Parameter optimization system, method and user interface for adaptive search space adjustment

Abstract

The present disclosure provides a parameter optimization system, method and user interface for adaptive search space adjustment. The parameter optimization system includes a data acquisition unit, an adaptive adjustment unit and an optimization search unit. The data acquisition unit captures multiple sets of test values ​​of multiple operating parameters and a target parameter. The adaptive adjustment unit includes a parameter space converter and a search range definer. The parameter space converter performs a space conversion on a parameter space of these operating parameters based on these test values. The search range definer defines a parameter search range with the converted parameter space based on these test values. The optimization search unit searches for a set of recommended values ​​of these operating parameters with the parameter search range as a restriction condition and with the optimization of the target parameter as the goal.

Description

Technical Field

[0001] The present disclosure relates to a parameter optimization system, method and user interface for adaptive search space adjustment. Background Art

[0002] In the industry, it is necessary to search for recommended values ​​of operating parameters with the goal of optimizing target parameters. Since the operating parameters are highly correlated, it is not easy to define an effective parameter search range. If the defined parameter search range is too narrow, it may converge to a regional solution that is not the best solution. If the defined parameter search range is too wide, the number of trials must be quite large to converge to the best solution. Many industries have few process equipment and the materials are quite expensive. If the number of trials is too large, it will not only affect the product delivery time, but also cause a lot of cost waste.

[0003] Therefore, in cases where operating parameters are highly correlated, researchers are actively studying how to define an effective parameter search range to accelerate parameter optimization. Summary of the invention

[0004] The present disclosure relates to a parameter optimization system, method and user interface for adaptive search space adjustment.

[0005] According to one embodiment of the present disclosure, a parameter optimization system with adaptive search space adjustment is proposed. The parameter optimization system with adaptive search space adjustment includes a data capture unit, an adaptive adjustment unit, and an optimization search unit. The data capture unit is used to capture multiple sets of test values ​​of multiple operating parameters and a target parameter. The adaptive adjustment unit includes a parameter space converter and a search range definer. The parameter space converter is used to perform a space conversion on a parameter space of these operating parameters based on these test values. The search range definer is used to define a parameter search range in the converted parameter space based on these test values. The optimization search unit searches for a set of recommended values ​​of these operating parameters with the parameter search range as a restriction condition and with the optimization of the target parameter as the goal.

[0006] According to another embodiment of the present disclosure, a parameter optimization method for adaptive search space adjustment is proposed. The parameter optimization method for adaptive search space adjustment includes the following steps: capturing multiple sets of test values ​​of multiple operating parameters and a target parameter. Based on these test values, a parameter space of these operating parameters is subjected to a spatial transformation. Based on these test values, a parameter search range is defined in the transformed parameter space. With the parameter search range as a constraint condition and with the goal of optimizing the target parameter, a set of recommended values ​​of these operating parameters is searched.

[0007] According to another embodiment of the present disclosure, a user interface is provided. The user interface includes a classification window, a filtering window, a parameter search range window and a recommendation window. The classification window is used to display multiple groups classified by multiple operating parameters. The filtering window is used to display the filtering results of these operating parameters. The parameter search range window is used to display a parameter search range. The recommendation window is used to display a set of recommended values ​​of these operating parameters.

[0008] In order to better understand the above and other aspects of the present disclosure, embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A block diagram of a parameter optimization system for adaptive search space adjustment in one embodiment is shown.

[0010] Figure 2 Different parameter search ranges are shown.

[0011] Figure 3 A flow chart of a parameter optimization method for adaptive search space adjustment in one embodiment is shown.

[0012] Figure 4 A schematic diagram of step S110 in one embodiment is shown.

[0013] Figure 5 A schematic diagram of step S121 is shown.

[0014] Figure 6 A schematic diagram of step S122 is shown.

[0015] Figure 7 A Gaussian process model of conversion parameters in one embodiment is shown.

[0016] Figure 8 The expected improvement in the sampling function of the conversion parameters is shown.

[0017] Fig. 9 The test value range and parameter search range in one embodiment are shown.

[0018] Fig.10 The parameter search range obtained by executing the process twice is shown.

[0019] Fig.11 The figure shows the comparison between the traditional filtering Bayesian optimization (FBO) technique and the technique disclosed in the present invention for parameter optimization.

[0020] Fig.12 A user interface is shown.

