search using the Monte Carlo method. In this embodiment, the root in the tree structure is made to correspond to the entire space, and each node is made to correspond to a partial space. In this embodiment, space division is expressed using a tree structure. The space limiting unit 13 classifies the data into two classes in consideration of the value of the evaluation function (objective function). Further, the space limiting unit 13 divides the entire space or the partial space using, for example, a supervised classification method using machine learning. The space limiting unit 13 then preferentially selects a space with a high evaluation function value and a small number of samples. Note that examples of the supervised classification method using machine learning in the space limiting unit 13 include SVM (support vector machine) and logistic regression.

The space limiting unit 13 repeatedly executes the above classification, division, and selection, and gradually limits the size of the partial space. The space limiting unit 13 ends limiting the size of the partial space when the size of the partial space is equal to or less than the threshold, for example, based on a comparison result between a predetermined threshold and the size of the partial space.

Note that the space limiting unit 13 uses UCB as a numerical value serving as a reference for classification and branch selection. UCB is calculated using the formula below.

UCB=J+α√{(log(np)/n)}

Here, J is the average evaluation function value. n is the number of samples. np is the number of samples of the parent node. α is a coefficient determined based on empirical rules.

Furthermore, based on the results of the optimization process performed by the optimization unit 14, the space limiting unit 13 adds the newly searched sample points, searches the subspace from the entire space again, and determines a new search space. do.

The optimization unit 14 uses the partial space or the entire space limited by the space limiting unit 13 as a search space as a starting point, and limits the search space to a hypercube centered on the best solution. In this embodiment, the limited search space is referred to as a limited search space. The optimization unit 14 determines the optimal solution by repeating the process of limiting the search space to a hypercube centered on the best solution. In this embodiment, the optimization unit 14 searches for an optimal solution using the above-mentioned TuRBO as a base, which is a method of evaluating a response surface within a confidence region (hypercube) centered on the best solution and searching for an optimal solution. . TuRBO is a method of black box optimization. Further, TuRBO is a Bayesian optimization method that uses a Gaussian process and Thompson sampling.

In addition, with black box optimization, even if the shape of the objective function to be minimized or maximized is not known, if the output for the input is known, the appropriate input is determined as the next search point based on that relationship, and the output is obtained again. This method searches for the optimal point by repeating the process of searching for the next search point. Furthermore, Bayesian optimization is a black box optimization method that uses uncertainty to determine the next point to search. Thompson sampling is a sampling method in which the posterior probabilities of options are determined, sampling is performed on the probability distribution, and the best option is selected.

(Example of operation of design variable optimization device)
An example of the operation of the design variable optimization apparatus 1 in the first embodiment will be described with reference to FIGS. 2 to 5. 2 and 3 are flowcharts illustrating an example of the operation of the design variable optimization device according to the first embodiment of the present disclosure. 4 and 5 are schematic diagrams for explaining an example of the operation of the design variable optimization device according to the first embodiment of the present disclosure.

FIG. 2 shows an example of the operation of the design variable optimization device 1 as a whole. FIG. 3 shows the flow of processing in step S11 and step S12 in FIG. FIG. 4 schematically shows an example of a search space. FIG. 5 schematically shows the flow of setting the search limited space.

As shown in FIG. 2, the optimization unit 14 first uses the partial space or the entire space limited by the space limiting unit 13 as a search space as a starting point, and performs initial sampling at multiple points at random within the search space (S11). . Next, the optimization unit 14 limits the search space to the vicinity of the best solution and performs sampling using Bayesian optimization (S12). The optimization unit 14 updates the best solution and the search space (S13). Next, the control unit 11 determines whether the search space has converged (S14). Here, the state in which the search space has converged is, for example, a state in which the search space is no longer updated. If the search space has not converged (S14: No), the optimization unit 14 executes the processes of steps S12 and S13 again.

In steps S11 and S12, the optimization unit 14 creates a response surface using only the design solutions for which the analytical solver calculation was successful, as shown in FIG. 3 (S21). Next, the optimization unit 14 determines a calculation failure region from the analytical solver calculation failure points, for example, by a supervised classification method using machine learning (S22). Note that examples of the supervised classification method using machine learning in the optimization unit 14 include the k-nearest neighbor method, one class SVM, and the like. Next, when generating initial sampling and generating candidate points in the TuRBO optimization calculation, the optimization unit 14 performs filtering (region elimination) and repeat until the required number is obtained (S23). Next, the optimization unit 14 evaluates the posterior distribution on the created response surface with respect to the obtained candidate point group, and calculates the next sampling point (S24).

