Parameter generation apparatus, system, method, and program

Abstract

An input means 81 receives an input of a constraint and a first objective function defining a combination of factors relating to the production of a material. An objective function generation means 82 generates a second objective function that defines a stochastic fluctuation for a parameter of the first objective function. An optimisation processing means 83 optimises a model containing the second objective function and the constraint. The output means 84 sets the values of variables of the second objective function obtained by the optimisation as a parameter set and outputs the same.

Description

[Technical Field]

[0001] The present invention relates to a parameter generation device, a parameter generation system, a parameter generation method, and a parameter generation program for generating desired parameters. [Background technology]

[0002] In research into new materials, researchers are trying out a vast number of combinations of factors such as material type and quantity, processing temperature, pressure, and time in order to discover manufacturing methods for new materials that exhibit the desired performance. However, the number of combinations is astronomical, making it impossible to try all patterns.

[0003] Generally, knowledge is accumulated from past experiences and simulation results regarding combinations of conditions that are expected to yield good results, and combinations of conditions that are expected to yield the opposite results. Therefore, in research settings, the process of determining new combinations of elements from within the range that satisfy the conditions indicated by this accumulated knowledge, and verifying the results through prototyping and simulations, is repeatedly carried out.

[0004] Furthermore, in order to derive a combination of elements that satisfies the desired conditions, a mathematical programming solver may be used to derive the optimal combination based on an objective function designed by engineers and other personnel, and constraints that define the conditions that must be satisfied (i.e., a mathematical optimization problem).

[0005] Furthermore, Patent Document 1 describes a design support system that reduces the number of numerical simulations required when examining design parameters to achieve design objectives. The design support system described in Patent Document 1 performs a forward analysis by providing initial settings for the design parameters, and then performs an inverse analysis based on the results of that analysis, thereby performing a sensitivity analysis to the design objectives. [Prior art documents] [Patent Documents]

[0006]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0007] In considering a method for manufacturing a desired new material, it is necessary to create a combination of elements for manufacturing the new material. As one such method, there is a method created based on the experience and intuition of experts. However, since this method is highly personalized, there is a problem that work cannot be efficiently advanced without a specific expert.

[0008] Also, as another method for creating a combination of elements, a method of selecting a combination that satisfies conditions from a randomly selected combination of elements can be considered. However, when the conditions become complex, the probability of a combination of elements that satisfies the conditions occurring becomes low, so there is also a problem of inefficiency.

[0009] On the other hand, by using a mathematical programming solver, it is possible to derive an optimal solution for a designed mathematical optimization problem. However, the obtained optimal solution is one for one designed objective function. Usually, in a scenario such as exploring a new material, a certain type of combination of candidates for elements for manufacturing the material is required. Therefore, it can be said that it is also inefficient for engineers to design a mathematical optimization problem every time a combination of elements is derived.

[0010] Also, the method described in Patent Document 1 aims to reduce the number of simulations when considering design parameters for achieving a design goal, and does not derive a combination of multiple elements.

[0011] Furthermore, there are cases where the objective function designed by engineers or the like is not necessarily accurate. In this case, there is also a problem that the obtained optimal solution may not necessarily be a solution for manufacturing the desired material. Therefore, it is desired to be able to efficiently discover a plurality of methods for manufacturing the desired material.

[0012] Therefore, an object of the present invention is to provide a parameter generation device, a parameter generation system, a parameter generation method, and a parameter generation program capable of discovering a plurality of methods for manufacturing a desired material.

Means for Solving the Problems

[0013] The parameter generation device according to the present invention A first objective function generation means generates a first objective function that includes a weighted linear sum of a predictive model in which the material is used as an explanatory variable, characteristic values ​​representing the material's properties are used as the objective variable, and parameters are set for the explanatory variable. First objective function of comprises an input means for receiving an input of constraint conditions, an objective function generation means for generating a second objective function in which a probabilistic fluctuation is set for the parameters of the first objective function, an optimization processing means for optimizing a model including the second objective function and the constraint conditions, and an output means for outputting the values of the variables of the second objective function obtained by the optimization as a parameter set. The objective function generation means accepts input from the user for weights to be modified for the generated second objective function, and updates the second objective function to reflect the accepted weights. It is characterized by this.

