JP2006293478A - Optimum value search device, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To search for a global optimum value with small number of computation and small number of trials. <P>SOLUTION: The optimum value search device searches for the optimum value by executing a production steps (S1to S3) for producing an orthogonal table allocated with a level value obtained from causes affecting the machine characteristic and the initial values of the causes, a selection steps (S4, S5) for selecting the optimum value out of the characteristic values obtained based on the combination of the orthogonal table, a re-production steps (S6 to S12) for producing the orthogonal table setting up the initial value of each cause with the combination of the level value of each cause for obtaining the selected optimum value. Successively, the system executes the selection steps by using the orthogonal table produced at the re-production steps, and then repeatedly the re-production steps. Therefore, if the optimum value selected through the selection steps is equal to the optimum value selected at the previous selection steps, the level value width to be used for the next selection steps is made different from that of the present selection steps. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an apparatus, a method, and a program for searching for an optimum value of a combination of factors that realizes a required characteristic to the maximum when designing, analyzing, or testing a device having a plurality of factors, for example. It is.

  In designing or analyzing or testing a device having a plurality of factors, various methods for searching for an optimum value of a combination of factors have been proposed in order to realize the required characteristics to the maximum. Methods proposed so far include methods for searching for optimal values by applying mathematical programming such as Newton's method or sequential quadratic programming (for example, Non-Patent Document 1), genetic algorithms (GA: Genetic: A method to search for the optimum value by applying a method based on random search such as Algorithm (Analysis) or annealing method (SA: Simulated Annealing) (for example, Non-Patent Document 1), and creating a response surface equation that estimates the characteristics by experimental design. A method of searching for an optimum value by mathematical programming (for example, Patent Document 1), a method of investigating a distribution state of characteristic values obtained from an orthogonal table and estimating a search direction based on the distribution state ( For example, Patent Document 2), a method of searching an optimum value by updating an orthogonal table for each search step (for example, Patent Document 3), and the like.

  That is, in general, when designing, analyzing, or testing a device, in order to achieve the required characteristics to the maximum, grasp multiple factors that affect the characteristics, and use an optimum value search device to Find the optimal level of factors. Hereinafter, a conventional example of the optimum value search apparatus will be described with reference to FIGS.

  FIG. 12 is a flow when searching for an optimum value by applying a mathematical programming method such as Newton's method or sequential quadratic programming, and corresponds to the contents of Non-Patent Document 1. After setting the initial value of the design factor (S61), the optimum solution search condition is set (S62), the optimum value is searched by mathematical programming (S63), and the obtained optimum value is evaluated (S64). Take.

FIG. 13 is a flow when searching for an optimum value by applying a method based on a random search such as a genetic algorithm or an annealing method, and corresponds to the contents of Non-Patent Document 1. Here, after setting the initial value of the design factor (S71), the optimum value search condition is set (S72), the optimum value is searched by the random search method (S73), and the obtained optimum value is evaluated (S74). ) Take the process. FIG. 14 is a flow when searching for an optimum value by applying the response surface method, and corresponds to the content of Patent Document 1. Here, after setting the initial value of the design factor (S81), the design factor and the number of levels are set (S82), and the orthogonal table is selected and assigned (S83). Thereafter, a characteristic value based on the combination of orthogonal tables is obtained (S84), and a response surface equation that approximates the relationship between the characteristic value and the design factor is created (S85). Further, after setting the optimum solution search condition (S86), the optimum value is searched on the response surface equation (S87), the device function calculation is performed with the optimum combination of the obtained design factors, and this effect is confirmed on the actual characteristics ( S88) Take a process.
“Basics and Applications of Optimization Theory” Corona Co., Ltd. (2000) Japanese Patent Laid-Open No. 10-207926 JP 2000-132535 A Japanese Patent Application No. 2004-223978

  However, such a conventional optimum value search method has the following problems.

  That is, the conventional optimum value search method has a problem that a huge amount of calculation is captured by the local optimum value. In the case of searching for an optimum value by applying a mathematical programming method such as Newton's method or sequential quadratic programming shown in FIG. 12, in order to search for the optimum value while obtaining the gradient of the characteristic using the initial value as a starting point, In the case of a characteristic in which a large number of optimum value peaks exist as shown in FIG.

