Constraint domain optimal Latin hypercube design method based on sample replacement

By adopting sample permutation technology in the optimal Latin hypercube design, the waste of sample resources and accuracy problems caused by irregular design domain irregularity in complex systems are solved, and high-quality sampling and good spatial fillability of irregular design domains are achieved.

CN118981904APending Publication Date: 2024-11-19GENERAL ENG RES INST CHINA ACAD OF ENG PHYSICS
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411175812.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

In the optimization design of complex systems, traditional experimental design methods are difficult to effectively deal with nonlinear or inequal constraints between design variables, resulting in waste of sample resources and poor model accuracy, especially in the case of irregular design domains.

Method used

Using the optimal Latin hypercube design method based on sample permutation, the initial sample set is divided into sample sets that satisfy and do not satisfy the constraints, random samples are used to generate candidate samples, and the unsatisfied constraints are replaced by minimizing the normalization distance. Samples, so that they meet constraints, and achieve good sampling of irregularly designed domains.

Benefits of technology

It improves sample quality and spatial fillability, is suitable for various irregular design domains, effectively enhancing the universality of sampling quality and engineering applications.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981904A_ABST
    Figure CN118981904A_ABST
Patent Text Reader

Abstract

The invention discloses a constraint domain optimal Latin hypercube design method based on sample replacement, and relates to the field of test design, and the method comprises the steps: S1, obtaining an initial sample according to nv and ns; s2, dividing the initial sample set into xin and xout; s3, selecting one sample from the xout, and generating a large number of random xrand; s4, xcandidate is selected from the xrand, and the normalized distance between each candidate sample and the xin is calculated; s5, a new replacement sample xnew is obtained from the xcandidate, and a new replacement sample xnew is obtained from the xcandidate; s5, adding the xnew into the xin to obtain xin ', and removing the sample selected in the S3 from the xout to obtain xout'; s6, judging whether xout'is a null set or not, if so, ending, and otherwise, entering S7; according to the method, by fixing a certain variable dimension of an initial optimal Latin hypercube design result, an initial sample which does not meet constraints is replaced so that the initial sample can meet the target of minimum distance maximization in a constraint space, and the purpose that irregular design domain sampling has good space filling performance and projection uniformity is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of experimental design, and in particular to a constrained domain optimal Latin hypercube design method based on sample replacement. Background Art

[0002] The optimization design of complex systems often requires thousands of high-fidelity simulations, which is a time-consuming and costly process. Therefore, rapid prediction technology based on small sample data has broad application prospects. Experimental design is a key link in rapid prediction technology. The selection of initial samples largely determines the accuracy of the model and subsequent design analysis, and thus has a great impact on the accuracy and efficiency of computer simulation analysis. Based on the assumption that the input variables are independent of each other, the optimal Latin hypercube design is widely used due to its good low-dimensional projection properties and space filling properties. However, in actual engineering applications, there are many design variables involved and the relationship between design objectives and design constraints is relatively complex. When there is a physical dependency between design variables, the design domain is usually irregular due to the linear or nonlinear inequality constraints of the mutual coupling between variables. Therefore, the experimental samples selected using traditional experimental design methods are prone to non-informative simulations, which in turn leads to waste of sample resources or poor approximation accuracy of the prediction model. Therefore, when conducting experimental design, the inequality constraints between variables must be observed to avoid a large amount of resource waste of high-fidelity numerical models and irrelevant physical relationships of the input set. Summary of the invention

[0003] The purpose of the present invention is to design a constrained domain optimal Latin hypercube design method based on sample replacement in order to solve the above problems.

[0004] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0005] A constrained domain optimal Latin hypercube design method based on sample replacement, comprising:

[0006] S1, based on sample dimension n v and the number of samples required n s , using the optimal Latin hypercube experimental design to obtain the initial sample , where x i is the column vector of the sample in each dimensional space, x∈ ;

[0007] S2. Divide the initial sample set into two categories according to the variable constraint relationship. The sample set that satisfies the constraint is x in , the sample set that does not satisfy the constraint is x out ;

[0008] S3, from the sample set x outSelect a sample from the sample and fix a certain x in the sample i The value of the axis remains unchanged, and the other x values ​​are changed j Axis coordinates, generate a large number of random samples x rand , j≠i;

[0009] S4, from the random sample x rand Select all samples that meet the variable constraints as candidate sample set x candidate , calculate each candidate sample and sample set x in The normalized distance between i , expressed as: , where n in For the sample set x in The number of samples;

[0010] S5, from x candidate Get the new replacement sample x new ;

[0011] S5. x new Add to sample set x in In the above example, we get the sample set x in ’ , and select the sample from S3 from the sample set x out Remove from the sample set x out ’ ;

[0012] S6. Determine sample set x out ’ Is it an empty set? If so, end. Otherwise, enter S7;

[0013] S7, let the sample set x in ’ As a sample set x in , sample set x out ’ As a sample set x out , and returns to S3.