Note that the k-nearest neighbor method is a classification method using a supervised machine learning model. Further, SOBOL sampling is a method of experimental design using a quasi-random number method.

FIG. 4 schematically shows the hypercube search space SP1 on a two-dimensional plane. The search space SP1 includes regions with different response surface model evaluation values, which are indicated by black or hatched horizontal lines with different densities. Furthermore, the search space SP1 includes a sampling point (design solution) P1 where an analysis result was obtained and a sampling point (design failure point) P2 where an analysis result was not obtained, which is indicated by a broken line. Moreover, point P3 is the next sampling point. Furthermore, the area A1 indicated by diagonal hatching is a calculation failure area.

FIG. 5 shows an example of determining the search limited space in steps S11 to S14. Among the sampling points SP, sampling point SP2 is the best solution. The search space includes a constraint violation space and a constraint satisfaction space. In step ST1-1, a sampling point SP2, which is the best solution, is selected from a plurality of sampled sampling points SP1. Next, in step ST1-2, a limited search space is set near the sampling point SP2, which is the best solution, and the next sampling point SP3 is calculated within the limited search space. In step ST1-3, a limited search space is set near the sampling point SP2 which is the best solution, and the next sampling point SP3 is calculated within the limited search space. The limited search space is gradually limited in size. In this way, by limiting the search space to a hypercube centered on the best solution, it is possible to efficiently search for the optimal solution. Note that the best solution is a sampling point where the evaluation function value is the best when there is a constraint-satisfying solution, and a sampling point where the amount of constraint violation is the minimum when there is no constraint-satisfying solution.

On the other hand, if the search space has converged (S14 in FIG. 2: Yes), the control unit 11 determines whether the termination condition is satisfied (S15). The termination condition is, for example, that the evaluation function value, the number of searches, the search time, etc. exceed a predetermined threshold. If the termination condition is not satisfied (S15: No), the space limiting unit 13 repeatedly updates the Monte Carlo tree starting from the search space including the sampling points newly selected in the optimization process (S16). Next, the space limitation unit 13 ends the limitation of the subspace, for example, when the size of the subspace becomes equal to or less than a predetermined threshold, and determines a search space based on the Monte Carlo tree (S17). Next, the optimization unit 14 executes the process of step S11 again.

Further, if the termination condition is satisfied (S15: Yes), the control unit 11 terminates the process shown in FIG. 2.

As described above, in this embodiment, data is labeled depending on whether the analytical solver calculation is successful or not, and a response surface model is generated using only data for which the analytical solver calculation has been successful. In addition, a region covering the solution for which the analytical solver calculation has failed is determined using a supervised classification method using machine learning such as the k-nearest neighbor method, and points within that region are excluded to search for the optimal solution. For example, when generating candidate points for evaluating response surfaces and constraint values during optimization, the randomly generated candidate points are filtered for analysis solver calculation failures, the remaining candidate points are evaluated, and the next It is possible to determine sampling points, etc.

(Actions and effects of the first embodiment)
Optimization can now be applied even when analysis solver calculations fail and no analysis results are obtained, expanding the scope of optimization technology.

<Second embodiment>
A design variable optimization device according to a second embodiment of the present disclosure will be described with reference to FIGS. 6 to 9. FIG. 6 is a block diagram illustrating a configuration example of a design variable optimization device according to a second embodiment of the present disclosure. FIG. 7 is a flowchart illustrating an example of the operation of the design variable optimization device according to the second embodiment of the present disclosure. 8 and 9 are schematic diagrams for explaining an example of the operation of the design variable optimization device according to the second embodiment of the present disclosure.

As shown in FIG. 6, the design variable optimization device 1a according to the second embodiment of the present disclosure includes a control section 11, an input/output section 12, a space limiting section 13a, and an optimization section 14. In the design variable optimization device 1a shown in FIG. 6, the configuration of a space limiting section 13a, which corresponds to the space limiting section 13 shown in FIG. 1, is partially different from the space limiting section 13. Further, the basic operation of the design variable optimization device 1a shown in FIG. 6 is the same as the basic operation of the design variable optimization device 1 shown in FIG. However, in the operation of the design variable optimization device 1a shown in FIG. 6, compared to the operation of the design variable optimization device 1 shown in FIG. 2, the contents of step S16 shown in FIG. 2 are partially different.

