Optimization device, optimization method, and program

Abstract

This optimization device (2) is configured to comprise: a condition acquisition unit (11) that acquires a condition to be used for optimization calculation; a prediction information acquisition unit (12) that acquires prediction information, which is information predicted on the basis of the condition; a restriction information acquisition unit (13) that acquires restriction information for imposing a restriction on the optimization calculation on the basis of the prediction information; and an optimization unit (14) that optimizes a multi-objective function on the basis of the condition and the restriction information.

Description

Optimization device, optimization method, and program

[0001] The present disclosure relates to an optimization device, an optimization method, and a program.

[0002] There are optimization devices that optimize a multi-objective function, which is a plurality of objective functions. For example, Patent Document 1 discloses an optimization device that optimizes a multi-objective function based on a constraint function.

[0003] JP 2013-120521 A

[0004] The constraint functions used in the device disclosed in Patent Document 1 are designed in advance and are not necessarily suitable for the conditions used in the optimization calculations, which can lead to a problem of a deterioration in the accuracy of the optimization calculations.

[0005] The present disclosure has been made to solve the above-mentioned problems, and aims to provide an optimization device that is capable of more appropriately optimizing a multi-objective function.

[0006] The optimization device according to the present disclosure optimizes a multi-objective function, which is a set of multiple objective functions. The optimization device includes a condition acquisition unit that acquires conditions to be used in optimization calculations, a prediction information acquisition unit that acquires prediction information that is information predicted based on the conditions, a restriction information acquisition unit that acquires restriction information for imposing restrictions on the optimization calculations based on the prediction information, and an optimization unit that optimizes the multi-objective function based on the conditions and the restriction information.

[0007] According to the present disclosure, it becomes possible to more appropriately optimize a multi-objective function.

[0008] FIG. 1 is a configuration diagram showing an optimization device 2 according to a first embodiment. FIG. 2 is a hardware configuration diagram showing hardware of the optimization device 2 according to the first embodiment. FIG. 3 is a hardware configuration diagram of a computer when the optimization device 2 is realized by software, firmware, or the like. FIG. 4 is an explanatory diagram showing an example of a sewage treatment plant where prediction of inflow water and prediction of water quality are performed. １ (x) and the objective function f ２8A is an explanatory diagram showing an example of a solution showing the optimization result of a multi-objective function by the device disclosed in Patent Document 1, and FIG. 8B is an explanatory diagram showing an example of a solution showing the optimization result of a multi-objective function by the optimization device 2 shown in FIG. 1. A configuration diagram showing another optimization device 2 according to the first embodiment. An explanatory diagram showing an example of restriction information when the variable of the objective function is a predicted value of water level, and an example of restriction information when the objective function has a function value of pumping volume. An explanatory diagram showing the optimization result of a multi-objective function under the constraint that the water level, etc., is changed within a restricted range. A configuration diagram showing an optimization device 2 according to a third embodiment. A hardware configuration diagram showing the hardware of the optimization device 2 according to the third embodiment. An explanatory diagram showing an objective function space in which prediction information is arranged. An explanatory diagram showing preference range information acquired based on a reference point. A configuration diagram showing an optimization device 2 according to a fourth embodiment. A hardware configuration diagram showing the hardware of the optimization device 2 according to the fourth embodiment. A flowchart showing the processing procedure of an optimization unit 16. FIG. 10 is a configuration diagram showing an optimization device 2 according to embodiment 5. FIG. 11 is a hardware configuration diagram showing the hardware of the optimization device 2 according to embodiment 5. FIG. 12 is a configuration diagram showing an optimization device 2 according to embodiment 6. FIG. 13 is a hardware configuration diagram showing the hardware of the optimization device 2 according to embodiment 6. FIG. 14 is a configuration diagram showing an optimization device 2 according to embodiment 7. FIG. 15 is a hardware configuration diagram showing the hardware of the optimization device 2 according to embodiment 7.

[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. Fig. 1 is a configuration diagram showing an optimization device 2 according to embodiment 1. Fig. 2 is a hardware configuration diagram showing hardware of the optimization device 2 according to embodiment 1. In Fig. 1, a learning model 1 is a trained model in which, during learning, conditions to be used in optimization calculations and prediction information, which is information predicted based on the conditions, are given as learning data, and the correspondence between the conditions and the prediction information is learned. When conditions to be used in optimization calculations are given during inference, the learning model 1 outputs prediction information corresponding to the conditions. The optimization device 2 optimizes a multi-objective function and includes a condition acquisition unit 11, a prediction information acquisition unit 12, a restriction information acquisition unit 13, and an optimization unit 14.

[0011] The condition acquisition unit 11 is realized by, for example, the condition acquisition circuit 31 shown in Fig. 2. The condition acquisition unit 11 acquires conditions to be used in the optimization calculation. The condition acquisition unit 11 outputs the conditions to both the prediction information acquisition unit 12 and the optimization unit 14.

[0012] The prediction information acquisition unit 12 is realized by, for example, the prediction information acquisition circuit 32 shown in FIG. 2 . The prediction information acquisition unit 12 acquires conditions to be used in optimization calculations from the condition acquisition unit 11. The prediction information acquisition unit 12 acquires prediction information, which is information predicted based on the conditions. Specifically, the prediction information acquisition unit 12 provides the conditions to the learning model 1 and acquires the prediction information from the learning model 1. The prediction information acquisition unit 12 outputs the prediction information to the restriction information acquisition unit 13.

[0013] The restriction information acquisition unit 13 is realized by, for example, the restriction information acquisition circuit 33 shown in FIG. 2 . The restriction information acquisition unit 13 acquires prediction information from the prediction information acquisition unit 12. The restriction information acquisition unit 13 acquires restriction information for imposing restrictions on the optimization calculation based on the prediction information. The restriction information acquisition unit 13 outputs the restriction information to the optimization unit 14.

[0014] The optimization unit 14 is realized by, for example, the optimization circuit 34 shown in Fig. 2. The optimization unit 14 acquires conditions from the condition acquisition unit 11 and acquires restriction information from the restriction information acquisition unit 13. The optimization unit 14 optimizes the multi-objective function based on the conditions and the restriction information. The optimization unit 14 outputs the optimization result of the multi-objective function to, for example, an external device.

[0015] 1, it is assumed that each of the components of the optimization device 2, the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 13, and the optimization unit 14, is realized by dedicated hardware as shown in FIG. 2. That is, it is assumed that the optimization device 2 is realized by a condition acquisition circuit 31, a prediction information acquisition circuit 32, a restriction information acquisition circuit 33, and an optimization circuit 34. Each of the condition acquisition circuit 31, the prediction information acquisition circuit 32, the restriction information acquisition circuit 33, and the optimization circuit 34 is, for example, a single circuit, a composite circuit, a processor that executes processing based on instructions from a program, a processor that executes processing based on instructions from a parallel program, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0016] The components of the optimization device 2 are not limited to those realized by dedicated hardware, and the optimization device 2 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. A 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).

[0017] 3 is a hardware configuration diagram of a computer when the optimization device 2 is realized by software, firmware, etc. When the optimization device 2 is realized by software, firmware, etc., programs for causing the computer to execute the respective processing procedures of the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 13, and the optimization unit 14 are stored in memory 51. Then, a processor 52 of the computer executes the programs stored in memory 51.