[0021] Description of reference numerals:

[0022] 100: parameter optimization system;

[0023] 110: data acquisition unit;

[0024] 120: Adaptive adjustment unit;

[0025] 121: parameter filter;

[0026] 122: parameter space transformer;

[0027] 123: search scoper;

[0028] 130: Optimize search unit;

[0029] 140: user interface;

[0030] 900: Process equipment;

[0031] A1: Accurate range;

[0032] C1: expected improvement sampling function;

[0033] comp1, comp2: conversion parameters;

[0034] DT: test value;

[0035] G1, G2: groups;

[0036] GP1: Gaussian process model;

[0037] P1, P2, P3, P4, P5: points;

[0038] R0: test value range;

[0039] R1, R2, R4, R6, R7, R8: parameter search range;

[0040] R11, R12: search range;

[0041] Ra, Rb, Rc, RG: parameter search range;

[0042] RV: recommended value;

[0043] S110, S121, S122, S123, S130: steps;

[0044] W1: classification window;

[0045] W2: Filter window;

[0046] W3: Parameter search range window;

[0047] W4: Recommendation window;

[0048] x1, x2, x3: operation parameters;

[0049] y: target parameter. DETAILED DESCRIPTION

[0050] Please refer to Figure 1 , which shows a block diagram of a parameter optimization system 100 for adaptive search space adjustment in one embodiment. The parameter optimization system 100 includes a data acquisition unit 110, an adaptive adjustment unit 120, an optimization search unit 130 and a user interface 140. The adaptive adjustment unit 120 includes a parameter filter 121, a parameter space converter 122 and a search range definer 123. The data acquisition unit 110, the adaptive adjustment unit 120, the parameter filter 121, the parameter space converter 122, the search range definer 123, and the optimization search unit 130 are, for example, a circuit, a chip, a circuit board or a storage device storing program code. The user interface 140 is, for example, a display screen, an input and output device, or a smart phone. The parameter optimization system 100 is used to optimize multiple operating parameters (such as pressure, temperature, flow) for the process equipment 900 (or production line) in order to optimize the target parameter (such as yield). During the optimization process, the parameter space converter 122 of the adaptive adjustment unit 120 can convert the parameter space, and the search range definer 123 of the adaptive adjustment unit 120 can define a suitable parameter search range to improve the efficiency of the optimization process. Figure 2 , which shows different parameter search ranges Ra, Rb, Rc. Figure 2 As shown on the left side of , the traditional parameter search range Ra is fixed to the global, and a considerable number of trials are required to converge to the optimal solution (square point). Figure 2 As shown on the right side of the figure, the present embodiment dynamically adjusts the parameter search range during iteration (for example, the parameter search range Rb is adjusted to the parameter search range Rc after multiple iterations), and can quickly converge to the optimal solution (square point) without too many trials. The operation of each component is also described in detail with a flow chart below.

[0051] Please refer to Figure 3 , which shows a flow chart of a parameter optimization method for adaptive search space adjustment in one embodiment. In step S110, the data acquisition unit 110 acquires multiple sets of test values ​​of multiple operating parameters and target parameters. For example, please refer to Figure 4, which shows a schematic diagram of step S110 in one embodiment. The process equipment 900, for example, has operating parameters x1, x2, x3 and a target parameter y. The operating parameter x1 is, for example, flow rate, the operating parameter x2 is, for example, temperature, and the operating parameter x3 is, for example, pressure. The target parameter y is, for example, yield. The number of operating parameters is not limited to 3, and the number of operating parameters can be 2 or greater than 3. The subsequent steps are explained using the operating parameters x1, x2, x3 and the target parameter y as examples, but it should be emphasized that the present embodiment is not limited to the number of operating parameters. Multiple groups of test values ​​DT (indicated in Figure 1 ) is shown in Table 1 below. The number of test values ​​is not limited to 3, as long as the values ​​that have been tested in the iteration process are included, and thus continuously increased, so that the subsequent optimization process can move towards the best solution.

