Good solution identification device and good solution identification method

Abstract

This good solution identification device (3) comprises: an actual index value acquisition unit (11) that gives, to a black box (1), a portion of combinations among a plurality of combinations of values that can be taken by each of a plurality of parameters indicating a plurality of configuration degrees of freedom for an evaluation target, and acquires index values of the evaluation target corresponding to the portion of combinations as actual index values from the black box (1); a sub-index value acquisition unit (12) that gives, to each of a plurality of sub-black boxes (2-1 to 2-N), values of some parameters among the parameters indicating the plurality of configuration degrees of freedom, the some parameters being in different combinations for each sub-black box, and that acquires index values of the evaluation target corresponding to the given values of the some parameters as sub-index values from each of the sub-black boxes (2-n); and a good solution identification unit (13) that identifies a combination that is a good solution, which is a solution expected to be good, from among the plurality of combinations on the basis of the actual index values acquired by the actual index value acquisition unit (11) and the sub-index values acquired by the sub-index value acquisition unit (12).

Description

Apparatus for identifying good solutions and method for identifying good solutions

[0001] This disclosure relates to a device for identifying good solutions and a method for identifying good solutions.

[0002] There is a good solution identification device (hereinafter referred to as the "conventional good solution identification device") that provides a black box with all possible combinations of values ​​for each of several parameters indicating the degrees of freedom for setting multiple parameters for the object to be evaluated, obtains an index value for the object to be evaluated corresponding to each combination from the black box, and identifies a combination that is a good solution based on the index value of the object to be evaluated. A good solution is a solution that is expected to be good. Among good solutions, in addition to the optimal solution, which is the most suitable solution, there may be, for example, a second-best solution and a third-best solution. Non-patent document 1 discloses a device that identifies a combination among multiple possible values ​​for each of the parameters W and B that is a good solution, such as a combination that reduces the SNR (Signal Noise Ratio) and reduces energy consumption.

[0003] “Energy-Efficient Implementation of Carrier Phase Recovery for Higher-Order Modulation Formats”, JOURNAL OF LIGHT WAVE TECHNOLOGY,VOL.39,NO.2,JANUARY15,2021

[0004] Conventional good solution identification devices have the problem that the processing load for identifying good solution combinations increases exponentially as the number of parameters indicating the degrees of freedom of setting increases. Furthermore, the device disclosed in Non-Patent Document 1 also exhibits an exponential increase in processing load for identifying good solution combinations as the number of parameters increases.

[0005] This disclosure was made to solve the above-mentioned problems, and aims to provide a good solution identification device that can suppress the exponential increase in processing load for identifying good solution combinations, even when the number of parameters indicating the degrees of freedom of setting increases.

[0006] The good solution identification device according to this disclosure includes: an actual index value acquisition unit that provides some combinations of values ​​from among multiple combinations of values ​​that each of the multiple parameters indicating multiple degrees of freedom for setting the object to be evaluated can take to a black box, and acquires the index values ​​of the object to be evaluated corresponding to the some combinations from the black box as actual index values; and a sub-index value acquisition unit that provides values ​​for some of the parameters from among the multiple parameters indicating multiple degrees of freedom for setting, which are different combinations for each sub-black box, to each of the multiple sub-black boxes, and acquires the index values ​​of the object to be evaluated corresponding to the values ​​of the given some parameters as sub-index values ​​from each sub-black box. Furthermore, the good solution identification device includes a good solution identification unit that identifies a combination of multiple combinations that is expected to be a good solution, based on the actual index values ​​acquired by the actual index value acquisition unit and the sub-index values ​​acquired by the sub-index value acquisition unit.

[0007] According to this disclosure, even if the number of parameters indicating the degrees of freedom of setting increases, the exponential increase in the processing load required to identify a good solution combination can be suppressed.

[0008] This is a configuration diagram showing the good solution identification device 3 according to Embodiment 1. This is a hardware configuration diagram showing the hardware of the good solution identification device 3 according to Embodiment 1. This is a hardware configuration diagram of a computer when the good solution identification device 3 is implemented by software or firmware, etc. This is a flowchart showing the good solution identification method, which is the processing procedure of the good solution identification device 3. This is a configuration diagram showing the good solution identification device 3 according to Embodiment 2. This is a hardware configuration diagram showing the hardware of the good solution identification device 3 according to Embodiment 2. This is a configuration diagram showing the good solution identification device 3 according to Embodiment 3. This is a hardware configuration diagram showing the hardware of the good solution identification device 3 according to Embodiment 3. This is an explanatory diagram showing an example of the function of the digital circuit to be evaluated, parameters indicating multiple degrees of freedom for setting the evaluation target, and values ​​that the parameters can take. This is an explanatory diagram showing an example of the cost function CF used by the combination extraction unit 15. This is a configuration diagram showing a system including the good solution identification device 3 and the circuit design device 51. This is a configuration diagram showing a system including the good solution identification device 3 and the warehouse management device 52. This is a configuration diagram showing a system including the good solution identification device 3 and the robot control device 53.

[0009] To provide a more detailed explanation of this disclosure, the forms for implementing this disclosure will be described below with reference to the attached drawings.

[0010] Embodiment 1. Figure 1 is a configuration diagram showing a good solution identification device 3 according to Embodiment 1. Figure 2 is a hardware configuration diagram showing the hardware of the good solution identification device 3 according to Embodiment 1. The good solution identification device 3 shown in Figure 1 comprises a real index value acquisition unit 11, a sub-index value acquisition unit 12, and a good solution identification unit 13. In the example of Figure 1, each of the black box 1 and sub-black boxes 2-1 to 2-N is provided outside the good solution identification device 3. However, this is merely an example, and each of the black box 1 and sub-black boxes 2-1 to 2-N may be provided inside the good solution identification device 3. N is an integer of 2 or more.