[0018] 2 shows an example in which each of the components of the optimization device 2 is realized by dedicated hardware, while Fig. 3 shows an example in which the optimization device 2 is realized by software, firmware, etc. However, this is merely an example, and some of the components in the optimization device 2 may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0019] As shown in FIG. 4 , examples of targets for which optimization by the optimization device 2 is applied include optimization of an objective function that predicts inflow water into a sewage treatment plant and an objective function that predicts the water quality of water discharged from the sewage treatment plant into a river or the like. FIG. 4 is an explanatory diagram showing an example of a sewage treatment plant where inflow water and water quality are predicted. Targets for which optimization by the optimization device 2 is applied are not limited to those shown in FIG. 4 , and include optimization of an objective function that predicts the amount of power consumed by an air conditioner and an objective function that predicts the room temperature regulated by the air conditioner. Furthermore, examples of targets for which optimization by the optimization device 2 is applied include optimization of an objective function that predicts the integration level of multiple components to be arranged on a board and an objective function that predicts the cost of wiring multiple components to be arranged on the board.

[0020] FIG. 5 shows the objective function f １ (x) and the objective function f ２ 5 is an explanatory diagram showing an example of optimization with the objective function f １ (x) is a function that predicts risk, and the objective function f ２(x) is a function that predicts the cost. In FIG. 5, a circle indicates the objective function f when x is a certain value. １ The value of (x) and the objective function f ２ In the example of FIG. 5, the objective function f １ The value of (x) and the objective function f ２ (x) and the value of the objective function f １ As the value of (x) increases, the objective function f ２ On the other hand, the value of the objective function f １ When the value of (x) becomes smaller, the objective function f ２ 5, the solid line represents the Pareto front, and the circles on the Pareto front represent the Pareto solutions, which are the optimization values ​​of the multi-objective function.

[0021] Next, the operation of the optimization device 2 shown in FIG. 1 will be described. FIG. 6 is a flowchart showing an optimization method, which is a processing procedure of the optimization device 2 shown in FIG. 1. The condition acquisition unit 11 acquires, for example, conditions at multiple times as conditions to be used in the optimization calculation (step ST1 in FIG. 6). If the target of optimization by the optimization device 2 is, for example, an objective function for predicting inflow water and an objective function for predicting water quality, as shown in FIG. 4, the condition acquisition unit 11 acquires, for example, observation values ​​at multiple times as conditions to be used in the optimization calculation: - Observation values ​​of a rainfall sensor installed in a river, etc., upstream of the sewage treatment plant; - Observation values ​​of a pumping rate sensor that observes the pumping rate of a pump that pumps water from a pump well at the sewage treatment plant to a water treatment tank; - Observation values ​​of a water level sensor that observes the water level in the water treatment tank. The condition acquisition unit 11 outputs the acquired conditions to the prediction information acquisition unit 12 and the optimization unit 14, respectively.

[0022] The prediction information acquisition unit 12 acquires the conditions to be used in the optimization calculation from the condition acquisition unit 11. The prediction information acquisition unit 12 acquires prediction information based on the conditions (step ST2 in FIG. 6 ). Specifically, the prediction information acquisition unit 12 provides the conditions to the learning model 1. When the learning model 1 receives the conditions from the prediction information acquisition unit 12, it outputs prediction information corresponding to the conditions to the prediction information acquisition unit 12. The prediction information acquisition unit 12 acquires the prediction information output from the learning model 1. As the prediction information, for example, time-series information is acquired, as shown in FIG. 7 . FIG. 7 is an explanatory diagram showing an example of prediction information and an example of a solution indicating the optimization result. FIG. 7 illustrates prediction information for the pump discharge rate and prediction information for the water level. The prediction information acquisition unit 12 outputs the prediction information to the restriction information acquisition unit 13.

[0023] The restriction information acquisition unit 13 acquires prediction information from the prediction information acquisition unit 12. The restriction information acquisition unit 13 acquires restriction information for imposing restrictions on the optimization calculation based on the prediction information (step ST3 in FIG. 6). The restriction information acquisition method may, for example, acquire restriction information from a regression model by providing the prediction information to the regression model having a formula indicating the correspondence between the prediction information and the restriction information, or acquire restriction information corresponding to the prediction information from a database indicating the correspondence between the prediction information and the restriction information. The restriction information corresponding to the prediction information may, for example, be the range of change of the predicted value at each time indicated by the prediction information. The range of change of the predicted value indicates the range between the upper limit and lower limit of the predicted value at each time. The restriction information acquisition unit 13 outputs the restriction information to the optimization unit 14.

[0024] The optimization unit 14 acquires conditions from the condition acquisition unit 11 and acquires restriction information from the restriction information acquisition unit 13. The optimization unit 14 optimizes the multi-objective function based on the conditions and restriction information (step ST4 in FIG. 6 ). Specifically, the optimization unit 14 extracts conditions included in the range of change of the predicted value from the conditions for multiple times acquired from the condition acquisition unit 11. Conditions included in the range of change of the predicted value are likely to be suitable conditions for use in the optimization calculation. On the other hand, conditions not included in the range of change of the predicted value are unlikely to be suitable conditions for use in the optimization calculation. The optimization unit 14 then optimizes the multi-objective function using the extracted conditions. The optimization unit 14 outputs the optimization result of the multi-objective function to, for example, an external device.

[0025] Fig. 8A is an explanatory diagram showing an example of a solution indicating the optimization result of the multi-objective function by the device disclosed in Patent Document 1. Fig. 8B is an explanatory diagram showing an example of a solution indicating the optimization result of the multi-objective function by the optimization device 2 shown in Fig. 1. The solutions indicating the optimization result of the multi-objective function by the optimization device 2 shown in Fig. 1 are narrowed down to solutions with higher calculation accuracy than the solutions indicating the optimization result of the multi-objective function by the device disclosed in Patent Document 1.

[0026] In the first embodiment described above, the optimization device 2 is configured to include a condition acquisition unit 11 that acquires conditions to be used in the optimization calculation, a prediction information acquisition unit 12 that acquires prediction information, which is information predicted based on the conditions, a restriction information acquisition unit 13 that acquires restriction information for imposing restrictions on the optimization calculation based on the prediction information, and an optimization unit 14 that optimizes the multi-objective function based on the conditions and the restriction information. Therefore, the optimization device 2 can more appropriately optimize the multi-objective function. Note that optimization based on the restriction information enables optimization suited to the current problem, which is similar to the driving of an expert or the like.

[0027] In the first embodiment, the optimization device 2 is configured so that the prediction information is information output from the learning model 1 to which conditions are input. Therefore, the optimization device 2 can obtain prediction information corresponding to the conditions. Furthermore, the optimization device 2 can handle unknown restriction information by making predictions through machine learning.

[0028] In the first embodiment, the optimization device 2 is configured so that the prediction information is time-series information. Therefore, the optimization device 2 can optimize a multi-objective function whose value changes over time. This makes it possible to apply the optimization device 2 to sewage treatment plants and other places where time-series prediction is highly necessary.

[0029] In the optimization device 2 shown in FIG. 1 , the prediction information acquisition unit 12 acquires, as prediction information, information output from the learning model 1 to which conditions are input. However, this is merely an example, and the prediction information acquisition unit 12 may acquire prediction information based on conditions from the storage unit 3, as shown in FIG. 9 . FIG. 9 is a configuration diagram showing another optimization device 2 according to the first embodiment. In FIG. 9 , the same reference numerals as in FIG. 1 indicate the same or equivalent parts, and detailed description thereof will be omitted. The storage unit 3 stores prediction information corresponding to conditions. Even if the optimization device 2 is configured so that the prediction information acquisition unit 12 acquires prediction information based on conditions from the storage unit 3, prediction information corresponding to conditions can be acquired. Furthermore, by having the prediction information acquisition unit 12 acquire prediction information corresponding to past similar driving from the storage unit 3, explainability is improved.