[0014] The parameter generation system according to the present invention uses past experimental data as training data, uses the material as an explanatory variable, and uses the characteristic value indicating the characteristics of the material as an objective variable And parameters are set for those explanatory variables. a prediction model generation device for learning a prediction model, and a prediction model Includes a weighted linear sum a first objective function generation device for generating a first objective function, and a parameter generation device for generating a parameter set using the first objective function. The first objective function generation device generates a first objective function including a linear sum of the characteristic values indicated by the objective variable as a combination of elements and inputs it to the parameter generation device. The parameter generation device uses the first objective function ofThe system includes an input means for receiving constraint conditions, an objective function generation means for generating a second objective function in which stochastic fluctuations are set for the parameters of the first objective function, an optimization processing means for optimizing a model that includes the second objective function and constraint conditions, and an output means for outputting the values ​​of the variables of the second objective function obtained by optimization as a parameter set. Furthermore, the objective function generation means accepts input from the user for weights to be modified for the generated second objective function, and updates the second objective function reflecting the accepted weights. It is characterized by the following:

[0015] The parameter generation method according to the present invention is The computer generates a first objective function that includes a weighted linear sum of predictive models, where the material is the explanatory variable, characteristic values ​​representing the material's properties are the dependent variable, and parameters are set for the explanatory variable. The computer, the first objective function of The computer accepts constraints as input, generates a second objective function with stochastic fluctuations set for the parameters of the first objective function, optimizes the model including the second objective function and constraints, and outputs the values ​​of the variables of the second objective function obtained through optimization as a parameter set. The computer then accepts input from the user for weights to modify the generated second objective function, and updates the second objective function to reflect the accepted weights. It is characterized by the following:

[0016] The parameter generation program according to the present invention is provided to a computer, A first objective function generation process that generates a first objective function that includes a weighted linear sum of a predictive model in which the material is used as an explanatory variable, characteristic values ​​representing the material's properties are used as the dependent variable, and parameters are set for the explanatory variable. First objective function of The process involves an input process that accepts constraint conditions, an objective function generation process that generates a second objective function by setting stochastic fluctuations for the parameters of the first objective function, an optimization process that optimizes the model including the second objective function and constraint conditions, and an output process that outputs the values ​​of the variables of the second objective function obtained through optimization as a parameter set. In the objective function generation process, the user is asked to input weights to be modified for the generated second objective function, and the second objective function is updated to reflect the accepted weights. It is characterized by the following: [Effects of the Invention]

[0017] According to the present invention, multiple methods for manufacturing a desired material can be discovered. [Brief explanation of the drawing]

[0018] [Figure 1] This is a block diagram showing an example configuration of one embodiment of the simulation system of the present invention. [Figure 2] This is an explanatory diagram showing an example of the operation of the parameter generation device. [Figure 3]This is a block diagram illustrating the parameter generation device according to the present invention. [Figure 4] This is a block diagram outlining the simulation system according to the present invention. [Figure 5] This is a schematic block diagram showing the configuration of a computer according to at least one embodiment. [Modes for carrying out the invention]

[0019] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0020] Figure 1 is a block diagram showing an example configuration of one embodiment of the simulation system of the present invention. The simulation system 100 of this embodiment includes a prediction model generation device 10, a first objective function generation device 20, a parameter generation device 30, an optimization processing device 40, and a simulator 50.

[0021] The predictive model generation device 10 is a device that generates a predictive model that predicts the effect of the type and quantity of material on the properties of a product (for example, a material) based on past experimental data. Specifically, the predictive model generation device 10 learns a predictive model that predicts values ​​indicating the properties of a material (hereinafter referred to as "characteristic values") based on past experimental data. These characteristic values ​​can also be called performance indicators.

[0022] The predictive model generation device 10 includes a storage unit 11, a learning unit 12, and a model output unit 13.

[0023] The memory unit 11 stores training data that the learning unit 12 uses for learning. The training data is, for example, data that associates multiple materials used in the manufacture of a material with characteristic values ​​that show the material properties such as hardness, toughness, and heat resistance when those materials are used. The memory unit 11 is implemented, for example, by a magnetic disk.

[0024] The learning unit 12 uses past experimental data as training data to learn a predictive model with material as the explanatory variable and characteristic values ​​as the dependent variable. The method by which the learning unit 12 learns the predictive model is arbitrary, and any method such as machine learning may be used.

[0025] The model output unit 13 outputs the prediction model generated by the learning unit 12. The model output unit 13 may also input the prediction model to the first objective function generator 20.

[0026] The first objective function generation device 20 generates an objective function that defines a combination of elements related to material manufacturing in order to obtain a target characteristic value. Here, elements related to material manufacturing refer to the details that should be specified in the material manufacturing method, and specifically include the type and quantity of material, as well as processing temperature, pressure, and processing time. The objective function is defined using parameters such as weights (coefficients) and biases set for each element.

[0027] In other words, the first objective function generation device 20 generates an objective function (hereinafter referred to as the first objective function) used to derive the optimal combination of material type, quantity, processing method, etc., that achieve the target value, using elements related to material manufacturing and the parameters described above.