[0011] Black box 1 is implemented, for example, by a neural network. When black box 1 receives parameter values ​​indicating multiple degrees of freedom for the evaluation target from the good solution identification device 3, it outputs the index value of the evaluation target corresponding to the parameter values ​​indicating multiple degrees of freedom as an actual index value to the good solution identification device 3. Sub-black boxes 2-1 to 2-N are multiple black boxes included in black box 1. In other words, each of sub-black boxes 2-1 to 2-N is a black box resulting from the division of black box 1. Each of sub-black boxes 2-1 to 2-N is implemented, for example, by a neural network. When sub-black box 2-n (n=1, ..., N) receives parameter values ​​corresponding to sub-black box 2-n from the good solution identification device 3, it outputs the index value of the evaluation target corresponding to the parameter value as a sub-index value to the good solution identification device 3. Furthermore, sub-black box 2-n is not limited to those included in black box 1; it is sufficient if it outputs the evaluation target index values ​​corresponding to some of the values ​​of some of the parameters that indicate multiple degrees of freedom of configuration, when those parameters are given.

[0012] The evaluation target may include, for example, a digital circuit under development or a digital circuit under modification. Multiple degrees of freedom in settings include, for example, functional indicators such as communication quality, as well as circuit-related indicators such as circuit size, power consumption, or timing margin. Examples of applications of the good solution identification device 3 shown in Figure 1 include, for example, optimization of warehouse logistics, optimization of mobility operation plans, optimization of robot control, and optimization of the layout of building equipment or power equipment. Since the optimization problems using the good solution identification device 3 shown in Figure 1 can also be applied to dynamically changing objects, another example of application of the good solution identification device 3 is the optimization of operating conditions.

[0013] The actual index value acquisition unit 11 is implemented, for example, by the actual index value acquisition circuit 21 shown in Figure 2. The actual index value acquisition unit 11 provides the black box 1 with some of the multiple combinations of values ​​that each of the multiple parameters indicating the set degrees of freedom for the evaluation target can take. The actual index value acquisition unit 11 acquires the index values ​​of the evaluation target corresponding to some of the combinations from the black box 1 as actual index values. The actual index value acquisition unit 11 outputs the actual index values ​​to the good solution identification unit 13.

[0014] The sub-index value acquisition unit 12 is implemented, for example, by the sub-index value acquisition circuit 22 shown in Figure 2. The sub-index value acquisition unit 12 provides each of the multiple sub-black boxes 2-1 to 2-N with values ​​for some of the parameters that represent multiple degrees of freedom in setting, which are different combinations for each sub-black box. Specifically, the sub-index value acquisition unit 12 provides sub-black box 2-n with the values ​​of the parameters corresponding to sub-black box 2-n (n=1, ..., N) from among the multiple degrees of freedom in setting. The sub-index value acquisition unit 12 acquires the index values ​​of the evaluation target corresponding to the values ​​of the provided some parameters from sub-black box 2-n as sub-index values. The sub-index value acquisition unit 12 outputs the sub-index values ​​to the good solution identification unit 13.

[0015] The good solution identification unit 13 is implemented, for example, by the good solution identification circuit 23 shown in Figure 2. The good solution identification unit 13 includes a sub-index value correction unit 14, a combination extraction unit 15, a good candidate solution acquisition unit 16, and a good solution identification processing unit 17. The good solution identification unit 13 acquires actual index values ​​from the actual index value acquisition unit 11 and sub-index values ​​from the sub-index value acquisition unit 12. Based on the actual index values ​​and sub-index values, the good solution identification unit 13 identifies a combination that is a good solution from among multiple combinations. A good solution is a solution that is expected to be good. The number of combinations identified by the good solution identification unit 13 is not limited to one, but may be multiple. In addition to the optimal solution, which is the most suitable solution, good solutions may include the second most suitable solution, the third most suitable solution, and so on. For example, if there is a prior agreement that up to the third most suitable solution is included in good solutions, then good solutions will include the optimal solution, the second most suitable solution, and the third most suitable solution. Furthermore, if there is a prior agreement that the first four best solutions are included in the good solutions, then the good solutions will include the optimal solution, the second best solution, the third best solution, and the fourth best solution.

[0016] The sub-index value correction unit 14 includes a correction function calculation unit 14a and a corrected index value acquisition unit 14b. The sub-index value correction unit 14 acquires sub-index values ​​from the sub-index value acquisition unit 12. The sub-index value correction unit 14 corrects the sub-index values ​​and outputs the corrected sub-index values ​​to the combination extraction unit 15.

[0017] The correction function calculation unit 14a obtains actual index values ​​from the actual index value acquisition unit 11 and sub-index values ​​from the sub-index value acquisition unit 12. The correction function calculation unit 14a calculates a correction function to correct the sub-index values ​​obtained by the sub-index value acquisition unit 12 so that the error between the actual index values ​​and the corrected sub-index values ​​is reduced. The corrected index value acquisition unit 14b provides the sub-index values ​​obtained by the sub-index value acquisition unit 12 to the correction function and obtains the corrected sub-index values ​​from the correction function. The corrected index value acquisition unit 14b outputs the corrected sub-index values ​​to the combination extraction unit 15.

[0018] The combination extraction unit 15 obtains the corrected sub-index values ​​from the sub-index value correction unit 14. The combination extraction unit 15 extracts combinations corresponding to each of the multiple good candidate solutions from among the multiple combinations of values ​​that each of the multiple setting degrees of freedom parameters can take, based on the corrected sub-index values. The combination extraction unit 15 outputs the extracted combinations to the good candidate solution acquisition unit 16.

[0019] The good candidate solution acquisition unit 16 acquires combinations corresponding to each of the multiple good candidate solutions from the combination extraction unit 15. The good candidate solution acquisition unit 16 provides each combination extracted by the combination extraction unit 15 to the black box 1. The good candidate solution acquisition unit 16 acquires the evaluation target index value corresponding to each combination from the black box 1 as a good candidate solution. The good candidate solution acquisition unit 16 outputs the multiple good candidate solutions to the good solution identification processing unit 17.