  In the case of searching for an optimum value by applying a method based on a random search such as the genetic algorithm of FIG. 13, a calculation with a large number of combinations is required, and therefore the global optimum value G can be obtained. Is expensive, but may require a large amount of calculation.

  In the case of searching for the optimum value by applying the response surface method of FIG. 14, the calculation time is short because the optimum value is searched on the response surface equation, but the accuracy of the response surface equation may be a problem. . In particular, when there are many design factors and there are many interactions between design factors, the prediction accuracy of the response surface equation tends to decrease.

  Patent Document 2 proposes a method of investigating a distribution state of characteristic values obtained from an orthogonal table, and searching for an optimum value by estimating a search direction based on the distribution state. In this method, it is necessary to determine which level value is selected for improvement for each factor. For the two-level problem shown as an example in Patent Document 2, it is considered that the direction of improvement can be determined because the characteristic of the factor is linear, but in many design problems, there is an interaction between the factors. Since it has non-linear characteristics, it is necessary to use an orthogonal table of three levels or more. In such a case, the improvement direction of each factor is determined only by the distribution of characteristic values described in Patent Document 2. It is extremely difficult to do.

  Patent Document 3 proposes a method for searching for an optimum value by updating an orthogonal table for each search step. FIG. 15 shows a state in which all 27 combinations in the case of three factors and three levels are assigned to the vertexes and midpoints of a cube to search for an optimum value. In the case of this method, there is a high possibility that the global optimal solution G can be obtained with a small number of searches, but when it is captured by the local optimal value K, it is difficult to get out of it.

  The present invention has been made in view of such circumstances, and an object thereof is to provide an apparatus, method, and program capable of searching for a global optimum value with a small number of calculations or trials.

  In order to achieve the above object, the present invention takes the following measures.

  That is, the optimum value search apparatus, method, and program of the present invention are obtained from factors that affect the characteristics of devices and the initial values of the factors when performing design, analysis, or testing related to devices having multiple factors. Step to create an orthogonal table with assigned standard values, a selection step to select the optimum value with the best characteristics from the characteristic values obtained based on the combination of orthogonal tables, and the optimum selected in the selection step A re-creation step to create an orthogonal table in which the combination of the level values of each factor from which values are obtained is reset to the initial value of each factor, and then a selection step using the orthogonal table created in the re-creation step Is executed, and then the re-creation step is repeatedly executed to search for the optimum value.

  If the optimal value selected in the current selection step is equal to the optimal value selected in the previous selection step, the level width that is the difference between adjacent level values in the orthogonal table is set in the next selection step. The values in the orthogonal table to be used are different from those in the orthogonal table used in this selection step.

In this case, for example, the level width is set to a different value based on the number of equality establishments by each optimum value selected in a predetermined number of selection steps. More specifically, the level width in the orthogonal table used in the n-th (n is an integer of 2 or more) selection step is expressed as a function W (n), and the first time based on each optimum value selected in each selection step. When the equal sign is established, W (n) = A × W (n−1), A> 1.0 When the second equal sign is established, W (n) = B × W (n−2), B < 1.0. Or, conversely, the level width in the orthogonal table used in the n-th selection step (n is an integer of 2 or more) is expressed by a function W (n), and the first time by each optimum value selected in each selection step. W (n) = B × W (n−1), B <1.0 when the equal sign is established, and W (n) = A × W (n−2), A> when the second equal sign is established. 1.0. Alternatively, the level width in the orthogonal table used in the n-th selection step (n is an integer equal to or greater than 2) is represented by a function W (n), depending on any of the optimum values selected in each selection step, etc. After the issue is established, until a predetermined number of selection steps,
W (n) = C × W (n−1), C <1.0 or C> 1.0.

  ADVANTAGE OF THE INVENTION According to this invention, the apparatus, method, and program which can search a global optimal value with few calculation frequency or trial frequency are realizable.

  The best mode for carrying out the present invention will be described below with reference to the drawings.

  FIG. 1 is a functional block diagram showing a configuration example of an optimum value search apparatus to which an optimum value search method according to an embodiment of the present invention is applied. The optimum value search apparatus includes an input device unit 21, an output device unit 23, and a recording medium 24 that are connected to the computer main body 22.