[0028] The first objective function generator 20 may generate the first objective function using the prediction model generated by the learning unit 12. For example, the i-th characteristic value y i A predictive model that predicts the following uses j elements x as a combination of elements, as exemplified by Equation 1 below. j Let's assume it is represented as a linear combination of . Here, the x bar (superscript x) is the mean of the elements, and σ is the standard deviation of the elements, which are values ​​calculated when creating a predictive model.

[0029]

number

[0030] At this time, the first objective function generation device 20 may generate a first objective function including a linear sum of each characteristic value y i For example, when the weight for each characteristic value y i is W i , the first objective function generation device 20 may generate a first objective function as shown in Equation 2 exemplified below. Equation 2 is a linear sum of the squares of the differences from the target median value of the characteristic values. Here, Lmed i is the target median value of the characteristic value y i . Also, the weight W i is determined by a technician or the like. Note that the designation of W i may be accepted by the first objective function generation device 20 or may be accepted by the parameter generation device 30 described later.

[0031]

Equation

[0032] Also, as exemplified in Equation 3 below, the first objective function may be represented in a form obtained by expanding the above Equation 2.

Equation

[0033] In the examples of Equations 1 to 3 above, for example, a ij , b i , Lmed i , W i , Q ij , L i and the like are the parameters described above. That is, the parameters shown in this embodiment include not only the parameters at the time of formulation but also the parameters obtained during the formulation.

[0034] In the above explanation, we used the example of a case where the first objective function is composed of a linear sum of the squared differences of the target variable (characteristic value) from the target median, but the contents of the first objective function are not limited to characteristic values. The first objective function may also include elements other than characteristic values ​​(for example, processing methods). The first objective function generator 20 inputs the generated first objective function to the parameter generator 30.

[0035] Furthermore, this embodiment illustrates a case where the first objective function generation device 20 is implemented as an independent device. However, the first objective function generation device 20 may be implemented as an integral part of other devices, for example, it may be included in the parameter generation device 30.

[0036] The parameter generation device 30 is a device that generates parameters to be input to the simulator 50, and is connected to the optimization processing device 40 and the simulator 50. The simulator 50 is a device that performs trials based on the generated parameters. The form of the simulator 50 is arbitrary, and any known device may be used.

[0037] Furthermore, the optimization processing unit 40 is a device that performs optimization processing based on the model generated by the parameter generation device 30. The optimization processing unit 40 may be implemented by a (classical) computer that runs a mathematical programming solver. Alternatively, the optimization processing unit 40 may be a dedicated device for finding the ground state of the Hamiltonian of the Ising model. In this case, the optimization processing unit 40 is implemented, for example, as a device that performs annealing based on the Ising model generated by the parameter generation device 30.

[0038] The parameter generation device 30 includes an input unit 31, an objective function generation unit 32, an optimization processing unit 33, and an output unit 34.

[0039] The input unit 31 accepts the input of the first objective function described above. The input unit 31 also accepts input of constraint conditions that indicate the constraints that each element must satisfy and the constraints when combining each element. The input unit 31 may accept input of the first objective function generated by the first objective function generation device 20, or it may accept input of the first objective function generated manually by another device (not shown) or by an engineer.

[0040] For example, constraints when manufacturing new materials may include specifications regarding the selection of material types (e.g., one from each material group), specifications regarding the proportion of materials (e.g., specifying the sum of the proportions of several materials, specifying the proportions of individual materials), and specifications for mutually exclusive materials. Other constraints when manufacturing new materials may include specifications regarding the processing of materials (e.g., restrictions on processing temperature (e.g., upper limit temperature) and pressure depending on the material).

[0041] The objective function generation unit 32 generates an objective function (hereinafter referred to as the second objective function) in which probabilistic fluctuations are set for the parameters of the input first objective function. Here, setting fluctuations for parameters means performing arithmetic operations such as addition, subtraction, multiplication, and division on the parameters based on the values ​​indicated by the fluctuations. Furthermore, the parameters to which fluctuations are set are those that appear in the final first objective function (for example, Q in equation 3 above). ij Yes, L i ) In addition to the parameters used in the formulation (for example, a in equation 1 above) ij Yeah, Lmed i ) is also included.

[0042] The parameters for which fluctuations are set are specified in advance. The method of specification is arbitrary; for example, the input unit 31 may accept input for the parameters for which fluctuations are set from an engineer or other person. Note that fluctuations are set for the parameters of the objective function, but not for the constraint conditions.

[0043] Specifically, the objective function generation unit 32 sets fluctuations for the parameters of the first objective function, which are represented by random variables following a predetermined probability distribution. Preferably, the objective function generation unit 32 sets fluctuations for the parameters of the first objective function, which are represented by random variables following a probability distribution with a mean of zero. Examples of probability distributions with a mean of zero include the normal distribution exemplified in Equation 4 and the uniform distribution exemplified in Equation 5 below.

