Search

KR-20260065306-A - System for optimizing values based on kriging and machine-learning and method thereof

KR20260065306AKR 20260065306 AKR20260065306 AKR 20260065306AKR-20260065306-A

Abstract

The present invention relates to a variable optimization system and a method using a surrogate model based on kriging and machine learning. According to the present invention, a variable optimization system using a surrogate model based on kriging and machine learning may include: a sensitivity analysis unit that divides a design variable into a first design variable and a second design variable according to sensitivity using a pre-set sensitivity analysis on the design variable of a set multivariate problem; a first variable optimization unit that applies a pre-set data extraction method to the first design variable to extract data respectively and generates a first surrogate model based on kriging to optimize the first design variable; and a second variable optimization unit that applies a pre-set data extraction method to the second design variable to extract data respectively and generates a second surrogate model based on machine learning to optimize the second design variable. As such, according to the present invention, the shapes of the rotor and stator of an electric motor can be optimally designed to achieve the design objectives.

Inventors

  • 임동국
  • 오승환

Assignees

  • 울산대학교 산학협력단

Dates

Publication Date
20260508
Application Date
20241101

Claims (12)

  1. A sensitivity analysis unit that divides a design variable into a first design variable and a second design variable according to sensitivity using a sensitivity analysis pre-set for the design variable of a set multivariate problem; A first variable optimization unit that optimizes the first design variable by applying a pre-set data extraction method to the first design variable to extract data respectively and generating a first surrogate model based on kriging; and A variable optimization system using a kriging and machine learning-based surrogate model, comprising a second variable optimization unit that applies a pre-set data extraction method to the second design variable to extract data respectively, generates a second surrogate model based on machine learning, and optimizes the second design variable.
  2. In paragraph 1, The above sensitivity analysis unit is, Extract data for the design variables by performing a pre-established method on the design variables of the established multivariate problem, and A variable optimization system using a surrogate model based on kriging and machine learning, which performs sensitivity analysis on the extracted data of the above design variables and divides the above design variables into a first design variable and a second design variable at a preset ratio according to the order of high sensitivity.
  3. In paragraph 1, The above-mentioned first variable optimization unit is, A data extraction method previously set for the first design variable is applied to extract multiple data of the first design variable, and Using the extracted data above, the first surrogate model based on the kriging is generated, and A variable optimization system using a Kriging and machine learning-based surrogate model that derives an optimal solution for the first design variable using the first surrogate model and a previously established optimization algorithm.
  4. In paragraph 3, The above-mentioned first variable optimization unit is, A variable optimization system using a kriging and machine learning-based surrogate model that sets the second design variable as an initial value and, when the optimal solution of the first design variable derived does not satisfy the first constraint condition, repeats the process of deriving the optimal solution for the first design variable using the first surrogate model and the pre-established optimization algorithm.
  5. In paragraph 1, The above second variable optimization unit is, A data extraction method previously set for the second design variable is applied to extract multiple data of the second design variable, and A second surrogate model based on machine learning is generated using the extracted data above, and A variable optimization system using a Kriging and machine learning-based surrogate model that derives an optimal solution for the second design variable using the second surrogate model and a previously established optimization algorithm.
  6. In paragraph 5, The above second variable optimization unit is, A variable optimization system using a kriging and machine learning-based surrogate model that sets the first design variable to an optimized value and, if the derived optimal solution of the second design variable does not satisfy the second constraint condition, repeats the process of deriving the optimal solution for the second design variable using the second surrogate model and a pre-established optimization algorithm.
  7. A step of dividing a design variable into a first design variable and a second design variable according to sensitivity using a sensitivity analysis pre-set on the design variable of a multivariate problem set by a sensitivity analysis unit; A first variable optimization unit applies a data extraction method previously set for the first design variable to extract data for each, and generates a first surrogate model based on kriging to optimize the first design variable; and A variable optimization method using a kriging and machine learning-based surrogate model, comprising the step of a second variable optimization unit applying a pre-set data extraction method to the second design variable to extract data, and generating a second surrogate model based on machine learning to optimize the second design variable.
  8. In Paragraph 7, The above-mentioned dividing step is, A step of extracting data of design variables by performing a pre-set method on the design variables of a set multivariate problem; and A method for optimizing variables using a surrogate model based on kriging and machine learning, comprising the step of performing a sensitivity analysis on the extracted data of the design variables and dividing the design variables into a first design variable and a second design variable at a preset ratio according to the order of high sensitivity.
  9. In Paragraph 7, The step of optimizing the first variable above is, A step of extracting multiple data of the first design variable by applying a data extraction method previously set to the first design variable; A step of generating a first surrogate model based on the kriging using the extracted data; and A method for optimizing variables using a Kriging and machine learning-based surrogate model, comprising the step of deriving an optimal solution for the first design variable using the first surrogate model and a previously established optimization algorithm.
  10. In Paragraph 9, The step of deriving an optimal solution for the first design variable above is, A variable optimization method using a kriging and machine learning-based surrogate model, wherein the second design variable is set as an initial value, and if the optimal solution of the first design variable derived does not satisfy the first constraint, the process of deriving the optimal solution for the first design variable using the first surrogate model and the previously established optimization algorithm is repeated.
  11. In Paragraph 7, The step of optimizing the second variable above is, A step of extracting multiple data of the second design variable by applying a data extraction method previously set to the second design variable; A step of generating a second surrogate model based on machine learning using the extracted data; and A method for optimizing variables using a Kriging and machine learning-based surrogate model, comprising the step of deriving an optimal solution for the second design variable using the second surrogate model and a previously established optimization algorithm.
  12. In Paragraph 11, The step of deriving an optimal solution for the above-mentioned second design variable is, A variable optimization method using a kriging and machine learning-based surrogate model, wherein the first design variable is set to an optimized value, and if the optimal solution of the derived second design variable does not satisfy the second constraint, the process of deriving the optimal solution for the second design variable using the second surrogate model and the previously established optimization algorithm is repeated.