[0030] In the optimization device 2 shown in FIG. 1 , the restriction information acquisition unit 13 acquires restriction information for imposing restrictions on optimization calculations based on prediction information. The restriction information acquisition unit 13 may accept changes to the prediction information acquired by the prediction information acquisition unit 12 and acquire restriction information based on the changed prediction information. In this case, the restriction information acquisition unit 13 includes a user interface that accepts changes made by the user or an input / output interface that receives changes to the prediction information from an external device, and changes the prediction information acquired by the prediction information acquisition unit 12. By configuring the optimization device 2 so that the restriction information acquisition unit 13 accepts changes to the prediction information acquired by the prediction information acquisition unit 12 and acquires restriction information based on the changed prediction information, optimization results that reflect the user's intentions can be obtained. As a result, it is possible to achieve operation that is closer to the operator's preferences.

[0031] Second Embodiment In the second embodiment, a case will be described in which the prediction information includes at least one of information predicting variables of an objective function, information predicting the objective function, and information predicting constraint functions that constrain optimization. The configuration of the optimization device 2 according to the second embodiment is the same as that of the optimization device 2 according to the first embodiment, and the configuration diagram showing the optimization device 2 according to the second embodiment is shown in FIG. 1 or FIG. 9.

[0032] When the prediction information acquisition unit 12 acquires prediction information including information predicting a variable of the objective function, the restriction information acquisition unit 13 acquires restriction information based on the information predicting the variable of the objective function. Examples of the objective function include an objective function whose function value is a water level or an objective function whose function value is a pumping volume. Examples of the variable of the objective function include a predicted value of the water level or a predicted value of the pumping volume. When the prediction information acquisition unit 12 acquires prediction information including information predicting a variable of the objective function, the restriction information acquisition unit 13 acquires restriction information based on the information predicting the variable of the objective function. If the variable of the objective function is a predicted value such as a water level, as shown in FIG. 10 , the restriction information may include, for example, information for limiting the predicted value of the water level or the like to a certain range of values ​​from the observed values ​​that are conditions provided to the optimization unit 14. FIG. 10 is an explanatory diagram showing an example of restriction information when the variable of the objective function is a predicted value of the water level and an example of restriction information when the variable of the objective function is a pumping volume.

[0033] When the prediction information acquiring unit 12 acquires prediction information including information for predicting the objective function, the restriction information acquiring unit 13 acquires restriction information based on the information for predicting the objective function. In the case of an objective function whose function value is a pumping amount or the like, as shown in Fig. 10, the restriction information may be information for restricting the predicted value of the pumping amount or the like to a certain range of function values ​​from among the observed values ​​that are conditions given to the optimization unit 14.

[0034] When the prediction information acquisition unit 12 acquires prediction information including information predicting a constraint function that constrains optimization, the restriction information acquisition unit 13 acquires restriction information based on the information predicting the constraint function. When the constraint function is a function that outputs a restriction range for restricting a predicted value of water level, which is a variable of the objective function, to a certain range of values, the restriction information may be, for example, information for extracting observed values ​​within the restriction range from observed values ​​that are conditions given to the optimization unit 14.

[0035] The restriction information acquired by the restriction information acquisition unit 13 may be a restriction condition that restricts the optimization calculation. The restriction condition is a condition that restricts the objective function or an explanatory variable that is a variable of the objective function to be included within a restriction range that is a range set based on the forecast information. In this case, the optimization unit 14 performs the optimization calculation while imposing a restriction that, for example, the water level or the pumping amount is changed within the restriction range, as shown in Figure 11. Figure 11 is an explanatory diagram showing the optimization result of the multi-objective function under the constraint of changing the water level, etc. within the restriction range.

[0036] In the second embodiment described above, the optimization device 2 is configured so that the prediction information includes at least one of information predicting variables of the objective function, information predicting the objective function, and information predicting constraint functions that constrain the optimization. Therefore, the optimization device 2 is able to more appropriately optimize the multi-objective function and also to acquire constraint information based on various pieces of prediction information. Since the information predicting variables of the objective function, information predicting the objective function, and information predicting constraint functions that constrain the optimization are all important information, if the prediction information includes one or more of these pieces of information, it is possible to acquire constraint information that is useful for optimization.

[0037] In the second embodiment, the optimization device 2 is configured so that the restriction information is a constraint condition that restricts the optimization calculation. Therefore, the optimization device 2 can impose restrictions on the optimization calculation.

[0038] In the second embodiment, the optimization device 2 is configured so that the limiting condition is a condition that constrains the objective function or the explanatory variables of the objective function to be included within a range set based on the prediction information. Therefore, the optimization device 2 can perform optimization calculations within the constrained range.

[0039] Third Embodiment In a third embodiment, a description will be given of a case where the restriction information is information indicating the ranges that a plurality of objective functions can take in the objective function space based on the prediction information.

[0040] Fig. 12 is a configuration diagram showing an optimization device 2 according to embodiment 3. In Fig. 12, the same reference numerals as in Fig. 1 or Fig. 9 indicate the same or corresponding parts, and detailed descriptions thereof will be omitted. Fig. 13 is a hardware configuration diagram showing the hardware of the optimization device 2 according to embodiment 3. In Fig. 13, the same reference numerals as in Fig. 2 indicate the same or corresponding parts, and detailed descriptions thereof will be omitted.

[0041] The restriction information acquisition unit 15 acquires prediction information from the prediction information acquisition unit 12. The restriction information acquisition unit 15 acquires, as restriction information, information indicating possible ranges of multiple objective functions in the objective function space based on the prediction information. The restriction information acquisition unit 15 outputs the restriction information to the optimization unit 14.

[0042] The optimization device 2 shown in Fig. 12 is configured such that the restriction information acquisition unit 15 is applied to the optimization device 2 shown in Fig. 1. However, this is merely an example, and the restriction information acquisition unit 15 may also be applied to the optimization device 2 shown in Fig. 9.

[0043] The restriction information acquisition unit 15 is realized, for example, by a restriction information acquisition circuit 35 shown in FIG. 12. In FIG. 12, it is assumed that the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 15, and the optimization unit 14, which are components of the optimization device 2, are each realized by dedicated hardware such as that shown in FIG. 13. That is, it is assumed that the optimization device 2 is realized by a condition acquisition circuit 31, a prediction information acquisition circuit 32, a restriction information acquisition circuit 35, and an optimization circuit 34. Each of the condition acquisition circuit 31, the prediction information acquisition circuit 32, the restriction information acquisition circuit 35, and the optimization circuit 34 corresponds to, for example, a single circuit, a composite circuit, a processor that executes processing based on instructions from a program, a processor that executes processing based on instructions from a parallel program, an ASIC, an FPGA, or a combination thereof.

[0044] The components of the optimization device 2 are not limited to those realized by dedicated hardware, and the optimization device 2 may be realized by software, firmware, or a combination of software and firmware. When the optimization device 2 is realized by software, firmware, or the like, programs for causing a computer to execute the respective processing procedures of the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 15, and the optimization unit 14 are stored in a memory 51 shown in Fig. 3. Then, a processor 52 shown in Fig. 3 executes the programs stored in the memory 51.

[0045] 13 shows an example in which each of the components of the optimization device 2 is realized by dedicated hardware, while Fig. 3 shows an example in which the optimization device 2 is realized by software, firmware, etc. However, this is merely an example, and some of the components in the optimization device 2 may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0046] Next, the operation of the optimization device 2 shown in Fig. 12 will be described. Except for the restriction information acquisition unit 15, the optimization device 2 is the same as the optimization device 2 shown in Fig. 1. The restriction information acquisition unit 15 acquires prediction information from the prediction information acquisition unit 12. The restriction information acquisition unit 15 acquires, as restriction information, information indicating the range that multiple objective functions can take in the objective function space based on the prediction information (hereinafter referred to as "preference range information"). The preference range information is set based on, for example, the predicted value closest to the Pareto front among the multiple predicted values ​​indicated by the prediction information. Specifically, the objective function f corresponding to the predicted value closest to the Pareto front is １ (x) and the objective function f corresponding to the predicted value ２ (x) and the function value of (x).