[0044]

number

[0045] To make the mean of the probability distribution zero, we can set μ=0 in the normal distribution in Equation 4 and a=-b (b>0) in the uniform distribution in Equation 5. In the case of the normal distribution, the standard deviation σ is an indicator of the magnitude of the fluctuation. In the case of the uniform distribution, the interval width ba is an indicator of the magnitude of the fluctuation. That is, the larger this parameter is, the greater the fluctuation, and conversely, the smaller this parameter is, the smaller the fluctuation. It can also be said that the similarity to the original objective function (optimization problem) changes depending on the magnitude of the fluctuation given.

[0046] The following describes how to set fluctuations for the parameters of the first objective function, using equations 1 to 3 as examples, and representing fluctuations expressed by a random variable that follows the probability distribution shown in equation 4 or 5 as an example. The fluctuations set in this embodiment are expressed by an equation for a random variable x that follows the probability distribution p(x) of the fluctuations.

[0047] For example, the probability distribution of fluctuations p(X ij Assume that ) is represented by the normal distribution shown in Equation 4 above. In this case, the fluctuation X for Equation 1 shown above is ij The second objective function, with the parameter set, is represented by Equation 6, which is exemplified below. As shown in Equation 6, the standard deviation σ is, for example, parameter a ij It is set to a constant c times.

[0048]

number

[0049] In Equation 6, the index representing the magnitude of the fluctuation is p(X ij This is the standard deviation σ of ). By increasing the positive constant c, the magnitude of fluctuations tends to increase (i.e., X ij (It tends to grow larger.)

[0050] Similarly, for equation 2 shown above, fluctuation X i The second objective function, with the setting, is represented by Equation 7, which is exemplified below. As shown in Equation 7, the index representing the magnitude of fluctuation is, for example, the parameter Lmed i It is set to a constant c times.

[0051]

number

[0052] Furthermore, for equation 3 shown above, fluctuation X ij And fluctuation X i The second objective function, with the setting, is represented by Equation 8, which is exemplified below. As shown in Equation 8, the index representing the magnitude of the fluctuation is, for example, the parameter Q ij and L i It is set to a constant c times the standard deviation of the non-zero values.

[0053]

number

[0054] The above examples illustrate the equation for fluctuations when the probability distribution is a normal distribution. The same applies when the probability distribution is a uniform distribution. For example, in the case of equation 1 above, the probability distribution is represented by equation 9, which is illustrated below.

[0055]

number

[0056] In this way, the objective function generation unit 32 generates a second objective function in which stochastic fluctuations are set for the parameters of the first objective function. Furthermore, the objective function generation unit 32 may output the generated second objective function (i.e., the objective function with fluctuations set) and accept modifications from engineers or other personnel.

[0057] For example, in Equation 1 illustrated above, parameter a ij Let's assume we set fluctuations for . In this case, the objective function generation unit 32 sets parameter a ij After setting fluctuations for this, the generated second objective function is output. Then, the engineer uses the W in equation 2 as exemplified above. i After determining the W, the objective function generation unit 32 then generates the W i The input was accepted as a correction, and the accepted W i Objective function H that reflects this O It is sufficient to generate the following. At this time, the objective function generation unit 32 determines the W i Instead, W i The objective function H reflects this. O You may accept this input as a correction.

[0058] In addition, for example, in equation 2 illustrated above, the parameter Lmed i Let's assume we set a fluctuation for . In this case, the objective function generation unit 32 sets the parameter Lmed i After setting fluctuations for this, the generated second objective function is output. Then, the engineer uses the W in equation 2 as exemplified above. i After determining the W, the objective function generation unit 32 then generates the W i The input was accepted as a correction, and the accepted W i Objective function H that reflects this O It is sufficient to generate the following. As described above, the objective function generation unit 32 generates the determined W i Instead, W i The objective function H reflects this. O You may accept this input as a correction.

[0059] In this way, by outputting the generated second objective function and accepting modifications from engineers, the objective function after the fluctuations have been set by engineers is verified, making it possible to generate a more favorable objective function (mathematical programming problem). In the following explanation, the objective function modified by engineers will also be referred to as the second objective function.

[0060] The optimization processing unit 33 optimizes the model that includes the second objective function and constraints generated by the objective function generation unit 32. Specifically, the optimization processing unit 33 sends the model to be optimized to the optimization processing unit 40 to execute the optimization process and receives the execution result.