Description

System for optimizing values based on kriging and machine-learning and method thereof The present invention relates to a variable optimization system and method using a Kriging and machine learning-based surrogate model. More specifically, it relates to a variable optimization system and method using a Kriging and machine learning-based surrogate model that sequentially optimizes variables of a multivariate problem using a Kriging and machine learning-based surrogate model. There are various structural design parameters for configuring the shapes of the rotor and stator of an electric motor, and the performance of the motor (e.g., average torque, torque ripple, back EMF THD, cogging torque, etc.) changes depending on the changes in these design parameters. In other words, since electric motor design is a multivariate problem with nonlinear characteristics, it is difficult for humans to directly optimize the structural design parameters that determine the shapes of the rotor and stator. On the other hand, while surrogate models using kriging have the advantage of higher accuracy compared to machine learning surrogate models when using the same amount of data, they have limitations in that generating the surrogate model requires a significant amount of time as the number of input variables increases, leading to reduced efficiency and heavy load. In addition, while machine learning-based surrogate models have the advantage of requiring less time and load than kriging-based models, they have the limitation of requiring significantly more data to achieve accuracy similar to kriging. Therefore, there is a need for techniques to optimize variables in multivariate problems. The technology forming the background of the present invention is described in Korean Published Patent No. 10-2023-0057171 (published April 28, 2023). FIG. 1 is a configuration diagram of a variable optimization system according to one embodiment of the present invention. FIG. 2 is a flowchart of a variable optimization method using a kriging and machine learning-based surrogate model according to another embodiment of the present invention. FIG. 3 is a drawing illustrating an initial model of an electric motor according to another embodiment of the present invention. FIG. 4 is a diagram illustrating the results of a sensitivity analysis according to another embodiment of the present invention. FIG. 5 is a drawing comparing the shape of an initial model and an optimal model according to another embodiment of the present invention. FIG. 6 is a flowchart of a genetic algorithm according to another embodiment of the present invention. FIG. 7 is a diagram illustrating the results of performing a sampling technique according to another embodiment of the present invention. FIG. 8 is a diagram comparing the torque performance, line back EMF, and cogging torque of an initial model and an optimal model according to another embodiment of the present invention. Preferred embodiments according to the present invention will be described in detail below with reference to the attached drawings. In this process, the thickness of lines or the size of components shown in the drawings may be exaggerated for clarity and convenience of explanation. Furthermore, the terms described below are defined in consideration of their functions within the present invention, and these may vary depending on the intent or practice of the user or operator. Therefore, the definitions of these terms should be based on the content throughout this specification. In the embodiments described below, the variable optimization system (100) is specifically described as optimizing a plurality of variables to implement the rotor and stator shapes of the electric motor. However, the present invention is not limited thereto and can be used for multivariate problems designed using multiple variables. FIG. 1 is a configuration diagram of a variable optimization system according to one embodiment of the present invention. As illustrated in FIG. 1, the variable optimization system (100) may include a sensitivity analysis unit (110), a first variable optimization unit (120), and a second variable optimization unit (130). First, the sensitivity analysis unit (110) can divide the design variable into a first design variable and a second design variable according to sensitivity by using a sensitivity analysis (e.g., signal-to-noise ratio (SNR)) that is set for the design variable of the set multivariate problem. Specifically, the sensitivity analysis unit (110) can perform a preset sensitivity analysis (e.g., signal-to-noise ratio) on the design variables of a set multivariate problem and divide the design variables into a first design variable (a design variable with a sensitivity of the top 30% and a design variable with a sensitivity of the bottom 70%) according to the order of high sensitivity and a preset ratio (e.g., a design variable with a sensitivity of the top 30% and a design variable with