[0020] The good solution identification processing unit 17 acquires multiple good candidate solutions from the good candidate solution acquisition unit 16. The good solution identification processing unit 17 compares the multiple good candidate solutions with each other and identifies a combination that constitutes a good solution based on the comparison results of the multiple good candidate solutions.

[0021] In Figure 1, it is assumed that the components of the good solution identification device 3, namely the actual index value acquisition unit 11, the sub-index value acquisition unit 12, and the good solution identification unit 13, are each implemented by dedicated hardware as shown in Figure 2. That is, it is assumed that the good solution identification device 3 is implemented by the actual index value acquisition circuit 21, the sub-index value acquisition circuit 22, and the good solution identification circuit 23. The actual index value acquisition circuit 21, the sub-index value acquisition circuit 22, and the good solution identification circuit 23 can each be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0022] The components of the optimal solution identification device 3 are not limited to those realized by dedicated hardware, and the optimal solution identification device 3 may be realized by software, firmware, or a combination of software and firmware. Software or firmware is stored in the memory of a computer as a program. A computer means hardware for executing a program, and for example, a CPU (Central Processing Unit), GPU (Graphics Processing Unit), central processing unit, processing device, arithmetic unit, microprocessor, microcomputer, processor, or DSP (Digital Signal Processor) is applicable.

[0023] FIG. 3 is a hardware configuration diagram of a computer when the optimal solution identification device 3 is realized by software, firmware, or the like. When the optimal solution identification device 3 is realized by software, firmware, or the like, a program for causing a computer to execute respective processing procedures in the actual index value acquisition unit 11, the sub-index value acquisition unit 12, and the optimal solution identification unit 13 is stored in the memory 31. Then, the processor 32 of the computer executes the program stored in the memory 31.

[0024] Also, FIG. 2 shows an example in which each component of the optimal solution identification device 3 is realized by dedicated hardware, and FIG. 3 shows an example in which the optimal solution identification device 3 is realized by software, firmware, or the like. However, this is only an example, and some components of the optimal solution identification device 3 may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, or the like.

[0025] Next, the operation of the good solution identification device 3 shown in FIG. 1 will be described. FIG. 4 is a flowchart showing a good solution identification method which is a processing procedure of the good solution identification device 3. The actual index value acquisition unit 11 acquires a plurality of parameters X[1] to X[I] as parameters indicating a plurality of degrees of freedom of setting for the evaluation target. I is an integer of 2 or more. X[i] (i = 1,..., I) includes J values x[i][1] to x[i][J] that the parameter X[i] can take, as shown in the following formula (1). J is an integer of 1 or more. However, if the value of i is different, the number J of values that the parameter X[i] can take is different.X[i] = {x[i][l],..., x[i][J]} (1)

[0026] The actual index value acquisition unit 11 selects some combinations from all combinations of values that each of the parameters X[1] to X[I] can take. All combinations of values that each can take are x[1][1],..., x[1][J], x[2][1],..., x[2][J],..., x[I][1],..., x[I][J]. The method of selecting some combinations by the actual index value acquisition unit 11 may be any method. For example, it may be a method of randomly selecting some combinations. Here, for convenience of explanation, the actual index value acquisition unit II is selecting some combinations from all combinations of values that each of the parameters X[1] to X[I] can take. However, this is only an example, and the actual index value acquisition unit 11 may select some combinations from two or more (a plurality) of combinations of values that each of the parameters X[1] to X[I] can take, rather than all combinations of values that each of the parameters X[1] to X[I] can take.

[0027] For the sake of explanation, here we assume that some of the combinations selected by the actual index value acquisition unit 11 are parameters X[m] (m = 1, ..., M). M is an integer of 2 or more. X[m] are two or more parameters from parameters X[1] to X[I]. For example, three parameters X[1], X[3], and X[I] are randomly selected from parameters X[1] to X[I], and then some of the possible values ​​that each of parameters X[1], X[3], and X[I] can take are randomly selected. In this case, X[m] contains some of the possible values ​​that each of parameters X[1], X[3], and X[I] can take. Alternatively, from the parameters X[1] to X[I], for example, five parameters X[2], X[4], X[5], X[I-2], and X[I] are randomly selected, and further, a portion of the possible values ​​for each of the parameters X[2], X[4], X[5], X[I-2], and X[I] are randomly selected. In this case, X[m] contains a portion of the possible values ​​for each of the parameters X[2], X[4], X[5], X[I-2], and X[I]. X[m] (m=1, ..., M) contains the K possible values ​​x[m][1] to x[m][K] for the parameter X[m], as shown in equation (2) below. X[m] = {x[m][1], ..., x[m][K]} (2) For example, if three parameters X[1], X[3], and X[I] are randomly selected, K is an integer less than or equal to the sum of the number of possible values ​​for each of the parameters X[1], X[3], and X[I].

[0028] The actual index value acquisition unit 11 provides the black box 1 with a parameter X[m] (m=1, ..., M) as a subset of combinations. When the black box 1 receives the parameter X[m] from the actual index value acquisition unit 11, it outputs the index values ​​to be evaluated corresponding to the K possible values ​​x[m][1] to x[m][K] of the parameter X[m] to the actual index value acquisition unit 11 as the actual index value Q[m][k] (m=1, ..., M: k=1, ..., K). The actual index value acquisition unit 11 acquires the actual index value Q[m][k] from the black box 1 (step ST1 in Figure 4). The actual index value acquisition unit 11 outputs the actual index value Q[m][k] to the good solution identification unit 13.

