Search device, method for operating a search device, and flow reaction apparatus

Abstract

The present application provides a search device capable of improving search speed, a search device operation method, a search device operation program, and a flow reaction apparatus. A prediction data set generation section generates a prediction data set composed of a plurality of prediction data obtained by associating an explanatory variable whose value of a target variable is unknown with a prediction value of the target variable, using a known data set. A first measured value acquisition section acquires a measured value of the target variable included in the prediction data whose prediction value is closest to a target value. An improvement rate calculation section calculates an improvement rate indicating a difference between the value of the target variable known and the measured value closest to the target value. In a case where the improvement rate is equal to or higher than a target improvement rate, a known data set update section adds the measured value and a value of the explanatory variable corresponding to the measured value to the known data set. In a case where the improvement rate is lower than the target improvement rate, a second measured value acquisition section acquires a measured value of the target variable with respect to a value of the explanatory variable included in the prediction data which is not used for acquisition of the measured value by the first measured value acquisition section.

Description

TECHNICAL FIELD

[0001] The technology of the present application relates to a search device, a method for operating a search device, a program for operating a search device, and a flow reaction apparatus. BACKGROUND

[0002] In a device such as a chemical reaction device and a semiconductor processing device that performs synthesis and processing of a substance and the like, a characteristic value of a product generated varies depending on various processing conditions of the device. In order to obtain a product having an optimal characteristic, it is necessary to search for input parameters from which a target output result can be obtained, with a plurality of processing conditions existing as input parameters and a characteristic value of the product as an output parameter. It is not realistic to evaluate output results for all combinations of a plurality of input parameters, and therefore a method based on machine learning such as a neural network is used to search for an optimal solution of the input parameters (for example, refer to Patent Literature 1).

[0003] In Patent Literature 1, it is disclosed that a prediction model representing a relationship between input parameters and output parameters is generated by machine learning, and an optimal solution is searched for from the prediction model. Specifically, in Patent Literature 1, it is proposed that a value of an input parameter from which a target output result can be obtained is calculated as a predicted value from the prediction model, and it is determined whether or not a measured value corresponding to the predicted value reaches a target value, and in the case where it does not reach, the prediction model is updated by adding the predicted value and the measured value to guide data to improve search efficiency. Further, in Patent Literature 1, it is proposed that in the case where the measured value does not approach the target value, the accuracy of reaching an optimal solution is improved by excluding a region in which a combination of a predicted value and a measured value in which the measured value does not approach the target value exists from a search region.

[0004] Prior Art Documents

[0005] Patent Literature

[0006] Patent Literature 1: Japanese Patent Application Publication No. 2019-40984 SUMMARY

[0007] Technical Problem to be Solved by the Invention

[0008] In Patent Literature 1, as described above, it is proposed that a new region is searched for to calculate an optimal solution in the case where a measured value does not approach a target value. However, in the search method described in Patent Literature 1, the search region is changed only based on a difference between the measured value and the target value, and therefore a region in which the difference between the measured value and the target value is only slightly reduced is continuously searched for, and it can take a long time to obtain an optimal solution.

[0009] An object of the technology of the present application is to provide a search device, a method for operating a search device, a program for operating a search device, and a flow reaction apparatus that can improve search speed.

[0010] Means for solving technical problems

[0011] To achieve the above object, the search device of the present application is provided with: a prediction data set generation section that generates a prediction data set composed of a plurality of prediction data obtained by associating an explanatory variable whose value is unknown with a predicted value of a target variable, using a known data set composed of a plurality of known data obtained by associating one or more explanatory variables with one or more target variables; a first measured value acquisition section that acquires a measured value of the target variable with respect to a value of the explanatory variable included in the prediction data in the prediction data set for which the predicted value is closest to a target value; an improvement rate calculation section that calculates an improvement rate indicating a difference between the value of the known target variable closest to the target value and the measured value, from the known data set; a known data set update section that adds the measured value and the value of the explanatory variable corresponding to the measured value to the known data set as known data, in the case where the improvement rate is equal to or greater than a target improvement rate; and a second measured value acquisition section that acquires a measured value of the target variable with respect to a value of the explanatory variable included in prediction data in the prediction data set that is not used by the first measured value acquisition section for the acquisition of the measured value, in the case where the improvement rate is less than the target improvement rate, the measured value being used for the recalculation of the improvement rate.

[0012] Preferably, a prediction data set determination section is provided that determines the prediction data set generated by the prediction data set generation section, from the known data set updated by the known data set update section immediately after the greatest improvement rate is obtained in the past, in the case where the improvement rate is less than the target improvement rate, the second measured value acquisition section acquires a measured value of the target variable with respect to a value of the explanatory variable included in prediction data in the prediction data set determined by the prediction data set determination section that is not used by the first measured value acquisition section for the acquisition of the measured value, the improvement rate calculation section calculates the improvement rate from the known data set updated by the known data set update section and the measured value acquired by the second measured value acquisition section, and the known data set update section adds the measured value acquired by the second measured value acquisition section and the value of the explanatory variable corresponding to the measured value to the known data set as known data, in the case where the improvement rate becomes equal to or greater than the target improvement rate.

[0013] Preferably, the second measured value acquisition section acquires the measured value from the prediction data including the value of the explanatory variable deviating from the threshold value or more, from the prediction data in the prediction data set determined by the prediction data set determination section that is used by the first measured value acquisition section.

[0014] Preferably, in the case where there are a plurality of prediction data including the value of the explanatory variable deviating from the threshold value or more, the second measured value acquisition section selects the prediction data for which the predicted value is closest to the target value to acquire the measured value.

[0015] Preferably, the target improvement rate changing section changes the target improvement rate when the known data set updating section updates the known data set.

[0016] Preferably, the prediction data set generating section generates a learned model representing a relationship between the explanatory variables and the target variable by machine learning from the known data set, and generates the prediction data set by the known data set, the learned model, and the search algorithm.

[0017] Preferably, the search algorithm is a genetic algorithm.

[0018] The operation method of the search device of the present application includes: a prediction data set generating step of generating a prediction data set composed of a plurality of prediction data obtained by associating explanatory variables whose values are unknown with predicted values of target variables, using a known data set composed of a plurality of known data obtained by associating one or more explanatory variables with one or more target variables; a first measured value acquiring step of acquiring measured values of the target variables with respect to values of the explanatory variables included in the prediction data in the prediction data set whose predicted values are closest to target values; an improvement rate calculating step of calculating an improvement rate representing a difference between the values of the known target variables closest to the target values and the measured values from the known data set; a known data set updating step of adding the measured values and the values of the explanatory variables corresponding to the measured values to the known data set as known data when the improvement rate is equal to or greater than a target improvement rate; and a second measured value acquiring step of acquiring measured values of the target variables with respect to values of the explanatory variables included in the prediction data in the prediction data set which are not used for the acquisition of the measured values in the first measured value acquiring step when the improvement rate is less than the target improvement rate, the measured values being used for the recalculation of the improvement rate.

[0019] The search device of the present application is operated as a computer by a program that functions as: a prediction data set generation section that generates a prediction data set composed of a plurality of prediction data obtained by associating an explanatory variable whose value is unknown with a predicted value of a target variable, using a known data set composed of a plurality of known data obtained by associating one or more explanatory variables with one or more target variables; a first measured value acquisition section that acquires a measured value of the target variable for a value of the explanatory variable included in the prediction data in the prediction data set for which the predicted value is closest to the target value; an improvement rate calculation section that calculates an improvement rate indicating a difference between the value of the known target variable closest to the target value and the measured value, from the known data set; a known data set update section that adds the measured value and the value of the explanatory variable corresponding to the measured value to the known data set as known data, in the case where the improvement rate is equal to or greater than a target improvement rate; and a second measured value acquisition section that acquires a measured value of the target variable for a value of the explanatory variable included in the prediction data in the prediction data set that is not used for the acquisition of the measured value by the first measured value acquisition section, in the case where the improvement rate is less than the target improvement rate, the measured value being used for the recalculation of the improvement rate.

[0020] The flow reaction apparatus of the present application includes the search device described above, a reaction section that performs a reaction while flowing a raw material, a recovery section that recovers a product, a system controller that controls the reaction section to a set reaction condition, and a detection section that detects a reaction result in the reaction section and outputs it. The system controller operates the search device with the reaction condition as an explanatory variable and the reaction result as a target variable, and sets the value of the explanatory variable corresponding to the measured value as the reaction condition in the reaction section in the case where the measured value reaches the target value.

[0021] Effects of the Invention

[0022] According to the technology of the present application, it is possible to provide a search device capable of improving search speed, a method of operating a search device, a program of operating a search device, and a flow reaction apparatus. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 FIG. 1 is a view showing a flow reaction apparatus having a reaction section with a T-shaped confluence portion.

[0024] Figure 2 FIG. 2 is a view showing a flow reaction apparatus having a reaction section with a cross-shaped confluence portion.

[0025] Figure 3 FIG. 3 is a block diagram showing the structure of a computer.

[0026] Figure 4 FIG. 4 is a block diagram showing the functional structure of a search device.

