Fermentative production process support method

Abstract

To provide a technique that can derive suitable culture conditions at each culture time point in a fermentative production process.SOLUTION: In a fermentative production process support method, at least one processor executes: a step for repeating a prediction cycle that includes a first prediction stage for acquiring a plurality of types of culture state indicators by inputting culture information at a certain culture time point into a trained model, and a second prediction stage for acquiring the culture information at the next culture time point; a step for acquiring control policy trend information, which is a collection of indicators to be optimized at every culture time point and in which at least one of the plurality of types of culture state indicators is designated as an indicator to be optimized at each culture time point; and a step for determining the optimal values of specific culture conditions at each culture time point so as to optimize the culture state indicator that is designated as the indicator to be optimized at each culture time point in the control policy trend information from among the plurality of types of culture state indicators acquired in the first prediction stage.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a technology for supporting a fermentation production process in which a target product is produced in a culture vessel in which microorganisms are cultured. [Background technology]

[0002] The following Patent Document 1 discloses a method for exploring a cell culture process that enables optimal culture conditions to be explored without repeating experiments in cell culture and bioproduction. This method includes a process condition generating step of generating a plurality of process conditions, a culture result predicting step of acquiring predicted cell culture results for the plurality of process conditions, and an optimized process condition acquiring step of finding optimal process conditions from the predicted culture results. The culture result predicting step includes a process condition acquiring step, an uptake constraint condition acquiring step, an optimization calculation step, and a concentration change calculation step. The optimization calculation step calculates a metabolic flow rate (consumption rate) using a mathematical model (metabolic circuit model) related to cell metabolism based on the medium composition (medium component concentration) and the uptake constraint conditions. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2021 / 166824 Summary of the Invention [Problem to be solved by the invention]

[0004] In the above-described method, a plurality of process conditions are generated, a culture prediction result is obtained for each process condition, and an optimal process condition is found from each culture prediction result. However, this method cannot obtain suitable process conditions throughout the entire fermentation production process, because the medium environment changes over time and the microorganisms change their metabolic balance successively in response to the external environment, and the above-mentioned method seems to ignore the time element of the process conditions. Even if the above-mentioned method were to be performed at each culture time point, the amount of calculation required would be enormous, making it an unrealistic method to implement.

[0005] The present invention has been made in consideration of these points, and provides a technique that makes it possible to obtain appropriate culture conditions for each culture point in a fermentation production process. [Means for solving the problem]

[0006] According to the present invention, there is provided a method for supporting a fermentation production process in which a target substance is produced in a culture vessel for culturing microorganisms, the method comprising: a culture behavior prediction step in which one or more processors capable of using a trained model that inputs culture information including at least specific culture conditions and outputs a plurality of culture state indexes, repeating a prediction cycle for each culture time point within a predetermined culture time, the prediction cycle including a first prediction step of acquiring the plurality of culture state indexes by inputting the culture information at a certain culture time point into the trained model and a second prediction step of acquiring the culture information at the next culture time point; a trend acquisition step of acquiring control policy trend information which is a set of optimization target indexes for each culture time point, in which at least one of the plurality of culture state indexes is specified as an optimization target index for each culture time point; and an optimal condition determination step of determining optimal values ​​of the specific culture conditions at each culture time point so that the culture state index specified as the optimization target index for each culture time point in the acquired control policy trend information among the plurality of culture state indexes acquired in the first prediction step is optimized.

[0007] Furthermore, according to the present invention, it is possible to provide a fermentation production process assistance apparatus that includes at least a memory and the one or more processors described above and is capable of executing the fermentation production process assistance method described above. Furthermore, it is possible to provide a program for causing a computer to execute the above-mentioned fermentation production process support method, and it is also possible to provide a computer-readable recording medium having such a program recorded thereon. This recording medium includes a non-transitory tangible medium. Effect of the Invention

[0008] According to the present invention, a technique for acquiring appropriate culture conditions for each culture point in a fermentation production process can be provided. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a schematic diagram showing an example of a culture system. [Diagram 2] 1 is a diagram conceptually illustrating an example of the hardware configuration of an information processing device capable of executing a fermentation production process support method according to the first and second embodiments. FIG. [Diagram 3] 1 is a flowchart showing a fermentation production process supporting method according to a first embodiment. [Figure 4] 10 is a flowchart showing a fermentation production process supporting method according to a second embodiment. [Diagram 5] FIG. 1 is a diagram showing a data flow for predicting culture behavior in an example. [Figure 6] 13 is a table showing the results of a demonstration experiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] A preferred embodiment of the present invention (hereinafter, referred to as the present embodiment) will be described below. Note that the present embodiment described below is merely an example, and the present invention is not limited to the configuration described below.

[0011] The present embodiment described below is a method for supporting a fermentation production process, and two fermentation production process supporting methods according to the present invention are exemplified below. The fermentation production process supporting method according to the first embodiment is referred to as a first supporting method, and the fermentation production process supporting method according to the second embodiment is referred to as a second supporting method.