[0052] Table 1

[0053] Operation parameters x1 Operation parameter x2 Operation parameters x3 Target parameter y 0.0085 2.1414 1.1703 6.0313 0.0573 0.2412 1.0046 3.0650 0.8442 1.2180 0.3925 4.1018

[0054] Next, in step S121, the parameter filter 121 of the adaptive adjustment unit 120 filters the operating parameters x1, x2, and x3 according to the group relationship of the operating parameters x1, x2, and x3. Figure 5 , which shows a schematic diagram of step S121. In this step, the user can classify these operating parameters x1, x2, x3 into groups G1, G2 (the number of groups is not limited to 2) according to the degree of mutual influence through the user interface 140. For example, the operating parameters x2 and x3 have a greater degree of mutual influence and are classified into the same group G2. The operating parameter x1 has a smaller degree of mutual influence with other operating parameters x2 and x3, and is classified into another group G1. Then, the parameter filter 121 filters the target parameter y according to the degree of correlation between each group G1, G2. For example, the parameter filter 121 calculates the operating parameters x1, x2, x3 of groups G1 and G2 through a group lasso regression algorithm (Group-Lasso Regularization). The values ​​are 0, 1.32, and 0.41 respectively. The more the absolute value of the value exceeds 0, the higher the correlation with the target parameter y is. Therefore, the parameter filter 121 can filter out the operation parameters x2 and x3 of the group G2 as the main adjustment objects in the parameter search range.

[0055] In addition to the Group Lasso Regression Algorithm, the parameter filter 121 may also filter the operation parameters x1, x2, x3 by a Sparse Group Lasso Algorithm, a Bayesian Group Lasso Algorithm, a Composite Absolute Penalty Algorithm or a Group Least Angle Regression Selection Algorithm.

[0056] In addition, the number of operating parameters in each group is not limited to 1 or 2 (may be greater than or equal to 3). Also, the number of operating parameters in each group does not need to be the same.

[0057] In this step, the operating parameters x1, x2, and x3 are classified according to the degree of mutual influence, so that the operating parameters x2 and x3 that are easy to influence each other can be considered together, and the mutual influence degree of the operating parameters x2 and x3 is reduced through subsequent spatial transformation. In addition, the groups G1 and G2 are screened according to the degree of association with the target parameter y to leave the group G2 that has the greatest influence on the target parameter y, so that the optimization process can converge more quickly.

[0058] Then, in step S122, the parameter space converter 122 converts the test values ​​DT (indicated in Figure 1 ), a space transformation is performed on the parameter space of these operation parameters x2 and x3. Figure 6 , which shows a schematic diagram of step S122. For example, Figure 6 The solid line coordinate axes are the operating parameters x2 and x3 of the original parameter space. Figure 6 The dotted coordinate axes are the conversion parameters comp1 and comp2 of the converted parameter space. The parameter space converter 122 performs space conversion by, for example, a partial least squares regression algorithm. In the partial least squares regression algorithm, as shown in the following formula (1), X is a matrix recording the operation parameters x2 and x3, S is a matrix recording the conversion parameters comp1 and comp2, and w is a conversion matrix.

[0059] S=Xw................................................(1)

[0060] in, X=[x2,x3].

[0061] In the partial least squares regression algorithm, the main requirement is to obtain w so that the correlation between the conversion parameter comp1 and the conversion parameter comp2 for the target parameter y is maximized, that is, the following formula (2).

[0062] max w {[cov(Xw,y)] 2}=max w {var(Xw)[corr(Xw,y)] 2 var(y)}... (2)

[0063] The conversion parameters comp1 and comp2 have an orthogonal relationship, so that the mutual influence of the parameters can be reduced. The conversion parameters comp1 and comp2 can maintain the variation degree of the original operation parameters x2 and x3, so that the recognition degree of the target parameter y can be improved.

[0064] The relationship between the conversion parameters comp1, comp2 and the operation parameters x2, x3 is as follows.

[0065]

[0066] in,

[0067] The above formula (3) can also be expressed as the following formulas (4) and (5).

[0068] comp1=0.987*x2+0.159*x3........................(4)

[0069] comp2=-0.159*x2+0.987*x3........................(5)

[0070] As shown in Table 2 and Table 3 below, Table 2 shows the operation parameters x2 and x3. After the spatial transformation, the transformation parameters comp1 and comp2 in Table 3 can be obtained.

[0071] Table 2

[0072] Operation parameter x2 Operation parameters x3 2.1414 1.1703 0.2412 1.0046 1.2180 0.3925

[0073] Table 3

[0074] Conversion parameter comp1 Conversion parameter comp2 2.300 0.815 0.398 0.953 1.265 0.194

[0075] In addition to the above-mentioned partial least squares regression algorithm, the parameter space transformer 122 can also perform space transformation through a principal component analysis (PCA) algorithm.

