Parameter adjustment device and parameter adjustment method

Abstract

A parameter adjustment device (3) is provided with: a data acquisition unit (11) that acquires, from an evaluation value calculation device (2) that acquires operation results of an adjustment target device (1), which performs operation using one parameter or a plurality of parameters, to calculate evaluation values for the operation results, pieces of data including each parameter value and an evaluation value relating to the parameter value; and an elite solution extraction unit (12) that extracts, from among the plurality of pieces of data acquired by the data acquisition unit (11), zero or more pieces of data as an elite solution on the basis of the evaluation value included in each piece of data. In addition, the parameter adjustment device (3) is provided with a parameter value determination unit (14) that determines, from among a plurality of parameter values existing in a parameter value search space, a parameter value to be used for subsequent operation of the adjustment target device (1) on the basis of the elite solution extracted by the elite solution extraction unit (12), and outputs the determined parameter value to the adjustment target device (1).

Description

Parameter adjustment device and parameter adjustment method

[0001] The present disclosure relates to a parameter adjustment device and a parameter adjustment method.

[0002] There is a parameter adjustment method for searching for multiple parameters to be used in operating an apparatus to be adjusted. For example, Non-Patent Document 1 discloses an example of such a parameter adjustment method in which multiple optimizers search for parameters. The method disclosed in Non-Patent Document 1 has a reception function for receiving a designation of the number of optimizers. Each optimizer is software or the like for setting one parameter or multiple parameters that can be used in operating an apparatus to be adjusted.

[0003] "Discovering Many Diverse Solutions with Bayesian Optimization", October 2022, 10953rd paper submitted to arXiv

[0004] The method disclosed in Non-Patent Document 1 has a problem in that, when setting parameter values ​​(hereinafter referred to as "parameter values"), if the appropriate number of parameter values ​​is unknown, it is difficult for the user to specify an appropriate number as the number of optimizers to be executed. The appropriate number is a number that makes it possible to find many diverse and highly evaluated parameter values ​​within an acceptable range of deterioration in search efficiency, which is the efficiency of searching for appropriate parameter values. If the number of optimizers specified is too few than the appropriate number, the diversity of parameter values ​​is lost. If the number of optimizers specified is too many than the appropriate number, the parameter search efficiency deteriorates.

[0005] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a parameter adjustment device that can search for diverse parameter values ​​within an acceptable range of degradation in search efficiency, even if the appropriate number of parameter values ​​to be found is unknown.

[0006] A parameter adjustment device according to the present disclosure includes a data acquisition unit that acquires data including values ​​of each parameter and evaluation values ​​associated with each parameter value from an evaluation value calculation device that acquires operation results of an adjusted device that operates using one parameter or multiple parameters and calculates an evaluation value for the operation results, and an elite solution extraction unit that extracts zero or more pieces of data as elite solutions from the multiple pieces of data acquired by the data acquisition unit based on the evaluation values ​​included in each piece of data.The parameter adjustment device also includes a parameter value determination unit that determines parameter values ​​to be used by the adjusted device for its next operation based on the elite solution extracted by the elite solution extraction unit from the multiple parameter values ​​present in a search space of parameter values, and outputs the determined parameter values ​​to the adjusted device.

[0007] According to the present disclosure, even if the appropriate number of parameter values ​​to be found is unknown, it is possible to search for a variety of parameter values ​​within an acceptable range of deterioration in search efficiency.

[0008] 1 is a configuration diagram showing a system including a parameter adjustment device 3 according to a first embodiment; FIG. 2 is a hardware configuration diagram showing the hardware of the parameter adjustment device 3 according to the first embodiment; FIG. 3 is a hardware configuration diagram of a computer when the parameter adjustment device 3 is realized by software, firmware, or the like; FIG. 4 is a flowchart showing a parameter adjustment method, which is a processing procedure of the parameter adjustment device 3; and FIG. 5 is a diagram showing an elite solution E by an elite solution extraction unit 12. ｊ10 is a flowchart showing an extraction process for (j=1, ..., J). FIG. 11 is a flowchart showing a process for selecting a determination method by a determination method selection unit 14a. FIG. 12 is a flowchart showing a process for determining parameter values ​​by a parameter value determination processing unit 14c. FIG. 13 is a configuration diagram showing a system including a parameter adjustment device 3 according to a second embodiment. FIG. 14 is a hardware configuration diagram showing the hardware of a parameter adjustment device 3 according to the second embodiment. FIG. 15 is a hardware configuration diagram showing the hardware of a parameter adjustment device 3 according to the third embodiment. FIG. 16 is a configuration diagram showing a system including a parameter adjustment device 3 for an air conditioner according to a fifth embodiment.

[0009] In order to explain the present disclosure in more detail, embodiments of the present disclosure will be described below with reference to the accompanying drawings.

[0010] Embodiment 1. Simultaneously finding diverse combinations of multiple parameter values ​​(hereinafter referred to as "parameter values") can improve development efficiency, etc. For example, to find simulation scenarios in which a vehicle control system under development falls into an unsafe state, the parameter values ​​of scenario parameters may be automatically adjusted using risk as an evaluation value. In this case, automatically extracting as many diverse and dangerous scenarios as possible can improve development efficiency. There may also be evaluation items or constraints that can only be evaluated after parameter adjustment. For example, a mechanical device simulator may be used to adjust the parameter values ​​of control parameters in advance, and then the extracted control parameter values ​​may be used to control a real device to check for defects that cannot be reproduced in simulation. If only one parameter value is extracted, automatic adjustment using the simulator must be restarted if a defect occurs in the real device. However, by extracting multiple diverse parameter value combinations, the likelihood that one of the parameter value combinations will pass testing using the real device increases, resulting in improved development efficiency. A parameter adjustment device 3 is described below that can search for diverse combinations of multiple parameter values ​​within an acceptable range of degradation in search efficiency, even when the appropriate number of parameter values ​​to be found is unknown.

[0011] In order to explain the present disclosure in more detail, embodiments for carrying out the present disclosure will be described below with reference to the accompanying drawings. Fig. 1 is a configuration diagram showing a system including a parameter adjustment device 3 according to embodiment 1. Fig. 2 is a hardware configuration diagram showing the hardware of the parameter adjustment device 3 according to embodiment 1. The system shown in Fig. 1 includes an adjusted device 1, an evaluation value calculation device 2, and a parameter adjustment device 3.

[0012] The adjusted device 1 operates using one parameter or multiple parameters and outputs the operation result to the evaluation value calculation device 2. The adjusted device 1 may be, for example, a simulator for an autonomous driving system or a simulator for an air conditioning device. If the adjusted device 1 is, for example, an air conditioning device simulator, the parameters used for operating the adjusted device 1 may be, for example, an operation pattern of the frequency of a compressor or an operation pattern of an expansion valve.

[0013] The evaluation value calculation device 2 is an evaluation value calculation device that acquires operation results of the adjusted device 1 related to each parameter value from the adjusted device 1 that has been operated using the parameter values ​​of each parameter, and calculates an evaluation value for the operation results. If the parameter used to operate the adjusted device 1 is, for example, a compressor frequency operation pattern, the evaluation value for the operation results is, for example, how long it takes for the temperature of the space to be air-conditioned by the air conditioner to reach a set temperature based on pre-set environmental conditions of the air conditioner. The shorter the time it takes for the temperature of the space to be air-conditioned to reach the set temperature, the higher the evaluation value for the operation results. The evaluation value calculation device 2 outputs data including each parameter value and an evaluation value related to each parameter value to the parameter adjustment device 3. In the system shown in FIG. 1 , the evaluation value calculation device 2 is provided external to the parameter adjustment device 3. However, this is merely an example, and the evaluation value calculation device 2 may be provided inside the parameter adjustment device 3.

[0014] The parameter adjustment device 3 includes a dataset storage unit 10, a data acquisition unit 11, an elite solution extraction unit 12, an evaluation value prediction unit 13, a learning model 13a, and a parameter value determination unit 14. The parameter adjustment device 3 is a device that searches for multiple parameter values ​​to be used in operating the adjusted device 1. The dataset storage unit 10 is realized, for example, by the dataset storage circuit 20 shown in FIG. 2. The dataset storage unit 10 acquires data including each parameter value and an evaluation value associated with each parameter value from the evaluation value calculation device 2, and stores a combination of all previously acquired data as a dataset. A dataset includes one or more pieces of data.