[0027] Figure 5This is a block diagram representing the functional structure of the prediction dataset generation unit.

[0028] Figure 6 This is a diagram illustrating the known dataset in the initial stage.

[0029] Figure 7 It is a diagram representing the structure of a neural network.

[0030] Figure 8 This is a flowchart showing the processing order of the prediction dataset generation section.

[0031] Figure 9 It means according to Figure 6 The graph shows the predicted dataset generated from the known dataset.

[0032] Figure 10 This is a schematic diagram illustrating the process of generating a prediction dataset from a known dataset and performing actual measurements to calculate the improvement rate of the measured values.

[0033] Figure 11 It means update Figure 6 The graph shows the updated known dataset obtained from the known dataset.

[0034] Figure 12 This is a schematic diagram illustrating the process of updating a known dataset to a new known dataset when the improvement rate is above the target improvement rate.

[0035] Figure 13 It means according to Figure 11 The graph shows the predicted dataset generated from the known dataset.

[0036] Figure 14 It means update Figure 11 The graph shows the updated known dataset obtained from the known dataset.

[0037] Figure 15 It means according to Figure 14 The graph shows the predicted dataset generated from the known dataset.

[0038] Figure 16 This is a schematic diagram illustrating the calculation method of the divergence rate.

[0039] Figure 17 This is a schematic diagram illustrating the process of selecting the predicted data used to obtain the measured value based on the deviation rate and the target value.

[0040] Figure 18 It means update Figure 14 The graph shows the updated known dataset obtained from the known dataset.

[0041] Figure 19 It means according to Figure 18A graph showing a prediction data set generated from the known data set shown.

[0042] Figure 20 is a flowchart showing a series of actions of the flow reaction apparatus.

[0043] Figure 21 is a graph explaining a problem caused by the conventional search method.

[0044] Figure 22 is a graph explaining the effect of the search method of the present application.

[0045] Figure 23 is a graph showing a prediction data set generated from the known data set shown. Figure 14

[0046] is a graph showing a prediction data set generated from the known data set shown. Figure 24 Figure 23 is a graph showing a prediction data set generated from the known data set shown.

[0047] Figure 25 is a block diagram showing the functional structure of the search device involved in the modification example.

[0048] Figure 26 is a flowchart showing a series of actions of the flow reaction apparatus involved in the modification example. DETAILED DESCRIPTION

[0049] [EMBODIMENT]

[0050] Hereinafter, a flow reaction apparatus as an embodiment of the present application will be described. The flow reaction apparatus is an apparatus that obtains a product by flow reaction processing.

[0051] Methods of causing a raw material containing a reactant to undergo a chemical reaction include batch processing in which the raw material is caused to undergo a reaction in a state in which the raw material is accommodated in a container, and continuous processing in which the raw material is caused to undergo a reaction while flowing. The continuous processing is a reaction in which the raw material is caused to flow, and is therefore referred to as flow reaction. Since the reaction is continuously caused, the flow reaction processing easily obtains a product having a uniform property. Moreover, the flow reaction processing has an advantage of high productivity as compared with the batch processing.

[0052] In the flow reaction apparatus 10, the flow reaction device 11, the search device 12, the system controller 13, the detection section 14, and the setting section 15 are provided. Figure 1

[0053] ​​In addition to the synthetic reaction of synthesizing a compound as a monomer and the polymerization reaction of generating a polymer by reacting monomers with each other, the flow reaction performed in the flow reaction apparatus 11 can also be a primary reaction such as an anionic polymerization reaction or a termination reaction. Therefore, the reactant that becomes the object of the flow reaction can be, for example, a compound in a growth stage that becomes the object of the termination reaction. In the present embodiment, the flow reaction apparatus 11 performs a termination reaction (anionic polymerization reaction) of terminating the growth of polystyryllithium with methanol by the flow reaction.

[0054] The flow reaction apparatus 11 includes a first raw material supply section 21, a second raw material supply section 22, a reaction section 23, a temperature adjustment section 24, and a recovery section 25. The first raw material supply section 21 and the second raw material supply section 22 are connected to the upstream end portion of the reaction section 23 via pipes not shown. The recovery section 25 is connected to the downstream end portion of the reaction section 23 via a pipe not shown.

[0055] The first raw material supply section 21 supplies the first raw material to the reaction section 23. The first raw material supply section 21 has a pump not shown, and adjusts the flow rate of the first raw material sent from the first raw material supply section 21 to the reaction section 23 by adjusting the rotation speed of the pump.

[0056] In the present embodiment, the first raw material is a solution obtained by dissolving polystyryllithium in a solvent. Polystyryllithium is an example of a reactant that becomes the object of the flow reaction. The solvent uses, for example, tetrahydrofuran (hereinafter referred to as THF), and a small amount of toluene and hexane are mixed. In addition, the first raw material of the flow reaction can be a solution obtained by dissolving a mixture of a reactant and other substances in a solvent, or a solution obtained by dissolving only the reactant in a solvent.

[0057] The second raw material supply section 22 supplies the second raw material to the reaction section 23. Like the first raw material supply section 21, the second raw material supply section 22 has a pump not shown, and adjusts the flow rate of the second raw material sent from the second raw material supply section 22 to the reaction section 23 by adjusting the rotation speed of the pump.

[0058] In the present embodiment, the second raw material is a mixture of methanol and water, that is, a methanol aqueous solution. Methanol is used as a terminator of the anionic polymerization reaction.

[0059] The reaction section 23 is a section for performing a flow reaction (anionic polymerization reaction in the present embodiment). The reaction section 23 has a confluence portion 31 and a reaction portion 32. The confluence portion 31 is composed of a first pipe portion 31a, a second pipe portion 31b, and a third pipe portion 31c connected in a manner that they communicate with each other. The first pipe portion 31a is connected to the second pipe portion 31b in a straight line shape. The third pipe portion 31c is connected to the connection portion of the first pipe portion 31a and the second pipe portion 31b in a manner that is orthogonal to the first pipe portion 31a and the second pipe portion 31b. That is, the confluence portion 31 has a T shape.

[0060] One end of the first pipe portion 31a is connected to the first raw material supply portion 21. One end of the second pipe portion 31b is connected to the second raw material supply portion 22. One end of the third pipe portion 31c is connected to the reaction portion 32. The first raw material is supplied from the first raw material supply portion 21 to the first pipe portion 31a. The second raw material is supplied from the second raw material supply portion 22 to the second pipe portion 31b. The first raw material supplied to the first pipe portion 31a and the second raw material supplied to the second pipe portion 31b are confluenced in the third pipe portion 31c, and are sent to the reaction portion 32 as a mixture of the first raw material and the second raw material.

[0061] The first flow rate sensor 33a that detects the flow rate of the first raw material passing through the first pipe portion 31a is provided on the first pipe portion 31a. Also, the second flow rate sensor 33b that detects the flow rate of the second raw material passing through the second pipe portion 31b is provided on the second pipe portion 31b. Furthermore, the third flow rate sensor 33c that detects the flow rate of the mixture of the first raw material and the second raw material passing through the third pipe portion 31c is provided on the third pipe portion 31c.

[0062] The respective detection signals of the first flow rate sensor 33a, the second flow rate sensor 33b, and the third flow rate sensor 33c are output to the system controller 13.

[0063] The reaction portion 32 is an elongated pipe in which a plurality of straight pipes of the same inner diameter are connected in a straight line shape. The reaction portion 32 can change the length L by changing at least one of the number of connected straight pipes and the length of the straight pipe. Also, the reaction portion 32 can change the inner diameter Φ by changing the inner diameter of the connected straight pipe.

[0064] The inside of the reaction portion 32 is a flow path in which the mixture of the first raw material and the second raw material flows, and is a place where a flow reaction is performed. The flow reaction is performed by causing the mixture to pass through the reaction portion 32, thereby generating a polystyrene solution. Also, although a flow reaction is slightly performed in the third pipe portion 31c of the confluence portion 31, the length of the third pipe portion 31c is extremely short relative to the length L of the reaction portion 32, and thus the reaction in the third pipe portion 31c can be ignored, and the length L of the reaction portion 32 can be regarded as the length of the place where the flow reaction is performed, that is, the reaction passage length. The inner diameter Φ of the reaction portion 32 is the diameter of the place where the flow reaction is performed, that is, the reaction passage diameter.

[0065] The temperature adjusting section 24 includes a heater and a cooler, and adjusts the temperature of the inside of the reaction section 32 (hereinafter, referred to as the reaction temperature). The downstream end portion of the reaction section 32 is provided with a temperature sensor 34 that detects the reaction temperature. The detection signal of the temperature sensor 34 is output to the system controller 13.

[0066] The recovery section 25 is a section that recovers the product of the flow reaction, i.e., polystyrene. The recovery section 25 is configured to include a precipitation section, a sampling section, a drying section, a waste section, and the like (all not shown). The precipitation section, the sampling section, and the drying section are connected in series, and function as a first recovery section that recovers the product. The waste section is connected to the reaction section 32 in parallel with the first recovery section, and functions as a second recovery section that recovers waste, i.e., a waste object. A switching valve not shown is provided on the branch portion of the waste section and the precipitation section. The liquid transported from the reaction section 23 is divided into the product and the waste object by the switching valve, and is guided to the first recovery section or the second recovery section.