[0029] The sub-index value acquisition unit 12 acquires multiple parameters X[1] to X[I] as parameters indicating multiple degrees of freedom in setting for the evaluation target. The sub-index value acquisition unit 12 identifies the value of the parameter corresponding to the sub-black box 2-n (n=1, ..., N) from among the multiple parameters X[1] to X[I]. When, for example, parameters X[1], X[2], and X[3] are selected by the actual index value acquisition unit 11 from among the parameters X[1] to X[I], it is assumed that parameter X[1] relates to functional indicators such as communication quality, and parameters X[2] and X[3] relate to circuit indicators such as circuit size. For example, when N=2, it is assumed that sub-black box 2-1 outputs a sub-index value related to functional indicators, and sub-black box 2-2 outputs a sub-index value related to circuit indicators.

[0030] In such cases, the values ​​of parameter X[1] from x[1][1] to x[1][J] are identified as the parameter values ​​corresponding to sub-black box 2-1. Similarly, the values ​​of parameter X[2] from x[2][1] to x[2][J] and x[3][1] to x[3][J] are identified as the parameter values ​​corresponding to sub-black box 2-2. For the sake of explanation, the parameter corresponding to sub-black box 2-n is assumed to be parameter X[n] (n=1, ..., N), and the values ​​of parameter X[n] are assumed to be x[n][1] to x[n][G]. In the above specific example, G=J or G=2J.

[0031] The sub-index value acquisition unit 12 provides the parameter values ​​x[n][1] to x[n][G] corresponding to the sub-black box 2-n (n=1, ..., N) to the sub-black box 2-n. When the sub-black box 2-n receives the parameter values ​​x[n][1] to x[n][G] from the sub-index value acquisition unit 12, it outputs the index value to be evaluated corresponding to the parameter values ​​x[n][1] to x[n][G] to the sub-index value acquisition unit 12 as the sub-index value Q[n][g] (n=1, ..., N: g=1, ..., G). The sub-index value acquisition unit 12 acquires the sub-index value Q[n][g] from the sub-black box 2-n (step ST2 in Figure 4). The sub-index value acquisition unit 12 outputs the sub-index value Q[n][g] to the good solution identification unit 13.

[0032] The good solution identification unit 13 obtains the actual index value Q[m][k] (m=1,..., M: k=1,..., K) from the actual index value acquisition unit 11 and the sub-index value Q[n][g] (n=1,..., N: g=1,..., G) from the sub-index value acquisition unit 12. Based on the actual index value Q[m][k] and the sub-index value Q[n][g], the good solution identification unit 13 identifies the combination that is a good solution from among all combinations. If the actual index value acquisition unit 11 has selected some combinations from two or more combinations, the good solution identification unit 13 identifies the combination that is a good solution from among those two or more combinations. The process of identifying the combination that is a good solution by the good solution identification unit 13 will be described in detail below.

[0033] The correction function calculation unit 14a obtains the actual index value Q[m][k] (m=1, ..., M: k=1, ..., K) from the actual index value acquisition unit 11 and the sub-index value Q[n][g] (n=1, ..., N: g=1, ..., G) from the sub-index value acquisition unit 12. The correction function calculation unit 14a calculates a correction function H for correcting the sub-index value Q[n][g] so that the error between the actual index value Q[m][k] and the corrected sub-index value hQ[n][g] becomes small. ｎ The correction function H is calculated (step ST3 in Figure 4). Specifically, the correction function calculation unit 14a calculates the correction function H ｎAs an example, an expected value (or average value) of a result obtained by dividing the actual index value Q[m][k] by the sub-index value Q[n][g] is calculated. In Equation (7) described later, r[i] corresponds to the correction function H ｎ and h ｈａｔ [i][j] corresponds to the corrected sub-index value hQ[n][g]. Here, the correction function calculation unit 14a calculates an expected value (or average value) of a result obtained by dividing the actual index value Q[m][k] by the sub-index value Q[n][g]. However, this is merely an example, and the correction function calculation unit 14a may fit the actual index value Q[m][k] with a linear function of the sub-index value Q[n][g] and calculate the correction function H ｎ based on the fitting result. In this case, the correction function H ｎ is represented by fitting coefficients. Since fitting itself is a known technique, a detailed description thereof is omitted. In Equation (8) described later, A[i] and B[i] correspond to the correction function H ｎ . ｎ

[0034] The corrected index value acquisition unit 14b acquires the correction function H ｎ from the correction function calculation unit 14a and acquires the sub-index values Q[n][g] (n = 1, ···, N: g = 1, ···, G) from the sub-index value acquisition unit 12. The corrected index value acquisition unit 14b gives the sub-index value Q[n][g] to the correction function H ｎ and acquires the corrected sub-index value hQ[n][g] from the correction function H ｎ ​​​​The combination extraction unit 15 acquires multiple parameters X[1] to X[I] as parameters indicating multiple degrees of freedom for setting the object to be evaluated. The combination extraction unit 15 acquires the corrected sub-index value hQ[n][g] (n=1, ..., N: g=1, ..., G) from the sub-index value correction unit 14. From all combinations of the J values ​​x[i][1] to x[i][J] that each of the multiple parameters X[1] to X[I] can take, the combination extraction unit 15 selects multiple good candidate solutions C based on the corrected sub-index value hQ[n][g]. １ ~C Ｐ The combination corresponding to each of these is extracted (step ST5 in Figure 4). If the actual index value acquisition unit 11 has selected some combinations from two or more combinations, the combination extraction unit 15 selects multiple good candidate solutions C from two or more combinations of J values ​​x[i][1] to x[i][J]. １ ~C Ｐ Extract the combinations corresponding to each of these.

[0036] Specifically, the combination extraction unit 15 provides the corrected sub-index values ​​hQ[n][g] to the cost function CF, and extracts multiple good candidate solutions C from the cost function CF. １ ~C Ｐ The combination corresponding to each of these is obtained. For the cost function CF, refer to equations (3) to (4) and Figure 10 described later. The cost function CF is a function that converts the corrected sub-index value hQ[n][g] into the cost of either a functional index or a circuit index, and identifies the combination that minimizes the cost among all combinations of J values ​​x[i][1] to x[i][J] that each of the parameters X[1] to X[I] can take. The combination extraction unit 15 obtains multiple good candidate solutions C １ ~C Ｐ The combinations corresponding to each of these are output to the good candidate solution acquisition unit 16.