[0015] The data acquisition unit 11 is realized by, for example, the data acquisition circuit 21 shown in Fig. 2. The data acquisition unit 11 acquires a dataset from the dataset storage unit 10. The data acquisition unit 11 outputs the dataset to each of the elite solution extraction unit 12 and the evaluation value prediction unit 13.

[0016] The elite solution extraction unit 12 is realized, for example, by the elite solution extraction circuit 22 shown in FIG. 2 . The elite solution extraction unit 12 acquires a dataset from the data acquisition unit 11. The elite solution extraction unit 12 extracts zero or more pieces of data from the dataset as elite solutions based on the evaluation values ​​included in each piece of data. Specifically, the elite solution extraction unit 12 extracts, from the data set, data whose evaluation values ​​are equal to or greater than a lower limit, as elite solutions. The elite solution extraction unit 12 outputs zero or more elite solutions to the parameter value determination unit 14.

[0017] The evaluation value prediction unit 13 is realized by, for example, an evaluation value prediction circuit 23 shown in FIG. 2. The evaluation value prediction unit 13 predicts an evaluation value of an operation result of the adjusted device 1 when the adjusted device 1 operates using each parameter value sampled by a sampling unit 14b (described later) of the parameter value determination unit 14. Specifically, upon receiving parameter values ​​from a parameter value determination processing unit 14c (described later) of the parameter value determination unit 14, the evaluation value prediction unit 13 provides the parameter values ​​to the learning model 13a and obtains, from the learning model 13a, an evaluation value corresponding to at least the parameter values ​​as a predicted result of the evaluation value of the operation result. The evaluation value prediction unit 13 outputs the predicted result of the evaluation value to the parameter value determination processing unit 14c.

[0018] The learning model 13a is realized by, for example, a Gaussian process regression model, a linear regression model, a neural network, a decision tree, a random forest, or a gradient boosting tree. During learning, the learning model 13a is given parameter values ​​contained in each data acquired by the data acquisition unit 11 and evaluation values ​​contained in each data, and learns evaluation values ​​corresponding to the respective parameter values. The evaluation values ​​are training data. During inference, when parameter values ​​are given from the evaluation value prediction unit 13, the learning model 13a outputs the evaluation value corresponding to the parameter value to the evaluation value prediction unit 13 as a predicted evaluation value. The parameter adjustment device 3 shown in FIG. 1 has the learning model 13a built in. However, this is merely an example, and the learning model 13a may be provided outside the parameter adjustment device 3.

[0019] The parameter value determiner 14 is realized by, for example, the parameter value determination circuit 24 shown in FIG. 2 . The parameter value determiner 14 includes a determination method selector 14a, a sampling unit 14b, and a parameter value determination processor 14c. The parameter value determiner 14 determines parameter values ​​to be used by the adjusted device 1 for the next operation based on the elite solution extracted by the elite solution extractor 12 from among a plurality of parameter values ​​present in the search space of parameter values. The parameter value determiner 14 outputs the determined parameter values ​​to the adjusted device 1.

[0020] The determination method selection unit 14a acquires zero or more elite solutions from the elite solution extraction unit 12. Based on the number of elite solutions extracted by the elite solution extraction unit 12, the determination method selection unit 14a selects either a global search method or a local search method as a method for determining parameter values ​​to be used in the next operation. The global search method is a search method that samples parameter values ​​according to a probability distribution that uniformly spreads throughout the entire search space of parameter values. The local search method is a search method that samples parameter values ​​from the search space of parameter values ​​according to a probability distribution centered on a certain elite solution. Specifically, if the number of elite solutions extracted by the elite solution extraction unit 12 is zero, the determination method selection unit 14a selects the global search method. If the number of elite solutions extracted by the elite solution extraction unit 12 is one or more, the determination method selection unit 14a selects either the global search method or the local search method based on the probability that the local search method will be selected. The probability that the local search method will be selected is calculated by the determination method selection unit 14 a based on the evaluation value included in the elite solution extracted by the elite solution extraction unit 12 .

[0021] If the determination method selection unit 14a selects the global search method, the sampling unit 14b samples one or more parameter values ​​according to a probability distribution that spreads uniformly throughout the entire search space of parameter values. If the determination method selection unit 14a selects the local search method, the sampling unit 14b samples one or more parameter values ​​according to a probability distribution centered on a certain elite solution within the search space of parameter values. The sampling unit 14b outputs the sampled parameter values ​​to the parameter value determination processing unit 14c.

[0022] The parameter value determination processing unit 14c acquires zero or more elite solutions from the elite solution extraction unit 12 and acquires one or more parameter values ​​from the sampling unit 14b. The parameter value determination processing unit 14c also acquires the predicted results of the evaluation values ​​from the evaluation value prediction unit 13. The parameter value determination processing unit 14c calculates the distance between each of the acquired parameter values ​​and the parameter values ​​included in each of the acquired elite solutions. Based on the calculated distance and the predicted results of the evaluation values, the parameter value determination processing unit 14c determines one parameter value to be used by the adjusted device 1 for the next operation from among the one or more parameter values ​​sampled by the sampling unit 14b.

[0023] 1, it is assumed that each of the components of the parameter adjustment device 3, namely, the dataset storage unit 10, the data acquisition unit 11, the elite solution extraction unit 12, the evaluation value prediction unit 13, and the parameter value determination unit 14, is realized by dedicated hardware as shown in Fig. 2. In other words, it is assumed that the parameter adjustment device 3 is realized by a dataset storage circuit 20, a data acquisition circuit 21, an elite solution extraction circuit 22, an evaluation value prediction circuit 23, and a parameter value determination circuit 24. Here, the data set storage circuit 20 may be, for example, a non-volatile or volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), or an EEPROM (Electrically Erasable Programmable Read Only Memory), a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, or a DVD (Digital Versatile Disc). Each of the data acquisition circuit 21, the elite solution extraction circuit 22, the evaluation value prediction circuit 23, and the parameter value determination circuit 24 may 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.

[0024] The components of the parameter adjustment device 3 are not limited to those realized by dedicated hardware, and the parameter adjustment device 3 may be realized by software, firmware, or a combination of software and firmware. Software or firmware is stored as a program in the memory of a computer. The computer refers to hardware that executes a program, and includes, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, a processor, or a DSP (Digital Signal Processor).

[0025] 3 is a hardware configuration diagram of a computer when the parameter adjustment device 3 is realized by software, firmware, or the like. When the parameter adjustment device 3 is realized by software, firmware, or the like, the dataset storage unit 10 is configured in the computer's memory 31. Programs for causing the computer to execute the respective processing procedures of the data acquisition unit 11, elite solution extraction unit 12, evaluation value prediction unit 13, and parameter value determination unit 14 are stored in the memory 31. The computer's processor 32 then executes the programs stored in the memory 31.

[0026] 2 shows an example in which each of the components of the parameter adjustment node 3 is implemented by dedicated hardware, while Fig. 3 shows an example in which the parameter adjustment node 3 is implemented by software, firmware, etc. However, this is merely an example, and some of the components in the parameter adjustment node 3 may be implemented by dedicated hardware, and the remaining components may be implemented by software, firmware, etc.

[0027] Next, the operation of the system shown in Fig. 1 will be described. Fig. 4 is a flowchart showing the parameter adjustment method, which is the processing procedure of the parameter adjustment device 3. The adjusted device 1 has I adjustable parameters p ｉThere exists a parameter p where i=1,...,I, and I is an integer equal to or greater than 1. ｉ The parameter value of x ｉ and as shown in the following equation (1), all parameter values ​​x ｉ The vector containing X is X = (x １ , x ２ , ..., x Ｉ ) (1) First, the parameter value determination unit 14 determines all the parameters p ｉ Initial value x (i=1,...,I) ｉ，ｉｎｉｔ is set in the adjusted device 1. ｉ The initial value x ｉ，ｉｎｉｔ may be determined using pseudo-random numbers, for example. ｉ The initial value x ｉ，ｉｎｉｔ is a parameter p that increases the evaluation value y of the operation result of the adjusted device 1. ｉ Parameter value x ｉ If known, such a parameter value may be used.