[0047] 14 is an explanatory diagram showing an objective function space in which prediction information is arranged. In FIG. 14, the horizontal axis represents the objective function f １ (x), and the vertical axis represents the objective function f ２ The left diagram of Figure 14 shows the function values ​​of the objective function f(x). The solid line is the Pareto front, the black dots are predicted values ​​indicated by the prediction information, and the circles are solutions existing on the Pareto front. １ Among the function values ​​of (x), the objective function f １ The function value of (x) is 0.4, and it exists on the Pareto front. ２ Among the function values ​​of (x), the objective function f ２ Therefore, in the example on the left side of FIG. 14, the function value of the objective function f １ The function value of (x) is in the range of about 0.1 to 0.4, and the objective function f ２ The function value of (x) is set in the range of approximately 0.1 to 0.5. The restriction information acquisition unit 15 outputs preference range information to the optimization unit 14 as restriction information. The optimization unit 14 optimizes the multi-objective function based on the conditions and the preference range information, which is restriction information. The optimization process by the optimization unit 14 is the same as in embodiment 1, so a detailed description will be omitted.

[0048] The restriction information acquisition unit 15 includes a man-machine interface that accepts changes to the preference range information by the user, or an input / output interface that receives change information for the preference range information from an external device. When the user wishes to change the preference range information, the restriction information acquisition unit 15 accepts the change to the preference range information and outputs the changed preference range information to the optimization unit 14 as restriction information.

[0049] The process of changing the preference range information by the restriction information acquisition unit 15 will be specifically described below. First, when the restriction information acquisition unit 15 acquires the predicted value indicated by the prediction information, it arranges the predicted value in the objective function space, for example, and displays the objective function space in which the predicted value is arranged on a touch panel (not shown) as shown in FIG. 14. If the preference area of ​​the objective function space displayed on the touch panel is not a desirable area, the user can set another area as a preference condition. For example, when the objective function f １ The function value of (x) is approximately 0.0 or more, and the objective function f １ When the preference condition is a region in which the function value of (x) is approximately 0.2 or less, if the user touches the region on the touch panel that corresponds to the preference condition, the restriction information acquisition unit 15 acquires preference range information indicating the region that corresponds to the preference condition. The right diagram of FIG. 14 shows that the preference region, which is the region that corresponds to the preference condition indicated by the preference range information, is a region that is approximately 0.2 or less in the range of the objective function f １ The function value of (x) is in the range of about 0.0 to 0.2, and the objective function f ２ This shows that the function value of (x) has been changed to a range of approximately 0.3 to 0.7. Here, the restriction information acquisition unit 15 accepts the change in preference range information and outputs the changed preference range information to the optimization unit 14 as restriction information. However, this is merely an example, and it goes without saying that the restriction information acquisition unit 15 may output the preference range information indicating the preference range shown in the left diagram of FIG. 14 directly to the optimization unit 14. Furthermore, the restriction information acquisition unit 15 may search for the preference range by an optimization process, rather than accepting a change in the preference range information.

[0050] FIG. 14 shows an example in which the restriction information is information indicating possible ranges of multiple objective functions in the objective function space based on the prediction information. However, this is merely an example, and the restriction information acquisition unit 15 may use a point in the objective function space corresponding to the prediction information as a reference point as the restriction information and acquire preference range information based on the reference point. In this case, the restriction information acquisition unit 15 may determine the possible ranges of each of the multiple objective functions based on one reference point, or may determine the possible range of one objective function or the possible ranges of each objective function based on multiple reference points. The process of acquiring preference range information by the restriction information acquisition unit 15 will be described in detail below.

[0051] The restriction information acquisition unit 15 selects one or more arbitrary predicted values ​​from the multiple predicted values ​​indicated by the prediction information, and sets each of the selected one or more predicted values ​​as a reference point. FIG. 15 is an explanatory diagram showing preference range information acquired based on the reference points. In the left diagram of FIG. 15, circles indicate predicted values ​​indicated by the prediction information. In the left diagram of FIG. 15, four predicted values ​​are illustrated, and of the four predicted values, the second predicted value from the left and the fourth predicted value from the left in the diagram are selected, and each selected predicted value is a reference point. Hereinafter, the second predicted value from the left will be referred to as reference point (1), and the fourth predicted value from the left will be referred to as reference point (2). Herein, for convenience, it is assumed that two reference points (1) and (2) are set, and the objective function f １ The range that (x) can take (indicated as the "preference interval" in FIG. 5) is determined, and the objective function f ２ An example of determining the range that (x) can take will be described.

[0052] The restriction information acquisition unit 15, for example, calculates the reference point (1) as a function of the objective function f １ The reference point (2) determines the range that (x) can take, and the reference point (2) is the objective function f ２ The restriction information acquisition unit 15 then calculates the objective function f １ In the function value of (x), the region to which the reference point (1) belongs is selected from among multiple selectable regions by the objective function f １The range in which (x) can be taken is determined. The multiple selectable regions include, for example, a region of 0.0 or more and less than 0.3, a region of 0.1 or more and less than 0.4, a region of 0.2 or more and less than 0.5, a region of 0.3 or more and less than 0.6, a region of 0.4 or more and less than 0.7, a region of 0.5 or more and less than 0.8, a region of 0.6 or more and less than 0.9, or a region of 0.7 or more and less than 1.0. In such a case, the restriction information acquisition unit 15 selects the target function f from the multiple selectable regions. １ The range that (x) can take is determined to be the range of 0.1 or more and less than 0.4, which is the range to which reference point (1) belongs.

[0053] Furthermore, the restriction information acquisition unit 15 calculates the objective function f ２ In the function value of (x), the region to which the reference point (2) belongs is selected from among multiple favorable regions by the objective function f ２ The range in which (x) can be taken is determined. The multiple selectable regions include, for example, a region of 0.0 or more and less than 0.3, a region of 0.1 or more and less than 0.4, a region of 0.2 or more and less than 0.5, a region of 0.3 or more and less than 0.6, a region of 0.4 or more and less than 0.7, a region of 0.5 or more and less than 0.8, a region of 0.6 or more and less than 0.9, or a region of 0.7 or more and less than 1.0. In such a case, the restriction information acquisition unit 15 selects the target function f from the multiple selectable regions. ２ The range that (x) can take is determined to be equal to or greater than 0.1 and less than 0.4.

[0054] The restriction information acquisition unit 15 calculates the objective function f １ Preference range information indicating the range that (x) can take, and the objective function f ２ The optimization unit 14 outputs the condition and preference range information indicating the range that (x) can take to the optimization unit 14. The optimization unit 14 optimizes the multi-objective function based on the conditions and the preference range information, which is restriction information. The right diagram in Fig. 15 shows the optimization result of the multi-objective function by the optimization unit 14, where the solution indicating the optimization result of the multi-objective function is restricted to the preference range.

[0055] Here, the reference point (1) is the objective function f １ The reference point (2) determines the range that (x) can take, and the reference point (2) is the objective function f ２The reference point (1) is used to determine the range that (x) can take. However, this is just an example. ２ The reference point (2) determines the range that (x) can take, and the reference point (2) is the objective function f １ The reference point (1) may be used to determine the range that (x) can take. １ (x) is the reference point that determines the range of possible values ​​of the objective function f ２ The reference point (2) may be used to determine the range that (x) can take. １ (x) is the reference point that determines the range of possible values ​​of the objective function f ２ It may be a reference point that determines the range that (x) can take.