[0037] The good candidate solution acquisition unit 16 extracts multiple good candidate solutions C from the combination extraction unit 15. １ ~C Ｐ The unit 16 obtains the combination corresponding to each of the following. The unit 16 obtains multiple good candidate solutions C１ ~C Ｐ The black box 1 is given the combination corresponding to each of the following. When the black box 1 receives the combination corresponding to each from the good candidate solution acquisition unit 16, it receives the good candidate solution C corresponding to each of the following combinations. １ ~C Ｐ The output is sent to the good candidate solution acquisition unit 16. The good candidate solution acquisition unit 16 receives the good candidate solution C corresponding to each combination from the black box 1. １ ~C Ｐ The good candidate solution acquisition unit 16 acquires a plurality of good candidate solutions C. １ ~C Ｐ This is output to the good solution identification processing unit 17.

[0038] The good solution identification processing unit 17 receives a plurality of good candidate solutions C from the good candidate solution acquisition unit 16. １ ~C Ｐ The good solution identification processing unit 17 obtains multiple good candidate solutions C. １ ~C Ｐ We compare them with each other and select multiple good candidate solutions C １ ~C Ｐ Based on the comparison results, a combination that results in a good OS is identified (step ST7 in Figure 4). The process of identifying a good OS by the good solution identification processing unit 17 is, for example, a good candidate solution C. １ ~C Ｐ The distance between each of these points and a certain reference point is calculated, and a good candidate solution C is obtained. １ ~C Ｐ Within this process, there is a step in identifying good candidate solutions whose distance from the reference point is below a threshold as the good solution OS.

[0039] In the above embodiment 1, the good solution identification device 3 is configured to include: a real index value acquisition unit 11 that provides some of the multiple combinations of values ​​that each of the multiple parameters indicating the multiple degrees of freedom for setting the object to be evaluated can take to the black box 1, and acquires the index values ​​of the object to be evaluated corresponding to the some combinations from the black box 1 as real index values; and a sub-index value acquisition unit 12 that provides each of the multiple sub-black boxes 2-1 to 2-N with values ​​for some of the parameters that are among the multiple parameters indicating the multiple degrees of freedom for setting, and which are different combinations for each sub-black box, and acquires the index values ​​of the object to be evaluated corresponding to the values ​​of the given some parameters from each sub-black box 2-n (n=1, ..., N) as sub-index values. Therefore, the good solution identification device 3 can suppress the exponential increase in processing load for identifying good solution combinations, even when the number of parameters indicating the degrees of freedom of setting increases.

[0040] In Embodiment 1, the good solution identification device 3 is configured such that the good solution identification unit 13 includes a sub-index value correction unit 14 that corrects the sub-index value acquired by the sub-index value acquisition unit 12, a combination extraction unit 15 that extracts combinations corresponding to each of the multiple good candidate solutions from among a plurality of combinations based on the sub-index value corrected by the sub-index value correction unit 14, a good candidate solution acquisition unit 16 that provides each of the combinations extracted by the combination extraction unit 15 to the black box 1 and acquires the index value to be evaluated corresponding to each combination from the black box 1 as a good candidate solution, and a good solution identification processing unit 17 that compares the plurality of good candidate solutions acquired by the good candidate solution acquisition unit 16 with each other and identifies a combination that is a good solution based on the comparison result of the plurality of good candidate solutions. Therefore, even if the number of parameters indicating the degree of freedom of setting increases, the processing load of the process of identifying a combination that is a good solution is suppressed in the good solution identification device 3.

[0041] Embodiment 2. Embodiment 2 describes a good solution identification device 3 that includes a black box acquisition unit 18 and a black box division unit 19.

[0042] Figure 5 is a configuration diagram showing the good solution identification device 3 according to Embodiment 2. In Figure 5, the same reference numerals as in Figure 1 indicate the same or corresponding parts, so a detailed explanation is omitted. Figure 6 is a hardware configuration diagram showing the hardware of the good solution identification device 3 according to Embodiment 2. In Figure 6, the same reference numerals as in Figure 2 indicate the same or corresponding parts, so a detailed explanation is omitted. The good solution identification device 3 shown in Figure 5 includes a black box acquisition unit 18, a black box division unit 19, a real index value acquisition unit 11, a sub-index value acquisition unit 12, and a good solution identification unit 13.

[0043] The black box acquisition unit 18 is implemented, for example, by the black box acquisition circuit 28 shown in Figure 6. The black box acquisition unit 18 provides the Artificial Intelligence (AI) 40 with multiple functions of the object under evaluation and acquires the black box 1 from the Artificial Intelligence (AI) 40. The black box acquisition unit 18 outputs the black box 1 to the black box division unit 19.

[0044] The generative AI 40 is implemented using, for example, Llama3 or Claude3. Llama3 is an open-source large-scale language model announced by Meta on April 18, 2024. Claude3 is an interactive generative AI announced by Anthropic on March 14, 2023. When the generative AI 40 is given multiple functions of the evaluation target from the black box acquisition unit 18, it generates a black box 1 based on the multiple functions and outputs the black box 1 to the black box acquisition unit 18.

[0045] The black box division unit 19 is implemented, for example, by the black box division circuit 29 shown in Figure 6. The black box division unit 19 acquires black box 1 from the black box acquisition unit 18. The black box division unit 19 divides black box 1 into N sub-black boxes 2-1 to 2-N.