[0028] The adjusted device 1 operates using a parameter value X. The adjusted device 1 outputs the parameter value X and an operation result corresponding to the parameter value X to the evaluation value calculation device 2. If the adjusted device 1 is, for example, a simulator for an autonomous driving system, the operation result of the adjusted device 1 includes, for example, the positions (or relative positions) of the host vehicle and other vehicles, or the speeds (or relative speeds) of the host vehicle and other vehicles at multiple times included in a simulation period. If the adjusted device 1 is, for example, a simulator for an air conditioning device, the operation result of the adjusted device 1 includes, for example, a state quantity of the air conditioning. Examples of the state quantity of the air conditioning include the refrigerant temperature, the refrigerant pressure, the heating capacity, the cooling capacity, and the energy-saving efficiency.

[0029] The evaluation value calculation device 2 acquires a parameter value X and an operation result corresponding to the parameter value X from the adjusted device 1. The evaluation value calculation device 2 calculates an evaluation value y for the operation result. The calculation process of the evaluation value y by the evaluation value calculation device 2 is a known technique, and therefore a detailed description thereof will be omitted. If the adjusted device 1 is a simulator for an autonomous driving system, for example, when the subject vehicle approaches another vehicle, the shorter the distance between the subject vehicle and the other vehicle, the lower the evaluation value y for the operation result. If the adjusted device 1 is an air conditioning device simulator, for example, the time it takes for the temperature of a space to be air-conditioned by the air conditioning device to reach a set temperature is evaluated based on the environmental conditions of the air conditioning device that are set in advance. The shorter the time it takes for the temperature of the space to be air-conditioned to reach the set temperature, the higher the evaluation value for the operation result. Alternatively, the evaluation value may be calculated such that the higher the heating capacity or cooling capacity, the higher the evaluation value y for the operation result.

[0030] The evaluation value calculation device 2 outputs data D including a parameter value X and an evaluation value y to the parameter adjustment device 3 as shown in the following equation (2): D=(X, y) (2)

[0031] The data acquisition unit 11 of the parameter adjustment device 3 receives the stored M pieces of data D from the evaluation value calculation device 2. １ ~D Ｍ (Step ST1 in FIG. 4). The data acquisition unit 11 acquires M pieces of data D １ ~D Ｍ are output to the elite solution extraction unit 12 and the evaluation value prediction unit 13, respectively.

[0032] The elite solution extraction unit 12 receives M pieces of data D １ ~D Ｍ The elite solution extraction unit 12 obtains M pieces of data D １ ~D Ｍ From among them, Data D ｍ (m=1, . . . , M) ｍ Based on this, J data are divided into the elite solution E １ ~E Ｊ(Step ST2 in FIG. 4). j=1, ..., J, and j does not indicate a power of E but indicates an index. J is an integer between 0 and M. The elite solution extraction unit 12 extracts J elite solutions E １ ~E Ｊ to the parameter value determination unit 14.

[0033] Hereinafter, the elite solution E by the elite solution extraction unit 12 ｊ The extraction process of the elite solution E (j=1, . . . , J) will be specifically described. ｊ 1 is a flowchart showing the extraction process of M pieces of data D １ ~D Ｍ Let K be the elite solution candidates Ec １ ~Ec Ｋ K is an integer equal to or greater than 0, and at this stage, K is equal to M. The elite solution candidates are grouped together and called the elite solution candidate set Ec = {Ec １ , ..., Ec Ｋ Next, the elite solution extraction unit 12 extracts the elite solution E ｊ 5. The list E including the above is initialized (step ST11 in FIG. 5).

[0034] In formula (3), [□] indicates an empty list.

[0035] The elite solution extraction unit 12 extracts K elite solution candidates Ec １ ~Ec Ｋ The evaluation value y included in １ ~y Ｋ The elite solution extraction unit 12 compares the K evaluation values ​​y １ ~y Ｋ Among them, the highest evaluation value y ｍａｘ (Step ST12 in FIG. 5). The elite solution extraction unit 12 extracts the highest evaluation value y ｍａｘ and the lower limit of the evaluation value y Ｌ Compare with the lower limit y Ｌmay be stored in the internal memory of the elite solution extraction unit 12, or may be provided from outside the parameter adjustment device 3. The elite solution extraction unit 12 calculates the evaluation value y ｍａｘ is the lower limit y Ｌ If it is less than (NO in step ST13 of FIG. 5), the elite solution E ｊ The elite solution extraction unit 12 ends the extraction process of the evaluation value y ｍａｘ is the lower limit y Ｌ If it is equal to or greater than this (step ST13 in FIG. 5: YES), the evaluation value y ｍａｘ and the evaluation value y ｍａｘ and the parameter value X corresponding to the elite solution candidate Ec ｋ Elite solution E ｊ and add it to the end of the list E (step ST14 in FIG. 5). ｊ The parameter value X included in the elite solution candidate Ec ｋ (k=1, . . . , K) to distinguish the parameter values ​​X from the elite solution E ｊ The parameters included in X Ｅ ｊ Also, the elite solution E ｊ and the elite solution candidate Ec ｋ In order to distinguish the evaluation value y included in ｊ The evaluation value included in y Ｅ ｊ Here, we will write the elite solution E ｊ is the parameter value X ｊ and the evaluation value y ｊ However, this is just one example, and the elite solution E ｊ is the parameter value X ｊ It may also contain only

[0036] The elite solution extraction unit 12 updates the elite solution candidate set Ec with a new elite solution candidate set Ec′ (step ST15 in FIG. 5). １ ~Ec Ｋ Among them, a certain elite solution candidate Ec ｋ Regarding Ecｋ Included parameter value X ｋ and all elite solutions E included in list E ｊ The parameter value X included in Ｅ ｊ Distance d (X ｋ , X Ｅ ｊ The elite solution extraction unit 12 calculates all the calculated distances d(X ｋ , X Ｅ ｊ ) the minimum distance min d(X ｋ , X Ｅ ｊ ) is equal to or greater than the threshold τ, the elite solution candidate Ec ｋ The elite solution extraction unit 12 includes the minimum distance min d(X ｋ , X Ｅ ｊ ) is less than the threshold τ, the elite solution candidate Ec ｋ The elite solution extraction unit 12 extracts each elite solution candidate Ec ｋ , a new elite solution candidate set Ec' is generated, and this Ec' is then designated as the elite solution candidate set Ec. The threshold value τ may be stored in the internal memory of the elite solution extraction unit 12, or may be provided from outside the parameter adjustment device 3. The elite solution extraction unit 12 then performs the processes of steps ST12 to ST15 in FIG. 5.

[0037] In formula (4), d(X ｋ , X Ｅ ｊ ) is the parameter value X ｋ and parameter value X Ｅ ｊ This is a function that calculates the distance between

[0038] The function d(X ｋ , X Ｅ ｊ) is, for example, a function that uses distance in parameter space. Specifically, for example, the distance between parameters can be the Euclidean distance in a space spanned by the parameter vector X, a vector obtained by normalizing the parameter vector X, or a vector obtained by nonlinearly transforming the parameter vector X using the Kernel method, or the Manhattan distance in a space spanned by any of the vectors. Alternatively, the distance may be defined in a similar manner using one or more values ​​from among the parameter value, the driving result, and the evaluation value.

[0039] Here, the evaluation value y ｍａｘ is the lower limit y Ｌ If it is less than 1, the elite solution extraction unit 12 extracts the elite solution E ｊ The elite solution extraction unit 12 extracts the elite solution E included in the list E. ｊ Even when the number of exceeds a predetermined number, the elite solution E ｊ In this case, the extraction process of the elite solution E ｊ This can prevent a decrease in search efficiency due to an excessive increase in the number of

[0040] Here, the lower limit y of the evaluation value Ｌ are stored in the internal memory of the elite solution extraction unit 12. The elite solution extraction unit 12 extracts M evaluation values ​​y １ ~y Ｍ The highest evaluation value y ｍａｘ Based on this, the lower limit y Ｌ y may be calculated. Ｌ = α × y ｍａｘ +β (5) In equation (5), α and β are constants.