In step S16 shown in FIG. 2, the space limitation unit 13a according to the second embodiment executes space limitation in consideration of constraint conditions, as shown in FIGS. 7 and 8. In step S16 shown in FIG. 2, the Monte Carlo tree is repeatedly updated by repeatedly executing the process shown in FIG.

In the process shown in FIG. 7, the space limiting unit 13a labels data that violates a constraint within a node, and classifies it into normal data and abnormal (constraint violation) data as follows using the k-nearest neighbor method (S31 ).

The abnormal data is data that satisfies labn<δabn and lnor>δnor and lnor>labn. Here, labn is the distance to the abnormal data. δabn is a predetermined threshold value. lnor is the distance to normal data. δnor is a predetermined threshold value.

Normal data is data other than the above.

Next, the space limiting unit 13a classifies the normal data within the node into two classes based on the UCB value (S32).

Next, the space limiting unit 13a divides the area using a supervised classification method using machine learning such as linear SVM (S33).

FIG. 8 schematically shows the flow of processing for limiting the entire space A to the partial space J. In step ST2-1, the entire space A includes a plurality of sample points SP, and a label SP1 is attached to an abnormal sample point SP indicated by a broken line circle. In step ST-2, space A is divided into space B, space C, and space D. Space B is a space for data classified as constraint violations. Space C is a space of data classified as having a good UCB value. Space D is a space of data classified as having a bad UCB value. Further, in the tree structure T1, the root Ra corresponds to the space A. Node Nb corresponds to space B. Node Nc corresponds to space C. Node Nd corresponds to space D. The space limiting unit 13a repeatedly performs classification, division, and selection, and ends the search space limitation when a predetermined condition is satisfied. In step ST-3, the space J is limited to the search space.

Note that in the tree structure T1, the node Ne and the node Nh correspond to the space of data classified as violating a constraint. Node Nf and node Nj correspond to a space of data classified as having a good UCB value. Node Ng and node Ni correspond to a space of data classified as having a bad UCB value.

FIG. 9 shows the overall process flow in the design variable optimization device 1a according to the second embodiment. Note that the basic process flow in space limitation is the same as the process flow shown in FIG. Further, the basic process flow in optimization is the same as the process flow shown in FIG. In the example shown in FIG. 9, the optimization process is executed starting from a search space specified by spatial limitation (ST20). Further, when the search space becomes sufficiently small in the optimization process (ST10), space limitation is performed again from the entire space A by adding newly obtained sampling points.

Note that in the second embodiment, the filtering process (S22 and S23 in FIG. 3) in the calculation failure area may be performed in the same manner as in the first embodiment, or may not be performed. That is, consideration of the constraint conditions may be limited to only space-limited processing.

In the second embodiment, a solution that violates a constraint is labeled, and an area that covers the solution that violates the constraint is determined by a k-nearest neighbor method or the like. In addition, by dividing the remaining solution after excluding the inside of the region based on the evaluation of the UCB value, it is possible to determine an unsearched region with a high evaluation value while taking constraints into consideration.

(Actions and effects of the second embodiment)
In the second embodiment, it is now possible to determine a region with a high UCB value in a large-scale design space while taking constraints into consideration, making it possible to search for an optimal solution more efficiently.

<Third embodiment>
With reference to FIG. 10, a design variable optimization device according to a third embodiment of the present disclosure will be described. FIG. 10 is a flowchart illustrating an example of the operation of the design variable optimization device according to the third embodiment of the present disclosure. As shown in FIG. 10, in the operation of the design variable optimization apparatus according to the third embodiment, between the processing of step S13 and step S14 of the flow shown in FIG. Processing is provided. The other configurations and operations are the same as the design variable optimization device 1 of the first embodiment or the design variable optimization device 1a of the second embodiment.

As shown in FIG. 10, in the design variable optimization device according to the third embodiment, during the optimization process (steps S11 to S13), it is determined whether the search space is just before convergence (S13-2). ), if the search space is about to converge (S13-2: Yes), the acquisition function is optimized (S13-3). Note that the processing in step S13-2 and the processing in step S13-3 may be executed by the optimization unit 14 or the control unit 11. In the flow shown in FIG. 10, the criterion for determining whether or not it is just before convergence is the same as, for example, the criterion for determining whether the search space has converged in step S14 (however, the criterion value is different). Can be set.

