Parameter optimization device and parameter optimization method using it

Abstract

The parameter optimization device includes a data extraction module, a sampling function calculation module, a clustering module, and a parameter recommendation module. The data extraction module extracts multiple input parameter values ​​and their corresponding multiple measurement output values. The sampling function calculation module generates multiple sampling function values ​​based on the input parameter values ​​and the measurement output values. The clustering module generates multiple parameter value clusters based on the input parameter values ​​and the sampling function values. The parameter recommendation module generates multiple recommended parameter values ​​from at least one of the parameter value clusters.

Description

Technical Field

[0001] This invention discloses a parameter optimization device and a parameter optimization method using the same. Background Technology

[0002] In current parameter optimization methods, a process parameter is output one at a time, input into the process, and the corresponding process result is measured. If the process result does not meet the process target value, the next process parameter can be recommended until the process result corresponding to that parameter meets the process target value. However, this method of process parameter recommendation is quite time-consuming and inefficient. Therefore, proposing a new process parameter recommendation mechanism is one of the goals of those skilled in the art. Summary of the Invention

[0003] According to an embodiment of this disclosure, a parameter optimization apparatus is proposed. The parameter optimization apparatus includes a data extraction module, a sampling function calculation module, a clustering module, and a parameter recommendation module. The data extraction module is used to extract multiple input parameter values ​​and corresponding multiple measurement output values. The sampling function calculation module is used to generate multiple sampling function values ​​based on the input parameter values ​​and the measurement output values. The clustering module is used to generate multiple parameter value groups based on the input parameter values ​​and the sampling function values. The parameter recommendation module is used to generate multiple recommended parameter values ​​from at least one of the parameter value groups.

[0004] According to another embodiment of this disclosure, a parameter optimization method is proposed. The parameter optimization method includes the following steps: extracting multiple input parameter values ​​and corresponding multiple measurement output values; generating multiple sampling function values ​​based on the input parameter values ​​and the measurement output values; generating multiple parameter value groups based on the input parameter values ​​and the sampling function values; and generating multiple recommended parameter values ​​from at least one of the parameter value groups.

[0005] To provide a better understanding of the above and other aspects of this disclosure, specific embodiments are described below in conjunction with the accompanying drawings: Attached Figure Description

[0006] Figure 1 This is a functional block diagram of a parameter optimization device according to an embodiment of the present disclosure.

[0007] Figure 2 for Figure 1 The flowchart of the parameter optimization device.

[0008] Figure 3 This is a schematic diagram of a sampling function according to an embodiment of the present disclosure.

[0009] [Explanation of Labels in the Attached Image]

[0010] 1-7: punctuation marks;

[0011] 10: Process equipment;

[0012] 100: Parameter optimization device;

[0013] 110: Data extraction module;

[0014] 120: Sampling function calculation module;

[0015] 130: Clustering module;

[0016] 140: Parameter recommendation module;

[0017] EI(x) j : Sampling function value;

[0018] EI: Sampling function;

[0019] G j(1) Group 1;

[0020] G j(2) Group 2;

[0021] G j(3) Group 3;

[0022] G j : Group of parameter values;

[0023] M1: Optimization completion information;

[0024] p: Improvement tolerance;

[0025] S110~S160: Steps;

[0026] T1: Threshold value;

[0027] V j : Neighbor variation value;

[0028] W j Sample weights;

[0029] x: numerical value;

[0030] X' j+1 Recommended parameter values;

[0031] X j : Input parameter value;

[0032] Y' j+1 Recommended output value;

[0033] Y j : Measurement output value;

[0034] μ(x): The average function;

[0035] ymax : Maximum measured output value;

[0036] Gaussian cumulative probability distribution function;

[0037] σ(x): Standard deviation function. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0039] Please refer to Figure 1 This is a functional block diagram of a parameter optimization device 100 according to an embodiment of the present disclosure.

[0040] The parameter optimization device 100 includes a data extraction module 110, a sampling function calculation module 120, a clustering module 130, and a parameter recommendation module 140.