  The recording medium 24 stores a program such as, but not limited to, a magnetic disk, a floppy (registered trademark) disk, a hard disk, an optical disk (CD-ROM, DVD, etc.), a magneto-optical disk (MO, etc.), and a semiconductor memory. The format may be any form as long as it can be read by a computer.

  The recording medium 24 may be a part of hardware such as a hard disk or a memory, or may be an independent medium (not shown). Further, the recording medium 24 may be downloaded and stored or temporarily stored via a LAN or the Internet. It may be a stored medium.

  In addition, the number of recording media is not limited to one, and the case where the program for optimal value search support in the embodiment of the present invention is executed from a plurality of media is also included in the recording media here, and the medium configuration is any It may be configured as follows.

  The recording medium 24 includes a device function calculation unit 25 and an optimum value search calculation unit 26. The device function calculation unit 25 acquires an initial value of the device design input from the input device unit 21 via the computer main body 22, and performs a calculation related to the function of the device based on the acquired initial value. The optimum value search calculation unit 26 performs an optimum value search calculation applied to, for example, device design, analysis, or test, by operating a program according to the optimum value search method of the present invention.

  The output device unit 23 outputs the calculation results of the device function calculation unit 25 and the optimum value search calculation unit 26.

  In the following first and second embodiments, the optimum value search method according to the embodiment of the present invention will be specifically described. However, the optimum value search apparatus shown in FIG. This is a common configuration applicable also to the embodiment.

(First embodiment)
FIG. 2 is a flowchart showing a process flow of the optimum value search method according to the first embodiment of the present invention.

  The flowchart shown in FIG. 2 is based on a combination of step S1 for setting an initial value of a factor affecting the characteristics of the device, step S2 for setting a level width, step S3 for assigning to an orthogonal table, and an orthogonal table. Step S4 for obtaining a value, Step S5 for selecting the optimum value having the best characteristic from the obtained characteristic values, Steps S6 to S8 for checking the number of searches, and the level width of the next step from the obtained optimum value Steps S9 to S11, Step S12 for reassigning to the orthogonal table in order to obtain a repetition characteristic value, and Step S13 for outputting an optimum value.

  First, in step S1, an initial value of each factor when a design, analysis, or test regarding a device having a plurality of factors is performed is set. Next, in step S2, a level width is set. Further, in step S3, an orthogonal table suitable for the number of factors and the number of levels is selected, and each factor is assigned to the orthogonal table. In this case, an orthogonal table suitable for allocation is selected in consideration of the number of design factors and the number of levels. FIG. 3 shows an example in which seven types of design factors a to g are assigned using the L18 orthogonal table as an example. In this case, the number of levels is 3 levels for each design factor, and the level values are displayed as 1, 2, 3 in the orthogonal table. The orthogonal table is a suitable method for grasping the characteristics of a device having a plurality of factors with a small number of combinations because the factors are combined in a balanced manner.

Next, in step S4, a characteristic value is obtained for each combination of orthogonal tables, and in step S5, an optimum value having the best characteristic is selected from the obtained characteristic values. As shown in FIG. 3A, (Characteristic-1) to (Characteristic-18) are obtained for each combination of orthogonal tables, and the most characteristic is obtained from the obtained characteristic values (Characteristic-1) to (Characteristic-18). Select the optimum value of. For example, as shown in FIG. 3B, the combination of (characteristic-8)
(A, b, c, d, e, f, g) = (3, 2, 3, 2, 1, 3, 1)
Is the most suitable value for the first factor.

Next, as shown in FIG. 3C, the optimum value of the obtained factor is
(A, b, c, d, e, f, g) = (3, 2, 3, 2, 1, 3, 1)
= (2,2,2,2,2,2,2)
Thus, a new orthogonal table is created by setting the median level, and characteristic values (characteristic-1) to (characteristic-18) are obtained based on the combinations of the orthogonal tables.

For example, as shown in FIG. 3D, it is a combination of (characteristic -15).
(A, b, c, d, e, f, g) = (2, 3, 1, 2, 3, 2, 1)
3 has the most excellent characteristic, this combination is set as the optimum value of the second factor as shown in FIG. Next, in step S6, as shown in FIG.
(A, b, c, d, e, f, g) = (2, 3, 1, 2, 3, 2, 1)
= (2,2,2,2,2,2,2)
Thus, the optimum value of the obtained factor is set to the median value of the level, a new orthogonal table is repeatedly created, and the optimum value is searched within the limit of the number of searches.