[0046] In Figure 5, it is assumed that each of the components of the good solution identification device 3—the black box acquisition unit 18, the black box division unit 19, the actual index value acquisition unit 11, the sub-index value acquisition unit 12, and the good solution identification unit 13—is implemented by dedicated hardware as shown in Figure 6. That is, it is assumed that the good solution identification device 3 is implemented by the black box acquisition circuit 28, the black box division circuit 29, the actual index value acquisition circuit 21, the sub-index value acquisition circuit 22, and the good solution identification circuit 23. Each of the black box acquisition circuit 28, the black box division circuit 29, the actual index value acquisition circuit 21, the sub-index value acquisition circuit 22, and the good solution identification circuit 23 can be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a combination thereof.

[0047] The components of the good solution identification device 3 are not limited to those implemented by dedicated hardware; the good solution identification device 3 may also be implemented by software, firmware, or a combination of software and firmware. When the good solution identification device 3 is implemented by software or firmware, a program for causing a computer to execute the respective processing procedures in the black box acquisition unit 18, the black box division unit 19, the actual index value acquisition unit 11, the sub-index value acquisition unit 12, and the good solution identification unit 13 is stored in the memory 31 shown in Figure 3. Then, the processor 32 shown in Figure 3 executes the program stored in the memory 31.

[0048] Furthermore, Figure 6 shows an example in which each component of the good solution identification device 3 is implemented by dedicated hardware, while Figure 3 shows an example in which the good solution identification device 3 is implemented by software or firmware, etc. However, this is only one example, and some components of the good solution identification device 3 may be implemented by dedicated hardware, while the remaining components may be implemented by software or firmware, etc.

[0049] Next, the operation of the good solution identification device 3 shown in Figure 5 will be explained. Except for the black box acquisition unit 18 and the black box division unit 19, it is the same as the good solution identification device 3 shown in Figure 1. Therefore, only the operation of the black box acquisition unit 18 and the black box division unit 19 will be explained here.

[0050] The black box acquisition unit 18 provides the generation AI 40 with multiple functions of the object under evaluation. For example, if the object under evaluation is a filter circuit included in a digital circuit, the functions of the object under evaluation may include, for example, the processing time per clock cycle, the number of samples per clock cycle, the bit width for representing one sample, the number of delay stages (number of taps) of the filter, the cutoff frequency, the amplitude ripple of the passband, or the attenuation of the stopband. When the generation AI 40 is given multiple functions of the object under evaluation, it generates a black box 1 based on the multiple functions and outputs the black box 1 to the black box acquisition unit 18. The generation process of black box 1 by the generation AI 40 is a known technique, so a detailed explanation is omitted. The black box acquisition unit 18 acquires the black box 1 from the generation AI 40. The black box acquisition unit 18 outputs the black box 1 to the black box division unit 19.

[0051] The black box division unit 19 acquires black box 1 from the black box acquisition unit 18. The black box division unit 19 divides black box 1 into N sub-black boxes 2-1 to 2-N. Black box 1 is generally generated by stacking multiple sub-black boxes, and the stacking structure of multiple sub-black boxes in black box 1 is clear. For this reason, the black box division unit 19 divides black box 1 into N sub-black boxes 2-1 to 2-N based on the stacking structure of multiple sub-black boxes in black box 1.

[0052] In the above embodiment 2, the good solution identification device 3 is configured to include a black box acquisition unit 18 that provides the generation AI 40 with multiple functions of the object to be evaluated and acquires the black box 1 from the generation AI 40. Therefore, even if the number of parameters indicating the degree of freedom of setting increases, the good solution identification device 3 can suppress the increase in processing load for identifying combinations that result in a good solution, and can also acquire the black box 1.

[0053] In the second embodiment, the good solution identification device 3 is configured to include a black box division unit 19 that divides the black box acquired by the black box acquisition unit 18 into sub-black boxes 2-1 to 2-N. Therefore, the good solution identification device 3 can acquire sub-black boxes 2-1 to 2-N.

[0054] Embodiment 3. Embodiment 3 describes a good solution identification device 3 equipped with a sub-black box acquisition unit 20.

[0055] Figure 7 is a configuration diagram showing the good solution identification device 3 according to Embodiment 3. In Figure 7, the same reference numerals as in Figures 1 and 5 indicate the same or corresponding parts, so a detailed explanation is omitted. Figure 8 is a hardware configuration diagram showing the hardware of the good solution identification device 3 according to Embodiment 3. In Figure 8, the same reference numerals as in Figures 2 and 6 indicate the same or corresponding parts, so a detailed explanation is omitted. The good solution identification device 3 shown in Figure 7 includes a black box acquisition unit 18, a sub-black box acquisition unit 20, a real index value acquisition unit 11, a sub-index value acquisition unit 12, and a good solution identification unit 13.

[0056] The sub-black box acquisition unit 20 is implemented, for example, by the sub-black box acquisition circuit 30 shown in Figure 8. The sub-black box acquisition unit 20 provides some of the functions of the object under evaluation to the generation AI 40 and acquires sub-black boxes 2-n (n=1, ..., N) from the generation AI 40. Some of the functions correspond to sub-black boxes 2-n.

[0057] In Figure 7, it is assumed that each of the components of the good solution identification device 3—the black box acquisition unit 18, the sub-black box acquisition unit 20, the actual index value acquisition unit 11, the sub-index value acquisition unit 12, and the good solution identification unit 13—is implemented by dedicated hardware as shown in Figure 8. That is, it is assumed that the good solution identification device 3 is implemented by the black box acquisition circuit 28, the sub-black box acquisition circuit 30, the actual index value acquisition circuit 21, the sub-index value acquisition circuit 22, and the good solution identification circuit 23. Each of the black box acquisition circuit 28, the sub-black box acquisition circuit 30, the actual index value acquisition circuit 21, the sub-index value acquisition circuit 22, and the good solution identification circuit 23 can be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a combination thereof.