[0041] The data extraction module 110, sampling function calculation module 120, clustering module 130, and / or parameter recommendation module 140 are, for example, software, firmware, and / or hardware. In one embodiment, the data extraction module 110, sampling function calculation module 120, clustering module 130, and / or parameter recommendation module 140 are, for example, physical circuits formed using semiconductor processes. At least two of the data extraction module 110, sampling function calculation module 120, clustering module 130, and parameter recommendation module 140 can be integrated into a single module, or at least one of the data extraction module 110, sampling function calculation module 120, clustering module 130, and parameter recommendation module 140 can be integrated into a processor or a controller.

[0042] Data extraction module 110 is used to extract multiple input parameter values ​​X. j and the corresponding multiple measurement output values ​​Y j The sampling function calculation module 120 is used to calculate the input parameter values ​​X based on these input parameter values. j and these measurement output values ​​Y j This generates multiple sampling function values ​​EI(x). j The grouping module 130 is used to group based on these input parameter values ​​X. j and the sampling function values ​​EI(x) j This generates multiple parameter value groups G. j The parameter recommendation module 140 is used to select from the parameter value group G j At least one of them generates multiple recommended parameter values ​​X' j+1 In this embodiment, multiple recommended parameter values ​​X' are generated at once. j+1Therefore, multiple recommended parameter values ​​X' can be generated through measurement in a single measurement batch. j+1 Multiple measurement output values ​​(process results) can be obtained, which can reduce the number of measurement batches and thus reduce the time of the entire optimization process, and quickly generate the best recommended parameter values ​​that meet the output target value (process target).

[0043] The subscript j indicates the batch (or iteration number). For example, the input parameter value X j and measurement output value Y j This represents the value for the current batch (or the current iteration), while the recommended parameter value is X'. j+1 This indicates the value to be used in the next batch (or the next iteration).

[0044] In the embodiment, each input parameter value X j For example, it could be a numerical value representing a process parameter type. Process parameter types might include various process parameters such as flow rate, pressure, temperature, humidity, and additive ratios. The process itself could be, for example, mechanical design, mechanical manufacturing, semiconductor manufacturing, chemical manufacturing, or any other process that requires or can perform parameter optimization. The input parameter value X... j The measured value of the process result is the measurement output value Y. j Examples of indicators that represent process quality include epitaxial layer thickness, process yield, minimum cost, maximum efficiency, and productivity. Recommended parameter value X' j+1 The recommended parameter values ​​after executing the current batch optimization process can be added to the calculations of the next batch optimization process. Additionally, the input parameter value X... j The number and / or recommended parameter value X' j+1 The number can be determined based on the measurement capability of the process, for example, when the process equipment 10 (process equipment 10 in) Figure 1 When capable of measuring up to N output values ​​in a batch, the input parameter value X... j The number and / or recommended parameter value X' j+1 The number can be equal to or less than N, where N is any positive integer equal to or greater than 2. Furthermore, the input parameter value X... j The number and / or recommended parameter value X' j+1 The number can be equal. In one embodiment, the process equipment 10 itself has measurement function; in another embodiment, the process equipment 10 itself does not have measurement function, but can also have measurement function through external measurement equipment.

[0045] Please refer to Figure 2 This is a flowchart of the parameter optimization device 100 in Figure 1.

[0046] In step S110, the data extraction module 110 extracts multiple input parameter values ​​X.j and the corresponding multiple measurement output values ​​Y j In each batch (iteration), there are at least two input parameter values ​​X. j and at least two measurement output values ​​Y j The number of input parameter values ​​and measurement output values ​​participating in the optimization calculation is not limited in this embodiment. The more input parameter values ​​there are, the faster the optimization process can be completed (i.e., the fewer iterations).

[0047] In step S120, the sampling function calculation module 120 calculates the input parameter values ​​X based on these input parameter values ​​X. j and these measurement output values ​​Y j This generates multiple sampling function values ​​EI(x). j In this embodiment, the sampling function value is illustrated using the "expected improvement value" as an example.

[0048] like Figure 3 The diagram illustrates a sampling function EI according to an embodiment of this disclosure. The horizontal axis represents the distribution of the numerical values ​​x in the input parameter value space, while the vertical axis represents the sampling function value EI(x) corresponding to the numerical value x. j The numerical value x can be any integer. The sampling function EI, for example, is based on multiple sampling function values ​​EI(x). j The curve that was drawn.