[0014] The beneficial effects of the present invention are as follows: the method fixes a certain variable dimension of the initial optimal Latin hypercube design result, permutes the initial samples that do not meet the constraints so that they meet the goal of maximizing the minimum distance in the constraint space, and achieves the purpose of irregular design domain sampling with good space filling and projection uniformity; and can effectively enhance the sampling quality, has strong versatility for engineering problems, is suitable for various irregular design domain sampling, and solves related engineering technical problems of irregular design domain sampling in the field of experimental design technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1This is the flow chart of the optimal Latin hypercube design method for constrained domain based on sample replacement;

[0016] Figure 2 It is a schematic diagram of the initial design domain sampling in the sampling process of the present invention;

[0017] Figure 3 A schematic diagram of sample classification in the sampling process of the present invention;

[0018] Figure 4 A schematic diagram of random sample acquisition in the sampling process of the present invention;

[0019] Figure 5 A schematic diagram of sample replacement results during the sampling process of the present invention;

[0020] Figure 6 It is a schematic diagram of two-dimensional linear constraints in the sampling results of the present invention;

[0021] Figure 7 It is a schematic diagram of two-dimensional nonlinear constraints in the sampling results of the present invention;

[0022] Figure 8 It is a schematic diagram of three-dimensional linear constraints in the sampling results of the present invention;

[0023] Fig. 9 It is a schematic diagram of the three-dimensional nonlinear constraints in the sampling results of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0025] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0027] In the description of the present invention, it should be understood that the terms "upper", "lower", "inside", "outside", "left", "right", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, or are directions or positional relationships in which the product of the invention is usually placed when in use, or are directions or positional relationships commonly understood by those skilled in the art. These directions or positional relationships are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present invention.

[0028] Furthermore, the terms “first”, “second”, etc. are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0029] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms such as "setting" and "connection" should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0030] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0031] like Figure 1 As shown, a constrained domain optimal Latin hypercube design method based on sample replacement includes:

[0032] S1, based on sample dimension n v and the number of samples required n s , using the optimal Latin hypercube experimental design to obtain the initial sample , where x i is the column vector of the sample in each dimensional space, x∈ .

[0033] S2. Divide the initial sample set into two categories according to the variable constraint relationship. The sample set that satisfies the constraint is x in , the sample set that does not satisfy the constraint is x out ; The specific variable constraint relationship is: f(x)>0.

[0034] S3, from the sample set x out Select a sample from the sample and fix a certain x in the sample i The value of the axis remains unchanged, and the other x values ​​are changed j Axis coordinates, generate a large number of random samples x rand , j≠i.

[0035] S4, from the random sample xrand Select all samples that meet the variable constraints as candidate sample set x candidate , calculate each candidate sample and sample set x in The normalized distance between i , expressed as: , where n in For the sample set x in The number of samples; the normalized distance is expressed as: , where x i N For x i The normalized vector of , each component is expressed as: , where UB and LB are the upper and lower limits of the variable, respectively, and k is the corresponding dimension.

[0036] S5, according to the minimum distance maximization criterion from x candidate Get the new replacement sample x new , the distance metric of the new permuted sample satisfies d maxmin =max{min(D1),min(D2),...,min(D i )}.

[0037] S5. x new Add to sample set x in In the above example, we get the sample set x in ’ , and select the sample from S3 from the sample set x out Remove from the sample set x out ’ .

[0038] S6. Determine sample set x out ’ Is it an empty set? If so, end; otherwise, enter S7.

[0039] S7, let the sample set x in ’ As a sample set x in , sample set x out ’ As a sample set x out , and returns to S3.

[0040] Example:

[0041] A1. Taking two-dimensional linear constraints as an example, the sample dimension of the embodiment is n v =2, the number of samples is n s =10, the constraint is expressed as: . Perform optimal Latin square sampling according to the number of samples in the initial design domain, and record the obtained initial sample points as the set , x∈ .

[0042] A1 specifically includes:

[0043] A1.1 Determine the initial experimental design domain: Based on the constraints , the initial design domain is a two-dimensional cube, and the domains of both variables are [0,5];

[0044] A1.2 Obtaining initial sample points: Use the optimal Latin hypercube to obtain n in the initial design domain. s = 10 initial samples, the resulting initial sample set is ,like Figure 2 shown.

[0045] A2. According to the linear constraint relationship , classify the initial sample set, and the results are as follows Figure 3 As shown, the sample set that satisfies the constraints is , the sample set that does not satisfy the constraints is .