  FIG. 4 is a diagram showing a state in which a global optimum value is searched for in the present embodiment, and the design factor is set to one factor and three levels for simplicity. In the figure, # 1 is the start point, # 15 is the reaching point of the optimum value, and the length of the horizontal bar in each search indicates the level width (difference between adjacent level values). The local optimum value is captured at # 4, but the level width is increased at # 5 to escape from the local optimum value. Further, at # 8 and # 9, it stops near the global optimum value, but at # 10, the level width is reduced and the search is restarted to reach the global optimum value.

  Next, a method for changing the level width of the orthogonal table that greatly affects the search for the optimum value in the present embodiment will be described.

This applies to the flow after the determination formula “Opt (n−1) −Opt (n−2)” shown in step S7 of FIG. Here, Opt (n-1) represents the (n-1) th optimum value, and Opt (n-2) represents the (n-2) th optimum value. First, the difference between the optimum values obtained at the (n-1) th and (n-2) th is determined. When ΔOpt = Opt (n−1) −Opt (n−2)> 0 (step S7:> 0), since the search for the optimum value is in progress, in step S11, (n−1) and ( Using the optimum value of (n-2) as a function, the level width W (n) of the next search is obtained by the following equation (1). Then, reassignment to the orthogonal table is performed using this level width (step S12), and the process returns to step S4.
W (n) = f (Opt (n−1), Opt (n−2)) × W (n−1) (1)
In the function f of the above equation (1), the level width is larger than the previous step when the change amount of the optimum value is large, and the level width is larger than that of the previous step when the change amount of the optimum value is small. Is set to be smaller. This is because, in a search process that shows a characteristic with a large amount of change in the optimum value, a larger level width can approach the optimum value with a smaller number of searches, while a characteristic with a small amount of change in the optimum value. This is because, in the search process, it is considered that a higher level optimal value can be reached if the level width is set smaller.

If ΔOpt = 0, that is, if the previous and current optimum values do not change (step S7: = 0), the search range may be larger than the previous search because there is a possibility of being captured by the local optimum value, or Make small changes. Then, in step S8, when the optimum value equality is established for the first time (step S8: = 1), the level width W (n) of the next step is determined by the following equation (2). Then, reassignment to the orthogonal table is performed using this level width (step S12), and the process returns to step S4.
W (n) = A × W (n−1) (2)
A in the formula (2) is a constant arbitrarily set within the range of A> 1.0. This search process corresponds to # 5 in FIG. Although it is captured by the local optimum value at # 4, the peak of the global optimum value G at a position distant from the current optimum value is searched for by changing the level width greatly. If the slope of the mountain containing the global optimum value can be searched by changing the level width greatly according to the above equation (2), then the global optimum value will be reached based on the change in the level width according to the above equation (1). Can do.

  If the optimum value equality is established for the second time (step S8: = 2), that is, if the optimum value does not change even if the level width is greatly changed, a global optimum value exists in the vicinity of the current optimum value. Since there is a possibility, the level width is changed small as shown in the following equation (3) (step S9). Then, reassignment to the orthogonal table is performed using this level width (step S12), and the process returns to step S4.

W (n) = B × W (n−2) (3)
B in the above equation (3) is a constant arbitrarily set within a range of B <1.0. This search process corresponds to # 10 in FIG. Although it stops near the global optimum value at # 8 and # 9, the search is resumed by reducing the level width based on the above equation (3) at # 10 to reach the global optimum value. In addition, the order of the above formulas (2) and (3) is switched, that is, when the equality of the optimum value is established for the first time, the level width is changed small, and the establishment of the equality of the optimum value is established for the second time. However, the level range can be changed greatly.

  If the optimal value is equal to or more than 3 times (step S8: ≧ 3), that is, if the optimal value does not change even if the level width is increased or decreased, the current optimal value is the global optimal value. It is determined that the value is G, and this value is output (step S13), and the calculation is terminated. In step S6, even when the number of searches is not within the limit (step S6: No), it is determined that the current optimum value is the global optimum value G, and this value is output (step S13). End the calculation.

  It should be noted that the method of changing the level width and the number of times the convergence determination equality is established can be changed according to the characteristics of the target problem.