[0076] Next, in step S123, the search range definer 123 of the adaptive adjustment unit 120 determines the search range according to the test value DT (indicated in Figure 1 ), the converted parameter space defines a parameter search range RG (indicated by Figure 1 ). This step includes the following procedures. Please refer to Figure 7 , which shows a Gaussian process model GP1 of the conversion parameter comp1 in one embodiment. Figure 7 As shown, the search range definer 123 uses the test value DT to establish the Gaussian process model GP1 of the conversion parameter comp1. The Gaussian process model GP1 shows inaccuracy. The closer to the test value DT, the lower the inaccuracy; the farther away from the test value DT, the higher the inaccuracy.

[0077] Next, the search range definer 123 reserves an accurate range A1, such as -15 to 25, according to the Gaussian process model GP1.

[0078] Then, please refer to Figure 8 , which shows the expected improvement sampling function (ExpectedImprovement function, EI function) C1 of the conversion parameter comp1. In the expected improvement sampling function C1, the vertical axis is the expected improvement value (ExpectedImprovement, EI), and the horizontal axis is the accurate range of the conversion parameter comp1. The search range definer 123 then defines the search range R11 in the accurate range A1 according to the expected improvement value. Figure 8 As shown, when the conversion parameter comp1 is in the interval of 0.40 to 2.30 where the test value DT is located, the expected improvement value is 0. When the conversion parameter comp1 moves from 0.40 to -2.50, the expected improvement value can be greatly increased to 0.62. When the conversion parameter comp1 moves from 2.3 to 25, the expected improvement value can only be slightly increased to 0,004. It can be seen that a greater improvement can be obtained by moving the conversion parameter comp1 in the negative direction. Therefore, the search range definer 123 can define the interval of the conversion parameter comp1 in the range of -2.50 to 0.40 as the search range R11, that is, the following formula (6).

[0079] -2.50≤0.987*x2+0.159*x3≤0.40........................ (6)

[0080] Likewise, in a similar manner, the search range definer 123 may define the conversion parameter comp2 in the range of 0.19 to 0.95 as a search range R12 (as shown in FIG. 9 ).

[0081] 0.19≤-0.159*x2+0.987*x3≤0.95........................ (7)

[0082] Please refer to Fig. 9 , which shows the test value range R0 and the parameter search range R1 in one embodiment. The tested test value DT includes point P1, point P2 and point P3. Point P1, point P2 and point P3 are located in the test value range R0. The search range definer 123 further defines a search range R11 in the conversion parameter comp1 based on the test value range R0, and defines a search range R12 in the conversion parameter comp2 to form the parameter search range R1. The parameter search range R1 is a range that has a high chance of converging toward the optimal solution.

[0083] Next, in step S130, the optimization search unit 130 uses the parameter search range R1 as a constraint condition and optimizes the target parameter y as a target to search for a set of recommended values ​​RV (indicated in FIG. Figure 1 ), for example, point P4 in FIG. 9. For example, the optimization search unit 123 may search for the recommended values ​​RV of the operating parameters x1, x2, and x3 by a restricted Bayesian optimization algorithm, as shown in the following formula (8).

[0084]

[0085] After obtaining the recommended values ​​RV of the operating parameters x1, x2, and x3, they can be provided to the process equipment 900 (or production line) to obtain the corresponding target parameter y. If the search for the best solution has not yet converged, this set of values ​​is added to the test value DT (indicated in Figure 1 ), and re-execute steps S110, S121, S122, S123, and S130. During the re-execution, the grouping result and the screening result of step S121 may change or remain unchanged. The operating parameters of the space conversion and parameter search range in steps S122 and S123 may also change or remain unchanged.

[0086] Please refer to Fig.10 The process shown in FIG. 1 is executed twice to obtain the parameter search range R1 and the parameter search range R2. Fig.10As shown, the initial test values ​​DT are point P1, point P2, and point P3, and points P1, point P2, and point P3 are within the test value range R0. After the process is executed once, the parameter search range R1 is obtained, and point P4 is recommended in the parameter search range R1. After the process is executed for the second time, the parameter search range R2 is obtained (step S121 also screens out the operating parameters x2 and x3), and point P5 is recommended in the parameter search range R2. Fig.10 In the example, the first execution process and the second execution process both filter out the operating parameters x2 and x3, so the parameter search range R1 and the parameter search range R2 can be drawn on the same graph. If the first execution process and the second execution process do not both filter out the operating parameters x2 and x3, the parameter search range R1 and the parameter search range R2 cannot be drawn on the same graph.

[0087] In another embodiment, the step S121 may be omitted, and all the operating parameters x1, x2, and x3 are used as adjustment objects of the parameter search range.