[0049] The sampling function EI can be generated by the following equation (1). EI(x) in equation (1) j μ(x) represents the sampling function value corresponding to the numerical value x in the input parameter value space, and μ(x) represents the value of the sampling function based on these input parameter values ​​x. j and these measurement output values ​​Y j The resulting mean function, y max This indicates multiple measurement output values ​​Y in the current batch. j The largest of the three, p represents the improvement tolerance value (any integer, not limited by the embodiments of this disclosure), and Z is (μ(x)-y). max ) / σ(x), This indicates that based on these input parameter values ​​X j and these measurement output values ​​Y j The resulting Gaussian cumulative probability distribution function, This indicates that based on these input parameter values ​​X j and these measurement output values ​​Y j The resulting Gaussian probability distribution function, σ(x), represents the probability distribution based on the input parameter values ​​X. j and these measurement output values ​​Y j The resulting standard deviation function. In addition, the mean function μ(x) and the Gaussian probability distribution function. The standard deviation function σ(x) can be obtained using general mathematical and / or statistical methods, and is not limited to the embodiments disclosed herein.

[0050]

[0051] The sampling function calculation module 120 can first use appropriate mathematical and / or statistical methods to calculate the input parameter values ​​X. j and these measurement output values ​​Y j To generate the mean function μ(x) and the Gaussian cumulative probability distribution function. Gaussian probability distribution function And the standard deviation function σ(x), and then according to equation (1), calculate the mean function μ(x) and the Gaussian cumulative probability distribution function. Gaussian probability distribution function Standard deviation function σ(x), maximum measured output value y max And improve the tolerance value p to generate the sampling function EI.

[0052] In step S130, the grouping module 130 determines the grouping module based on the input parameter values ​​X. j and the sampling function values ​​EI(x) j This generates multiple parameter value groups G. j For example, such as Figure 3 As shown, the parameter optimization device 100 or the clustering module 130 can select multiple different numerical points as "punctuation marks" from the sampling function EI, and perform clustering on these "punctuation marks". This embodiment does not limit the number or value of the punctuation marks. In one embodiment, the clustering module 130 can determine (or select) these punctuation marks based on a specific number (e.g., random selection), or determine the punctuation marks based on the intervals of the numerical value x (e.g., taking one punctuation mark at each fixed interval of the numerical value x).

[0053] In this embodiment, please also refer to Figure 3 And Table 1 below, which lists the 7 punctuation marks in the j-th batch and the numerical value x and sampling function value EI(x) for each punctuation mark. j Is it a candidate point? Neighbor variation value V j Sample weights W j And the corresponding subgroup, etc. Table 1 is only one example of this disclosure. The number of punctuation marks and / or the values ​​of each parameter in different embodiments will vary depending on the actual situation and are not limited to Table 1. Neighbor variation value V j EI(x) represents the sampling function value of a punctuation mark. j Compared to the sampled function value EI(x) of multiple neighboring (K-value) points. j The degree of variation. The greater the degree of variation, the more likely it is to be a neighboring value of variation V. j The larger the value, the smaller the value; conversely, the smaller the value, the smaller the value. These neighboring variation values ​​V jFor example, the sampling function calculation module 120 uses, for example, a statistical method for calculating variance, based on the sampling function values ​​EI(x). j The resulting sample weights W for punctuation. j This represents the sampling function value EI(x). j Compared with neighboring variation values ​​V j The product value (i.e., W) j =EI(x) j ×V j ).

[0054] Table 1

[0055]

[0056] There are several ways to determine the group of parameter values; some of them are described below.

[0057] The first clustering method uses the "sampling function value" as the weight for clustering. For example, the clustering module 130 can use, for instance, the K-means method, based on the input parameter value X of each point. j and the sampling function value EI(x) j These punctuation marks are then grouped. Specifically, in this embodiment of the disclosure, the sampling function value EI(x) is used. j Using the difference in numerical values ​​x as weights, we determine whether two punctuation marks belong to the same group. The difference between two punctuation marks can be interpreted as their distance; a larger distance indicates a greater degree of difference, and vice versa. This is further determined by the difference in numerical values ​​x and the sampling function value EI(x). j The method employs a dual-judgment approach to group parameters, generating diverse parameter value clusters. "Diversity" in this paper refers to multiple clusters exhibiting a certain degree of difference; for example, the parameter values ​​within a cluster show little difference, while the parameter values ​​between clusters show significant difference. Any two of the recommended parameter values ​​determined from at least one of these parameter value clusters exhibit a certain degree of difference.