[0061] Specifically, first, the optimization processing unit 33 generates a model to be optimized from the second objective function and constraints, according to the optimization processing unit 40. For example, as described above, suppose the optimization processing unit 40 is implemented by a computer that runs a mathematical programming solver. In this case, the optimization processing unit 33 generates a mathematical optimization problem including the second objective function and constraints as the model to be optimized, and then has the computer run the generated model.

[0062] Furthermore, for example, let's assume that the optimization processing unit 40 is implemented by a device that performs annealing (annealing machine), as described above. In this case, the optimization processing unit 33 only needs to generate the Ising model to be optimized based on the second objective function and constraints. Since the method for generating the Ising model from the objective function and constraints is widely known, a detailed explanation is omitted here.

[0063] The output unit 34 outputs the variable values ​​of the second objective function obtained by optimization as a parameter set. Specifically, the variable values ​​here are information indicating the specific values ​​and settings of each element (for example, material type and quantity, processing temperature, pressure, time, etc.). The output unit 34 may output the parameter set directly to the simulator 50, or it may output it in a file format (for example, CSV (Comma Separated Value) format). Furthermore, if the optimization processing device 40 is an annealing machine, the optimization results are obtained as binary variables, so the output unit 34 may output a parameter set converted from the optimization results.

[0064] The input unit 31, the objective function generation unit 32, the optimization processing unit 33, and the output unit 34 are all implemented by a computer processor (e.g., a CPU (Central Processing Unit)) that operates according to a program (parameter generation program).

[0065] For example, the program may be stored in the memory unit (not shown) of the parameter generation device 30, and the processor may read the program and operate as the input unit 31, objective function generation unit 32, optimization processing unit 33, and output unit 34 according to the program. Alternatively, the functions of the parameter generation device 30 may be provided in SaaS (Software as a Service) format.

[0066] Furthermore, the input unit 31, the objective function generation unit 32, the optimization processing unit 33, and the output unit 34 may each be implemented with dedicated hardware. Also, some or all of the components of each device may be implemented by general-purpose or dedicated circuits, processors, etc., or combinations thereof. These may be configured by a single chip or by multiple chips connected via a bus. Some or all of the components of each device may be implemented by a combination of the above-mentioned circuits, etc., and programs.

[0067] Furthermore, if some or all of the components of the parameter generation device 30 are realized by multiple information processing devices or circuits, these multiple information processing devices or circuits may be centrally located or distributed. For example, the information processing devices or circuits may be realized in a form in which each is connected via a communication network, such as a client-server system or a cloud computing system.

[0068] Next, the operation of the parameter generation device 30 in this embodiment will be described. Figure 2 is a flowchart showing an example of the operation of the parameter generation device 30.

[0069] The input unit 31 accepts the input of the first objective function and constraints (step S11). As described above, the first objective function is a function that defines the combination of elements related to the manufacturing of the material. The objective function generation unit 32 generates a second objective function by setting stochastic fluctuations for the parameters of the first objective function (step S12). The optimization processing unit 33 optimizes the model including the second objective function and constraints (step S13). More specifically, the optimization processing unit 33 causes the optimization processing unit 40 to perform the optimization process. The output unit 34 then outputs the values ​​of the variables of the second objective function obtained by optimization as a parameter set (step S14).

[0070] As described above, in this embodiment, the input unit 31 receives input of the first objective function and constraints, and the objective function generation unit 32 generates a second objective function in which stochastic fluctuations are set for the parameters of the first objective function. Then, the optimization processing unit 33 optimizes the model including the second objective function and constraints, and the output unit 34 outputs the values ​​of the variables of the second objective function obtained by optimization as a parameter set.

[0071] With the above configuration, it is possible to obtain a combination of elements (i.e., a parameter set) that represents multiple methods for manufacturing the desired material. Then, by performing a simulation based on this parameter set, it is possible to determine whether or not the desired material was obtained. As a result, it becomes possible to discover multiple methods for manufacturing the desired material.

[0072] Furthermore, in this embodiment, the optimization processing unit 33 causes the computer running the mathematical programming solver to perform the optimization process, making it possible to quickly determine the parameter set that achieves the desired properties.

[0073] Furthermore, in this embodiment, the objective function generation unit 32 sets fluctuations only for the objective function without changing the constraints, thereby obtaining a diverse set of parameters that satisfy the constraints even in a mathematical programming solver. In this case, the objective function generation unit 32 sets the fluctuations to be set for the objective function using a probability distribution centered on the original model (objective function), so it can obtain a set of parameters that is close to the optimal solution for the original model.

[0074] Furthermore, in this embodiment, the objective function generation unit 32 can continuously change the degree of fluctuation by setting the fluctuation based on a probability distribution, so a variety of parameter sets can be obtained, from parameter sets close to the optimal solution to parameter sets relatively far from it.