[0041] The parameter value determination unit 14 receives J elite solutions E １ ~E Ｊ The parameter value determination unit 14 obtains J elite solutions E from among a plurality of parameter values ​​present in the search space of parameter values. １ ~E ＪThe parameter value determining unit 14 determines the parameter values ​​to be used by the adjusted device 1 for the next operation based on the parameter values ​​(step ST3 in FIG. 4).

[0042] The parameter determination process by the parameter value determination unit 14 will be specifically described below. Fig. 6 is a flowchart showing the process of selecting a determination method by the determination method selection unit 14a. The determination method selection unit 14a selects J elite solutions E from the elite solution extraction unit 12. １ ~E Ｊ The determination method selection unit 14a obtains the elite solution E １ ~E Ｊ Based on the number of the elite solutions E extracted by the elite solution extracting unit 12, the determination method selecting unit 14a selects either the global search method or the local search method as a method for determining the parameter values ​​to be used in the next operation. ｊ If the number of is 0 (step ST21 in FIG. 6: YES), the global search method is selected (step ST22 in FIG. 6).

[0043] The determination method selection unit 14a selects the elite solution E extracted by the elite solution extraction unit 12. ｊ If the number of the local search methods is equal to or greater than 1 (step ST21 in FIG. 6: NO), the probability p that the local search method will be selected is calculated (step ST23 in FIG. 6). The calculation method of the probability p that the local search method will be selected may be any method, but the calculation method of the probability p may be, for example, a method that uses a predetermined probability. ｊ Alternatively, a method may be used in which the probability p that the local search will be selected is calculated according to the number of elite solutions E ｊ The probability p is calculated so that the larger the number of the local search parameters, the higher the probability p that the local search is selected. In this case, it is possible to obtain an effect that it is easier to find highly evaluated parameters. ｊ The evaluation value y included in ｊSpecifically, the determination method selection unit 14a may use a method of calculating the probability p that the local search method will be selected based on the J evaluation values ​​y １ ~y Ｊ or the sum of J evaluation values ​​y １ ~y Ｊ The probability p is calculated so that the higher the average value of , the higher the probability p that the local search method will be selected.

[0044] The determination method selection unit 14a selects either the global search method or the local search method based on the probability p (step ST24 in FIG. 6). If the determination method selection unit 14a selects the global search method (NO in step ST25 in FIG. 6), the determination method selection unit 14a ends the determination method selection process. If the determination method selection unit 14a selects the local search method (YES in step ST25 in FIG. 6), the determination method selection unit 14a selects the J elite solutions E １ ~E Ｊ The probability p of each of these being the target of local search e ｅ ｊ (Step ST26 in FIG. 6). Specifically, the determination method selection unit 14a determines, for example, the evaluation value y ｊ Low elite solution E ｊ The probability that e will be the target of local search is p ｅ ｊ So that probability p ｅ ｊ By doing so, the effect of increasing the evaluation value of the entire elite solution can be obtained. ｊ High elite solution E ｊ The probability that e will be the target of local search is p ｅ ｊ So that probability p ｅ ｊ In this way, the effect of increasing the maximum evaluation value of the elite solution can be obtained. In addition, the determination method selection unit 14a determines the probability p ｅ ｊ For example, each elite solution E ｊ From N ｃｌｏｓｅThe greater the distance to the nearest data, the greater the probability p ｅ ｊ So that probability p ｅ ｊ By doing this, elite solutions with sparse data distribution in the neighborhood are more likely to be selected, making it easier to explore unexplored areas.

[0045] The decision method selection unit 14a selects J elite solutions E １ ~E Ｊ The probability p of each of these being the target of local search e ｅ ｊ Based on this, J elite solutions E １ ~E Ｊ Among them, the elite solution E that is the target of local search is ｊ Specifically, the determination method selection unit 14a selects J elite solutions E by using pseudo-random numbers. １ ~E Ｊ Among them, the elite solution E that is the target of local search is ｊ with probability p ｅ ｊ The process of using pseudorandom numbers to select one option from a plurality of options according to a predetermined probability is a well-known technique, and therefore a detailed description thereof will be omitted.

[0046] If the global search method is selected by the determination method selection unit 14a, the sampling unit 14b samples the parameter value X hat from a uniform distribution over the entire search space of the parameter value, for example, using pseudorandom numbers. The process of sampling the parameter value X hat using pseudorandom numbers is a well-known technique, and therefore a detailed description thereof will be omitted. In the text of the specification, the letter "x" cannot be preceded by a "^" symbol due to electronic filing, so it is written as "x hat". The search space of the parameter value is a space that includes a plurality of parameter values ​​that can be used to operate the adjusted device 1. The search space of the parameter value may be any search space, but it is preferable that the search space of the parameter value is, for example, a two-dimensional space, with the first dimension of the two-dimensional space being x. １ , the second dimension is x２ If so, then [x １Ｌ , x ２Ｌ ], [x １Ｌ , x ２Ｈ ], [x １Ｈ , x ２Ｈ ], [x １Ｈ , x ２Ｌ The rectangle with vertices x becomes the search space. １Ｌ is the parameter value x １ The lower limit of x １Ｈ is the parameter value x １ The upper limit of x ２Ｌ is the parameter value x ２ The lower limit of x ２Ｈ is the parameter value x ２ In this case, x １Ｌ ≦x １ ≦x １Ｈ , and x ２Ｌ ≦x ２ ≦x ２Ｈ The values ​​that satisfy both of the above are the parameter values ​​included in the search space.

[0047] If the local search method is selected by the determination method selection unit 14a, the sampling unit 14b selects an elite solution E that is a target e of the local search in the search space of the parameter values. ｊ One or more parameter values ​​X are sampled from a normal distribution centered on x by using, for example, pseudo-random numbers. The timing at which the sampling unit 14b samples the parameter values ​​X is the timing shown in step ST2 of FIG. 7, which will be described later. A predetermined covariance matrix showing the normal distribution may be used, or the sampling unit 14b may adjust the covariance matrix of the normal distribution depending on the search situation. When the sampling unit 14b adjusts the covariance matrix, for example, the elite solution E, which is the target e of the local search, ｊ The evaluation value y ｊ The larger the value of X is, the smaller each element of the covariance matrix is ​​adjusted. The sampling unit 14b outputs the sampled parameter value X to the parameter value determination processing unit 14c.

[0048] 7 is a flowchart showing the parameter value determination process and the like performed by the parameter value determination processing unit 14c. The parameter value determination processing unit 14c initializes the candidate point set C with an empty set (step ST31 in FIG. 7). If the elite solution list E is empty, i.e., if the number J of elite solutions is 0, the parameter value determination processing unit 14c adds the parameter value X sampled by the sampling unit 14b to the candidate point set C. If the number J of elite solutions is 1 or greater, the parameter value determination processing unit 14c determines a prioritized elite solution (step ST33 in FIG. 7).

[0049] That is, if the local search method is selected by the determination method selection unit 14a, the parameter value determination processing unit 14c determines the J elite solutions E １ ~E Ｊ In this case, the elite solution E ｊ The elite solution E is sorted earlier than １ ~E ｊ－１ The preferred elite solution Ep １ ~Ep N N is the number of prioritized elite solutions, which is equal to (j-1) in this case. The elite solution E, which is the target e of the local search, is determined as follows: ｊ is the j-th elite solution E from the top of the list E ｊ The following equation (6) is the preferred elite solution Ep １ ~Ep Ｎ Ep = [Ep １ , ..., Ep Ｎ ] (6) The elite solution E, which is the target of the local search ｊ is at the top of the list E, the number N of prioritized elite solutions becomes 0, and the list Ep becomes empty. If the global search method is selected, the parameter value determination processing unit 14c determines the J elite solutions E included in the list E. １ ~E Ｊ The preferred elite solution Ep １ ~Ep N In this case, N is equal to J.