[0058] The components of the good solution identification device 3 are not limited to those implemented by dedicated hardware; the good solution identification device 3 may also be implemented by software, firmware, or a combination of software and firmware. When the good solution identification device 3 is implemented by software or firmware, a program for causing a computer to execute the respective processing procedures in the black box acquisition unit 18, the sub-black box acquisition unit 20, the actual index value acquisition unit 11, the sub-index value acquisition unit 12, and the good solution identification unit 13 is stored in the memory 31 shown in Figure 3. Then, the processor 32 shown in Figure 3 executes the program stored in the memory 31.

[0059] Furthermore, Figure 8 shows an example in which each component of the good solution identification device 3 is implemented by dedicated hardware, while Figure 3 shows an example in which the good solution identification device 3 is implemented by software or firmware. However, this is merely one example, and some components of the good solution identification device 3 may be implemented by dedicated hardware, while the remaining components may be implemented by software or firmware.

[0060] Next, the operation of the good solution identification device 3 shown in Figure 7 will be explained. Except for the sub-black box acquisition unit 20, it is the same as the good solution identification device 3 shown in Figure 5. Therefore, only the operation of the sub-black box acquisition unit 20 will be explained here.

[0061] The sub-black box acquisition unit 20 provides the generation AI 40 with some of the multiple functions of the object under evaluation. These functions correspond to sub-black boxes 2-n (n = 1, ..., N). When the generation AI 40 receives some of the functions from the sub-black box acquisition unit 20, it generates sub-black boxes 2-n based on these functions. The process of generating sub-black boxes 2-n by the generation AI 40 is based on publicly known technology, so a detailed explanation is omitted.

[0062] In the above embodiment 3, the good solution identification device 3 is configured to include a sub-black box acquisition unit 20 that provides some of the multiple functions of the object to be evaluated to the generation AI 40 and acquires a sub-black box 2-n (n=1, ..., N) from the generation AI 40. Therefore, even if the number of parameters indicating the degree of freedom of setting increases, the good solution identification device 3 can suppress the increase in processing load for identifying combinations that result in a good solution, and can also acquire the sub-black box 2-n.

[0063] <Examples> Figure 9 is an explanatory diagram showing an example of the function of the digital circuit to be evaluated, parameters indicating multiple degrees of freedom in setting the evaluation target, and possible values ​​for the parameters. Figure 9 exemplifies static equalization function, clock synchronization function, adaptive equalization function, frequency reproduction function, and phase reproduction function as functions of the digital circuit. For example, Figure 9 exemplifies the number of filter stages and signal resolution as parameters for the static equalization function, with possible values ​​for the number of filter stages being 36, 40, 44, 48, 52, 56, 60, 64, 68, and 72, and possible values ​​for the signal resolution being 10, 11, 12, 13, 14, 15, 16, 17, 18, and 19. For example, Figure 9 exemplifies the signal resolution as a parameter for the clock synchronization function, with possible values ​​for the signal resolution being 10, 11, 12, 13, 14, 15, 16, 17, 18, and 19.

[0064] In the example in Figure 9, parameters P1 to P6 are shown, and since each parameter can take 10 values, the total number of possible combinations of values ​​is 10 to the power of 6 (= 1 million). If it takes 3 hours to obtain the actual index value for one combination from black box 1, then it would take approximately 3 million hours to obtain the actual index value for 1 million combinations. Therefore, the time required for a conventional good solution identification device to obtain the actual index value would be approximately 3 million hours.

[0065] On the other hand, since the number of possible values ​​for each of the parameters P1 and P2 related to the static equalization function is 10, the total number of possible combinations of values ​​is 10 squared (=100). Since the number of possible values ​​for the parameter P3 related to the clock synchronization function is 10, the total number of possible combinations of values ​​is 10. Since the number of possible values ​​for the parameter P4 related to the adaptive equalization function is 10, the total number of possible combinations of values ​​is 10. Since the number of possible values ​​for the parameter P5 related to the frequency reproduction function is 10, the total number of possible combinations of values ​​is 10. Since the number of possible values ​​for the parameter P6 related to the phase reproduction function is 10, the total number of possible combinations of values ​​is 10. The time required to obtain sub-index values ​​for 140 (=100+10+10+10+10) combinations from sub-black box 2-n (n=1, ..., N) is approximately 100 hours. Therefore, the time required for the good solution identification device 3 according to Embodiments 1 to 3 to acquire sub-indicator values ​​is approximately 100 hours.

[0066] Figure 10 is an explanatory diagram showing an example of the cost function CF used by the combination extraction unit 15. In Figure 10, the cost function CF is illustrated when the sub-indicator values ​​are EVM (Error Vector Magnitude), LUT (Look Up Table), Registers, DSP (Digital Signal Processor), WNS (Worst Negative Slack), WHS (Worst Hold Slack), and ChipPower (Power Consumption). Specifically, for EVM, LUT, Registers, DSP, and ChipPower, the cost function CF is the same as the cost function C of Type 1. ｈａｔ [i][j], and for WNS and WHS, the cost function CF is the same as the cost function C of Type 2. ｈａｔ [i][j]. Cost function C of Type 1. ｈａｔ [i] and [j] can be expressed, for example, as shown in equation (3) below, and the cost function C of Type 2. ｈａｔ [i] and [j] can be expressed, for example, as shown in equation (4) below.

[0067]

[0068]

[0069] In equations (3) to (8), w[j] is the cost weight, M[j] is the maximum cost, α[j] is the low-cost limit, β[j] is the tolerance limit, r[j] is the correction ratio, A[j] is the linear regression slope, and B[j] is the linear regression intercept.