[0075] Next, an overview of the present invention will be described. Figure 3 is a block diagram illustrating an overview of the parameter generation device according to the present invention. The parameter generation device 80 (for example, the parameter generation device 30) according to the present invention includes an input means 81 (for example, an input unit 31) that accepts input of a first objective function that defines a combination of elements related to the manufacture of a material (for example, the type and amount of material, processing temperature, pressure, time, etc.) and constraints (for example, selection of material type, distribution of material quantities, specification of exclusive materials, processing method of materials, etc.), an objective function generation means 82 (for example, an objective function generation unit 32) that generates a second objective function in which probabilistic fluctuations are set for the parameters of the first objective function, an optimization processing means 83 (for example, an optimization processing unit 33) that optimizes a model including the second objective function and constraints, and an output means 84 (for example, an output unit 34) that outputs the values ​​of the variables of the second objective function obtained by optimization as a parameter set.

[0076] Such a configuration makes it possible to discover multiple methods for manufacturing the desired material. Specifically, the above configuration allows us to obtain a combination of elements (parameter set) representing multiple methods for manufacturing the desired material. By performing simulations based on this parameter set, we can determine whether or not the desired material was obtained. As a result, it becomes possible to discover multiple methods for manufacturing the desired material.

[0077] Furthermore, the objective function generation means 82 may generate a second objective function in which fluctuations, represented by a random variable following a predetermined probability distribution, are set as parameters.

[0078] Specifically, the objective function generation means 82 may generate a second objective function in which fluctuations represented by a random variable following a probability distribution with a mean of zero are set as parameters. With such a configuration, a parameter set close to the optimal solution for the original model can be obtained.

[0079] Furthermore, the objective function generation means 82 may generate a second objective function in which fluctuations, represented by a random variable following a normal distribution or a uniform distribution, are set as parameters.

[0080] Specifically, the objective function generation means 82 may generate a second objective function in which the parameter is a random variable that follows a normal distribution, where the standard deviation is a constant multiple of the parameter that gives the fluctuation.

[0081] Furthermore, the optimization processing means 83 may generate a model to be optimized for a mathematical optimization problem that includes a second objective function and constraints, and have a computer that runs a mathematical programming solver (e.g., the optimization processing unit 40) execute the generated model. Such a configuration makes it possible to quickly find a parameter set that realizes the desired properties.

[0082] On the other hand, the optimization processing means 83 may generate an Ising model to be optimized based on the second objective function and constraints, and have an annealing machine (e.g., the optimization processing device 40) execute the generated Ising model. Such a configuration makes it possible to obtain parameter sets with different properties from similar objective functions.

[0083] Furthermore, the objective function generation means 82 may output the generated second objective function and accept any modifications made by the user to that second objective function. The optimization processing means 83 may then optimize the model to be optimized, which includes the second objective function reflecting the modifications and the constraints. With such a configuration, the objective function after fluctuations have been set by engineers and others is verified, making it possible to generate a more favorable objective function (mathematical programming problem).

[0084] Figure 4 is a block diagram illustrating the overview of the simulation system according to the present invention. The simulation system 200 (for example, simulation system 100) according to the present invention includes a prediction model generation device 60 (for example, prediction model generation device 10) that uses past experimental data as training data to learn a prediction model in which the material is the explanatory variable and characteristic values ​​indicating the properties of the material are the objective variable; a first objective function generation device 70 (for example, first objective function generation device 20) that uses the prediction model to generate a first objective function that defines the combination of elements related to the manufacturing of the material; and a parameter generation device 80 (for example, parameter generation device 30) that uses the first objective function to generate a parameter set.

[0085] The first objective function generator 70 generates a first objective function that includes a linear sum of the characteristic values ​​represented by the objective variable as a combination of elements, and inputs it to the parameter generator 80.

[0086] The configuration of the parameter generation device 80 is the same as that of the parameter generation device 80 illustrated in Figure 3.

[0087] Even with such a configuration, it becomes possible to discover multiple methods for manufacturing the desired material.

[0088] Figure 5 is a schematic block diagram showing the configuration of a computer according to at least one embodiment. Computer 1000 comprises a processor 1001, main memory 1002, auxiliary memory 1003, and interface 1004. Computer 1000 may also be connected to a computer for running a mathematical programming solver, an annealing machine, a simulator, etc.

[0089] The parameter generation device 80 described above is implemented in the computer 1000. The operation of each processing unit described above is stored in the auxiliary storage device 1003 in the form of a program (parameter generation program). The processor 1001 reads the program from the auxiliary storage device 1003, loads it into the main memory 1002, and executes the above processing according to the program.