[0056] The restriction information acquisition unit 15 includes a man-machine interface that accepts a change in the reference point or an input / output interface that receives change information about the reference point from an external device. When a user desires to change a reference point (hereinafter referred to as a "first reference point"), the restriction information acquisition unit 15 may accept the change in the first reference point and acquire, as restriction information, preference range information indicating possible ranges of multiple objective functions based on a second reference point, which is the reference point after the change. Here, when a user desires to change the first reference point, the restriction information acquisition unit 15 accepts the change in the first reference point and acquires, as restriction information, preference range information based on the second reference point, which is the reference point after the change. However, this is merely an example. Even without the above-described man-machine interface or the above-described input / output interface, the restriction information acquisition unit 15 may automatically accept the change in the first reference point and acquire, as restriction information, preference range information based on the second reference point, which is the reference point after the change.

[0057] The process of changing the reference points performed by the restriction information acquisition unit 15 will be described in detail below. First, as shown in the left diagram of FIG. 15 , the restriction information acquisition unit 15 displays an objective function space, in which multiple predicted values ​​indicated by the prediction information are arranged, on a touch panel or the like (not shown). A user can change any of the multiple predicted values ​​displayed on the touch panel or the like, other than the predicted value set at an arbitrary reference point, to a reference point. For example, if, among the four predicted values, the second predicted value from the left in the diagram is set to the first reference point (1) and the fourth predicted value from the left is set to the first reference point (2), and the first predicted value from the left is set to the second reference point (1) and the third predicted value from the left is set to the second reference point (3), the user touches the first predicted value from the left and then the third predicted value from the left. When the user performs the above operation, the restriction information acquisition unit 15 changes the first predicted value from the left to the second reference point (1) and the third predicted value from the left to the second reference point (3). However, the above operation is merely an example, and the reference point may be changed by other operations.

[0058] The restriction information acquisition unit 15 selects, from among a plurality of selectable regions, a region to which the second reference point (1) belongs, for example, by using an objective function f １ The restriction information acquisition unit 15 also determines the range in which the second reference point (2) can be located from among the multiple selectable regions, using the objective function f ２ The restriction information acquisition unit 15 determines the range that the objective function f １ Preference range information indicating the range that (x) can take, and the objective function f ２The optimization unit 14 outputs preference range information indicating the range that (x) can take to the optimization unit 14. The optimization of the multi-objective function by the optimization unit 14 according to the first embodiment is optimization based on constraint information. Specifically, for example, the pumping volume is searched for based on the constraint information, and as a result, a solution that satisfies the water level is searched for. An example of such a search method is the constrained Non-Dominated Sorting Genetic Algorithm (NSGA) II. The optimization of the multi-objective function by the optimization unit 14 according to the third embodiment is optimization that also takes into account a reference point, etc. Specifically, it is optimization that approaches the reference point, and there is no constraint on the pumping volume. An example of such a search method is R-NSGA II. By utilizing a range-based prediction, it is possible to set multiple predicted values ​​to be passed to R-NSGA II and obtain solutions close to the multiple points.

[0059] In the third embodiment described above, the optimization device 2 is configured so that the restriction information is information that indicates the ranges that multiple objective functions can take on in the objective function space based on the prediction information. Therefore, the optimization device 2 can more appropriately optimize the multi-objective function and can obtain the restriction information based on the prediction information.

[0060] In the third embodiment, the optimization device 2 is configured so that the restriction information is information indicating possible ranges of multiple objective functions based on reference points, with points in the objective function space corresponding to the prediction information as reference points. Therefore, the optimization device 2 can more appropriately optimize the multi-objective function and obtain restriction information based on the reference points corresponding to the prediction information.

[0061] In the third embodiment, the optimization device 2 is configured so that a plurality of reference points are set in the objective function space based on prediction information. Therefore, the optimization device 2 can more appropriately optimize the multi-objective function and can obtain restriction information based on the plurality of reference points.

[0062] In the third embodiment, the optimization device 2 is configured so that the restriction information is information indicating possible ranges of multiple objective functions based on a second reference point changed from the first reference point by the user. Therefore, the optimization device 2 can more appropriately optimize the multi-objective function and can obtain restriction information based on the reference point desired by the user.

[0063] Fourth Embodiment In a fourth embodiment, an optimization device 2 in which the optimization unit 16 optimizes a multi-objective function by evolutionary computation will be described.

[0064] Fig. 16 is a configuration diagram showing an optimization device 2 according to embodiment 4. In Fig. 16, the same reference numerals as those in Fig. 1, Fig. 9, or Fig. 12 indicate the same or corresponding parts, and detailed descriptions thereof will be omitted. Fig. 17 is a hardware configuration diagram showing the hardware of the optimization device 2 according to embodiment 4. In Fig. 17, the same reference numerals as those in Fig. 2 or Fig. 13 indicate the same or corresponding parts, and detailed descriptions thereof will be omitted.

[0065] The optimization unit 16 acquires conditions from the condition acquisition unit 11 and acquires restriction information indicating constraint conditions and preference range information from the restriction information acquisition unit 15. The optimization unit 16 optimizes the multi-objective function using evolutionary computing. Examples of evolutionary computing include genetic algorithms. Specifically, the optimization unit 16 performs optimization processing on a first group of solution candidates, which are one or more solution candidates, based on constraint conditions included in the restriction information that restrict the optimization calculation. The optimization unit 16 also performs optimization processing on a second group of solution candidates different from the first solution candidates, based on preference conditions, which are information indicating the range in the objective function space where multiple objective functions can be taken, based on prediction information included in the restriction information. During the optimization processing using evolutionary computing, the optimization unit 16 swaps a portion of the first group of solution candidates that have been optimized based on the constraint conditions with a portion of the second group of solution candidates that have been optimized based on the preference conditions. The optimization unit 16 then performs optimization processing on a portion of the first group of solution candidates based on the preference conditions, and performs optimization processing on a portion of the second group of solution candidates based on the constraint conditions. The optimization unit 16 outputs the optimization result of the multi-objective function to, for example, the outside.

[0066] The optimization unit 16 is realized, for example, by an optimization circuit 36 ​​shown in FIG. 17. In FIG. 17, it is assumed that the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 15, and the optimization unit 16, which are components of the optimization device 2, are each realized by dedicated hardware as shown in FIG. 17. That is, it is assumed that the optimization device 2 is realized by a condition acquisition circuit 31, a prediction information acquisition circuit 32, a restriction information acquisition circuit 35, and an optimization circuit 36. Each of the condition acquisition circuit 31, the prediction information acquisition circuit 32, the restriction information acquisition circuit 35, and the optimization circuit 36 ​​corresponds to, for example, a single circuit, a composite circuit, a processor that executes processing based on instructions from a program, a processor that executes processing based on instructions from a parallel program, an ASIC, an FPGA, or a combination thereof.

[0067] The components of the optimization device 2 are not limited to those realized by dedicated hardware, and the optimization device 2 may be realized by software, firmware, or a combination of software and firmware. When the optimization device 2 is realized by software, firmware, or the like, programs for causing a computer to execute the respective processing procedures of the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 15, and the optimization unit 16 are stored in a memory 51 shown in Fig. 3. Then, a processor 52 shown in Fig. 3 executes the programs stored in the memory 51.

[0068] 17 shows an example in which each of the components of the optimization device 2 is realized by dedicated hardware, while Fig. 3 shows an example in which the optimization device 2 is realized by software, firmware, etc. However, this is merely an example, and some of the components in the optimization device 2 may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0069] Next, the operation of the optimization device 2 shown in Fig. 16 will be described. Except for the optimization unit 16, the optimization device 2 is the same as the optimization device 2 shown in Fig. 12. Therefore, only the operation of the optimization unit 16 will be described here. Fig. 18 is a flowchart showing the processing procedure of the optimization unit 16.