In TuRBO used in Non-Patent Document 1 and Non-Patent Document 2, when the best solution is not updated, the region is reduced to improve the convergence of the solution. However, in cases where constrained optimization problems or analytical solver calculations fail, it is difficult to update the best solution compared to optimization without constraints, and the evaluation value of the solution obtained with one TuRBO may not be good. sell.

Furthermore, TuRBO targets large-scale problems with approximately 50 dimensions of design parameters. Therefore, in order to reduce calculation costs, rather than strictly optimizing the response surface, the response surface is evaluated on candidate points randomly sampled within the limited search space, and the best point is determined as the next sampling point. Thompson sampling is used.

Therefore, the method is improved so that optimization of the acquisition function in consideration of the constraints is performed immediately before it is determined that the solution by TuRBO has converged (the size of the confidence region becomes less than or equal to a specified value). This increases the probability that the evaluation value of the solution obtained by one TuRBO will improve.

Note that the acquisition function is a function used as an index for searching for the next sampling point in Bayesian optimization. For example, there is a UCB value that uses the sum of the mean value and variance of a response surface model (Gaussian process regression model). In Bayesian optimization, the next sampling point is calculated by strictly optimizing this function using a gradient method or the like.

(Actions and effects of the third embodiment)
Even when constraints and analysis solver calculation failures are taken into account, it is possible to increase the possibility that the evaluation value of the solution obtained by one TuRBO will improve.

(effect)
The design variable optimization devices 1 and 1a configured as described above include a space limiting unit 13 or 13a that limits a search space to be searched in a multidimensional space including many design variables, and Using the combination of design variables that optimizes the objective function as a solution, the optimal solution is found by repeatedly executing calculation processing to search for a new best solution by limiting the search space to the vicinity of the best solution using the search space as the starting point. and an optimization unit 14 for searching. Further, the space limiting unit 13 relimits the search space including the solution searched by the optimization unit, and the optimization unit 14 excludes from the search space a region near the combination of design variables for which the calculation process has failed. . Therefore, the design variable optimization device, design variable optimization method, and program of the present disclosure can be applied to problems in which analysis results may not be obtained.

The design variable optimization device 1a having the above configuration also includes a space limiting section 13a that limits a search space to be searched in a multidimensional space including many design variables, and a space limiting section 13a that limits a search space to be searched, and a space limiting section 13a that satisfies a predetermined constraint condition and performs a predetermined objective function. The optimal solution is searched for by repeatedly executing arithmetic processing to search for a new best solution by using the combination of design variables that optimizes the solution as a starting point and limiting the search space to the vicinity of the best solution. and an optimization section 14. Further, the space limiting unit 13a relimits the search space including the solution searched by the optimization unit, and excludes from the search space a region near a combination of design variables that does not satisfy the constraint conditions. Therefore, the design variable optimization device, design variable optimization method, and program of the present disclosure can be easily applied to problems in which analysis results may not be obtained due to failure to satisfy constraint conditions.

<Other embodiments>
Although the embodiment of the present disclosure has been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes design changes within the scope of the gist of the present disclosure. .

<Computer configuration>
FIG. 11 is a schematic block diagram showing the configuration of a computer according to at least one embodiment.
Computer 90 includes a processor 91, main memory 92, storage 93, and interface 94.
The above-described design variable optimization devices 1 and 1a are implemented in a computer 90. The operations of each processing section described above are stored in the storage 93 in the form of a program. The processor 91 reads the program from the storage 93, expands it into the main memory 92, and executes the above processing according to the program. Further, the processor 91 reserves storage areas corresponding to each of the above-mentioned storage units in the main memory 92 according to the program.

The program may be for realizing a part of the functions to be performed by the computer 90. For example, the program may function in combination with other programs already stored in storage or in combination with other programs installed in other devices. Note that in other embodiments, the computer may include a custom LSI (Large Scale Integrated Circuit) such as a PLD (Programmable Logic Device) in addition to or in place of the above configuration. Examples of PLDs include PAL (Programmable Array Logic), GAL (Generic Array Logic), CPLD (Complex Programmable Logic Device), and FPGA (Field Programmable Gate Array). In this case, some or all of the functions implemented by the processor may be implemented by the integrated circuit.