[0067] The precipitation section precipitates the product, i.e., polystyrene. For example, a container provided with a stirrer can be used as the precipitation section. Methanol is contained in the container and stirred, and the polystyrene solution guided from the reaction section 23 is introduced into the stirred methanol, whereby the polystyrene can be precipitated.

[0068] The sampling section samples the polystyrene precipitated by the precipitation section from the mixed liquid of methanol and THF or the like. For example, a filter can be used as the sampling section.

[0069] The drying section dries the polystyrene sampled by the sampling section. For example, a constant-temperature tank having a reduced-pressure function can be used as the drying section. By heating the inside of the constant-temperature tank in a state where the pressure is reduced, the dried polystyrene can be obtained as the product.

[0070] The waste section is, for example, a tank. The waste section stores the liquid guided from the reaction section 23 via the switching valve as the waste object.

[0071] The detection section 14 is included in the recovery section 25, and is connected to the first recovery section that recovers the product. The detection section 14 detects the processing result of the flow reaction, i.e., the reaction result, and outputs it to the search device 12 via the system controller 13. As a parameter indicating the reaction result (hereinafter, referred to as the result parameter), the molecular weight dispersity (hereinafter, simply referred to as the dispersity), the purity and the molecular weight of the product, and the like can be given, and in addition to these, the yield and the like can be given.

[0072] Also, in a case where the product is obtained in a solution state dissolved in a solvent in the recovery section 25, the concentration (molar concentration or the like) of the product in the solution can be used as the result parameter. In addition, the detection section 14 can detect, as the result parameter, various properties such as the yield and purity of the byproduct in addition to various properties of the product and the yield. The result parameter can be plural, but in the present embodiment, the result parameter is set to the dispersity.

[0073] The dispersity is a value obtained by dividing the weight average molecular weight by the number average molecular weight. The detection section 14, for example, uses a polystyrene solution obtained by dissolving the product, i.e., polystyrene, in tetrahydrofuran to measure the dispersity by gel permeation chromatography (hereinafter, referred to as GPC).

[0074] The GPC is measured, for example, under the following conditions.

[0075] Apparatus: HLC-8220 GPC (manufactured by TOSOH CORPORATION)

[0076] Detector: differential refractometer (RI (Refractive Index) detector)

[0077] Pre-column: TSK GUARD COLUMN HXL-L 6 mm x 40 mm (manufactured by TOSOH CORPORATION)

[0078] Sample-side column: three (all manufactured by TOSOH CORPORATION) of the following (1) to (3) are directly connected in this order

[0079] (1) TSK-GEL GMHXL 7.8 mm x 300 mm

[0080] (2) TSK-GEL G4000HXL 7.8 mm x 300 mm

[0081] (3) TSK-GEL G2000HXL 7.8 mm x 300 mm

[0082] Reference-side column: TSK-GEL G1000HXL 7.8 mm x 300 mm

[0083] Thermostatic bath temperature: 40°C

[0084] Mobile phase: tetrahydrofuran

[0085] Sample-side mobile phase flow rate: 1.0 ml / min

[0086] Reference-side mobile phase flow rate: 1.0 ml / min

[0087] Sample concentration: 0.1% by mass

[0088] Sample injection volume: 100 μl

[0089] Data sampling time: 5 to 45 minutes after sample injection

[0090] Sampling interval: 300msec

[0091] The setting unit 15 receives settings from the operator of the flow reaction apparatus 11 for reaction conditions related to product production. The reaction conditions received by the setting unit 15 are registered as condition data in the system controller 13. For example, the condition data includes the following items: the concentration (mol / L) and flow rate (ml / min) of the first raw material, the concentration (mol / L) and flow rate (ml / min) of the second raw material, the shape of the confluence section, the diameter Φ of the reaction channel (mm), the length L of the reaction channel (m), and the reaction temperature (°C). Furthermore, the concentrations of the first and second raw materials are pre-adjusted.

[0092] The system controller 13 provides unified control over the operation of the entire flow reaction device 11. The system controller 13 is connected to the first raw material supply unit 21, the second raw material supply unit 22, the temperature control unit 24, the first flow rate sensor 33a, the second flow rate sensor 33b, the third flow rate sensor 33c, the temperature sensor 34, and the aforementioned switching valve included in the recovery section 25.

[0093] The system controller 13 controls the rotational speed of the pump in the first raw material supply unit 21 based on the flow rate of the first raw material detected by the first flow rate sensor 33a, thereby regulating the flow rate of the first raw material. Similarly, the system controller 13 controls the rotational speed of the pump in the second raw material supply unit 22 based on the flow rate of the second raw material detected by the second flow rate sensor 33b, thereby regulating the flow rate of the second raw material. Furthermore, the system controller 13 drives the temperature regulating unit 24 based on the reaction temperature detected by the temperature sensor 34. Additionally, the system controller 13 controls the switching valve included in the recovery section 25.

[0094] The search device 12 performs a condition search process to determine the optimal reaction conditions for the flow reaction treatment carried out by the flow reaction device 11. In this embodiment, the search device 12 searches for the values ​​(optimal solutions) of each item of the reaction conditions in which the dispersion of the result parameter reaches a target value. The search device 12 registers the optimal reaction conditions obtained through the search into the system controller 13.

[0095] It can be used Figure 2 The reaction section 23a shown is used instead of reaction section 23. Additionally, in Figure 2 In the middle, to and Figure 1The same components are denoted by the same reference symbols, and the description will not be repeated.

[0096] In Figure 2 In this embodiment, the reaction section 23a has a confluence portion 41 and a reaction portion 42. The confluence portion 41 is composed of a first pipe portion 41a, a second pipe portion 41b, a third pipe portion 41c, and a fourth pipe portion 41d connected in a manner that they communicate with each other. The first pipe portion 41a and the second pipe portion 41b are connected in a straight line. The third pipe portion 41c and the fourth pipe portion 41d are connected in a straight line. Further, the first pipe portion 41a and the second pipe portion 41b are connected in a manner that they are orthogonal to the third pipe portion 41c and the fourth pipe portion 41d. That is, the confluence portion 41 has a shape of a cross.

[0097] The first raw material supply portion 21 is connected to the first pipe portion 41a and the second pipe portion 41b. The second raw material supply portion 22 is connected to the third pipe portion 41c. The fourth pipe portion 41d is connected to the reaction portion 32. The first raw material is supplied from the first raw material supply portion 21 to the first pipe portion 41a and the second pipe portion 41b. The second raw material is supplied from the second raw material supply portion 22 to the third pipe portion 41c. The first raw material supplied to the first pipe portion 41a and the second pipe portion 41b and the second raw material supplied to the third pipe portion 41c are confluenced in the fourth pipe portion 41d, and are sent to the reaction portion 32 as a mixture of the first raw material and the second raw material.

[0098] The first pipe portion 41a is provided with a first flow rate sensor 43a that detects the flow rate of the first raw material passing through the first pipe portion 41a. The second pipe portion 41b is provided with a second flow rate sensor 43b that detects the flow rate of the first raw material passing through the second pipe portion 41b. The third pipe portion 41c is provided with a third flow rate sensor 43c that detects the flow rate of the second raw material passing through the third pipe portion 41c. Further, the fourth pipe portion 41d is provided with a fourth flow rate sensor 43d that detects the flow rate of the mixture of the first raw material and the second raw material passing through the fourth pipe portion 41d.

[0099] The first flow rate sensor 43a, the second flow rate sensor 43b, the third flow rate sensor 43c, and the fourth flow rate sensor 43d output respective detection signals to the system controller 13.

[0100] In this case, the system controller 13 controls the rotational speed of the pump of the first raw material supply portion 21 based on the average of the flow rate of the first raw material detected by the first flow rate sensor 43a and the flow rate of the first raw material detected by the second flow rate sensor 43b, thereby adjusting the flow rate of the first raw material. Further, the system controller 13 controls the rotational speed of the pump of the second raw material supply portion 22 based on the flow rate of the second raw material detected by the third flow rate sensor 43c, thereby adjusting the flow rate of the second raw material.

[0101] In addition to the above, the setting section 15 accepts the shape of the confluence portion of the reaction section as a setting item, and registers it as an item of the condition data in the system controller 13. For example, in the case of using the reaction section 23 shown in FIG. 7, information indicating that the shape of the confluence portion is "T shape" is registered in the system controller 13 via the setting section 15. Also, in the case of using the reaction section 23 shown in FIG. 8, information indicating that the shape of the confluence portion is "cross shape" is registered in the system controller 13 via the setting section 15. Figure 1 Figure 2

[0102] In the case of using the reaction section 23, the system controller 13 controls the rotation speed of the pump of the first raw material supply section 21 to adjust the flow rate of the first raw material so that the flow rate of the first raw material detected by the first flow rate sensor 33a coincides with the registered flow rate of the first raw material. Similarly, the system controller 13 controls the rotation speed of the pump of the second raw material supply section 22 to adjust the flow rate of the second raw material so that the flow rate of the second raw material detected by the second flow rate sensor 33b coincides with the registered flow rate of the second raw material.