[0058] The second grouping method uses "nearest neighbor variation" as the weight for grouping. For example, the grouping module 130 can use, for instance, the K-means method, based on the input parameter value X of each point. j and neighboring variation value V j These punctuation marks are then grouped. Specifically, in this embodiment, the nearest neighbor variation value V is used. j The sample weight is determined based on the difference between the values ​​x of any two punctuation points and the sampling function value EI(x). j The difference in value x is used to determine whether the two punctuation marks belong to the same group. This is based on the difference in numerical value x and the sampling function value EI(x). j The recommended parameter values ​​generated by grouping based on multiple judgments of differences are quite diverse.

[0059] The third grouping method uses the "sampling function value" and the "neighbor variation value" as weights for grouping. For example, the grouping module 130 can use, for instance, the K-means method, based on the input parameter value X of each point. j Sampling function value EI(x) j Neighbor variation value V j These punctuation marks are then grouped. Specifically, in this embodiment of the disclosure, the sampling function value EI(x) is used. j Compared with neighboring variation values ​​V j The product value is weighted based on the difference between the values ​​x of any two punctuation points and the sampling function value EI(x). j Differences and neighboring variances V j Determine whether these two punctuation marks belong to the same group. This is based on the difference in numerical values ​​x and the sampling function value EI(x). j Differences and neighboring variances V j The recommended parameter values ​​generated by clustering are quite diverse.

[0060] The grouping method in this embodiment is illustrated using the third method described above as an example. For example... Figure 3 As shown, according to the third grouping method, the grouping module 130 divides these punctuation marks into three-parameter value groups, such as the first group G. j(1) Group G, the second group j(2) and the third group G j(3) .

[0061] Furthermore, the clustering module 130 can cluster some (not all) of these punctuation marks. As shown in Table 1 above, the clustering module 130 clusters multiple sampled function values ​​EI(x) that meet the threshold value T1. j (Referred to as "candidate points") are grouped. For example, grouping module 130 only groups points 1-2, 4, and 6-7 (candidate points) that are equal to or higher than the threshold value T1, but excludes points 3 and 5 (non-candidate points) that are below the threshold value T1. After the operation, points 1 and 2 belong to the first group G. j(1) Punctuation mark 4 belongs to the second group G. j(2) punctuation marks 6-7 belong to the third group G. j(3) Different numbers of punctuation marks and their values ​​x may produce different clustering results. In another embodiment, the clustering module 130 may also cluster all punctuation marks. Furthermore, this disclosure does not limit the specific value of the threshold T1.

[0062] In step S140, the parameter recommendation module 140 can select from the parameter value group G j From at least one of them, multiple recommended parameter values ​​X' are generated. j+1 Because the input parameter value X for the j-th batch (current batch) j Since the quantity is 3, the recommended parameter value X' for the (j+1)th batch (the next batch) is...j+1 It can also be three, however this is not intended to limit the embodiments of this disclosure. In another embodiment, the recommended parameter value X' for the (j+1)th batch (next batch) is... j+1 It can also be less than or more than 3.

[0063] In one embodiment, the parameter recommendation module 140 can select from each parameter value group G j In this process, multiple sampling function values ​​EI(x) are selected. j The value x corresponding to one of them is used as the recommended parameter value X' j+1 For example, taking Table 1 as an example, the parameter recommendation module 140 can select from the first group G. j(1) Select multiple sampling function values ​​EI(x) j The value x corresponding to the largest (e.g., punctuation mark 2) is used as the recommended parameter value X'. j+1 From the second group G j(2) Select the value x corresponding to the largest sampling function value EI(x)j (e.g., punctuation 4) as the recommended parameter value X'. j+1 And from the third group G j(3) Select the sampling function value EI(x) j The value x corresponding to the largest (e.g., punctuation mark 6) is used as the recommended parameter value X' for the next batch. j+1 For example, the parameter recommendation module 140 can select from each parameter value group G j In this process, multiple sampling function values ​​EI(x) are selected. j The closest to a mean (e.g., the sample function value EI(x)). j The value x corresponding to the average value is used as the recommended parameter value X' for the next batch. j+1 Alternatively, the parameter recommendation module 140 can select from these parameter value groups G j From some (not all) of them, select one or more sampling function values ​​EI(x) )j The corresponding value x is used as the recommended parameter value X' j+1 .