[0046] A3. Sample set x that never satisfies the constraints out Select a sample [0.56,3.89], keep its x unchanged, change y, and generate a large number of random samples x rand ,like Figure 4 shown.

[0047] A4. From a random sample x rand Select all samples that meet the constraints as candidate sample set x candidate , calculate the normalized distance between each candidate sample and the sample set that satisfies the constraints.

[0048] A4 specifically includes:

[0049] A4.1 x in and x candidate Normalized, we get .

[0050] A4.2 Calculation of x candidate Each sample in x in The normalized distance of the candidate samples [0.56, 3.10] and [0.56, 1.50] is taken as an example, and their normalized values ​​are [0.11, 0.62] and [0.11, 0.30] respectively. According to the normalized distance With x inThe normalized distance is D1=[0.31, 0.52, 0.23, 0.76, 0.58, 0.87] and D2=[0.12, 0.22, 0.34, 0.54, 0.57, 0.78]. Then min(D1)=0.23, min(D2)=0.12. The other candidate samples can be calculated in the same way. According to the calculation results of the other samples, the maximum and minimum distances are obtained at the sample [0.56, 3.10], which is d maxmin =0.23.

[0051] A5. According to the calculation results of A4, in this iteration, x new =[0.56,3.10], and x new Add to sample set x in and from the sample set x out Remove the samples selected in step 3 and update x in sequence in and x out , respectively and , due to x out If it is not an empty set, perform A3 and iterate three times in sequence until all samples that do not meet the constraints are replaced. The final sample replacement result is shown in tu 5.

[0052] The following four constraint domain sampling examples with typical representative characteristics are used to verify the effectiveness of this method without loss of generality.

[0053] Two-dimensional linear constraint example: The constraint condition is , the number of samples is n s =50, the sampling results are as follows Figure 6 shown.

[0054] Two-dimensional nonlinear constraint study: The constraint condition is As shown, the number of samples is n s =50, the sampling results are as follows Figure 7 shown.

[0055] Three-dimensional linear constraint example: The constraint condition is , the number of samples is n s =100, the sampling results are as follows Figure 8 shown.

[0056] Three-dimensional nonlinear constraint study: The constraint conditions are , the number of samples is n s =100, the sampling results are as follows Fig. 9 shown.

[0057] For both linear constraints and nonlinear constraints, the method of the present invention can obtain samples with good filling uniformity and projection uniformity. It can be seen that the method can achieve high-quality sampling in irregular constrained design domains.

[0058] The technical solution of the present invention is not limited to the above-mentioned specific embodiments. All technical variations made according to the technical solution of the present invention fall within the protection scope of the present invention.

Claims

1. A constrained domain optimal Latin hypercube design method based on sample replacement, characterized in that: include: S1, based on sample dimension n v and the number of samples required n s , using the optimal Latin hypercube experimental design to obtain the initial sample , where x i is the column vector of the sample in each dimensional space, x∈ ; S2. Divide the initial sample set into two categories according to the variable constraint relationship. The sample set that satisfies the constraint is x in , the sample set that does not satisfy the constraint is x out ; S3, from the sample set x out Select a sample from the sample and fix a certain x in the sample i The value of the axis remains unchanged, and the other x values ​​are changed j Axis coordinates, generate a large number of random samples x rand , j≠i; S4, from the random sample x rand Select all samples that meet the variable constraints as candidate sample set x candidate , calculate each candidate sample and sample set x in The normalized distance between i , expressed as: , where n in For the sample set x in The number of samples; S5, from x candidate Get the new replacement sample x new ; S5. x new Add to sample set x in In the above example, we get the sample set x in ’ , and select the sample from S3 from the sample set x out Remove from the sample set x out ’ ; S6. Determine sample set x out ’ Is it an empty set? If so, end. Otherwise, enter S7; S7, let the sample set x in ’ As a sample set x in , sample set x out ’ As a sample set x out , and returns to S3.

2. The constrained domain optimal Latin hypercube design method based on sample replacement according to claim 1, characterized in that: In S2, the variable constraint relationship is: f(x)>0.

3. The constrained domain optimal Latin hypercube design method based on sample replacement according to claim 1, characterized in that: In S4, the normalized distance is expressed as: , where x i N For x i The normalized vector of , each component is expressed as: , where UB and LB are the upper and lower limits of the variable, respectively, and k is the corresponding dimension.

4. The constrained domain optimal Latin hypercube design method based on sample replacement according to claim 1, characterized in that: In S5, according to the minimum distance maximization criterion, from x candidate Get the new replacement sample x new , the minimum distance maximization criterion is expressed as: d maxmin =max{min(D1),min(D2),...,min(D i )}.