[0103] In the case of using the reaction section 23a, the system controller 13 calculates the average value of the flow rate of the first raw material detected by the first flow rate sensor 43a and the flow rate of the first raw material detected by the second flow rate sensor 43b. The system controller 13 controls the rotation speed of the pump of the first raw material supply section 21 to adjust the flow rate of the first raw material so that the calculated average value coincides with the registered flow rate of the first raw material. Similarly, the system controller 13 controls the rotation speed of the pump of the second raw material supply section 22 to adjust the flow rate of the second raw material so that the flow rate of the second raw material detected by the third flow rate sensor 43c coincides with the registered flow rate of the second raw material.

[0104] Also, the system controller 13 drives the temperature adjustment section 24 so that the reaction temperature detected by the temperature sensor 34 coincides with the registered reaction temperature.

[0105] In the case where the deviation between the detection value of each sensor and each value registered in the system controller 13 exceeds a range set in advance, the system controller 13 controls the switching valve included in the recovery section 25 to guide the waste to the second recovery section.

[0106] The system controller 13 and the search device 12 are each constituted by a computer. In addition, the setting section 15 can be a part of the computer constituting the system controller 13.

[0107] In the case where the deviation between the detection value of each sensor and each value registered in the system controller 13 exceeds a range set in advance, the system controller 13 controls the switching valve included in the recovery section 25 to guide the waste to the second recovery section. Figure 3 ​​The computer constituting the system controller 13 and the search device 12 includes, for example, a storage device 50, a memory 51, a CPU (Central Processing Unit) 52, a communication unit 53, a display 54, and an input device 55. They are connected to each other via a bus 56.

[0108] Storage device 50 is a hard disk drive built into a computer or connected via cable or network. Storage device 50 may also be a disk array containing multiple hard disk drives. Storage device 50 stores control programs such as operating systems, various application programs, and various data associated with these programs. Alternatively, a solid-state drive (SSD) may be used instead of a hard disk drive, or a SSD may be used in addition to a hard disk drive.

[0109] Memory 51 is the working memory for CPU 52 to perform processing. CPU 52 loads the program stored in storage device 50 into memory 51 and performs program-based processing, thereby controlling all parts of the computer in a unified manner.

[0110] The communication unit 53 is a network interface for controlling the transmission of various information via the network connecting the system controller 13 and the search device 12. The display 54 is a display device for displaying various screens. The computer receives operation instructions from the input device 55 through these screens. The input device 55 may be a keyboard, mouse, touch panel, etc.

[0111] Alternatively, the system controller 13 and the search device 12 can be integrated into a single computer.

[0112] Next, the functional structure of the search device 12 will be explained. Figure 4 The search device 12 includes a prediction dataset generation unit 60, a first measured value acquisition unit 61, a first determination unit 62, an improvement rate calculation unit 63, a second determination unit 64, a known dataset update unit 65, a prediction dataset determination unit 66, and a second measured value acquisition unit 67. Furthermore, the search device 12 is composed of a first storage unit 71, a second storage unit 72, and a third storage unit 73. These functional units are implemented based on the operating program of the search device 12 stored in the computer's storage device 50 and the hardware resources on which the operating program runs.

[0113] And, as Figure 5 As shown, the prediction dataset generation unit 60 consists of a learning unit 60a, a learned model storage unit 60b, and a search unit 60c.

[0114] The first storage unit 71 stores a known dataset NDSa, which serves as pre-acquired guidance data. The known dataset NDSa consists of multiple known data obtained by associating the values ​​of each item (condition parameter) of the above-mentioned reaction conditions with the values ​​of the reaction results (result parameters). The prediction dataset generation unit 60 generates a prediction dataset PDS based on the known dataset NDSa, representing the predicted values ​​of the reaction results for unknown reaction conditions that do not exist in the known dataset NDSa.

[0115] Figure 6 The example shown is a known dataset, NDSa. Figure 6 In the example shown, the known dataset NDSa consists of, for instance, 10 sets of known data (No. 1 to 10). Each set of known data is obtained by correlating the values ​​of each item of the reaction conditions with the measured values ​​of the reaction results. The measured values ​​of the reaction results are the dispersion values ​​detected by the detection unit 14. The known dataset NDSa is pre-created and stored in the storage device 50, which serves as the data storage unit, before the condition search processing is performed by the search device 12. The number of known data constituting the known dataset NDSa can be appropriately changed.

[0116] in addition, Figure 6 The “NDS1” shown here is the identification information for identifying each known dataset NDSa and NDSb. The known dataset NDSb is the known dataset updated by the known dataset update unit 65 described later. Regarding this identification information, the known datasets NDSa and NDSb are not distinguished, and are set as NDSn (n=1, 2, 3, ...) according to the order stored in the first storage unit 71.

[0117] The prediction dataset generation unit 60, through the learning unit 60a, performs machine learning based on the learning model M using the known dataset NDSa as guiding data, thereby generating a fully learned model TM. The fully learned model TM is a function representing the relationship between conditional parameters (explanatory variables) and outcome parameters (target variables). In this embodiment, the learning model M is composed of a neural network. However, the learning model M is not limited to a neural network; it can also use linear regression, Gaussian process regression, support vector regression, decision trees, ensemble methods, bagging methods, augmentation methods, or gradient augmentation methods, etc.

[0118] Neural networks can be constructed using commercially available neural network fitting applications. For example, the Matlab Neural Fitting tool from MathWorks can be used to construct neural networks. Alternatively, the keras package from RStudio, which runs on the R language, can also be used as a neural network fitting application.

[0119] like Figure 7As illustrated, the learning model M constituted by the neural network has, for example, a layer structure of an input layer Ll, an intermediate layer L2, and an output layer L3. The input layer Ll includes nodes that input the condition parameters X i (where i = 1, 2,..., 8) as explanatory variables. The output layer L3 includes nodes that represent the result parameter Z as a target variable. Y j (where j = 1, 2,..., 20) are parameters that represent each node of the intermediate layer L2.

[0120] For example, the condition parameters X i and the result parameter Z are represented by the following mathematical expression (1). In the mathematical expression (1), W XiYj , W YjZ are weight coefficients.

[0121] Z = W Y1Z / [1 + exp{- (W X1Y1 X1 + W X2Y1 X2 +... + W X8Y1 X8)}] + W Y2Z / [1 + exp{- (W X1Y2 X1 + W X2Y2 X2 +... + W X8Y2 X8)}] +... + W Y20Z / [1 + exp{- (W X1Y20 X1 + W X2Y20 X2 +... + W X8Y20 X8)}]... (1)

[0122] In addition, the number of nodes included in each layer of the neural network and the number of intermediate layers can be changed as appropriate.

[0123] The learning unit 60a repeatedly updates the weight coefficients W XiYj , W YjZ based on machine learning of the known data set NDSa, thereby obtaining optimal weight coefficients W XiYj , W YjZ . The learning unit 60a generates a learned model TM by applying the obtained optimal weight coefficients W XiYj , W YjZ to the learning model M.

[0124] The learned model storage unit 60b stores the learning model M generated by the learning unit 60a.

[0125] The search unit 60c generates a prediction data set PDS by executing a search algorithm based on the known data set NDSa and the learned model TM stored in the learned model storage unit 60b. The prediction data set PDS is a data set in which the values of the target variable (result parameter Z) are unknown for the explanatory variables (condition parameters Xi It consists of multiple predicted data obtained by establishing a correlation between the predicted value of the target variable and the target variable.

[0126] The search algorithm is, for example, a genetic algorithm. The search unit 60c generates individuals with high evaluation values ​​and combinations of evaluation values ​​(predictive datasets) by repeatedly selecting, crossing over, and mutating individuals with explanatory variable values ​​and evaluation values ​​with target variable values.

[0127] Specifically, the prediction dataset generation unit 60, for example, according to... Figure 8 The calculations are performed in the order shown. First, the learning unit 60a reads the known dataset NDSa from the first storage unit 71 (step S10), and generates a learned model TM by performing machine learning using the known dataset NDSa (step S11). The learned model TM generated by the learning unit 60a is stored in the learned model storage unit 60b (step S12).

[0128] Next, the search unit 60c reads the known dataset NDSa from the first storage unit 71 (step S13). Then, the search unit 60c selects multiple known data points with high evaluation values ​​from the known dataset NDSa based on the value of the target variable (dispersion in this embodiment) used as the evaluation value (step S14). Since a smaller dispersion as the evaluation value is better, the search unit 60c selects known data points from the known dataset NDSa with a dispersion value below a predetermined value.

[0129] Next, the search unit 60c performs a cross-validation process, swapping the values ​​of some explanatory variables between any two sets of known data selected from the plurality of known data (step S15). Next, the search unit 60c triggers a mutation that alters the values ​​of some explanatory variables (step S16). Next, the search unit 60c calculates the dispersion as an evaluation value by inputting the values ​​of the explanatory variables generated through selection, cross-validation, and mutation into the learned model TM stored in the learned model storage unit 60b (step S17).