[0064] In step S150, the data extraction module 110 determines multiple recommended parameter values ​​X' j+1 The corresponding recommended output value Y' j+1 Does it meet the stopping optimization condition? Recommended output value Y' j+1 The generation method is the same as the above measurement output value Y. j This will not be elaborated upon further. If the recommended output value is Y' j+1 The optimization stop condition has not yet been met, indicating that the recommended output value Y' is... j+1 There is still room or opportunity to reach the target output value. The process proceeds to step S160 to continue optimizing the process. If the recommended output value Y' is... j+1The optimization stops when the measurement output value Y is met. j There is no longer any space or opportunity to reach the target output value, therefore the process can be terminated, and the data extraction module 110 can issue an optimized termination message M1 (in Figure 1 ), and use the recommended parameter value X' for this batch. j+1 The recommended parameter value is considered the optimal one. Furthermore, the aforementioned stopping optimization condition is, for example: (1). Recommended parameter value X' j+1 The output target value has been reached or exceeded; (2). Recommended parameter value X' j+1 The difference (or absolute value) between the output target value and the target value is within an acceptable range; or (3). The number of iterations (repeating steps S110 to S160 once is one iteration) has reached a preset number. This embodiment does not limit the specific values ​​of the aforementioned acceptable range and preset number. In addition, the output target value can be pre-stored in the data extraction module 110.

[0065] In step S160, the recommended parameter value X' is... j+1 (Input parameter value for the next batch) Add the input parameter value X from the current batch. j And will correspond to these recommended parameter values ​​X' j+1 Multiple recommended output values ​​Y' j+1 (The measurement output values ​​of the next batch) are added to the measurement output values ​​Y of the current batch. j This makes the recommended parameter values ​​X' j+1 These input parameter values ​​X j Members and those recommended output values ​​Y' j+1 These measurement output values ​​Y become j The members. In other words, in the next batch of calculations, the input parameter value X that participates in the calculation. j The members include the input parameter values ​​X for the current batch (i.e., the j-th batch). j and the recommended parameter value X' for the next batch (i.e., the (j+1)th batch). j+1 The members of the measurement output values ​​Yj involved in the calculation include the measurement output values ​​Yj of the current batch (i.e., the j-th batch). j and the recommended output value Y' for the next batch (i.e., the (j+1)th batch). j+1 .

[0066] Then, the process returns to step S110, using the updated input parameter value X. j (Including the input parameter value X for the current batch) j and the recommended parameter value X' for the next batch j+1 and updated measurement output value Y j (Including the measurement output value Y of the current batch) j and the recommended output value Y' for the next batch j+1), and continue to generate the next batch of recommended parameter values ​​X' j+1 Repeat steps S110 to S150 according to this principle until the optimal recommended parameter values ​​are obtained.

[0067] The verification results of applying the optimized process of this disclosure to a chemical process are presented in Table 2 below. The conditions for the chemical process are: (1) The feed rate is 50 cubic meters per hour (m³ / h). 3 / hr) and the feed concentration is 82%; (2). The parameter type of the input parameter value is, for example, parameter types including additive ratio, temperature, etc.; and (3). The "average number of iterations" is used as the verification index (for example, the average of the total number of iterations in 10 experiments (the sum of the number of iterations required to produce the best recommended parameter value in each experiment). As shown in Table 2, the optimization process using the existing clustering method requires an average of 28 iterations to produce the best recommended parameter value, the optimization process using the first clustering method of this disclosure only requires an average of 24 iterations to produce the best input parameter value (the improvement rate is 14.2% compared to the existing clustering method), the optimization process using the second clustering method of this disclosure only requires an average of 17.8 iterations to produce the best recommended parameter value (the improvement rate is 36.4% compared to the existing clustering method), and the optimization process using the third clustering method of this disclosure only requires an average of 18.5 iterations to produce the best recommended parameter value (the improvement rate is 33.9% compared to the existing clustering method). In comparison, the three grouping methods adopted in the embodiments of this disclosure can effectively reduce the number of iterations or roughly maintain the number of iterations.

[0068] Table 2

[0069]

[0070] In summary, the present disclosure provides a parameter optimization device and a parameter optimization method using the same, which groups parameters based on at least two parameters, such as input parameter values ​​and sampling function values. This increases the difference between multiple groups, making the recommended parameter values ​​determined from at least one of these groups more diverse.