[0012] First, common points between the first support method and the second support method will be described. In a fermentation production process, a target product is produced in a culture vessel in which microorganisms are cultured. FIG. 1 is a schematic diagram showing an example of a culture system. FIG. 1 shows an example of a culture tank for liquid culture, in which microorganisms are cultured in a culture liquid in the culture tank (culture vessel). In such a culture system, the temperature inside the culture tank or the culture solution, the hydrogen ion exponent (pH) of the culture solution, etc. are measured, and the input of substrates, which are the nutrient sources necessary for microbial cultivation, the supply of oxygen, etc. are controlled to promote the production of the target substance (target product) in the fermentation production process. The first and second support methods provide support for optimizing a fermentation production process by acquiring appropriate culture conditions for each culture point in such a fermentation production process. The substrate is introduced into the culture tank as a solution, and in this specification, the solution is referred to as a substrate introduction solution. Oxygen is supplied by constantly flowing air at a set flow rate from the air supply pipe, which is installed at the bottom of the culture tank and the exhaust pipe at the top. Furthermore, the oxygen supply efficiency can be increased by using an agitator blade to agitate at a set speed. In addition, in this culture behavior, in addition to the target product, other substances (by-products, carbon dioxide, etc.) are also produced, as shown in FIG.

[0013] However, the first and second support methods are not limited to the liquid culture exemplified in FIG. 1, but can also be applied to solid culture. Furthermore, the type of microorganism and the culture medium may be determined according to the target product to be produced, and this support method does not limit the microorganisms, substrates, target products, etc. to be cultured. Microorganisms include, for example, strains belonging to bacterial strains, fungal strains, undifferentiated cell and tissue cultures of animals or plants. The medium may contain various components that are generally contained in microbial media, such as a carbon source, a nitrogen source, metal salts such as magnesium salts and zinc salts, sulfates, phosphates, pH adjusters, surfactants, and antifoaming agents. Examples of substrates include sugars such as glucose, sucrose, and fructose; alcohols such as ethanol and methanol; organic acids such as citric acid, malic acid, and succinic acid; and blackstrap molasses. The target product or by-product may be a substance produced by the metabolism of a microorganism, such as organic acids, amino acids, alcohols, antibodies, enzymes, coenzymes, vitamins, carbohydrates, fatty acids, and nucleic acids.

[0014] [Hardware configuration example] The first and second support methods are executed by one or more processors included in one or more information processing devices. FIG. 2 is a diagram conceptually illustrating an example of a hardware configuration of an information processing device 10 capable of executing the first and second support methods. The information processing device 10 is a so-called computer, and includes a CPU 11, a memory 12, an input / output interface (I / F) 13, a communication unit 14, etc. The information processing device 10 may be a stationary PC (Personal Computer), or may be a portable terminal such as a portable PC, a smartphone, or a tablet.

[0015] The CPU 11 is a so-called processor, and may include not only a general CPU (Central Processing Unit), but also an application specific integrated circuit (ASIC), a DSP (Digital Signal Processor), a GPU (Graphics Processing Unit), etc. The memory 12 is a RAM (Random Access Memory), a ROM (Read Only Memory), and an auxiliary storage device (such as a hard disk). The input / output I / F 13 can be connected to user interface devices such as a display device 15 and an input device 16. The display device 15 is a device that displays a screen corresponding to drawing data processed by the CPU 11 or the like, such as an LCD (Liquid Crystal Display) or CRT (Cathode Ray Tube) display. The input device 16 is a device that accepts input of user operations, such as a keyboard, a mouse, etc. The display device 15 and the input device 16 may be integrated and realized as a touch panel. The communication unit 14 communicates with other computers via a communication network and exchanges signals with other devices such as a printer, etc. A portable recording medium, etc. can also be connected to the communication unit 14.

[0016] The hardware configuration of the information processing device 10 is not limited to the example in Fig. 2. The information processing device 10 may include other hardware elements not shown. Furthermore, the number of each hardware element is not limited to the example in Fig. 2. For example, the information processing device 10 may have multiple CPUs 11. Furthermore, the information processing device 10 may be realized by multiple computers consisting of multiple housings.

[0017] The information processing device 10 can execute the first and second support methods by the CPU 11 executing a computer program stored in the memory 12. It can also be expressed as the CPU 11 being able to execute the first and second support methods through execution of the computer program stored in the memory 12. This computer program is installed, for example, from a portable recording medium such as a CD (Compact Disc) or a memory card, or from another computer on a network, via the input / output I / F 13 or the communication unit 14, and stored in the memory 12.

[0018] [Pre-trained model] The information processing device 10 (CPU 11) can use a trained model that inputs culture information and outputs multiple culture state indices. This trained model is obtained by machine learning using training data including multiple combinations of culture information and multiple culture state indices. For this reason, the trained model can also be referred to as a culture behavior prediction model. The information processing device 10 can also be described as a fermentation production process assistance device that is an apparatus that can use a trained model and executes a fermentation production process assistance method.