[0130] Next, the search unit 60c determines whether the termination condition is met based on each evaluation value (step S18). Specifically, the search unit 60c determines the evaluation value that is closest to the target value DT in the known dataset NDSa as the baseline value DK, and determines the termination if a predetermined number or more evaluation values ​​exceeding the baseline value DK are obtained.

[0131] In this embodiment, the evaluation value is dispersion. Dispersion is taken as a positive value, and the smaller the value, the better (higher evaluation). In this embodiment, for example, the target value DT is set to 1.04. Figure 6In the example shown, the dispersion of the known data No. 4 is 1.20, which is closest to the target value DT, and thus the reference value DK is 1.20. In step S18, the determination ends in a case where a dispersion less than the reference value DK is obtained for a prescribed number or more.

[0132] In a case where it is determined in step S18 that the end condition is not satisfied, the search section 60c returns the processing to step S14 again. In step S14, selection is made in accordance with the selection, crossover, and mutation of the evaluation values obtained in step S17.

[0133] In a case where it is determined in step S18 that the end condition is satisfied, the search section 60c outputs the prediction data set PDS including a plurality of data in which the evaluation values (dispersions) exceeding the determined evaluation value are associated with the explanatory variables (reaction conditions) (step S19).

[0134] Figure 9 The prediction data set PDS generated by the prediction data set generation section 60 is illustrated in Figure 6 The prediction data set PDS generated by the prediction data set generation section 60 is illustrated in Figure 9 In the example shown, the prediction data set PDS is composed of eight prediction data (Nos. 1 to 8). Each prediction data is obtained by associating the values of the items of the reaction conditions with the predicted values of the reaction results.

[0135] In addition, the "PDS1" shown in Figure 9 The identification information for identifying each prediction data set PDS is illustrated in

[0136] In Figure 4 In the example shown, the prediction data set PDS is composed of eight prediction data (Nos. 1 to 8). Each prediction data is obtained by associating the values of the items of the reaction conditions with the predicted values of the reaction results.

[0137] Specifically, the first measured value acquisition section 61 registers the values EV of the reaction conditions included in the prediction data closest to the target value DT in the system controller 13, and operates the flow reaction device 11 with the values EV of the reaction conditions, thereby acquiring the measured value DE of the dispersion from the detection section 14. For example, in Figure 9 In the example shown, the dispersion of the prediction data No. 1 is 1.12, which is the closest to the target value DT, and thus the values EV of the reaction conditions included in the prediction data No. 1 are registered in the system controller 13 to operate the flow reaction device 11.

[0138] The first determination unit 62 compares the measured value DE obtained by the first measured value acquisition unit 61 with the target value DT, and provides the measured value DE to the improvement rate calculation unit 63 if the measured value DE is greater than the target value DT. If the measured value DE is less than or equal to the target value DT (the target value has been reached), the first determination unit 62 registers the value EV of the reaction condition obtained from the measured value DE into the system controller 13 for operation of the flow reaction device 11.

[0139] The improvement rate calculation unit 63 uses the following formula (2) to calculate the improvement rate R (unit: percentage) of the reaction result (dispersion) based on the baseline value DK and the measured value DE.

[0140] R=(DK-DE)×100 / DK……(2)

[0141] For example, if the measured value DE obtained by the first measured value acquisition unit 61 is 1.16, the improvement rate R is approximately 3.3%. This means that, through... Figure 8 The measured value DE obtained under the new reaction conditions obtained by the conditional search process shown is about 3.3% closer to the target value DT than the latest measured value closest to the target value DT, i.e., the baseline value DK.

[0142] The third storage unit 73 stores the improvement rate R calculated by the improvement rate calculation unit 63. The improvement rate R calculated by the improvement rate calculation unit 63 is input to the second determination unit 64 via the third storage unit 73.

[0143] Figure 10 The diagram illustrates the process of generating a prediction dataset PDS from a known dataset NDSa and performing real-world tests to calculate the improvement rate R of the measured value DE.

[0144] exist Figure 4 In the process, the second determination unit 64 compares the improvement rate R with the target improvement rate RT and inputs the comparison result into the known dataset update unit 65 and the prediction dataset determination unit 66.

[0145] When the improvement rate R is greater than or equal to the target improvement rate RT, the known dataset update unit 65 updates the known dataset NDSa to a new known dataset NDSb by adding the measured value DE and the value EV of the explanatory variable (response condition) corresponding to the measured value DE to the known dataset NDSa.

[0146] For example, if the target improvement rate RT is set to 3%, the aforementioned improvement rate R is approximately 3.3%, which is above the target improvement rate RT. In this case, such as Figure 11As shown, the known data set updating section 65 adds the measured value DE of the prediction value DS closest to the target value DT, i.e., 1.16, and the value EV of the reaction condition corresponding to the prediction value DS to the known data set NDSa as known data No. 11. Thus, a new known data set NDSb is generated. In this new known data set NDSb, the measured value of the added known data No. 11 becomes the reference value DK closest to the target value DT.

[0147] Figure 12 The process in which the known data set NDSa is updated to the new known data set NDSb when the improvement rate R is the target improvement rate RT or more is schematically shown in FIG. 6.

[0148] In Figure 4 In FIG. 6, the new known data set NDSb generated by the known data set updating section 65 is stored in the first storage section 71. The prediction data set generating section 60 generates a new prediction data set PDS by reading the known data set NDSb updated by the known data set updating section 65 and performing the same process. The new prediction data set PDS generated by the prediction data set generating section 60 is stored in the second storage section 72.

[0149] The known data set NDSa and the known data set NDSb generated by the known data set updating section 65 are stored in the first storage section 71. The prediction data set PDS generated by the prediction data set generating section 60 is stored in the second storage section 72.

[0150] Figure 13 In FIG. 6, the new known data set NDSb generated by the known data set updating section 65 is stored in the first storage section 71. The prediction data set generating section 60 generates a new prediction data set PDS by reading the known data set NDSb updated by the known data set updating section 65 and performing the same process. The new prediction data set PDS generated by the prediction data set generating section 60 is stored in the second storage section 72. Figure 11 The new prediction data set PDS shown in FIG. 6 is generated by the prediction data set generating section 60 from the known data set NDSb. In this new prediction data set PDS, the prediction value DS closest to the target value DT is the prediction value of prediction data No. 4. Therefore, the measured value DE is acquired by the first measured value acquiring section 61 using the value EV of the reaction condition included in prediction data No. 4.

[0151] The measured value DE acquired by the first measured value acquiring section 61 is, for example, 1.09. This measured value DE is greater than the target value DT, and therefore the improvement rate R is calculated by the improvement rate calculating section 63, and the calculated improvement rate R is about 6%. In this case, the improvement rate R is the target improvement rate RT or more, and therefore a new known data set NDSb is newly generated by the known data set updating section 65.

[0152] Figure 14 The new known data set NDSb newly generated by the known data set updating section 65 is exemplified in FIG. 7. In this new known data set NDSb, the measured value of the added known data No. 12 becomes the reference value DK closest to the target value DT.

[0153] FromFigure 11 The new known dataset NDSb shown is regenerated. Figure 14 The process up to the new known dataset NDSb shown is the same as... Figure 10 and Figure 12 The process shown is the same.

[0154] Figure 15 The example shown is based on Figure 14 The known dataset NDSb shown is a new prediction dataset PDS generated by the prediction dataset generation unit 60. In this new prediction dataset PDS, the prediction value DS that is closest to the target value DT is the prediction value of prediction data No.4. Therefore, the first measured value acquisition unit 61 uses the value EV of the reaction condition included in prediction data No.4 to acquire the measured value DE.

[0155] In this case, the measured value DE acquired by the first measured value acquisition unit 61 is, for example, 1.06. Since this measured value DE is greater than the target value DT, the improvement rate calculation unit 63 calculates the improvement rate R. The improvement rate R calculated by the improvement rate calculation unit 63 is approximately 2.8%. Thus, when the calculated improvement rate R is less than the target improvement rate RT, the prediction dataset determination unit 66 is run.

[0156] exist Figure 4 In the process, the prediction dataset determination unit 66 refers to the past improvement rate R data stored in the third storage unit 73, and determines the prediction dataset PDS generated by the prediction dataset generation unit 60 based on the known dataset NDSb generated by the known dataset update unit 65 immediately after the past maximum improvement rate R.

[0157] For example, in Figure 9 , Figure 11 , Figures 13-15 In the example shown, the maximum improvement rate R is approximately 6%, and the next known dataset updated after that, NDSb, is... Figure 14 The known dataset NDSb is shown. And, according to... Figure 14 The known dataset NDSb shown above generates the prediction dataset PDS. Figure 15 The prediction dataset PDS is shown. Therefore, the prediction dataset PDS determined by the prediction dataset determination unit 66 is... Figure 15 The prediction dataset PDS is shown.

[0158] The prediction dataset determination unit 66 provides the identification information of the determined prediction dataset PDS to the second measured value acquisition unit 67. For example, in the above example, the prediction dataset determination unit 66 provides the identification information of the determined prediction dataset PDS to the second measured value acquisition unit 67. Figure 15 The identification information (PDS3) of the prediction dataset PDS is shown.