  As described above, according to the optimum value search method according to the present embodiment, for each search step, by changing the level range that affects the search range based on the trend of the optimum value, the global value can be reduced with a small number of searches. It is possible to increase the possibility of reaching the target optimum value.

(Second Embodiment)
FIG. 5 is a flowchart showing a process flow of the optimum value search method according to the second embodiment of the present invention.

  The optimum value search method according to the present embodiment is different from the first embodiment only in how to determine the level width W (n) of the next step when the equality of the optimum value is established. Therefore, here, steps that perform the same processing as in the first embodiment are given the same step numbers, and different points are described to avoid redundant description.

  That is, in the present embodiment, in step S8 shown in the flowchart of FIG. 5, a maximum value of the number of equality establishment is set in advance, and before the number of equality establishment reaches the maximum value (step S8: <max ), The level width is determined according to the following equation (4).

W (n) = C × W (n−1) (4)
In the above formula (4), C is a constant set arbitrarily within the range of C <1.0 and C> 1.0. This search process corresponds to # 4 to # 7 in FIG. Although # 4 is captured in the vicinity of the local optimum value, the slope of the mountain including the global optimum value can be searched by gradually increasing the level width based on the above equation (4). Thereafter, the global optimum value can be reached based on the change in the level width according to the above equation (1). If the optimum value does not change even when the preset maximum number of equality is reached (step S8: ≧ max), it is determined that the current optimum value is the global optimum value G and this The value is output (step S13), and the calculation is terminated.

(Effects of the present invention)
In order to verify the effectiveness of the above-described optimal value search method (hereinafter referred to as “the present method”) according to the first and second embodiments, two types of functions used in the benchmark of the optimization method For the optimal value search problem, the search results obtained by this method were compared with the search results obtained by other methods. The objective function and constraint conditions that simulate unimodality are shown in the following formulas (5) and (6), and the 3D characteristics are shown in FIG.

f (x ₁ , x ₂ ) = cos (x ₁ ) cos (x ₂ ) exp (− (x ₁ −π) ² − (x ₂ −π) ² )
... (5)
−20 <x ₁ , x ₂ <20 (6)
The global optimum G of this function is f (x ₁ , x ₂ ) = 1.0 when x ₁ = x ₂ = π. Since the optimum value exists in a very narrow range, the point is how the optimum value can be reached with a small number of searches. In this method, since the number of factors is _two , x ₁ and x ₂ , the L9 orthogonal table of two factors and three levels as shown in FIG. 8 is used, and a program is created and calculated according to the flowchart of FIG. Went. FIG. 9 shows a comparison of search results for optimum values of this method (OA: Orthogonal Array), genetic algorithm (GA), and annealing method (SA). The horizontal axis is the number of calculations, and the vertical axis is f (x ₁ , x ₂ ). In all three methods, f (x ₁ , x ₂ ) = 1.0 is reached, but it can be seen that there is a large difference in the number of calculations until reaching. In the present method (OA), the optimum value is reached in 100 times or less, whereas in the genetic algorithm (GA), about 1000 times are required, and in the annealing method (SA), about 1500 times are required.

Next, objective functions and constraints that simulate multimodality are shown in the following formulas (7) and (8), and 3D characteristics are shown in FIG.
f (x ₁ , x ₂ )
= 10 × exp (−0.01 (x ₁ −10) ² −0.01 (x ₂ −15) ² ) sin (x ₁ )
... (7)
−10 <x ₁ , x ₂ <40 (8)
The global optimum G of this function is f (x ₁ , x ₂ ) = 9.56 when x ₁ = 7.90 and x ₂ = 15.0. Since there are a plurality of local optimum values K as shown in the figure, the point is how to reach the global optimum value. In this case as well, since the number of factors is _two , x ₁ and x ₂ , this method uses L9 which is an orthogonal table of two factors and three levels as shown in FIG. 9, and the program is executed according to the flowchart of FIG. Created and calculated. The method compared with this method is the genetic algorithm (GA) and the annealing method (SA) as well as the search problem of the unimodal function. FIG. 11 shows the optimum value search results of this method (OA), genetic algorithm (GA), and annealing method (SA). The horizontal axis is the number of calculations, and the vertical axis is f (x ₁ , x ₂ ). As can be seen from the figure, although the three methods reach f (x ₁ , x ₂ ) = 9.56, there is a difference in the number of calculations. In this method (OA), the optimum value is reached in about 100 times, whereas the genetic algorithm (GA) and the annealing method (SA) require about 350 calculation times. Thus, it was confirmed that the optimal value search method according to the present embodiment has a search capability superior to that of the conventional method in the unimodal and multimodal optimal value search problems.