[0088] Please refer to Fig.11 , which shows the comparison between the traditional filtering Bayesian optimization (FBO) technique and the disclosed technique for parameter optimization. The researchers initially performed the FBO technique and the disclosed technique with the same experimental value DT. Fig.11 It can be found that the recommended values ​​of the FBO technology in the optimization process are scattered everywhere, and it takes a considerable number of iterations to converge to the best solution. The disclosed technology quickly searches for the best solution after adjusting from the parameter search range R6 to the parameter search range R7 and then to the parameter search range R8.

[0089] The researchers also compared the disclosed technology with several other traditional technologies under the same experimental conditions, and summarized the number of iterations that converged to the optimal solution as shown in Table 4. Method 1 is the traditional Bayesian optimization (BO), method 2 is the PCA-BO technology, method 3 is the FBO technology, method 4 is the disclosed technology with step S121 omitted, and method 5 is the disclosed technology with step S121 not omitted.

[0090] Table 4

[0091] Method 1 Method 2 Method 3 Method 4 Method 5 Experimental condition 1 59 times 79 times 43 times 27 times 16 times Experimental condition 2 67 times >100 times 84 times 48 times 41 times

[0092] It is obvious that no matter what the experimental conditions are, the disclosed technology can quickly converge to the optimal solution.

[0093] In the above process, the user can learn about the operation of the parameter optimization system 100 through the user interface 140. Fig.12, which shows a user interface 140. The user interface 140 includes a classification window W1, a screening window W2, a parameter search range window W3 and a recommendation window W4. The classification window W1 is used to display a plurality of groups G1, G2 (indicated in FIG. 1 ) into which the operation parameters x1, x2, x3 are classified. Figure 5 ).like Fig.12 As shown, in the classification window W1, "1" represents group G1 and "2" represents group G2. The screening window W2 is used to display the screening results of the operation parameters x1, x2, and x3. Fig.12 As shown, the filter window W2 displays the operation parameters x1, x2, x3. The values ​​are 0, 1.7, and 0.5 respectively. The more the absolute value of the value exceeds 0, the higher the correlation with the target parameter y is. Therefore, the operating parameters x2 and x3 of group G2 can be selected as the main adjustment objects of the parameter search range. The parameter search range window W3 is used to display the parameter search range R4. The parameter search range R4 is the range after spatial conversion, rather than the individual simple range of the operating parameters x2 and x3. Fig.12 As shown, the parameter search range R4 is a range limit of a linear combination of the operating parameters x2 and x3. In addition, when the operating parameters x1, x2, and x3 have been filtered, the parameter search range R4 displayed in the parameter search range window W3 only corresponds to part of the operating parameters x2 and x3. The recommendation window W4 is used to display a set of recommended values ​​of the operating parameters x1, x2, and x3, such as "0.01, 0.22, and 0.83".

[0094] Through the above embodiments, the parameter search range can be adaptively adjusted through the space conversion technology, and can quickly converge to the optimal solution without too many trials. In addition, when there are too many operating parameters, the operating parameters that need to be space converted and the search range restricted can be appropriately screened, which can also improve the efficiency of parameter optimization search.

[0095] In summary, although the present disclosure has been described above with embodiments, it is not intended to limit the present disclosure. A person skilled in the art of the present disclosure can make various modifications and improvements without departing from the spirit and scope of the present disclosure. Therefore, the scope of protection of the present disclosure shall be subject to the scope of protection of the claims.

Claims

1. A parameter optimization system for adaptive search space adjustment, characterized in that: The parameter optimization system includes a process equipment, a processing circuit and a storage device, wherein the processing circuit is connected to the process equipment, and the storage device stores program codes, which are executed by the processing circuit to optimize a plurality of process parameters of the process equipment, wherein the processing circuit executes the program codes to perform: a data acquisition unit for acquiring a plurality of test values ​​of a plurality of operating parameters and a target parameter; An adaptive adjustment unit, comprising: a parameter space converter for performing a space conversion on a parameter space of the operating parameters according to the test values ​​to obtain the converted parameter space, wherein a plurality of conversion parameters of the converted parameter space have an orthogonal relationship; and a search range definer for defining a parameter search range in the converted parameter space according to the test values; and an optimization search unit, which uses the parameter search range as a constraint condition and optimizes the target parameter to search for a set of recommended values ​​of the operating parameters; The set of recommended values ​​is transmitted to the process equipment to obtain the target parameter.