[0159] The second measured value acquisition unit 67 extracts the predicted dataset PDS corresponding to the identification information provided by the predicted dataset determination unit 66 from the multiple predicted datasets PDS stored in the second storage unit 72. Then, the second measured value acquisition unit 67 acquires the measured value DE of the target variable (dispersion) for the values ​​of the explanatory variables (reaction conditions) included in the predicted data in the extracted predicted dataset PDS that were not used by the first measured value acquisition unit 61 to acquire the measured value DE.

[0160] For example, in determining the prediction dataset, section 66 determined... Figure 15 In the case of the prediction dataset PDS shown, prediction data No.4 has been used by the first measured value acquisition unit 61 to acquire the measured value DE. Therefore, the measured value DE is acquired based on one of the prediction data No.1 to No.3.

[0161] Specifically, the second measured value acquisition unit 67 acquires the measured value DE based on the predicted data, which includes the values ​​of explanatory variables included in the predicted data used by the first measured value acquisition unit 61 that deviate from the threshold by more than a certain value. The second measured value acquisition unit 67 calculates the deviation rate δ for each of the explanatory variables (reaction conditions). When the values ​​of the explanatory variables included in the predicted data used by the first measured value acquisition unit 61 to acquire the measured value DE are set as Xa, and the values ​​of the explanatory variables included in the predicted data not used by the first measured value acquisition unit 61 to acquire the measured value DE are set as Xb, the deviation rate δ is calculated using the following formula (3).

[0162] δ=|Xa-Xb|×100 / Xa……(3)

[0163] For example, in Figure 15 In the case of the prediction dataset PDS shown, such as Figure 16 As shown, the second measured value acquisition unit 67 calculates the deviation rate δ of each item of the predicted data No. 1 to 3 relative to each item of the predicted data No. 4. The deviation rates δ of the predicted data No. 1 to 3 related to the reaction channel diameter are 3.3%, 10%, and 16.7%, respectively. With the threshold set to 5%, the deviation rate δ of the reaction channel diameter of the predicted data No. 2 and 3 is above the threshold.

[0164] Furthermore, when there are multiple predicted data points containing values ​​of explanatory variables (response conditions) that deviate from the threshold, the second measured value acquisition unit 67 selects the predicted data whose predicted value is closest to the target value DT to acquire the measured value DE. For example, in Figure 15 In the case of the prediction dataset PDS shown, among the prediction data No.2 and No.3 where the deviation rate δ is above the threshold, the predicted value of prediction data No.2 is close to the target value DT. Therefore, the second measured value acquisition unit 67 selects prediction data No.2 to acquire the measured value DE.

[0165] Figure 17 The diagram schematically illustrates the process of selecting predictive data for obtaining measured values ​​based on the deviation rate δ and the target value DT. The second measured value acquisition unit 67 uses the value EV of the reaction condition included in the predictive data selected based on the deviation rate δ and the target value DT to obtain the measured value DE.

[0166] The acquisition process for other measured values ​​DE by the second measured value acquisition unit 67 is the same as that by the first measured value acquisition unit 61. The measured values ​​DE acquired by the second measured value acquisition unit 67 are input to the first determination unit 62, and the same determination process is performed. If the measured value DE is greater than the target value DT, the improvement rate R is calculated by the improvement rate calculation unit 63 based on the measured value DE. In addition, the reference value DK used for calculating the improvement rate R at this time is based on the known dataset NDSb updated by the known dataset update unit 65 immediately after the largest improvement rate R obtained in the past.

[0167] For example, if the measured value DE obtained by the second measured value acquisition unit 67 is 1.05, and the baseline value DK is 1.09, then the improvement rate R calculated by the improvement rate calculation unit 63 is approximately 3.7%. In this case, the improvement rate R is above the target improvement rate RT, therefore, the known dataset update unit 65 regenerates a new known dataset NDSb. Specifically, as... Figure 18 As shown, by using the measured value DE obtained by the second measured value acquisition unit 67, i.e., 1.05 and Figure 15 The predicted data No.2 shows the value of the reaction condition EV attached to... Figure 14 Use the known dataset NDSb shown to generate a new known dataset NDSb.

[0168] Figure 19 The example shown is based on Figure 18 The known dataset NDSb shown is a new prediction dataset PDS generated by the prediction dataset generation unit 60. In this new prediction dataset PDS, the prediction value DS that is closest to the target value DT is the prediction value of prediction data No.3. The first measured value acquisition unit 61 uses the value EV of the reaction condition included in prediction data No.3 to acquire the measured value DE.

[0169] The measured value DE acquired by the first measured value acquisition unit 61 is, for example, 1.04. In this case, the measured value DE reaches the target value DT, and therefore the reaction conditions for acquiring the measured value DE are set in the flow reaction apparatus 11 by the system controller 13 for the operation of the flow reaction apparatus 11.

[0170] Next, refer to Figure 20The flowchart shown explains a series of actions of the flow reaction apparatus 10. First, the search device 12 acquires a known data set NDSa obtained by the system controller 13 operating the flow reaction apparatus 11 according to the reaction conditions registered in the system controller 13 (step S20).

[0171] Next, in the search device 12, a prediction data set PDS whose reaction result (dispersion) is closer to the target value DT than the known data set NDSa is generated from the known data set NDSa by the prediction data set generation section 60 (step S21). The generation of this prediction data set PDS is performed in the order shown. Figure 8

[0172] Next, the measured value DE of the reaction result (dispersion) of the flow reaction apparatus 11 using the reaction conditions included in the prediction data whose predicted value is closest to the target value DT in the prediction data set PDS is acquired by the 1st measured value acquisition section 61 (step S22). Next, a determination is made by the 1st determination section 62 as to whether the acquired measured value DE reaches the target value DT (step S23).

[0173] In the case where the measured value DE is greater than the target value DT and does not reach the target value DT (step S23: "No"), the improvement rate R is calculated from the above-described equation (2) by the improvement rate calculation section 63 (step S24). Next, a determination is made by the 2nd determination section 64 as to whether the improvement rate R is the target improvement rate RT or more (step S25).

[0174] In the case where the improvement rate R is the target improvement rate RT or more (step S25: "Yes"), the known data set NDSa is updated to a new known data set NDSb by the known data set update section 65 (step S26). The generation of this new known data set NDSb is performed by adding the measured value DE and the value EV of the reaction conditions corresponding to the measured value DE to the known data set NDSa. Then, the process returns to step S21, and the prediction data set PDS is generated using the updated new known data set NDSb. Then, the same processing is performed.

[0175] On the other hand, in the case where the improvement rate R is less than the target improvement rate RT (step S25: "No"), the prediction data set PDS is determined based on the past improvement rate R by the prediction data set determination section 66 (step S27). At this time, the prediction data set determination section 66 determines the prediction data set PDS generated by the prediction data set generation section 60 from the known data set NDSb updated by the known data set update section 65 immediately after the past improvement rate R was obtained.

[0176] ​Next, the second measured value acquisition unit 67 acquires the measured value DE based on the determined prediction dataset PDS (step S28). At this time, the second measured value acquisition unit 67 acquires the measured value DE based on prediction data that includes values ​​of reaction conditions whose values ​​EV, included in the prediction data already used by the first measured value acquisition unit 61, deviate from a threshold or higher. Furthermore, if multiple prediction data sets exist that include values ​​of reaction conditions deviating from the threshold or higher, the second measured value acquisition unit 67 selects the prediction data whose predicted value is closest to the target value DT to acquire the measured value DE.

[0177] If a measured value DE is obtained by the second measured value acquisition unit 67, the process returns to step S23, where the first determination unit 62 makes a new determination. Then, the same process is repeated. Then, if in step S23 it is determined that the measured value DE obtained by the first measured value acquisition unit 61 or the second measured value acquisition unit 67 is below the target value DT and the target value DT has been reached (step S23: "Yes"), the reaction conditions for obtaining the measured value DE are set in the flow reaction apparatus 11 via the system controller 13. The flow reaction apparatus 11 operates according to the set reaction conditions.

[0178] In addition, steps S21, S22, S24, S26 and S28 are examples of the prediction dataset generation step, the first measured value acquisition step, the improvement rate calculation step, the known dataset update step and the second measured value acquisition step.

[0179] As described above, when the improvement rate R since the previous response result is less than the target improvement rate RT, the search device 12 acquires, via the second measured value acquisition unit 67, the measured value DE of the target variable (response result), which includes the values ​​of the explanatory variables (response conditions) in the prediction data included in the prediction dataset PDS that were not used by the first measured value acquisition unit 61 to acquire the measured value DE. Then, the search device 12 calculates the improvement rate R again based on the acquired measured value DE. The improvement rate R does not simply represent the difference between the measured value DE and the target value DT, but rather the rate at which the difference between the measured value DE and the target value DT decreases (i.e., the search speed).