[0050] When the number N of prioritized elite solutions is 0, the parameter value determination processing unit 14c adds the parameter value X sampled by the sampling unit 14b to the candidate point set C. When the number N of prioritized elite solutions is 1 or more, the parameter value determination processing unit 14c compares the extracted parameter value X with the prioritized elite solutions Ep included in the list Ep. １ ~Ep Ｎ The parameter values ​​X contained in each ｐ ｎ (n=1,...,N) and the distance s(X hat, X ｐ ｎ ) is calculated (step ST34 in FIG. 7). ｐ ｎ ) is the above-mentioned distance d(X ｋ , X Ｅ ｊ Similarly to the above, the distance in the parameter space can be used. Alternatively, the evaluation value or the driving result may be predicted by using machine learning, and the distance may be calculated using the predicted result.

[0051] The parameter value determination processing unit 14c calculates the extracted parameter value X and each prioritized elite solution E ｐ ｎ Parameter value X included in ｐ ｎ (n=1, . . . , N) ｐ ｎ ) and a threshold value δ. The threshold value δ may be stored in an internal memory of the parameter value determination processing unit 14c, or may be provided from outside the parameter adjustment device 3. The parameter value determination processing unit 14c calculates the minimum distance min s(X hat, X ｐ ｎ If X is smaller than the threshold value δ, the parameter value X is added to the candidate point set C (step ST35 in FIG. 7).

[0052]

[0053] The parameter value determination processing unit 14c determines whether the number of parameter values ​​X included in the candidate point set C is N. ｃｎｄIf it has not reached (NO in step ST36 in FIG. 7), the processes of steps ST32 to ST36 are repeatedly executed. However, the process of step ST32 is a process performed by the sampling unit 14b. ｃｎｄ may be stored in an internal memory of the parameter value determination processing unit 14c, or may be provided from outside the parameter adjustment device 3. The parameter value determination processing unit 14c determines whether the number of parameter values ​​X included in the candidate point set C is N ｃｎｄ If it has reached (step ST36 in FIG. 7: YES), each parameter value X hat included in the candidate point set C is output to the evaluation value prediction unit 13.

[0054] The evaluation value prediction unit 13 acquires each parameter value X included in the candidate point set C from the parameter value determination processing unit 14c. The evaluation value prediction unit 13 provides each parameter value X to the learning model 13a and acquires, from the learning model 13a, an evaluation value y corresponding to at least each parameter value X. When the learning model 13a receives the parameter x from the evaluation value prediction unit 13 during inference, it outputs, as a predicted evaluation value, an evaluation value y corresponding to at least the parameter value X to the evaluation value prediction unit 13. The evaluation value prediction unit 13 outputs, as a predicted evaluation value, an evaluation value y corresponding to each parameter value X to the parameter value determination processing unit 14c. In the parameter adjustment device 3 shown in FIG. 1 , the evaluation value prediction unit 13 provides the parameter value X to the learning model 13a and acquires, from the learning model 13a, an evaluation value y corresponding to the parameter value X. However, the evaluation value prediction unit 13 only needs to be able to predict the evaluation value y corresponding to the parameter value X, and may predict the evaluation value y corresponding to the parameter value X without using the learning model 13a.

[0055] The parameter value determination processing unit 14c acquires at least an evaluation value y corresponding to each parameter value x from the evaluation value prediction unit 13 (step ST37 in FIG. 7). The parameter value determination processing unit 14c calculates an acquisition function value based on at least each evaluation value y (step ST38 in FIG. 7). The parameter value determination processing unit 14c uses an acquisition function to calculate the acquisition function value. Examples of acquisition functions include UCB (Upper Confidence Bound) and EI (Expected Improvement). Common acquisition functions such as UCB and EI assume that a distribution of evaluation values, such as a mean and variance, can be obtained when making predictions using machine learning, and are often adopted when Gaussian process regression is used as the machine learning method. When only a single deterministic value is obtained in prediction, an acquisition function that returns the predicted evaluation value as is can be used.

[0056] The parameter value determination processing unit 14c selects the parameter value X with the highest calculated acquisition function value from among the one or more parameter values ​​X acquired from the sampling unit 14b. The parameter value determination processing unit 14c outputs the parameter value X with the highest acquisition function value to the adjusted device 1 as the parameter value to be used by the adjusted device 1 for its next operation (step ST39 in FIG. 7). In the parameter adjustment device 3 shown in FIG. 1, the parameter value determination processing unit 14c outputs the parameter value X with the highest acquisition function value to the adjusted device 1 as the parameter value to be used by the adjusted device 1 for its next operation. However, the parameter value to be used by the adjusted device 1 for its next operation is not limited to the parameter value X with the highest acquisition function value. A parameter value X with a lower acquisition function value than the parameter value X with the highest acquisition function value may be output to the adjusted device 1 as the parameter value to be used by the adjusted device 1 for its next operation, within a practically acceptable range.

[0057] The adjusted device 1 operates using the parameter value X output from the parameter value determination processing unit 14c, and outputs the operation result to the evaluation value calculation device 2. Thereafter, the parameter adjustment device 3 repeatedly performs the processes of steps ST1 to ST3 shown in FIG.

[0058] In the first embodiment described above, the parameter adjustment device 3 is configured to include: a data acquisition unit 11 that acquires data including parameter values ​​and evaluation values ​​associated with each parameter value from an evaluation value calculation device 2 that acquires operation results of the adjusted device 1, which operates using one parameter or multiple parameters, and calculates an evaluation value for the operation results; and an elite solution extraction unit 12 that extracts zero or more pieces of data as elite solutions from the multiple pieces of data acquired by the data acquisition unit 11 based on the evaluation values ​​included in each piece of data. The parameter adjustment device 3 also includes a parameter value determination unit that determines parameter values ​​to be used by the adjusted device 1 for its next operation based on the elite solution extracted by the elite solution extraction unit 12 from multiple parameter values ​​present in a search space of parameter values, and outputs the determined parameter values ​​to the adjusted device 1. Therefore, even if the appropriate number of parameter values ​​to be found is unknown, the parameter adjustment device 3 can search for a variety of parameter values ​​within an acceptable range of degradation in search efficiency.

[0059] 1 , the evaluation value y calculated by the evaluation value calculation device 2 is a scalar value. However, this is merely an example, and the evaluation value calculation device 2 may calculate an evaluation vector Y with L dimensions as shown in the following equation (8), where L is an integer equal to or greater than 2.

[0060]

[0061] Then, the evaluation value calculation device 2 calculates the evaluation value y by calculating, for example, L elements y included in the evaluation vector Y as shown in the following equation (9): １ ~y ＬAs described above, the evaluation value calculation device 2 calculates the evaluation vector Y having the number of dimensions L, which makes it possible to reflect each element of the evaluation vector in the distance d, and to search for a variety of parameter values ​​even in the evaluation vector space.

[0062] In equation (9), w １ , ..., w Ｍ Each of these is a weighting factor.

[0063] Second Embodiment In a second embodiment, a parameter adjustment device 3 will be described in which the parameter value determiner 15 includes a determination method selector 15a that selects a parameter determination method based on the evaluation value included in the elite solution and the upper limit of the evaluation value.

[0064] Fig. 8 is a configuration diagram showing a system including a parameter adjustment device 3 according to embodiment 2. In Fig. 8, the same reference numerals as in Fig. 1 indicate the same or corresponding parts, and detailed description thereof will be omitted. Fig. 9 is a hardware configuration diagram showing the hardware of the parameter adjustment device 3 according to embodiment 2. In Fig. 9, the same reference numerals as in Fig. 2 indicate the same or corresponding parts, and detailed description thereof will be omitted. The system shown in Fig. 8 includes an adjusted device 1, an evaluation value calculation device 2, and a parameter adjustment device 3.

[0065] The parameter value determiner 15 is realized by, for example, a parameter value determination circuit 25 shown in FIG. 9 . The parameter value determiner 15 includes a determination method selector 15a, a sampling unit 14b, and a parameter value determination processor 14c. The parameter value determiner 15 determines parameter values ​​to be used by the adjusted device 1 for the next operation based on the elite solution extracted by the elite solution extractor 12 from among a plurality of parameter values ​​present in the search space of parameter values. The parameter value determiner 15 outputs the determined parameter values ​​to the adjusted device 1.