[0070] The optimization unit 16 acquires conditions from the condition acquisition unit 11. The optimization unit 16 also acquires restriction information indicating constraint conditions for restricting the optimization calculation, as well as restriction information indicating preference range information, from the restriction information acquisition unit 15. The optimization unit 16 performs an optimization process on a first group of solution candidates, which are one or more solution candidates, based on the conditions and constraint conditions acquired from the condition acquisition unit 11 (step ST11 in FIG. 18 ). This optimization process is similar to the optimization process performed by the optimization unit 14 according to the first embodiment. The optimization unit 16 performs an optimization process on a second group of solution candidates, which are one or more solution candidates, based on the conditions acquired from the condition acquisition unit 11 and preference conditions that are the preference range indicated by the preference range information (step ST12 in FIG. 18 ). This optimization process is similar to the optimization process performed by the optimization unit 14 according to the third embodiment.

[0071] Before completing the optimization processes in steps ST11 and ST12, the optimization unit 16 performs a swap process between a part of the first solution candidate group that has been subjected to the optimization process based on the constraint conditions and a part of the second solution candidate group that has been subjected to the optimization process based on the preference conditions (step ST13 in FIG. 18). The swap process is a process equivalent to so-called crossover, and since the swap process itself is a known technique, a detailed description thereof will be omitted.

[0072] After performing the swap process, the optimization unit 16 performs an optimization process based on the preference conditions on a part of the first solution candidate group (step ST14 in FIG. 18), and performs an optimization process based on the constraint conditions on a part of the second solution candidate group (step ST15 in FIG. 18). The optimization unit 16 outputs the optimization results of the multi-objective function obtained by each process to, for example, an external device.

[0073] In the fourth embodiment described above, the optimization unit 16 optimizes the multi-objective function using evolutionary computing. The optimization unit 16 performs an optimization process on a first group of solution candidates, which are solution candidates, based on constraint conditions included in the restriction information that constrain the optimization calculation. The optimization unit 2 also performs an optimization process on a second group of solution candidates different from the first solution candidates, based on preference conditions, which are information indicating possible ranges of objective functions in the objective function space, based on prediction information included in the restriction information. During the optimization process using evolutionary computing, the optimization unit 2 swaps a portion of the first group of solution candidates optimized based on the constraint conditions with a portion of the second group of solution candidates optimized based on the preference conditions. The optimization unit 2 then performs an optimization process based on the preference conditions on a portion of the first group of solution candidates, and an optimization process based on the constraint conditions on a portion of the second group of solution candidates. Therefore, the optimization unit 2 can more appropriately optimize the multi-objective function and can improve the accuracy of the optimization results of the multi-objective function. In the fourth embodiment, optimization can be performed while achieving both the effectiveness of the optimization process based on the constraint conditions and the effectiveness of the optimization process based on the preference conditions.

[0074] Fifth Embodiment In a fifth embodiment, an optimization device 2 including an output unit 17 that outputs a plurality of optimization results including the results optimized by the optimization unit 14 will be described.

[0075] Fig. 19 is a configuration diagram showing an optimization device 2 according to embodiment 5. In Fig. 19, the same reference numerals as those in Fig. 1, Fig. 9, Fig. 12, or Fig. 16 indicate the same or corresponding parts, and detailed descriptions thereof will be omitted. Fig. 20 is a hardware configuration diagram showing the hardware of the optimization device 2 according to embodiment 5. In Fig. 20, the same reference numerals as those in Fig. 2, Fig. 13, or Fig. 17 indicate the same or corresponding parts, and detailed descriptions thereof will be omitted.

[0076] The output unit 17 outputs a plurality of optimization results, including the result optimized by the optimization unit 14, to the processing unit 18. The processing unit 18 performs processing based on an optimization result selected by a user from the plurality of optimization results output from the output unit 17.

[0077] In the optimization device 2 shown in Fig. 19, the output unit 17 and the processing unit 18 are respectively applied to the optimization device 2 shown in Fig. 1. However, this is merely an example, and the output unit 17 and the processing unit 18 may respectively be applied to the optimization device 2 shown in Fig. 9, the optimization device 2 shown in Fig. 12, or the optimization device 2 shown in Fig. 16.

[0078] The output unit 17 is realized, for example, by an output circuit 37 shown in FIG. 20 , and the processing unit 18 is realized, for example, by a processing circuit 38 shown in FIG. 19 . In FIG. 19 , it is assumed that the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 13, the optimization unit 14, the output unit 17, and the processing unit 18, which are components of the optimization device 2, are each realized by dedicated hardware as shown in FIG. 20 . That is, it is assumed that the optimization device 2 is realized by a condition acquisition circuit 31, a prediction information acquisition circuit 32, a restriction information acquisition circuit 33, an optimization circuit 34, an output circuit 37, and a processing circuit 38. Each of the condition acquisition circuit 31, the prediction information acquisition circuit 32, the restriction information acquisition circuit 33, the optimization circuit 34, the output circuit 37, and the processing circuit 38 corresponds to, for example, a single circuit, a composite circuit, a processor that executes processing based on instructions from a program, a processor that executes processing based on instructions from a parallel program, an ASIC, an FPGA, or a combination thereof.

[0079] The components of the optimization device 2 are not limited to those realized by dedicated hardware, and the optimization device 2 may be realized by software, firmware, or a combination of software and firmware. When the optimization device 2 is realized by software, firmware, or the like, a program for causing a computer to execute the respective processing procedures of the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 13, the optimization unit 14, the output unit 17, and the processing unit 18 is stored in a memory 51 shown in Fig. 3. Then, a processor 52 shown in Fig. 3 executes the program stored in the memory 51.

[0080] 20 shows an example in which each of the components of the optimization device 2 is realized by dedicated hardware, while Fig. 3 shows an example in which the optimization device 2 is realized by software, firmware, etc. However, this is merely an example, and some of the components in the optimization device 2 may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0081] Next, the operation of the optimization device 2 shown in Fig. 19 will be described. Except for the output unit 17 and the processing unit 18, the optimization device 2 is the same as that shown in Fig. 1. Therefore, only the operations of the output unit 17 and the processing unit 18 will be described here.

[0082] The output unit 17 not only acquires the optimization result of the multi-objective function from the optimization unit 14, but also acquires one or more Pareto solutions that exist on the Pareto front and are different from the solution indicated by the optimization result, as optimization results. The output unit 17 outputs the multiple optimization results to the processing unit 18.

[0083] The processing unit 18 is equipped with a man-machine interface that accepts the selection of an optimization result, or an input / output interface that receives optimization result selection information from the outside. The processing unit 18 acquires multiple optimization results from the output unit 17. The processing unit 18 accepts the selection of one of the multiple optimization results based on the user's operation to select an optimization result. The processing unit 18 performs processing based on the optimization result selected by the user. Examples of processing performed by the processing unit 18 include controlling the amount of water pumped by a pump if the optimization is applied to a sewage treatment plant, and controlling the room temperature of the air conditioning device if the optimization is applied to an air conditioning device.

[0084] In the above-described fifth embodiment, the optimization device 2 is configured to include an output unit 17 that outputs multiple optimization results, including the result optimized by the optimization unit 14, and a processing unit 18 that performs processing based on an optimization result selected by a user from the multiple optimization results output from the output unit 17. Therefore, the optimization device 2 can more appropriately optimize the multi-objective function and can perform processing based on optimization results that reflect the user's intentions. The user can select a desired optimization result from the multiple optimization results output from the output unit 17, thereby selecting a solution from the multiple optimization results.

[0085] Sixth Embodiment In a sixth embodiment, an optimization device 2 will be described that includes a reception unit 19 that receives a selection of whether to optimize a multi-objective function using restriction information or to optimize a multi-objective function without using restriction information.