[0180] Typically, the relationship between explanatory variables and the target variable is non-linear. Therefore, within the space of explanatory variables, besides the optimal solution, there exist multiple local solutions. Thus, as... Figure 21 As shown, in conventional search methods described in Patent Document 1, the search area is changed based on the difference between the measured value DE and the target value DT. Therefore, when the search reaches a local solution other than the optimal solution for the target value DT, the difference between the measured value DE and the target value DT only slightly decreases, resulting in a decrease in search speed and potentially taking a long time to obtain the optimal solution.

[0181] In contrast, in the search device 12 of this embodiment, as described above, when the improvement rate R is small, the measured value DE is obtained using another predicted data from the predicted dataset PDS that was not used to obtain the measured value DE, and the improvement rate R is obtained again. This is equivalent to the following processing: Figure 22 As shown, when the improvement rate R is less than the target improvement rate RT (when the search speed is low), the search region is significantly changed from the vicinity of the local solution. Therefore, by changing the search region to the area where the optimal solution exists, the improvement rate R is increased (the search speed is increased), which may shorten the time until the optimal solution is obtained.

[0182] Furthermore, when the improvement rate R is small, it is possible to obtain a larger improvement rate R by using another predicted data from the prediction dataset PDS generated immediately after the largest improvement rate R in the past that was not used to obtain the measured value DE.

[0183] Furthermore, when selecting another prediction data, by obtaining the measured value DE based on prediction data including values ​​of explanatory variables whose values ​​deviate from the threshold or higher from the prediction data already used by the first measured value acquisition unit 61, the search direction changes towards the optimal solution, potentially resulting in a larger improvement rate R. Moreover, when multiple prediction data exist that include values ​​of explanatory variables deviating from the threshold or higher, by selecting the prediction data whose predicted value is closest to the target value DT to obtain the measured value DE, the search direction changes towards the optimal solution, potentially resulting in an even greater improvement rate R.

[0184] [Comparative Example]

[0185] Next, as a comparative example, the following situation is illustrated: when the improvement rate R is small, the known dataset NDSb is updated based on the prediction dataset PDS with the smaller improvement rate R, instead of determining the prediction dataset PDS based on the past improvement rate R as in the above implementation.

[0186] Figure 23 The example shown is based on Figure 14 The known datasets shown are NDSb and Figure 15 If the improvement rate R obtained from the predicted dataset PDS is less than the target improvement rate RT, the predicted dataset determination unit 66 is not run, and the known dataset NDSb is updated by the known dataset update unit 65. Figure 24 The example shown is based on Figure 23 The known dataset NDSb shown is a new predicted dataset PDS generated by the predicted dataset generation unit 60.

[0187] In this case, the measured value DE acquired by the 1st measured value acquisition section 61 is, for example, 1.06, and the improvement rate R calculated by the improvement rate calculation section 63 is 0%. Thus, if the update of the known data set is performed in the same manner as when the improvement rate R is large in the case where the improvement rate R is small, the improvement rate R will not increase due to the continued search in the vicinity of the local solution other than the optimal solution, and it can take a long time to acquire the optimal solution.

[0188] [Modified Example]

[0189] Next, a modified example of the above-described embodiment will be shown. In the above-described embodiment, the search device 12 has the structure of the search device 12a of the modified example. Figure 25 In the modified example, the search device 12a has the structure of the search device 12 of the above-described embodiment, and further has a target improvement rate changing section 68. In the case where the known data set is updated by the known data set update section 65, the target improvement rate changing section 68 changes the target improvement rate RT used for the determination by the 2nd determination section 64.

[0190] Specifically, it is considered that the improvement rate R decreases as the measured value DE approaches the target value DT, and thus in the case where the known data set is updated by the known data set update section 65, the target improvement rate changing section 68 changes the target improvement rate RT in the direction of decreasing the target improvement rate RT. Thus, by changing the target improvement rate RT, in the case where the improvement rate R decreases as the measured value DE approaches the target value DT, it is possible to suppress the search region from being changed unintentionally.

[0191] Figure 26 is a flowchart showing a series of actions of the flow reaction apparatus according to the modified example. As shown in Figure 26 In the present action, after the known data set is updated (step S26), the target improvement rate RT is changed by the target improvement rate changing section 68 (step S30). As for the other actions of the flow reaction apparatus according to the modified example, the actions of the flow reaction apparatus 10 according to the above-described embodiment are the same. Further, the change of the target improvement rate RT by the target improvement rate changing section 68 can be performed during the period from when the determination is made "Yes" in step S25 to when the determination is made again in step S25.

[0192] <Other Modified Examples>

[0193] Further, in the above-described embodiment, the dispersion degree is used as the value of the target variable. The value of the measured value DE of the dispersion degree is closer to the target value DT as it is smaller, and thus by defining the improvement rate R by the above-described equation (2), the improvement rate R is set to a positive value. In contrast to this, in the case where a parameter whose value is closer to the target value as it is larger, like the yield in the flow synthesis, is used as the target variable, the improvement rate R can be defined by the following equation (2a).

[0194] R = (DE - DK) x 100 / DK (2a)

[0195] Also, in a case where a parameter that converges to a specific value like a molecular weight is used as a target variable, the improvement rate R can be defined by the following equation (2b).

[0196] R = |DK - DE| x 100 / DK (2b)

[0197] Also, in the above embodiment and modification example, one target variable is used, but the target variable can be two or more.

[0198] Also, in the above embodiment and modification example, the search device is applied to a flow reaction apparatus, but is not limited to a flow reaction apparatus. The present application can be applied to any apparatus as long as it is an apparatus that has a known data set that can be obtained by correlating one or more explanatory variables with one or more target variables.

[0199] In the above embodiment and modification example, for example, as a hardware structure of a processing unit (Processing Unit) that performs various processes such as the predicted data set generation unit 60, the first measured value acquisition unit 61, the first determination unit 62, the improvement rate calculation unit 63, the second determination unit 64, the known data set update unit 65, the predicted data set determination unit 66, the second measured value acquisition unit 67, and the target improvement rate change unit 68, various processors (Processor) shown below can be used. As described above, the various processors include the CPU 52 as a general-purpose processor that functions as the various processing units by executing software (running program), and in addition thereto, include a PLD (Programmable Logic Device) such as an FPGA (Field Programmable Gate Array) as a processor whose circuit structure can be changed after manufacture, an ASIC (Application Specific Integrated Circuit) as a dedicated circuit having a circuit structure that is specially designed in order to execute a specific process, and the like.

[0200] One processing unit can be constituted by one of these various processors, or can be constituted by a combination of two or more processors of the same kind or different kinds (for example, a combination of a plurality of FPGAs and / or a combination of a CPU and an FPGA). Also, a plurality of processing units can be constituted by one processor.

[0201] As an example of a plurality of processing sections being constituted by one processor, first, there is a method in which, as typified by computers such as clients and servers, one processor is constituted by a combination of one or more CPUs and software, and the functions of a plurality of processing sections are exerted by the processor. Next, there is a method in which, as typified by a system on chip (SoC) or the like, a processor that realizes the functions of the entire system including a plurality of processing sections is used by one IC (integrated circuit) chip. In this way, various processing sections are constituted by one or more of the above various processors as hardware structures.

[0202] Further, as the hardware structures of these various processors, more specifically, a circuit (circuitry) constituted by combining circuit elements such as semiconductor elements can be used.

[0203] According to the above description, the inventions described in the following supplementary notes 1, 2, and 3 can be grasped.

[0204] [Supplementary note 1]

[0205] A search device including:

[0206] a prediction data set generation processor that generates a prediction data set composed of a plurality of prediction data obtained by associating an explanatory variable whose value is unknown with a predicted value of a target variable, using a known data set composed of a plurality of known data obtained by associating one or more explanatory variables with one or more target variables;

[0207] a first measured value acquisition processor that acquires a measured value of the target variable for a value of the explanatory variable included in the prediction data in the prediction data set in which the predicted value is closest to a target value;

[0208] an improvement rate calculation processor that determines a value of the target variable that is closest to the target value from the known data set as a reference value, and calculates an improvement rate that represents a difference between the reference value and the measured value;

[0209] a known data set update processor that, in a case where the improvement rate is equal to or higher than a target improvement rate, adds the measured value and the value of the explanatory variable corresponding to the measured value to the known data set as known data; and

[0210] a second measured value acquisition processor that, in a case where the improvement rate is lower than the target improvement rate, acquires a measured value of the target variable for a value of the explanatory variable included in the prediction data in the prediction data set in which the prediction data is not used for the acquisition of the measured value by the first measured value acquisition processor, the measured value being used to calculate the improvement rate again.

[0211] [Para 2]

[0212] The search device according to Para 1, wherein the prediction dataset determination processor determines the prediction dataset generated by the prediction dataset generation processor based on the known dataset updated by the known dataset update processor immediately after the improvement rate obtained the most in the past in a case where the improvement rate is less than the target improvement rate,

[0213] The 2nd measured value acquisition processor acquires a measured value of the target variable for a value of the explanatory variable included in the prediction data in the prediction dataset determined by the prediction dataset determination processor that is not used for acquisition of the measured value by the 1st measured value acquisition processor,

[0214] The improvement rate calculation processor calculates the improvement rate based on the known dataset updated by the known dataset update processor and the measured value acquired by the 2nd measured value acquisition processor,

[0215] In a case where the improvement rate is the target improvement rate or more, the known dataset update processor adds the measured value acquired by the 2nd measured value acquisition processor and the value of the explanatory variable corresponding to the measured value to the known dataset as known data.