[0071] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A parameter optimization device, comprising: A data extraction module is used to extract multiple input parameter values ​​and corresponding multiple measurement output values. The input parameter values ​​are numerical values ​​of process parameters of an industrial process. The industrial process is executed with the input parameter values, and the measurement value of the process result is used as the measurement output value. A sampling function calculation module is used to generate multiple sampling function values ​​based on the input parameter values ​​and the measurement output values; A grouping module is used to generate multiple parameter value groups based on the input parameter values ​​and the sampling function values; and A parameter recommendation module is used to generate multiple recommended parameter values ​​from at least one of the parameter value groups; The data extraction module is further configured to add the recommended parameter values ​​to the input parameter values ​​and add multiple recommended output values ​​corresponding to the recommended parameter values ​​to the measurement output values, so that the recommended parameter values ​​become members of the input parameter values ​​and the recommended output values ​​become members of the measurement output values. The sampling function calculation module is further configured to generate updated sampling function values ​​based on the updated input parameter values ​​and measurement output values. The clustering module is further configured to generate updated parameter value groups based on the updated input parameter values ​​and sampling function values. The parameter recommendation module is further configured to generate updated recommended parameter values ​​from at least one of the updated parameter value groups.

2. The parameter optimization apparatus according to claim 1, wherein the sampling function calculation module is further configured to generate a neighboring variation value for each of the sampling function values, and the clustering module is further configured to generate the parameter value groups based on the input parameter values, the sampling function values ​​and the neighboring variation values.

3. The parameter optimization apparatus according to claim 2, wherein the sampling function calculation module is further configured to perform a product operation between each sampling function value and the corresponding neighboring variation value to generate multiple sample weights, and the clustering module is further configured to generate the parameter value groups based on the input parameter values, the sampling function values, the neighboring variation values ​​and the sample weights.

4. The parameter optimization apparatus according to claim 1, wherein the data extraction module is further configured to determine whether the plurality of recommended output values ​​corresponding to the recommended parameter values ​​have reached a target output value; and if the recommended output values ​​have reached the target output value, to issue an optimization termination message.

5. The parameter optimization apparatus according to claim 1, wherein the clustering module is further configured to generate the parameter value clusters based on the sampling function values ​​and the input parameter values ​​that meet a threshold value.

6. A parameter optimization method, comprising: Extract multiple input parameter values ​​and corresponding multiple measurement output values. The input parameter values ​​are numerical values ​​of process parameter types of an industrial process. The industrial process is executed with the input parameter values, and the measurement value of the process result is used as the measurement output value. Based on these input parameter values ​​and these measurement output values, multiple sampling function values ​​are generated; Based on these input parameter values ​​and these sampling function values, multiple parameter value groups are generated; and Generate multiple recommended parameter values ​​from at least one of these parameter value groups; The recommended parameter values ​​are added to the input parameter values, and multiple recommended output values ​​corresponding to the recommended parameter values ​​are added to the measurement output values, so that the recommended parameter values ​​become members of the input parameter values ​​and the recommended output values ​​become members of the measurement output values. Based on the updated input parameter values ​​and measurement output values, updated sampling function values ​​are generated. Based on the updated input parameter values ​​and sampling function values, updated parameter value groups are generated. Updated recommended parameter values ​​are generated from at least one of the updated parameter value groups.

7. The parameter optimization method according to claim 6 further includes generating a neighboring variation value for each of the sampling function values; and generating a group of parameter values ​​based on the input parameter values, the sampling function values ​​and the neighboring variation values.

8. The parameter optimization method according to claim 7 further includes performing a product operation between each of the sampling function values ​​and the corresponding neighboring variant values ​​to generate a plurality of sample weights; and generating a group of parameter values ​​based on the input parameter values, the sampling function values, the neighboring variant values ​​and the sample weights.

9. The parameter optimization method according to claim 6 further includes determining whether the plurality of recommended output values ​​corresponding to the recommended parameter values ​​have reached a target output value; and if the recommended output values ​​have reached the target output value, issuing an optimization termination message.

10. The parameter optimization method according to claim 6 further includes generating a group of parameter values ​​based on the sampling function values ​​and the input parameter values ​​that meet a threshold value.

Images