[0086] Fig. 21 is a configuration diagram showing an optimization device 2 according to embodiment 6. In Fig. 21, the same reference numerals as those in Fig. 1, 9, 12, 16, or 19 indicate the same or corresponding parts, and detailed descriptions thereof will be omitted. Fig. 22 is a hardware configuration diagram showing the hardware of the optimization device 2 according to embodiment 6. In Fig. 22, the same reference numerals as those in Fig. 2, 13, 17, or 20 indicate the same or corresponding parts, and detailed descriptions thereof will be omitted.

[0087] The receiving unit 19 receives a selection of whether to optimize the multi-objective function using the restriction information or to optimize the multi-objective function without using the restriction information. The receiving unit 19 outputs the received selection to the optimization unit 20.

[0088] The optimization unit 20 acquires conditions from the condition acquisition unit 11 and acquires restriction information from the restriction information acquisition unit 13. The optimization unit 20 also acquires a selection acceptance result from the reception unit 19. The optimization unit 20 optimizes the multi-objective function based on the selection acceptance by the reception unit 19. Specifically, if the selection acceptance result by the reception unit 19 indicates a selection to optimize the multi-objective function using the restriction information, the optimization unit 20 optimizes the multi-objective function based on the conditions and the restriction information, similar to the optimization unit 14 shown in FIG. 1. If the selection acceptance result by the reception unit 19 indicates a selection to optimize the multi-objective function without using the restriction information, the optimization unit 20 optimizes the multi-objective function based on the conditions. The optimization unit 20 outputs the optimization result of the multi-objective function, for example, to the outside.

[0089] In the optimization device 2 shown in Fig. 21, the reception unit 19 and the optimization unit 20 are each applied to the optimization device 2 shown in Fig. 1. However, this is merely an example, and the reception unit 19 and the optimization unit 20 may each be applied to the optimization device 2 shown in Fig. 9, the optimization device 2 shown in Fig. 12, the optimization device 2 shown in Fig. 16, or the optimization device 2 shown in Fig. 19.

[0090] The reception unit 19 is realized, for example, by a reception circuit 39 shown in FIG. 22 , and the optimization unit 20 is realized, for example, by an optimization circuit 40 shown in FIG. 22 . In FIG. 21 , it is assumed that the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 13, the reception unit 19, and the optimization unit 20, which are components of the optimization device 2, are each realized by dedicated hardware as shown in FIG. 22 . That is, it is assumed that the optimization device 2 is realized by a condition acquisition circuit 31, a prediction information acquisition circuit 32, a restriction information acquisition circuit 33, a reception circuit 39, and an optimization circuit 40. Each of the condition acquisition circuit 31, the prediction information acquisition circuit 32, the restriction information acquisition circuit 33, the reception circuit 39, and the optimization circuit 40 corresponds to, for example, a single circuit, a composite circuit, a processor that executes processing based on instructions from a program, a processor that executes processing based on instructions from a parallel program, an ASIC, an FPGA, or a combination thereof.

[0091] The components of the optimization device 2 are not limited to those realized by dedicated hardware, and the optimization device 2 may be realized by software, firmware, or a combination of software and firmware. When the optimization device 2 is realized by software, firmware, or the like, programs for causing a computer to execute the processing procedures of the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 13, the reception unit 19, and the optimization unit 20 are stored in a memory 51 shown in Fig. 3. Then, a processor 52 shown in Fig. 3 executes the programs stored in the memory 51.

[0092] 22 shows an example in which each of the components of the optimization device 2 is realized by dedicated hardware, while Fig. 3 shows an example in which the optimization device 2 is realized by software, firmware, etc. However, this is merely an example, and some of the components in the optimization device 2 may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0093] Next, the operation of the optimization device 2 shown in Fig. 21 will be described. The optimization device 2 is the same as that shown in Fig. 1 except for the reception unit 19 and the optimization unit 20. Therefore, only the operation of the reception unit 19 and the optimization unit 20 will be described here.

[0094] The reception unit 19 includes a man-machine interface that receives a selection of whether to optimize the multi-objective function using the restriction information or to optimize the multi-objective function without using the restriction information, or an input / output interface that receives the above selection information from the outside. The reception unit 19 receives a selection of whether to optimize the multi-objective function using the restriction information or to optimize the multi-objective function without using the restriction information, and outputs the received selection result to the optimization unit 20.

[0095] The optimization unit 20 acquires conditions from the condition acquisition unit 11 and acquires restriction information from the restriction information acquisition unit 13. The optimization unit 20 also acquires a selection acceptance result from the reception unit 19. If the selection acceptance result from the reception unit 19 indicates a selection to optimize the multi-objective function using the restriction information, the optimization unit 20 optimizes the multi-objective function based on the conditions and the restriction information, similar to the optimization unit 14 shown in FIG. 1. If the selection acceptance result from the reception unit 19 indicates a selection to optimize the multi-objective function without using the restriction information, the optimization unit 20 optimizes the multi-objective function based on the conditions. The optimization unit 20 outputs the optimization result of the multi-objective function, for example, to the outside.

[0096] In the sixth embodiment described above, the optimization device 2 is configured to include a reception unit 19 that receives a selection of whether to optimize the multi-objective function using restriction information or to optimize the multi-objective function without using restriction information, and the optimization unit 20 optimizes the multi-objective function based on the selection received by the reception unit 19. Therefore, the optimization device 2 can more appropriately optimize the multi-objective function and can perform optimization using a calculation method that reflects the user's intentions. In the sixth embodiment, when optimization of the multi-objective function using restriction information is not suitable, normal optimization of the multi-objective function is possible.

[0097] Seventh Embodiment In a seventh embodiment, an optimization device 2 including a learning unit 21 that causes a learning model 1 to learn using the results of optimization by the optimization unit 14 as learning data will be described.

[0098] Fig. 23 is a configuration diagram showing an optimization device 2 according to embodiment 7. In Fig. 23, the same reference numerals as those in Fig. 1, 9, 12, 16, 19, or 21 indicate the same or corresponding parts, and detailed descriptions thereof will be omitted. Fig. 24 is a hardware configuration diagram showing the hardware of the optimization device 2 according to embodiment 7. In Fig. 24, the same reference numerals as those in Fig. 2, 13, 17, 20, or 22 indicate the same or corresponding parts, and detailed descriptions thereof will be omitted.

[0099] The learning unit 21 acquires the optimization result of the multi-objective function from the optimization unit 14. The learning unit 21 causes the learning model 1 to perform learning using the optimization result of the multi-objective function as learning data.

[0100] The optimization device 2 shown in Fig. 23 has a learning unit 21 applied to the optimization device 2 shown in Fig. 1. However, this is merely an example, and the learning unit 21 may also be applied to the optimization device 2 shown in Fig. 9, the optimization device 2 shown in Fig. 12, the optimization device 2 shown in Fig. 16, the optimization device 2 shown in Fig. 19, or the optimization device 2 shown in Fig. 21.

[0101] The learning unit 21 is realized, for example, by a learning circuit 41 shown in Fig. 24. In Fig. 23, it is assumed that each of the components of the optimization device 2, namely, the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 13, the optimization unit 14, and the learning unit 21, is realized by dedicated hardware as shown in Fig. 24. That is, it is assumed that the optimization device 2 is realized by a condition acquisition circuit 31, a prediction information acquisition circuit 32, a restriction information acquisition circuit 33, an optimization circuit 34, and a learning circuit 41. Each of the condition acquisition circuit 31, the prediction information acquisition circuit 32, the restriction information acquisition circuit 33, the optimization circuit 34, and the learning circuit 41 corresponds to, for example, a single circuit, a composite circuit, a processor that executes processing based on instructions from a program, a processor that executes processing based on instructions from a parallel program, an ASIC, an FPGA, or a combination thereof.