[0066] The decision method selection unit 15a selects J elite solutions E １ ~E Ｊ The determination method selection unit 15a obtains the elite solution E １ ~EＪ Based on the result, the determination method selection unit 15a selects either the global search method or the local search method as a method for determining parameter values ​​to be used in the next operation. Specifically, if the number J of elite solutions extracted by the elite solution extraction unit 12 is 0, the determination method selection unit 15a selects the global search method. １ ~E Ｊ The evaluation value y included in １ ~y Ｊ The lowest evaluation value y ｍｉｎ and the upper limit of the evaluation value y Ｈ The decision method selection unit 15a compares the evaluation value y ｍｉｎ is the upper limit y Ｈ If the evaluation value y is greater than or equal to the evaluation value y, the global search method is selected. ｍｉｎ is the upper limit y Ｈ If it is lower, the determination method selection unit 15a calculates the probability p that the local search method will be selected. The determination method selection unit 15a selects either the global search method or the local search method based on the probability p.

[0067] 8, it is assumed that the components of the parameter adjustment device 3, namely the data acquisition unit 11, the elite solution extraction unit 12, the evaluation value prediction unit 13, and the parameter value determination unit 15, are each realized by dedicated hardware such as that shown in Fig. 9. That is, it is assumed that the parameter adjustment device 3 is realized by a data acquisition circuit 21, an elite solution extraction circuit 22, an evaluation value prediction circuit 23, and a parameter value determination circuit 25. Each of the data acquisition circuit 21, the elite solution extraction circuit 22, the evaluation value prediction circuit 23, and the parameter value determination circuit 25 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a combination of these.

[0068] The components of the parameter adjustment device 3 are not limited to those realized by dedicated hardware, and the parameter adjustment device 3 may be realized by software, firmware, or a combination of software and firmware. When the parameter adjustment device 3 is realized by software or firmware, a program for causing a computer to execute the respective processing procedures of the data acquisition unit 11, the elite solution extraction unit 12, the evaluation value prediction unit 13, and the parameter value determination unit 15 is stored in a memory 31 shown in Fig. 3. Then, a processor 32 shown in Fig. 3 executes the program stored in the memory 31.

[0069] 9 shows an example in which each of the components of the parameter adjustment node 3 is implemented by dedicated hardware, while Fig. 3 shows an example in which the parameter adjustment node 3 is implemented by software or firmware, etc. However, this is merely an example, and some of the components in the parameter adjustment node 3 may be implemented by dedicated hardware, and the remaining components may be implemented by software or firmware, etc.

[0070] Next, the operation of the system shown in Fig. 8 will be described. However, apart from the determination method selection unit 15a of the parameter value determination unit 15, the system is the same as that shown in Fig. 1. Therefore, only the operation of the determination method selection unit 15a will be described here.

[0071] The decision method selection unit 15a selects J elite solutions E １ ~E Ｊ The determination method selection unit 15a obtains the elite solution E １ ~E Ｊ Based on the result, the determination method selection unit 15a selects either the global search method or the local search method as a method for determining parameter values ​​to be used in the next operation. Specifically, if the number J of elite solutions extracted by the elite solution extraction unit 12 is 0, the determination method selection unit 15a selects the global search method. １ ~E Ｊ The evaluation value y included in １ ~y ＪThe lowest evaluation value y ｍｉｎ and the upper limit of the evaluation value y Ｈ The decision method selection unit 15a compares the evaluation value y ｍｉｎ is the upper limit y Ｈ If the evaluation value y is greater than or equal to the evaluation value y, the global search method is selected. ｍｉｎ is the upper limit y Ｈ If the probability p is lower than the threshold, the determination method selection unit 15a calculates the probability p that the local search method will be selected. The determination method selection unit 15a selects either the global search method or the local search method based on the probability p. If the determination method selection unit 15a selects the global search method, it ends the determination method selection process. If the determination method selection unit 15a selects the local search method, it calculates the probability p that the local search method will be selected. １ ~E Ｊ The probability p of each of these being the target of local search e ｅ ｊ In the second embodiment, the elite solution E ｊ The evaluation value y included in ｊ is the upper limit y Ｈ The probability p corresponding to the elite solution is greater than or equal to ｅ ｊ is set to 0. The determination method selection unit 15a outputs the selection result of the parameter value determination method to the sampling unit 14b.

[0072] In the second embodiment described above, the parameter adjustment device 3 shown in FIG. 8 is configured so that the determination method selection unit 15a selects a parameter determination method based on the evaluation value included in the elite solution and the upper limit of the evaluation value. Therefore, like the parameter adjustment device 3 shown in FIG. 1, the parameter adjustment device 3 shown in FIG. 8 can search for diverse parameter values ​​within an acceptable range of degradation in search efficiency, even if the appropriate number of parameter values ​​to be found is unknown. Furthermore, the parameter adjustment device 3 shown in FIG. 8 has a higher probability of sampling parameter values ​​from a search space centered on elite solution candidates whose evaluation values ​​are equal to or less than the upper limit, compared to the parameter adjustment device 3 shown in FIG. 1. As a result, the parameter adjustment device 3 shown in FIG. 8 can search for diverse parameter values ​​more frequently than the parameter adjustment device 3 shown in FIG. 1.

[0073] Third Embodiment In the third embodiment, if the local search method is selected by the decision method selection unit 14a, M pieces of data D １ ~D Ｍ Among them, the elite solution E ｊ The parameter adjustment device 3 is provided with a learning model 16a that extracts data in the vicinity of the parameter value X as learning data and learns an evaluation value y corresponding to the parameter value X included in the learning data.

[0074] Fig. 10 is a configuration diagram showing a system including a parameter adjustment device 3 according to embodiment 3. In Fig. 10, the same reference numerals as in Figs. 1 and 8 indicate the same or corresponding parts, and detailed description thereof will be omitted. Fig. 11 is a hardware configuration diagram showing the hardware of the parameter adjustment device 3 according to embodiment 3. In Fig. 11, the same reference numerals as in Figs. 2 and 9 indicate the same or corresponding parts, and detailed description thereof will be omitted. The system shown in Fig. 10 includes an adjusted device 1, an evaluation value calculation device 2, and a parameter adjustment device 3.

[0075] The evaluation value prediction unit 16 is realized by, for example, an evaluation value prediction circuit 26 shown in Fig. 11. The evaluation value prediction unit 16 predicts an evaluation value of the operation result of the adjusted device 1 when the adjusted device 1 operates using each parameter value sampled by the sampling unit 14b of the parameter value determination unit 14. If the global search method is selected by the determination method selection unit 14a during learning of the learning model 16a, the evaluation value prediction unit 16 predicts an evaluation value of the operation result of the adjusted device 1 when the adjusted device 1 operates using each parameter value sampled by the sampling unit 14b of the parameter value determination unit 14. １ ~D Ｍ If the local search method is selected by the decision method selection unit 14a during learning of the learning model 16a, the evaluation value prediction unit 16 provides each of the M pieces of data D １ ~D Ｍ Among them, the elite solution E ｊ The data in the vicinity of the vector .theta. is extracted as learning data, and the learning data is provided to the learning model 16a.

[0076] The learning model 16a is realized by, for example, a Gaussian process regression model, a linear regression model, a neural network, a decision tree, a random forest, or a gradient boosting tree. If the global search method is selected by the decision method selection unit 14a during learning, the learning model 16a receives the data D acquired by the data acquisition unit 11 from the evaluation value prediction unit 16. ｍ (m=1, . . . , M). The learning model 16a acquires data D ｍ (m=1, . . . , M) is used to learn the evaluation value y corresponding to the parameter value X included in the data D. １ ~D Ｍ Among them, the elite solution E ｊ The learning model 16a acquires data in the vicinity of the parameter value X as learning data. The learning model 16a learns an evaluation value y corresponding to the parameter value X included in the learning data. When the parameter value X is given from the evaluation value prediction unit 16 during inference, the learning model 16a outputs the evaluation value y corresponding to the parameter value X to the evaluation value prediction unit 16 as a prediction result of the evaluation value.