  The best mode for carrying out the present invention has been described above with reference to the accompanying drawings, but the present invention is not limited to such a configuration. Within the scope of the invented technical idea of the scope of claims, a person skilled in the art can conceive of various changes and modifications. The technical scope of the present invention is also applicable to these changes and modifications. It is understood that it belongs to.

The functional block diagram which shows the structural example of the optimal value search apparatus to which the optimal value search method which concerns on embodiment of this invention is applied. The flowchart which shows the flow of a process of the optimal value search method which concerns on 1st Embodiment. The figure which shows the example which searches an optimal value by applying an orthogonal table. The figure which shows a mode that a global optimal value is searched in 1st Embodiment. The flowchart which shows the flow of a process of the optimal value search method which concerns on 2nd Embodiment. The figure which shows a mode that the optimal value is searched by changing the level width of an orthogonal table | surface continuously. The 3D characteristic view of the unimodal function used in order to verify the effect of the optimal value search method of this invention. The L9 orthogonal table used in order to verify the effect of the optimal value search method of this invention. The comparison figure of the optimal value search result with respect to a unimodal function. The 3D characteristic figure of the multimodal function used in order to verify the effect of the optimal value search method of this invention. The comparison figure of the optimal value search result with respect to a multimodal function. The flowchart which shows the flow of the process which searches the optimal value by applying the conventional mathematical programming. The flowchart which shows the flow of the process which searches the optimal value by applying the conventional random search method. The flowchart which shows the flow of the process which searches the optimal value by applying the conventional response surface method. The figure which shows the example which allocates all 27 combinations in the case of 3 factors 3 levels, and searches for an optimal value by assigning to the vertex and midpoint of a cube. The figure which shows the example in which many peaks of the optimal value exist and are captured by the local optimal value.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 21 ... Input device part, 22 ... Computer main body, 23 ... Output device part, 24 ... Recording medium, 25 ... Device function calculation part, 26 ... Optimum value search calculation part

Claims (15)