[0102] The components of the optimization device 2 are not limited to those realized by dedicated hardware, and the optimization device 2 may be realized by software, firmware, or a combination of software and firmware. When the optimization device 2 is realized by software, firmware, or the like, programs for causing a computer to execute the processing procedures of the condition acquisition unit 11, the prediction information acquisition unit 12, the restriction information acquisition unit 13, the optimization unit 14, and the learning unit 21 are stored in a memory 51 shown in Fig. 3. Then, a processor 52 shown in Fig. 3 executes the programs stored in the memory 51.

[0103] 24 shows an example in which each of the components of the optimization device 2 is realized by dedicated hardware, while Fig. 3 shows an example in which the optimization device 2 is realized by software, firmware, etc. However, this is merely an example, and some of the components in the optimization device 2 may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0104] Next, the operation of the optimization device 2 shown in Fig. 23 will be described. Except for the learning unit 21, the optimization device 2 is the same as that shown in Fig. 1. Therefore, only the operation of the learning unit 21 will be described here.

[0105] The learning unit 21 acquires the optimization result of the multi-objective function from the optimization unit 14. The learning unit 21 causes the learning model 1 to learn using the optimization result of the multi-objective function as learning data. Specifically, when the optimization result of the multi-objective function is obtained, the learning unit 21 causes the learning model 1 to learn the correspondence between the conditions output from the condition acquisition unit 11 and the prediction information output from the prediction information acquisition unit 12. As a result of this learning, if the condition acquisition unit 11 subsequently acquires conditions similar to the conditions under which the optimization result of the multi-objective function was obtained, the learning model 1 to which such conditions are given will output prediction information similar to the prediction information under which the optimization result of the multi-objective function was obtained.

[0106] In the seventh embodiment described above, the optimization device 2 is configured to include a learning unit 21 that causes the learning model 1 to learn using the results of optimization by the optimization unit 14 as learning data. Therefore, the optimization device 2 improves the optimization accuracy of the optimization unit 14. In other words, by causing the learning model 1 to learn using the optimized results as learning data, it is possible to avoid a situation where correct inference cannot be made due to a lack of past data.

[0107] 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.

[0108] The present disclosure is suitable for an optimization device, an optimization method, and a program.

[0109] 1 Learning model, 2 Optimization device, 3 Memory unit, 11 Condition acquisition unit, 12 Prediction information acquisition unit, 13 Restriction information acquisition unit, 14 Optimization unit, 15 Restriction information acquisition unit, 16 Optimization unit, 17 Output unit, 18 Processing unit, 19 Reception unit, 20 Optimization unit, 21 Learning unit, 31 Condition acquisition circuit, 32 Prediction information acquisition circuit, 33 Restriction information acquisition circuit, 34 Optimization circuit, 35 Restriction information acquisition circuit, 36 Optimization circuit, 37 Output circuit, 38 Processing circuit, 39 Reception circuit, 40 Optimization circuit, 41 Learning circuit, 51 Memory, 52 Processor.

Claims

1. An optimization device that optimizes a multi-objective function, which is a set of multiple objective functions, A condition acquisition unit that acquires conditions to be used in the aforementioned optimization calculation, A prediction information acquisition unit acquires prediction information, which is information predicted based on the aforementioned conditions. A restriction information acquisition unit acquires restriction information to impose restrictions on the optimization calculation based on the prediction information, The system includes an optimization unit that optimizes the multi-objective function based on the aforementioned conditions and the aforementioned limitation information. An optimization device characterized by the following features.

2. The aforementioned prediction information is information output from a trained model into which the aforementioned conditions have been input. The optimization apparatus according to claim 1, characterized by the features described above.

3. The aforementioned predictive information acquisition unit, The predictive information is obtained from the memory unit based on the conditions described above. The optimization apparatus according to claim 1, characterized in that it is a feature of the present invention.

4. The aforementioned forecast information is time-series information. The optimization apparatus according to any one of claims 1 to 3.

5. The aforementioned restriction information acquisition unit, The unit that acquires the prediction information accepts changes to the prediction information and acquires the restriction information based on the changed prediction information. The optimization apparatus according to any one of claims 1 to 3.

6. The prediction information includes at least one of the following: information predicting the variables of the objective function, information predicting the objective function, and information predicting the constraint functions that constrain the optimization. The optimization apparatus according to any one of claims 1 to 3.

7. The aforementioned restriction information is a constraint condition that imposes restrictions on the optimization calculation. The optimization apparatus according to any one of claims 1 to 3.

8. The aforementioned constraint is a condition that restricts the objective function or the explanatory variables of the objective function to be included within the range set based on the prediction information. The optimization apparatus according to claim 7.

9. The aforementioned restriction information is information that indicates the range in which the plurality of objective functions can take place on the objective function space, based on the prediction information. The optimization apparatus according to any one of claims 1 to 3.

10. The aforementioned restriction information is information that indicates the range that the plurality of objective functions can take, with reference points corresponding to the prediction information on the objective function space. The optimization apparatus according to feature 9.

11. The aforementioned reference points are set in the objective function space in multiple locations based on the prediction information. The optimization apparatus according to claim 10, characterized in that it is a feature of the present invention.

12. The aforementioned reference point is the first reference point, The aforementioned limitation information is information indicating the range that the multiple objective functions can take, based on a second reference point modified by the user from the first reference point. The optimization apparatus according to claim 10, characterized in that it is a feature of the present invention.

13. The optimization unit, The multi-objective function is optimized using evolutionary computation. For the first group of candidate solutions, which are candidates for the solution, an optimization process is performed based on constraint conditions included in the restriction information that impose constraints on the optimization calculation. For the second group of candidate solutions, which are different from the first group of candidate solutions, an optimization process is performed based on preference conditions, which are information indicating the range that the multiple objective functions can take in the objective function space, based on the prediction information included in the restriction information. During the optimization process by evolutionary computation, a swap process is performed between a part of the first group of candidate solutions that has undergone optimization based on the constraint conditions and a part of the second group of candidate solutions that has undergone optimization based on the preference conditions. An optimization process based on the preference conditions is performed on a part of the first group of candidate solutions, and an optimization process based on the constraint conditions is performed on a part of the second group of candidate solutions. The optimization apparatus according to any one of claims 1 to 3.

14. An output unit that outputs a plurality of optimization results, including the result optimized by the optimization unit, The system includes a processing unit that performs processing based on the optimization result selected by the user from among the plurality of optimization results output from the output unit. The optimization apparatus according to any one of claims 1 to 3.

15. The system includes a receiving unit that accepts a selection of whether to optimize the multi-objective function using the aforementioned restriction information or to optimize the multi-objective function without using the aforementioned restriction information. The optimization unit, Based on the selection received by the reception unit, the multi-objective function is optimized. The optimization apparatus according to any one of claims 1 to 3.

16. The system includes a learning unit that uses the results optimized by the optimization unit as training data to train the trained model. The optimization apparatus according to feature 2.

17. An optimization method that optimizes a multi-objective function, which is a set of multiple objective functions, The condition acquisition unit acquires the conditions to be used in the optimization calculation, The prediction information acquisition unit acquires prediction information, which is information predicted based on the above conditions. The restriction information acquisition unit acquires restriction information to impose restrictions on the optimization calculation based on the prediction information. The optimization unit optimizes the multi-objective function based on the conditions and the constraint information. An optimization method characterized by the following:

18. A program to be executed by a computer in an optimization device that optimizes multiple objective functions, which are a multi-objective function, To the aforementioned computer, A process to obtain the conditions used in the aforementioned optimization calculation, A process to obtain predictive information, which is information predicted based on the aforementioned conditions, A process to obtain constraint information to impose restrictions on the optimization calculation based on the aforementioned prediction information, A program for performing a process to optimize the multi-objective function based on the aforementioned conditions and the aforementioned limitation information.