Examples of the storage 93 include HDD (Hard Disk Drive), SSD (Solid State Drive), magnetic disk, magneto-optical disk, CD-ROM (Compact Disc Read Only Memory), and DVD-ROM (Digital Versatile Disc Read Only Memory). , semiconductor memory, etc. Storage 93 may be an internal medium connected directly to the bus of computer 90, or may be an external medium connected to computer 90 via an interface 94 or a communication line. Furthermore, when this program is distributed to the computer 90 via a communication line, the computer 90 that received the distribution may develop the program in the main memory 92 and execute the above processing. In at least one embodiment, storage 93 is a non-transitory, tangible storage medium.

<Additional notes>
The design variable optimization devices 1 and 1a described in each embodiment are understood as follows, for example.

(1) The design variable optimization devices 1 and 1a according to the first aspect include a space limiting unit 13 or 13a that limits a search space to be searched in a multidimensional space including a large number of design variables, and a predetermined Using the combination of design variables that satisfies the constraints and optimizes a predetermined objective function as a solution, the calculation process is repeated to search for a new best solution by limiting the search space to the vicinity of the best solution using the search space as a starting point. and an optimization unit 14 that searches for an optimal solution by executing the optimization unit 14, and the space limiting unit 13 or 13a relimits the search space including the solution searched by the optimization unit, and 14 excludes from the search space a region near the combination of design variables for which the arithmetic processing has failed. According to this aspect, it can be applied to problems in which analysis results may not be obtained.

(2) The design variable optimization device 1a according to the second aspect is the design variable optimization device 1a of (1), in which the space limiting unit 13a is configured to select combinations of the design variables that do not satisfy the constraint conditions. Excluding neighboring regions from the search space. According to this aspect, it can be more appropriately applied to problems in which analysis results may not be obtained.

(3) The design variable optimization device 1 or 1a according to the third aspect is the design variable optimization device 1 or 1a of (1) or (2), in which the optimization unit 14 performs the calculation processing. Immediately before the convergence condition is satisfied, the mode of the arithmetic processing is changed. According to this aspect, it can be more appropriately applied to problems in which analysis results may not be obtained.

(4) The design variable optimization device 1a according to the fourth aspect includes a space limiting unit 13a that limits a search space to be searched in a multidimensional space including a large number of design variables, and a space limiting unit 13a that limits a search space that satisfies predetermined constraint conditions. By using the combination of the design variables that optimizes a predetermined objective function as a solution, and repeatedly performing calculation processing to search for a new best solution by limiting the search space to the vicinity of the best solution using the search space as a starting point. , an optimization unit 14 that searches for an optimal solution, and the space limitation unit 13a relimits the search space including the solution searched by the optimization unit 14, and also includes the design that does not satisfy the constraint conditions. Exclude areas near the combination of variables from the search space. According to this aspect, it becomes easier to apply to problems in which analysis results may not be obtained due to failure to satisfy constraint conditions.

(5) The design variable optimization method according to the fifth aspect includes the steps of: limiting a search space to be searched in a multidimensional space including a large number of design variables; By using the combination of design variables that optimizes the solution as a solution, the optimal solution is found by repeatedly executing calculation processing to search for a new best solution by limiting the search space to the vicinity of the best solution using the search space as a starting point. relimiting the search space including the solution searched for by the calculation process, and excluding from the search space a region near the combination of design variables for which the calculation process failed.

(6) The program according to the sixth aspect includes the steps of limiting a search space to be searched in a multidimensional space including many design variables, and optimizing a predetermined objective function by satisfying predetermined constraints. using the combination of design variables as a solution and repeatedly performing calculation processing to search for a new best solution by limiting a search space to the vicinity of the best solution using the search space as a starting point, thereby searching for an optimal solution; A program for causing a computer to execute, the program relimiting the search space including the solution searched by the calculation process, and excluding from the search space a region in the vicinity of the combination of design variables for which the calculation process failed. do.

(7) The design variable optimization method according to the seventh aspect includes the steps of: limiting a search space to be searched in a multidimensional space including a large number of design variables; By using the combination of design variables that optimizes the solution as a solution, the optimal solution is found by repeatedly executing calculation processing to search for a new best solution by limiting the search space to the vicinity of the best solution using the search space as a starting point. relimiting the search space including the solution searched by the arithmetic processing, and excluding from the search space a region near the combination of design variables that does not satisfy the constraint conditions.