[0077] The parameter adjustment device 3 shown in Fig. 10 has a built-in learning model 16a. However, this is merely an example, and the learning model 16a may be provided externally to the parameter adjustment device 3. In the parameter adjustment device 3 shown in Fig. 10, the evaluation value prediction unit 16 and the learning model 16a are each applied to the parameter adjustment device 3 shown in Fig. 1. However, this is merely an example, and the evaluation value prediction unit 16 and the learning model 16a may each be applied to the parameter adjustment device 3 shown in Fig. 8.

[0078] 10, it is assumed that the components of the parameter adjustment device 3, namely the data acquisition unit 11, the elite solution extraction unit 12, the evaluation value prediction unit 16, and the parameter value determination unit 14, are each realized by dedicated hardware such as that shown in Fig. 11. That is, it is assumed that the parameter adjustment device 3 is realized by a data acquisition circuit 21, an elite solution extraction circuit 22, an evaluation value prediction circuit 26, and a parameter value determination circuit 24. Each of the data acquisition circuit 21, the elite solution extraction circuit 22, the evaluation value prediction circuit 26, and the parameter value determination circuit 24 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a combination thereof.

[0079] The components of the parameter adjustment device 3 are not limited to those realized by dedicated hardware, and the parameter adjustment device 3 may be realized by software, firmware, or a combination of software and firmware. When the parameter adjustment device 3 is realized by software or firmware, a program for causing a computer to execute the respective processing procedures of the data acquisition unit 11, the elite solution extraction unit 12, the evaluation value prediction unit 16, and the parameter value determination unit 14 is stored in a memory 31 shown in Fig. 3. Then, a processor 32 shown in Fig. 3 executes the program stored in the memory 31.

[0080] 11 shows an example in which each of the components of the parameter adjustment node 3 is implemented by dedicated hardware, while Fig. 3 shows an example in which the parameter adjustment node 3 is implemented by software, firmware, etc. However, this is merely an example, and some of the components in the parameter adjustment node 3 may be implemented by dedicated hardware, and the remaining components may be implemented by software, firmware, etc.

[0081] Next, the operation of the system shown in Fig. 10 will be described. However, apart from the evaluation value prediction unit 16 and the learning model 16a, the system is the same as the system shown in Fig. 1. Therefore, only the operation of the evaluation value prediction unit 16 and the learning model 16a will be described here.

[0082] The evaluation value prediction unit 16 acquires the selection result of the search method from the determination method selection unit 14 a. If the selection result of the search method indicates the global search method, the evaluation value prediction unit 16 acquires M pieces of data D １ ~D Ｍ and obtain M pieces of data D １ ~D Ｍ The learning model 16a receives M pieces of data D １ ~D Ｍ Given the data D ｍ The evaluation value y corresponding to the parameter value X included in (m=1, . . . , M) is learned.

[0083] If the result of the selection of the search method indicates the local search method, the evaluation value prediction unit 16 １ ~D Ｍ Among them, the elite solution E ｊ The learning model 16a extracts data in the vicinity of the parameter X and provides the data to the learning model 16a. The data in the vicinity may be, for example, data within the 3σ confidence interval of the normal distribution used by the sampling unit 14b when sampling. Some machine learning techniques allow weighting for each piece of data. Therefore, all M pieces of data may be used as learning data, and learning may be performed by assigning a weight to each piece of data that is closer to the elite solution. During learning, the learning model 16a learns an evaluation value y corresponding to a parameter value X included in the learning data.

[0084] When the evaluation value prediction unit 16 receives the parameter value X from the parameter value determination processing unit 14c, it provides the parameter value X to the learning model 16a and obtains the evaluation value y corresponding to the parameter value X from the learning model 16a. During inference, when the parameter value X is provided from the evaluation value prediction unit 16, the learning model 16a outputs the evaluation value y corresponding to the parameter value X to the evaluation value prediction unit 16 as the predicted evaluation value. The evaluation value prediction unit 16 outputs the evaluation value y corresponding to the parameter value X to the parameter value determination processing unit 14c as the predicted evaluation value.

[0085] When the local search is selected by the determination method selection unit 14a, the parameter value X included in the candidate point set C is the elite solution E ｊ Therefore, the elite solution E ｊ It is not a problem if the prediction accuracy is low at positions far from the target.

[0086] In the third embodiment, if the local search method is selected by the determination method selection unit 14a, M pieces of data D １ ~D Ｍ Among them, the elite solution E ｊ 10 is configured to include a learning model 16a that learns an evaluation value y corresponding to a parameter value X included in data in the neighborhood of . Therefore, the parameter adjustment device 3 shown in Fig. 10 can improve the prediction accuracy of the evaluation value when the local search method is selected by the determination method selection unit 14a compared to the parameter adjustment device 3 shown in Fig. 1.

[0087] Embodiment 4. The parameter adjustment device 3 according to embodiments 1 to 3 can be applied to, for example, a simulation scenario generation device for an autonomous driving system. When the parameter adjustment device 3 is applied to a simulation scenario generation device, it becomes possible to generate a scenario in which an autonomous driving system under development falls into an unsafe situation, thereby improving the efficiency of safety verification of the autonomous driving system.

[0088] The scenario parameter used when the autonomous driving system executes the above scenario is the parameter value X used for driving the adjusted device 1. For example, at least one of the risk level or the discomfort level is used as an evaluation value for the driving result of the adjusted device 1. The risk level is an index indicating the extent to which the autonomous driving system has caused the host vehicle to fall into a dangerous situation. For example, in the case of the risk level of a collision between the host vehicle and another vehicle, the risk level can be calculated by multiplying the closest inter-vehicle distance between the host vehicle and the other vehicle by -1. Furthermore, the risk level for breaking a traffic rule can be calculated by, for example, the importance of the broken traffic rule or the degree to which the traffic rule was broken. The risk level may be expressed as a scalar value or a vector with a value for each item.

[0089] In an autonomous driving system, it is necessary to avoid not only dangerous driving but also driving that makes occupants feel uncomfortable. The discomfort level indicates the degree to which occupants feel uncomfortable. Specific examples of the discomfort level include the acceleration of the host vehicle or the maximum absolute value of the jerk, which is the jerk of the host vehicle.

[0090] In the fourth embodiment, the elite solution extraction unit 12 can use, for example, the Euclidean distance in the space spanned by z shown in the following equation (10) as the distance d that defines the closeness between parameter values.

[0091] In equation (10), Tc is the time when the host vehicle is closest to the other vehicle, T is a constant, and z is an arrangement of two-dimensional relative positions (X(t), Y(t)) of the other vehicle as seen from the host vehicle.

[0092] Fifth Embodiment The parameter adjustment device 3 according to the first to third embodiments can be used, for example, as a control parameter adjustment device for an air conditioning system. Generally, the execution time of an air conditioning simulator is advantageously much shorter than the execution time of a real system. However, because an air conditioning simulator cannot reproduce some of the behavior of a real system, parameter values ​​that are highly evaluated on the simulator may be poorly evaluated on the real system, or may even cause problems. Compared to a single parameter value obtained using an air conditioning simulator, obtaining multiple diverse parameter values ​​increases the likelihood of obtaining parameter values ​​that achieve high evaluations on the real system. In the fifth embodiment, efficient parameter adjustment is enabled by using an air conditioning simulator in combination with a real system.

[0093] Fig. 12 is a configuration diagram showing a system including a parameter adjustment device 3 for an air conditioner according to embodiment 5. In Fig. 12, the same reference numerals as those in Figs. 1, 8, and 10 indicate the same or corresponding parts, and detailed description thereof will be omitted. The system shown in Fig. 12 includes an adjusted device 1, an evaluation value calculation device 2, a parameter adjustment device 3, a parameter transmission unit 41, an actual air conditioner 42, and an actual operation result storage unit 43.

[0094] In the system shown in FIG. 12 , the adjusted device 1 is an air conditioner simulator. The adjusted device 1 has a function of calculating state changes of the air conditioner according to an actuator operation pattern determined based on control parameters. State quantities of the air conditioner are obtained as operation results. Examples of state quantities include refrigerant temperature, refrigerant pressure, heating capacity, cooling capacity, energy saving efficiency, or the temperature, pressure, and enthalpy of each element constituting the air conditioner. The parameter adjustment device 3 shown in FIG. 12 is similar to the parameter adjustment device 3 shown in FIG. 1 . However, this is merely an example, and the parameter adjustment device 3 shown in FIG. 12 may be similar to either the parameter adjustment device 3 shown in FIG. 8 or the parameter adjustment device 3 shown in FIG. 10 .