[0090] In at least one embodiment, the auxiliary storage device 1003 is an example of a non-temporary tangible medium. Other examples of non-temporary tangible media include magnetic disks, magneto-optical disks, CD-ROMs (Compact Disc Read-only memory), DVD-ROMs (Read-only memory), and semiconductor memory connected via the interface 1004. Furthermore, if this program is distributed to the computer 1000 via a communication line, the computer 1000 that receives the program may expand it into the main memory 1002 and execute the above processing.

[0091] Furthermore, the program may be intended to implement some of the functions described above. In addition, the program may be a so-called differential file (differential program) that implements the functions described above in combination with other programs already stored in the auxiliary storage device 1003.

[0092] Some or all of the above embodiments may also be described as follows, but are not limited to the following:

[0093] (Note 1) An input means that accepts input of a first objective function and constraints that define a combination of elements related to the manufacture of materials, Objective function generation means for generating a second objective function in which a probabilistic fluctuation is set for the parameters of the first objective function, An optimization processing means for optimizing the model including the second objective function and the constraints, The system includes an output means for outputting the values ​​of the variables of the second objective function obtained by the optimization as a parameter set. A parameter generation device characterized by the following features.

[0094] (Note 2) The objective function generation means generates a second objective function in which fluctuations represented by a random variable following a predetermined probability distribution are set as parameters. The parameter generation device described in Appendix 1.

[0095] (Note 3) The objective function generation means generates a second objective function whose parameters are fluctuations represented by a random variable that follows a probability distribution with a mean of zero. Parameter generation device as described in Appendix 1 or Appendix 2.

[0096] (Note 4) The objective function generation means generates a second objective function whose parameters are fluctuations represented by random variables following a normal or uniform distribution. A parameter generation device as described in any one of the appendices 1 to 3.

[0097] (Note 5) The objective function generation means generates a second objective function in which the parameter is a fluctuation represented by a random variable that follows a normal distribution, the standard deviation of which is a constant multiple of the parameter that gives the fluctuation. A parameter generation device as described in any one of the appendices 1 through 4.

[0098] (Note 6) The optimization processing means generates a model of the mathematical optimization problem, including the second objective function and constraints, and causes a computer running a mathematical programming solver to execute the generated model. A parameter generation device as described in any one of the appendices 1 through 5.

[0099] (Note 7) The optimization processing means generates an Ising model to be optimized based on the second objective function and constraints, and causes the annealing machine to execute the generated Ising model. A parameter generation device as described in any one of the appendices 1 through 5.

[0100] (Note 8) The objective function generation means outputs the generated second objective function and accepts the modifications made by the user to the said second objective function. The optimization processing means optimizes the model to be optimized, which includes a second objective function and constraints that reflect the modifications. A parameter generation device as described in any one of the appendices 1 through 7.

[0101] (Note 9) A predictive model generation device that uses past experimental data as training data, with the material as the explanatory variable and characteristic values ​​indicating the material's properties as the dependent variable, A first objective function generation device that generates a first objective function that defines a combination of elements related to the manufacturing of materials using the aforementioned prediction model, The system includes a parameter generation device that generates a parameter set using the first objective function, The first objective function generating device generates a first objective function that includes a linear sum of the characteristic values ​​represented by the objective variable as a combination of the elements, The parameter generation device is An input means for receiving the first objective function and constraints, Objective function generation means for generating a second objective function in which a probabilistic fluctuation is set for the parameters of the first objective function, An optimization processing means for optimizing the model including the second objective function and the constraints, The system includes an output means for outputting the values ​​of the variables of the second objective function obtained by the optimization as a parameter set. A parameter generation system characterized by the following features.

[0102] (Note 10) The computer accepts input of a first objective function and constraints that define the combination of elements related to the manufacture of the material, The computer generates a second objective function in which stochastic fluctuations are set for the parameters of the first objective function. The computer optimizes the model including the second objective function and the constraints, The computer outputs the values ​​of the variables of the second objective function obtained through optimization as a parameter set. A parameter generation method characterized by the following:

[0103] (Note 11) To the computer, An input process that accepts input of a first objective function and constraints that define the combination of elements related to the manufacturing of materials. Objective function generation process that generates a second objective function by setting probabilistic fluctuations for the parameters of the first objective function, An optimization process that optimizes the model including the second objective function and the constraints, and Output process that outputs the values ​​of the variables of the second objective function obtained by the optimization as a parameter set. A program storage medium that stores a parameter generation program for executing a program.