A creation step of creating an orthogonal table in which a factor affecting the characteristics of the device and a level value obtained from the initial value of each factor are assigned when performing design, analysis, or testing on the device having a plurality of factors; A selection step for selecting the optimum value having the best characteristics from the characteristic values obtained on the basis of the combination of the orthogonal tables, and a combination of the level values of each factor for obtaining the optimum value selected in the selection step. A re-creating step for creating an orthogonal table reset to the initial value of the factor, and thereafter performing the selection step using the orthogonal table created in the re-creating step, and then performing the re-creating step. In an optimum value search device that searches for an optimum value by repeating execution,
If the optimal value selected in the current selection step is equal to the optimal value selected in the previous selection step, the level width that is the difference between adjacent level values in the orthogonal table is used in the next selection step. The optimum value search device is characterized in that the value is different from that of the orthogonal table used in this selection step and that of the orthogonal table used in this selection step. In the optimum value search apparatus according to claim 1,
An optimum value search device characterized in that the level range is set to a different value based on the number of equality establishments by each optimum value selected in a predetermined number of each selection step. In the optimum value search apparatus according to claim 2,
The level width in the orthogonal table used in the n-th (n is an integer of 2 or more) selection step is expressed by a function W (n), and when the first equality is established by each optimum value selected in each selection step. , W (n) = A × W (n−1), A> 1.0, when the second equality is established,
An optimum value search apparatus characterized by W (n) = B × W (n−2), B <1.0. In the optimum value search apparatus according to claim 2,
The level width in the orthogonal table used in the n-th (n is an integer of 2 or more) selection step is expressed by a function W (n), and when the first equality is established by each optimum value selected in each selection step. , W (n) = B × W (n−1), B <1.0, when the second equality is established,
W (n) = A × W (n−2), A> 1.0 In the optimum value search apparatus according to claim 2,
The level width in the orthogonal table used in the n-th selection step (n is an integer of 2 or more) is represented by a function W (n), and an equal sign by any one of the optimum values selected in each selection step. After the is established, until a predetermined number of selection steps,
W (n) = C × W (n−1), C <1.0 or C> 1.0 A creation step of creating an orthogonal table in which a factor affecting the characteristics of the device and a level value obtained from the initial value of each factor are assigned when performing design, analysis, or testing on the device having a plurality of factors; A selection step for selecting the optimum value having the best characteristics from the characteristic values obtained on the basis of the combination of the orthogonal tables, and a combination of the level values of each factor for obtaining the optimum value selected in the selection step. A re-creating step for creating an orthogonal table reset to the initial value of the factor, and thereafter performing the selection step using the orthogonal table created in the re-creating step, and then performing the re-creating step. In an optimum value search method for searching for an optimum value by repeating execution,
If the optimal value selected in the current selection step is equal to the optimal value selected in the previous selection step, the level width that is the difference between adjacent level values in the orthogonal table is used in the next selection step. The optimum value search method is characterized in that the values are different from those of the orthogonal table and those of the orthogonal table used in this selection step. The optimal value search method according to claim 6,
An optimum value search method characterized in that the level width is set to a different value based on the number of equal signs established by each optimum value selected in a predetermined number of each selection step. In the optimal value search method according to claim 7,
The level width in the orthogonal table used in the n-th (n is an integer of 2 or more) selection step is expressed by a function W (n), and when the first equality is established by each optimum value selected in each selection step. , W (n) = A × W (n−1), A> 1.0, when the second equality is established,
An optimal value search method characterized by W (n) = B × W (n−2), B <1.0. In the optimal value search method according to claim 7,
The level width in the orthogonal table used in the n-th (n is an integer of 2 or more) selection step is expressed by a function W (n), and when the first equality is established by each optimum value selected in each selection step. , W (n) = B × W (n−1), B <1.0, when the second equality is established,
An optimum value search method characterized by W (n) = A × W (n−2), A> 1.0. In the optimal value search method according to claim 7,
The level width in the orthogonal table used in the n-th selection step (n is an integer of 2 or more) is represented by a function W (n), and an equal sign by any one of the optimum values selected in each selection step. After the is established, until a predetermined number of selection steps,
W (n) = C × W (n−1), C <1.0 or C> 1.0 A program for searching for an optimum value when designing, analyzing, or testing a device having a plurality of factors that affect characteristics,
A creation function for creating an orthogonal table in which, when performing design, analysis, or testing on a device having a plurality of factors, the factor that affects the device characteristics and the level value obtained from the initial value of each factor are assigned.
A selection function for selecting an optimum value with the best characteristics from the characteristic values obtained based on the combination of the orthogonal tables,
A re-creation function for creating an orthogonal table in which the combination of the level values of each factor for obtaining the optimum value selected by the selection function is reset to the initial value of each factor;
Repetitive function of repeating the selection function using the orthogonal table created by the re-creation function, and then repeating the re-creation function;
If the optimal value selected in the current selection step is equal to the optimal value selected in the previous selection step, the level width that is the difference between adjacent level values in the orthogonal table is used in the next selection step. A program for causing a computer to realize a level width setting function that makes the values different from those in the orthogonal table and those in the orthogonal table used in this selection step. The program according to claim 11,
The level range setting function is a program in which the level range is set to a different value based on the number of equality established by each optimum value selected in a predetermined number of selection steps. The program according to claim 12,
The level width setting function represents the level width in the orthogonal table used in the n-th selection step (n is an integer of 2 or more) as a function W (n), and depends on each optimum value selected in each selection step. When the first equal sign is established, W (n) = A × W (n−1), A> 1.0, and when the second equal sign is established, W (n) = B × W (n−2), A program in which B <1.0. The program according to claim 12,
The level width setting function represents the level width in the orthogonal table used in the n-th selection step (n is an integer of 2 or more) as a function W (n), and depends on each optimum value selected in each selection step. When the first equal sign is established, W (n) = B × W (n−1), B <1.0, and when the second equal sign is established, W (n) = A × W (n−2), A program in which A> 1.0. The program according to claim 12,
The level width setting function expresses the level width in the orthogonal table used in the n-th selection step (n is an integer of 2 or more) as a function W (n), and sets each optimum value selected in each selection step. After the equal sign by either of them is established, until a predetermined number of selection steps,
A program in which W (n) = C × W (n−1), C <1.0 or C> 1.0.

Images