[0095] In the fifth embodiment, the control parameter that determines the actuator operation pattern is a parameter value X used for operating the adjusted device 1. As an evaluation value for the operation result of the adjusted device 1, for example, at least one of the refrigerant temperature, refrigerant pressure, heating capacity, cooling capacity, energy saving efficiency, and startup time is used. The startup time is the time required for the internal state of the air conditioner to reach a steady state.

[0096] In the fifth embodiment, the elite solution extraction unit 12 can use, for example, the Euclidean distance in the space spanned by z shown in the following equation (11) as the distance d that defines the closeness between parameter values.

[0097] In equation (11), Pd is the pressure at the outlet of the compressor of the air conditioner, and Ps is the pressure at the inlet of the compressor of the air conditioner. Δt is the discrete time width, and N is the number of simulation steps. z is a list of the pressures Pd and Ps at time t.

[0098] The parameter sending unit 41 sends the J elite solutions E extracted by the elite solution extracting unit 12 of the parameter adjusting device 3. １ ~E Ｊ The parameter transmission unit 41 obtains the elite solution E ｊ The parameter value X included in (j=1, . . . , J) is transmitted to the actual air conditioner 42 .

[0099] The actual air conditioner 42 receives the parameter value X from the parameter transmission unit 41. The actual air conditioner 42 performs air conditioning operation by setting an actuator operation pattern based on the parameter value X. The actual operation result storage unit 43 acquires the operating state of the actual air conditioner 42 and stores the operating state in association with the parameter value X.

[0100] According to the fifth embodiment, it is possible to use diverse control parameters with high evaluation values ​​to verify the operating state of the actual air conditioner 42. Furthermore, by adjusting the parameters using a simulator, it is possible to significantly reduce the adjustment time while increasing the probability of obtaining parameter values ​​with high evaluation values ​​when applied to the actual air conditioner 42.

[0101] In addition, the present disclosure allows for free combination of the respective embodiments, modification of any of the components of the respective embodiments, or omission of any of the components of the respective embodiments.

[0102] The present disclosure is suitable for a parameter adjustment device and a parameter adjustment method.

[0103] 1 Adjusted device, 2 Simulator (evaluation value calculation device), 3 Parameter adjustment device, 10 Data set storage unit, 11 Data acquisition unit, 12 Elite solution extraction unit, 13, 16 Evaluation value prediction unit, 13a, 16a Learning model, 14 Parameter value determination unit, 14a Determination method selection unit, 14b Sampling unit, 14c Parameter value determination processing unit, 15 Parameter value determination unit, 15a Determination method selection unit, 20 Data set storage circuit, 21 Data acquisition circuit, 22 Elite solution extraction circuit, 23, 26 Evaluation value prediction circuit, 24, 25 Parameter value determination circuit, 31 Memory, 32 Processor, 41 Parameter transmission unit, 42 Actual air conditioning device, 43 Actual operation result storage unit.

Claims

1. A parameter adjustment device comprising: a data acquisition unit that acquires data including each parameter value and an evaluation value related to each parameter value from an evaluation value calculation device that acquires operation results of an adjusted device that operates using one parameter or multiple parameters and calculates an evaluation value for the operation results; an elite solution extraction unit that extracts zero or more pieces of data as elite solutions from the multiple pieces of data acquired by the data acquisition unit based on the evaluation value included in each piece of data; and a parameter value determination unit that determines parameter values ​​to be used by the adjusted device for the next operation from the multiple parameter values ​​existing in a search space of parameter values ​​based on the elite solution extracted by the elite solution extraction unit, and outputs the determined parameter values ​​to the adjusted device.

2. The parameter value determination unit of the parameter adjustment device described in claim 1, characterized in that it comprises: a decision method selection unit that selects either a global search method or a local search method as a method for determining the values ​​of parameters to be used in the next operation based on the elite solution extracted by the elite solution extraction unit; a sampling unit that, if the global search method is selected by the decision method selection unit, samples the values ​​of one or more parameters according to a probability distribution that spreads uniformly throughout the entire search space, and, if the local search method is selected by the decision method selection unit, samples the values ​​of one or more parameters in the search space according to a probability distribution centered on a certain elite solution; and a parameter value determination processing unit that calculates a distance between each parameter value sampled by the sampling unit and a parameter value included in the elite solution extracted by the elite solution extraction unit, and determines, based on the distance, from the values ​​of one or more parameters sampled by the sampling unit, the values ​​of parameters to be used by the adjusted device for the next operation.

3. The parameter adjustment device described in claim 2, characterized in that the determination method selection unit selects either the global search method or the local search method based on the probability that the local search method will be selected if the number of elite solutions extracted by the elite solution extraction unit is one or more.

4. The parameter adjustment device described in claim 2, characterized in that if the number of elite solutions extracted by the elite solution extraction unit that contain an evaluation value lower than an upper limit value is one or more, the decision method selection unit selects either the global search method or the local search method based on the probability that the local search method will be selected.

5. A parameter adjustment device as described in claim 3 or 4, characterized in that the decision method selection unit calculates the probability that the local search method will be selected based on the evaluation value contained in the elite solution extracted by the elite solution extraction unit.

6. A parameter adjustment device as described in any one of claims 2 to 5, further comprising an evaluation value prediction unit which predicts an evaluation value for the operation result of the adjusted device when the adjusted device operates using the values ​​of each parameter sampled by the sampling unit, and the parameter value determination processing unit determines a parameter value to be used by the adjusted device for the next operation from among the values ​​of one or more parameters sampled by the sampling unit, based on the distance and the predicted result of the evaluation value by the evaluation value prediction unit.

7. The parameter adjustment device according to claim 6, characterized in that, when the evaluation value prediction unit is given the parameter values ​​contained in each data acquired by the data acquisition unit and the evaluation value contained in each data, it provides the parameter values ​​sampled by the sampling unit to a learning model that learns evaluation values ​​corresponding to each parameter value, and obtains from the learning model an evaluation value corresponding to the parameter values ​​sampled by the sampling unit as a predicted result of the evaluation value for the operating result of the adjusted device.

8. The parameter adjustment device according to claim 6, characterized in that, if a global search method is selected by the decision method selection unit, the evaluation value prediction unit, when given a parameter value contained in each data acquired by the data acquisition unit and an evaluation value contained in each data, provides the parameter value sampled by the sampling unit to a learning model that learns an evaluation value corresponding to each parameter value, and acquires, from the learning model, an evaluation value corresponding to the parameter value sampled by the sampling unit as a predicted result of an evaluation value for an operation result of the adjusted device; and, if a local search method is selected by the decision method selection unit, extracts a portion of the data acquired by the data acquisition unit as learning data, and, when given a parameter value contained in each learning data and an evaluation value contained in each learning data, provides the parameter value sampled by the sampling unit to a learning model that learns an evaluation value corresponding to each parameter value, and acquires, from the learning model, an evaluation value corresponding to the parameter value sampled by the sampling unit as a predicted result of an evaluation value for an operation result of the adjusted device.

9. A parameter adjustment device as claimed in any one of claims 1 to 8, characterized in that the elite solution extraction unit extracts, from the multiple data sets acquired by the data acquisition unit, data containing an evaluation value equal to or greater than a lower limit value as an elite solution.

10. A parameter adjustment method in which a data acquisition unit acquires operation results of an adjusted device that operates using one parameter or multiple parameters, and acquires data including each parameter value and an evaluation value related to each parameter value from an evaluation value calculation device that calculates an evaluation value for the operation results; an elite solution extraction unit extracts zero or more data as an elite solution from the multiple data acquired by the data acquisition unit based on the evaluation value included in each data; a parameter value determination unit determines parameter values ​​to be used by the adjusted device for its next operation from the multiple parameter values ​​existing in a search space of parameter values ​​based on the elite solution extracted by the elite solution extraction unit, and outputs the determined parameter values ​​to the adjusted device.

Images