[0070] The following describes specific application examples of the good solution identification device 3 according to Embodiments 1 to 3. <Application Example 1> As shown in Figure 11, the good solution identification device 3 outputs a combination that is a good solution to the circuit design device 51. The circuit design device 51 performs circuit design based on the combination that is a good solution. Figure 11 is a configuration diagram showing a system including the good solution identification device 3 and the circuit design device 51. <Application Example 2> As shown in Figure 12, the good solution identification device 3 outputs a combination that is a good solution to the warehouse management device 52. The warehouse management device 52 controls the flow of goods based on the combination that is a good solution. Specifically, the warehouse management device 52 may determine the location of the transport equipment that will transport goods, or the schedule of the transport equipment that will transport goods, based on the combination that is a good solution. Transport equipment includes, for example, mobility such as transport vehicles and transport robots, and fixed equipment such as conveyors. Examples of parameters include the number of units driven, the load capacity of each piece of equipment, the transport speed, and the transport route. Figure 12 is a configuration diagram showing a system including the good solution identification device 3 and the warehouse management device 52. <Application Example 3> As shown in Figure 13, the good solution identification device 3 outputs a combination that is a good solution to the robot control device 53. The robot control device 53 controls the robot's operation based on the combination that is a good solution. Specifically, the robot control device 53 determines the robot's action based on the combination that is a good solution and outputs a control signal to make the robot act. The robot control device 53 may also repeat such robot control. That is, when the robot control device 53 receives a command to the robot, it may acquire the current state of the robot, identify a combination that is a good solution as robot control from the acquired state until the command is completed, and perform an optimization loop process to determine the robot's next action. Examples of parameters include the current position of each joint of the robot, the direction of movement of each joint, the amount of movement, the amount of rotation, the drive sequence, etc. Figure 13 is a configuration diagram showing a system including the good solution identification device 3 and the robot control device 53.

[0071] Furthermore, this disclosure allows for free combination of each embodiment, modification of any component in each embodiment, or omission of any component in each embodiment.

[0072] The good solution identification device described herein can be used, for example, in robot motion control.

[0073] 1 Black box, 2-1 to 2-N sub-black boxes, 3 Good solution identification device, 11 Actual index value acquisition unit, 12 Sub-index value acquisition unit, 13 Good solution identification unit, 14 Sub-index value correction unit, 14a Correction function calculation unit, 14b Correction index value acquisition unit, 15 Combination extraction unit, 16 Good candidate solution acquisition unit, 17 Good solution identification processing unit, 18 Black box acquisition unit, 19 Black box division unit, 20 Sub-black box acquisition unit, 21 Actual index value acquisition circuit, 22 Sub-index value acquisition circuit, 23 Good solution identification circuit, 28 Black box acquisition circuit, 29 Black box division circuit, 30 Sub-black box acquisition circuit, 31 Memory, 32 Processor, 40 Generating AI, 51 Circuit design device, 52 Warehouse management device, 53 Robot control device.

Claims

1. A good solution identification device comprising: a real index value acquisition unit that provides some of the multiple combinations of values ​​that each of the multiple parameters indicating the multiple degrees of freedom for setting the object to be evaluated can take to a black box, and acquires the index value of the object to be evaluated corresponding to the given combination from the black box as a real index value; a sub-index value acquisition unit that provides each of the multiple sub-black boxes with values ​​for some of the parameters among the multiple parameters indicating the multiple degrees of freedom for setting, which are different combinations for each sub-black box, and acquires the index value of the object to be evaluated corresponding to the given values ​​of some of the parameters as a sub-index value from each sub-black box; and a good solution identification unit that identifies a combination of the multiple combinations that is expected to be a good solution, which is a good solution, based on the real index value acquired by the real index value acquisition unit and the sub-index value acquired by the sub-index value acquisition unit.

2. The good solution identification device according to claim 1, comprising: a sub-index value correction unit that corrects the sub-index values ​​obtained by the sub-index value acquisition unit; a combination extraction unit that extracts combinations corresponding to each of the multiple good candidate solutions from among the multiple combinations based on the sub-index values ​​corrected by the sub-index value correction unit; a good candidate solution acquisition unit that provides each of the combinations extracted by the combination extraction unit to the black box and obtains the index values ​​to be evaluated corresponding to each combination from the black box as good candidate solutions; and a good solution identification processing unit that compares the multiple good candidate solutions obtained by the good candidate solution acquisition unit with each other and identifies the combination that becomes the good solution based on the comparison result of the multiple good candidate solutions.

3. The good solution identification device according to claim 2, wherein the sub-index value correction unit comprises: a correction function calculation unit that calculates a correction function for correcting the sub-index value obtained by the sub-index value acquisition unit so that the error between the actual index value obtained by the actual index value acquisition unit and the corrected sub-index value is reduced; and a corrected index value acquisition unit that provides the sub-index value obtained by the sub-index value acquisition unit to the correction function and obtains the corrected sub-index value from the correction function.

4. The good solution identification device according to any one of claims 1 to 3, characterized in that it includes a black box acquisition unit that provides a plurality of functions of the object to be evaluated to a generating AI (Artificial Intelligence) and acquires the black box from the generating AI.

5. The good solution identification device according to claim 4, further comprising a black box division unit that divides the black box acquired by the black box acquisition unit into the sub-black boxes.

6. The good solution identification device according to any one of claims 1 to 4, characterized in that it includes a sub-black box acquisition unit that provides some of the multiple functions of the subject to evaluation to a generating AI (Artificial Intelligence) and acquires the sub-black box from the generating AI.

7. A method for identifying a good solution, wherein the actual index value acquisition unit provides a black box with some of the multiple combinations of values ​​that each of the multiple parameters indicating the set degrees of freedom for the object to be evaluated can take, and acquires the index value of the object to be evaluated corresponding to the some combination from the black box as an actual index value; the sub-index value acquisition unit provides each of the multiple sub-black boxes with the values ​​of some of the parameters among the multiple parameters indicating the set degrees of freedom, which are different combinations for each sub-black box, and acquires the index value of the object to be evaluated corresponding to the given values ​​of some of the parameters as a sub-index value; and the good solution identification unit identifies a combination of the multiple combinations that is expected to be a good solution, based on the actual index value acquired by the actual index value acquisition unit and the sub-index value acquired by the sub-index value acquisition unit.

Images