(8) The program according to the eighth aspect includes the steps of limiting a search space to be searched in a multidimensional space including many design variables, and optimizing a predetermined objective function by satisfying predetermined constraints. using the combination of design variables as a solution and repeatedly performing calculation processing to search for a new best solution by limiting a search space to the vicinity of the best solution using the search space as a starting point, thereby searching for an optimal solution; is a program that causes a computer to execute the following, which relimits the search space including the solution searched by the arithmetic processing, and extracts from the search space a region in the vicinity of the combination of design variables that does not satisfy the constraint. Exclude.

1, 1a... Design variable optimization device 11... Control section 12... Input/output section 13, 13a... Space limiting section 14... Optimization section

Claims (8)

a space limiter that limits a search space to be searched in a multidimensional space that includes many design variables;
Computation processing that searches for a new best solution by using the combination of design variables that satisfies predetermined constraints and optimizes a predetermined objective function as a solution, and limits the search space to the vicinity of the best solution using the search space as a starting point. It is equipped with an optimization section that searches for an optimal solution by repeatedly executing
The space limiting unit relimits the search space including the solution searched by the optimization unit,
The design variable optimization device, wherein the optimization unit excludes a region near the combination of design variables for which the calculation process has failed from the search space. The design variable optimization device according to claim 1, wherein the space limiting unit excludes from the search space a region in the vicinity of the combination of design variables that does not satisfy the constraint condition. The design variable optimization device according to claim 1 or 2, wherein the optimization unit changes the mode of the arithmetic processing immediately before a convergence condition for the arithmetic processing is satisfied. a space limiter that limits a search space to be searched in a multidimensional space that includes many design variables;
Computation processing that searches for a new best solution by using the combination of design variables that satisfies predetermined constraints and optimizes a predetermined objective function as a solution, and limits the search space to the vicinity of the best solution using the search space as a starting point. It is equipped with an optimization section that searches for an optimal solution by repeatedly executing
The space limiting unit relimits the search space including the solution searched by the optimization unit, and excludes from the search space a region in the vicinity of the combination of design variables that does not satisfy the constraint. Design variables Optimizer. a step of limiting a search space to be searched in a multidimensional space having many design variables as elements;
Computation processing that searches for a new best solution by using the combination of design variables that satisfies predetermined constraints and optimizes a predetermined objective function as a solution, and limits the search space to the vicinity of the best solution using the search space as a starting point. and searching for an optimal solution by repeatedly executing .
Relimiting the search space including the solution searched by the calculation process,
A design variable optimization method, wherein a region in the vicinity of the combination of design variables for which the arithmetic processing has failed is excluded from the search space. a step of limiting a search space to be searched in a multidimensional space having many design variables as elements;
Computation processing that searches for a new best solution by using the combination of design variables that satisfies predetermined constraints and optimizes a predetermined objective function as a solution, and limits the search space to the vicinity of the best solution using the search space as a starting point. A program that causes a computer to perform the steps of searching for an optimal solution by repeatedly executing .
Relimiting the search space including the solution searched by the calculation process,
A program for excluding from the search space a region in the vicinity of the combination of design variables for which the calculation process has failed. a step of limiting a search space to be searched in a multidimensional space having many design variables as elements;
Computation processing that searches for a new best solution by using the combination of design variables that satisfies predetermined constraints and optimizes a predetermined objective function as a solution, and limits the search space to the vicinity of the best solution using the search space as a starting point. and searching for an optimal solution by repeatedly executing .
A design variable optimization method, comprising: relimiting the search space including the solution searched for by the arithmetic processing, and excluding from the search space a region near the combination of the design variables that does not satisfy the constraint conditions. a step of limiting a search space to be searched in a multidimensional space having many design variables as elements;
Computation processing that searches for a new best solution by using the combination of design variables that satisfies predetermined constraints and optimizes a predetermined objective function as a solution, and limits the search space to the vicinity of the best solution using the search space as a starting point. A program that causes a computer to perform the steps of searching for an optimal solution by repeatedly executing .
A program for relimiting the search space including solutions searched by the arithmetic processing, and excluding from the search space a region in the vicinity of the combination of design variables that does not satisfy the constraint condition.

Images