[0216] [Para 3]

[0217] The search device according to Para 1 or 2, wherein the target improvement rate change processor changes the target improvement rate in a case where the known dataset update processor updates the known dataset.

[0218] The technology of the present application can also be appropriately combined with the various embodiments and various modifications described above. Also, of course, it is not limited to the above-described embodiments, and various structures can be employed within a range not deviating from the gist. Furthermore, the technology of the present application relates to a program, and in addition thereto, a storage medium that does not temporarily store the program.

[0219] The above-described contents and drawings are detailed descriptions of the technical aspects of the present application, and are only examples of the technical aspects of the present application. For example, the descriptions of the above-described structures, functions, actions, and effects are descriptions of examples of the structures, functions, actions, and effects of the technical aspects of the present application. Therefore, it is of course possible to delete unnecessary parts, add new elements, or make substitutions in the above-described contents and drawings without departing from the scope of the technical aspects of the present application. Also, in order to avoid trouble and facilitate understanding of the technical aspects of the present application, descriptions of technical common knowledge and the like that are not particularly necessary in implementing the technical aspects of the present application are omitted in the above-described contents and drawings.

[0220] In the present specification, the meaning of "A and / or B" is the same as that of "at least one of A and B". That is, "A and / or B" means that only A can be included, only B can be included, or both A and B can be included.

[0221] All of the literature, patent applications, and technical standards cited in the present specification are hereby expressly incorporated by reference in their entirety to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

[0222] Symbol Explanation

[0223] 10 - flow reaction apparatus, 11 - flow reaction device, 12, 12a - search device, 13 - system controller, 14 - detection section, 15 - setting section, 21 - first raw material supply section, 22 - second raw material supply section, 23, 23a - reaction section, 24 - temperature adjustment section, 25 - recovery section, 31, 41 - merging section, 31a, 41a - first pipe section, 31b, 41b - second pipe section, 31c, 41c - third pipe section, 32 - reaction section, 33a, 43a - first flow rate sensor, 33b, 43b - second flow rate sensor, 33c, 43c - third flow rate sensor, 34 - temperature sensor, 41d - fourth pipe section, 42 - reaction section, 43d - fourth flow rate sensor, 50 - storage device, 51 - memory, 53 - communication section, 54 - display, 55 - input device, 56 - bus, 60 - prediction data set generation section, 60a - learning section, 60b - learned model storage section, 60c - search section, 61 - first measured value acquisition section, 62 - first determination section, 63 - improvement rate calculation section, 64 - second determination section, 65 - known data set update section, 66 - prediction data set determination section, 67 - second measured value acquisition section, 68 - target improvement rate change section, 71 - first storage section, 72 - second storage section, 73 - third storage section, Φ - reaction channel diameter, δ - divergence rate, EV - value of reaction condition, DE - measured value, DK - reference value, DS - predicted value, DT - target value, L1 - input layer, L2 - intermediate layer, L3 - output layer, L - reaction channel length, M - learning model, NDSa, NDSb - known data set, PDS - prediction data set, R - improvement rate, RT - target improvement rate.

Claims

1. A search apparatus that performs a condition search process for determining a reaction condition related to production of a product at which a process result of a flow reaction, i.e., a reaction result, of a flow reaction process performed by a flow reaction apparatus reaches a target value, the search apparatus comprising: a prediction data set generation section that generates a prediction data set composed of a plurality of prediction data obtained by associating an explanatory variable whose value is unknown with a predicted value of a target variable, using a known data set composed of a plurality of known data obtained by associating one or more explanatory variables that are the reaction condition with one or more target variables that are the reaction result; a first measured value acquisition section that acquires a first measured value of the target variable with respect to a value of the explanatory variable included in the prediction data in the prediction data set in which the predicted value is closest to a target value; an improvement rate calculation section that calculates an improvement rate indicating a difference between a value of the target variable that is closest to the target value and the first measured value, from the known data set; a known data set update section that adds the first measured value and a value of the explanatory variable corresponding to the first measured value to the known data set as known data, in a case where the improvement rate is a target improvement rate or more; a second measured value acquisition section that acquires a second measured value of the target variable with respect to a value of the explanatory variable included in the prediction data in the prediction data set in which the first measured value acquisition section does not use the prediction data for acquisition of the first measured value, in a case where the improvement rate is less than the target improvement rate, the second measured value being used for recalculation of the improvement rate; and a prediction data set determination section that determines the prediction data set generated by the prediction data set generation section, in a case where the improvement rate is less than the target improvement rate, from the known data set that is updated by the known data set update section immediately after the improvement rate that is the largest in the past, wherein the second measured value acquisition section acquires the second measured value of the target variable with respect to a value of the explanatory variable included in the prediction data in the prediction data set determined by the prediction data set determination section, which is not used by the first measured value acquisition section for acquisition of the first measured value, the improvement rate calculation section calculates the improvement rate from the known data set updated by the known data set update section and the second measured value acquired by the second measured value acquisition section, in a case where the improvement rate becomes the target improvement rate or more, the known data set update section adds the second measured value acquired by the second measured value acquisition section and a value of the explanatory variable corresponding to the second measured value to the known data set as known data.

2. The search apparatus according to claim 1, wherein the second measured value acquisition section acquires the second measured value from the prediction data including a value of the explanatory variable deviating from a threshold value or more, from a value of the explanatory variable included in the prediction data determined by the prediction data set determination section, which has been used by the first measured value acquisition section. ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ 3. The search device according to claim 2, wherein In a case where a plurality of the prediction data including the value of the explanatory variable deviating from the threshold value or more exist, the 2nd measured value acquisition section acquires the 2nd measured value of the prediction data whose prediction value is closest to the target value.

4. The search device according to any one of claims 1 to 3, wherein The search device includes a target improvement rate changing section that changes the target improvement rate in a case where the known data set updating section updates the known data set.

5. The search device according to any one of claims 1 to 3, wherein The prediction data set generating section generates a learned model representing a relationship between the explanatory variable and the target variable by machine learning from the known data set, and generates the prediction data set by the known data set, the learned model, and a search algorithm.

6. The search device according to claim 5, wherein The search algorithm is a genetic algorithm.

7. A method of operating a search device that performs a condition search process for determining a reaction condition related to production of a product at which a process result of a flow reaction, i.e., a reaction result, of a flow reaction process performed by a flow reaction device reaches a target value, the method comprising: a prediction data set generating step of generating a prediction data set composed of a plurality of prediction data in which a value of an explanatory variable is associated with a prediction value of a target variable, using a known data set composed of a plurality of known data in which one or more explanatory variables as the reaction condition are associated with one or more target variables as the reaction result; a 1st measured value acquiring step of acquiring a 1st measured value of the target variable with respect to a value of the explanatory variable included in the prediction data of the prediction data set in which the prediction value is closest to the target value; an improvement rate calculating step of calculating an improvement rate representing a difference between a value of the target variable closest to the target value and the 1st measured value from the known data set; a known data set updating step of adding the 1st measured value and the value of the explanatory variable corresponding to the 1st measured value to the known data set as known data in a case where the improvement rate is a target improvement rate or more; a 2nd measured value acquiring step of acquiring a 2nd measured value of the target variable with respect to a value of the explanatory variable included in the prediction data of the prediction data set that is not used for acquisition of the 1st measured value in the 1st measured value acquiring step in a case where the improvement rate is less than the target improvement rate, the 2nd measured value being used to calculate the improvement rate again; and a prediction data set determining step of determining the prediction data set generated in the prediction data set generating step from the known data set that is updated immediately after the improvement rate that is the largest in the past in a case where the improvement rate is less than the target improvement rate. ​ ​ ​ ​ ​ ​ In the second measured value acquisition step, the second measured value of the target variable is acquired for the value of the explanatory variable included in the prediction data other than the prediction data used for the acquisition of the first measured value in the first measured value acquisition step in the prediction data set determination step, In the improvement rate calculation step, the improvement rate is calculated from the known data set updated in the known data set update step and the second measured value acquired in the second measured value acquisition step, In the case where the improvement rate is the target improvement rate or more, in the known data set update step, the second measured value acquired in the second measured value acquisition step and the value of the explanatory variable corresponding to the second measured value are added to the known data set as known data.

8. A flow reaction apparatus comprising: the search device according to any one of claims 1 to 6; a reaction section in which reaction is performed while at least two raw materials are flowing; a recovery section connected to a downstream side end portion of the reaction section, which recovers a product; a system controller which controls the reaction section to a set reaction condition; and a detection section which detects a reaction result in the reaction section and outputs it, the system controller operates the search device with the reaction condition as the explanatory variable and the reaction result as the target variable, and in the case where the first measured value or the second measured value reaches the target value, sets the value of the explanatory variable corresponding to the measured value as the reaction condition in the reaction section.

Images