[0104] (Note 12) To the computer, An input process that accepts input of a first objective function and constraints that define the combination of elements related to the manufacturing of materials. Objective function generation process that generates a second objective function by setting probabilistic fluctuations for the parameters of the first objective function, An optimization process that optimizes the model including the second objective function and the constraints, and Output process that outputs the values ​​of the variables of the second objective function obtained by the optimization as a parameter set. A parameter generation program for executing the program. [Industrial applicability]

[0105] The present invention is suitably applied to a parameter generation device that generates desired parameters. Specifically, the present invention is suitably applied in fields where prototyping and simulation are repeatedly performed in research settings for the discovery of new materials. [Explanation of Symbols]

[0106] 10 Predictive Model Generator 11 Storage section 12. Learning Department 13 Model Output Section 20. First Objective Function Generator 30 Parameter Generator 31 Input section 32 Objective Function Generation Unit 33 Optimization Processing Unit 34 Output section 40 Optimization Processing Unit 50 Simulators

Claims

1. A first objective function generation means generates a first objective function that includes a weighted linear sum of a predictive model in which the material is used as an explanatory variable, characteristic values ​​representing the properties of the material are used as the objective variable, and parameters are set for the explanatory variable. An input means for receiving the constraints of the first objective function, Objective function generation means for generating a second objective function in which a probabilistic fluctuation is set for the parameters of the first objective function, An optimization processing means for optimizing the model including the second objective function and the constraints, The system includes an output means that outputs the values ​​of the variables of the second objective function obtained by the optimization as a parameter set, The objective function generation means receives input from the user for the weights to be modified for the generated second objective function, and updates the second objective function to reflect the received weights. A parameter generation device characterized by the following features.

2. The objective function generation means generates a second objective function whose parameters are fluctuations represented by random variables following a predetermined probability distribution. The parameter generation device according to claim 1.

3. The objective function generation means generates a second objective function whose parameters are fluctuations represented by a random variable that follows a probability distribution with a mean of zero. The parameter generation device according to claim 1.

4. The objective function generation means generates a second objective function whose parameters are fluctuations represented by random variables following a normal or uniform distribution. The parameter generation device according to claim 1.

5. The objective function generation means generates a second objective function in which the parameter is a random variable that follows a normal distribution, where the standard deviation is a constant multiple of the parameter that gives the fluctuation. The parameter generation device according to claim 1.

6. The optimization processing means generates a model for optimizing a mathematical optimization problem that includes a second objective function and constraints, and then has a computer that runs a mathematical programming solver execute the generated model. The parameter generation device according to claim 1.

7. The optimization processing means generates an Ising model to be optimized based on the second objective function and constraints, and causes the annealing machine to execute the generated Ising model. The parameter generation device according to claim 1.

8. A predictive model generation device that uses past experimental data as training data, with the material as the explanatory variable and characteristic values ​​representing the material's properties as the dependent variable, and learns a predictive model in which parameters are set for the explanatory variable, A first objective function generator that generates a first objective function including a weighted linear sum of the prediction model, The system includes a parameter generation device that generates a parameter set using the first objective function, The first objective function generation device generates a first objective function that includes a linear sum of the characteristic values ​​indicated by the objective variable as a combination of elements, and inputs it to the parameter generation device. The parameter generation device is An input means for receiving the constraints of the first objective function, Objective function generation means for generating a second objective function in which a probabilistic fluctuation is set for the parameters of the first objective function, An optimization processing means for optimizing the model including the second objective function and the constraints, The system includes an output means that outputs the values ​​of the variables of the second objective function obtained by the optimization as a parameter set, The objective function generation means receives input from the user for the weights to be modified for the generated second objective function, and updates the second objective function to reflect the received weights. A parameter generation system characterized by the following features.

9. The computer generates a first objective function that includes a weighted linear sum of predictive models in which the material is used as an explanatory variable, characteristic values ​​representing the material's properties are used as the dependent variable, and parameters are set for the explanatory variable. The computer receives input of the constraints for the first objective function, The computer generates a second objective function in which stochastic fluctuations are set for the parameters of the first objective function. The computer optimizes the model including the second objective function and the constraints, The computer outputs the values ​​of the variables of the second objective function obtained by optimization as a parameter set. The computer receives input from the user for the weights to be modified for the generated second objective function, and updates the second objective function to reflect the received weights. A parameter generation method characterized by the following:

10. On the computer, A first objective function generation process that generates a first objective function including a weighted linear sum of a predictive model in which the material is used as an explanatory variable, characteristic values ​​representing the material's properties are used as the dependent variable, and parameters are set for the said explanatory variable. An input process that accepts the constraints of the first objective function, Objective function generation process that generates a second objective function by setting probabilistic fluctuations for the parameters of the first objective function, An optimization process that optimizes the model including the second objective function and the constraints, and The output process is executed to output the values ​​of the variables of the second objective function obtained by the optimization as a parameter set. In the objective function generation process, the system receives input from the user for the weights to be modified for the generated second objective function, and updates the second objective function to reflect the received weights. A parameter generation program for that purpose.

Images