2. The parameter optimization system for adaptive search space adjustment according to claim 1, characterized in that: The adaptive adjustment unit also includes: A parameter filter is used to filter the operation parameters according to the group relationship of the operation parameters to provide them to the parameter space converter for performing the space conversion.

3. The parameter optimization system for adaptive search space adjustment according to claim 2, characterized in that: The operating parameters are classified into multiple groups according to the degree of mutual influence.

4. The parameter optimization system for adaptive search space adjustment according to claim 3, characterized in that: The parameter filter filters the target parameter according to a correlation degree of each group.

5. The parameter optimization system for adaptive search space adjustment according to claim 2, characterized in that: The parameter filter selects the operating parameters by a grouped lasso regression algorithm, a sparse grouped lasso algorithm, a Bayesian grouped lasso algorithm, a comprehensive absolute penalty algorithm or a grouped minimum angle regression algorithm.

6. The parameter optimization system for adaptive search space adjustment according to claim 1, characterized in that: The parameter space converter utilizes the space conversion to reduce a parameter mutual influence degree of the operation parameters and improve a recognition degree of the target parameter of the operation parameters.

7. The parameter optimization system for adaptive search space adjustment according to claim 1, characterized in that: The parameter space transformer performs the space transformation by a partial least squares regression algorithm or a principal component analysis algorithm.

8. The parameter optimization system for adaptive search space adjustment according to claim 1, characterized in that: The search range definer uses the test values ​​to establish a Gaussian process model, retains an accurate range according to the Gaussian process model, and defines the parameter search range in the accurate range according to an expected improvement sampling function.

9. The parameter optimization system for adaptive search space adjustment according to claim 1, characterized in that: The optimization search unit searches for the set of recommended values ​​of the target parameters by using a constrained Bayesian optimization algorithm.

10. A parameter optimization method for adaptive search space adjustment, characterized in that: A processing circuit is connected to a process equipment, and a program code stored in a storage device is executed by the processing circuit to optimize a plurality of process parameters of the process equipment after executing the parameter optimization method. The parameter optimization method includes: Retrieving multiple test values ​​of multiple operating parameters and a target parameter; According to the test values, a parameter space of the operation parameters is spatially transformed to obtain the transformed parameter space, wherein a plurality of transformation parameters of the transformed parameter space have an orthogonal relationship; According to the test values, a parameter search range is defined in the converted parameter space; Using the parameter search range as a constraint condition and optimizing the target parameter as a goal, searching for a set of recommended values ​​of the operating parameters; and The set of recommended values ​​is transmitted to the process equipment to obtain the target parameter.

11. The parameter optimization method for adaptive search space adjustment according to claim 10, characterized in that: Also includes: The operation parameters are screened according to the group relationship of the operation parameters to perform the space conversion.

12. The parameter optimization method for adaptive search space adjustment according to claim 11, characterized in that: In the step of screening the operating parameters, the operating parameters are classified into a plurality of groups according to the mutual influence degree of the operating parameters.

13. The parameter optimization method for adaptive search space adjustment according to claim 12, characterized in that: In the step of screening the operating parameters, screening is performed according to a correlation degree of each group to the target parameter.

14. The parameter optimization method for adaptive search space adjustment according to claim 11, characterized in that: In the step of selecting the operating parameters, the operating parameters are selected by a grouping lasso regression algorithm, a sparse grouping lasso algorithm, a Bayesian grouping lasso algorithm, a comprehensive absolute penalty algorithm or a grouping minimum angle regression algorithm.

15. The parameter optimization method for adaptive search space adjustment according to claim 10, characterized in that: In the step of performing the spatial transformation on the operating parameters, the spatial transformation is utilized to reduce a mutual influence degree of the operating parameters and to improve a recognition degree of the target parameter by the operating parameters.

16. The parameter optimization method for adaptive search space adjustment according to claim 10, characterized in that: In the step of performing the spatial transformation on the operating parameters, the spatial transformation is performed by a partial least squares regression algorithm or a principal component analysis algorithm.

17. The parameter optimization method for adaptive search space adjustment according to claim 10, characterized in that: In the step of defining the parameter search range, a Gaussian process model is established using the test values, and an accurate range is retained according to the Gaussian process model. Then, the parameter search range is defined in the accurate range according to an expected improvement sampling function.

18. The parameter optimization method for adaptive search space adjustment according to claim 10, characterized in that: In the step of searching for the set of recommended values ​​of the target parameters, the set of recommended values ​​of the target parameters is searched by a constrained Bayesian optimization algorithm.

Images