[0019] "Culture information" is information related to the culture behavior of a microorganism, includes at least specific culture conditions, and is information input into the trained model. The culture information may include other information that is difficult to call culture conditions. For example, as described below, the culture information may include the concentration and change rate (production rate, consumption rate, etc.) and production state of specific components in the culture vessel, such as the target product, by-product, carbon dioxide, and substrate. The culture information may also include information indicating the bacterial activity state of the microorganisms in the culture vessel. For example, the bacterial activity state of the microorganisms in the culture vessel may be information indicating, as a categorical variable, whether the bacterial activity state is in the bacterial proliferation phase or the target product production phase, or information indicating the bacterial proliferation rate.

[0020] The "specific culture conditions" are culture conditions that may affect the fermentation production process, and are the culture conditions for which the optimal values ​​are sought in this support method. Examples of specific culture conditions include the temperature in the culture vessel, the hydrogen ion exponent (pH) of the culture solution in the culture vessel, the oxygen supply rate, the oxygen consumption rate, and the like. The specific culture conditions may also include the substrate concentration in the culture solution. However, the substrate concentration may not be treated as a specific culture condition, but may be treated as the concentration of the specific component in the culture vessel described above.

[0021] The "culture state index" is an index value that can indicate the culture state in the culture vessel, and is information output from the trained model. The culture state index includes, for example, the concentration or change rate of a specific component in the culture vessel, or both. The specific component here may be a component present in the culture vessel, and may be a component dissolved in the culture solution in the culture vessel, or may be a gas component in the culture vessel. The culture state index may also include a bacterial cell state index, such as a bacterial cell growth rate in the culture vessel. The rate of change of a specific component as an indicator of the culture state means the rate of change of the amount of a specific component per unit amount of bacterial cells (wet bacterial cell weight), and is, for example, all or part of the production rate of the target product, the production rate of a by-product, the production rate of carbon dioxide, or the consumption rate of the substrate, as described below.

[0022] In this way, the trained model in this embodiment needs only to be constructed so as to be capable of outputting multiple types of culture state indicators using culture information including at least specific culture conditions as input, and the specific content of the specific culture conditions or culture information, or the specific culture state indicators, are not limited.

[0023] In addition, the trained model in this embodiment is constructed to input only culture information and output only multiple culture state indices, and it is preferable that the culture information and the culture state indices are composed only of intensive variables or dimensionless numbers, or both. The specific culture conditions included in the culture information include temperature, hydrogen ion exponent, and oxygen supply rate (mol / m 3 / h) and oxygen consumption rate (mol / m 3 / h) is an intensive variable or a dimensionless number. The rate of change of a specific component as a culture state indicator is expressed, for example, by the rate of change of the amount of substance (mol / g / h) per unit amount of bacteria (per 1 g of bacteria), and can be said to be an intensive variable. Furthermore, the concentration of a characteristic component as a culture state indicator is also expressed, for example, by the weight of the component contained per unit volume of culture liquid (g / L), and can be said to be an intensive variable. In this way, by making the input and output of the AI ​​model into intensive variables or dimensionless numbers, it becomes possible to predict the culture behavior independent of the shape and volume of the culture vessel (culture tank). In other words, it is possible to predict with high accuracy the culture behavior even in a culture tank with a different shape and volume from the culture tank from which the training data was obtained.

[0024] Furthermore, the trained model in this embodiment may be a regression equation obtained by regression analysis, or a neural network model obtained by principal component analysis, deep learning, etc., and the data structure, learning algorithm, etc. of the model are not limited. For example, the trained model is realized by a combination of a computer program and parameters, a combination of multiple functions and parameters, etc. When the trained model is constructed with a neural network and the input layer, intermediate layer, and output layer are regarded as a unit of one neural network, the trained model may refer to one neural network or a combination of multiple neural networks. In addition, the trained model may be composed of a combination of multiple multiple regression equations, or may be composed of one multiple regression equation. The trained model used in the examples described below is a function that approximately represents the bacterial activity response, and is composed of a combination of multiple function models with culture information as an input variable and a culture state index as an output variable. These multiple function models are constructed using regression in supervised learning, and each function model has at least a part of the culture information that is input to the trained model as an input variable, and each function model has one of multiple culture state indexes that are outputs of the trained model as an output variable. The trained model may be stored in memory 12 within the information processing device 10, or may be stored in the memory of another computer that the information processing device 10 can access via communication. In addition, there are no limitations on the specific learning algorithm for constructing the trained model. Hereinafter, the trained model used in the first and second support methods will be referred to as an AI model. In addition, in the following description, the first and second support methods will be described as being executed by the CPU 11.

[0025] [First embodiment] The first supporting method will be described in detail below with reference to Fig. 3. Fig. 3 is a flowchart showing the fermentation production process supporting method according to the first embodiment. The first support method mainly includes a trend acquisition step (S31), an optimal condition determination step (S32), and a culture behavior prediction step (S33) (S34). Each step will now be described in detail.

[0026] In the trend acquisition step (S31), the CPU 11 acquires control policy trend information. Hereinafter, the control policy trend information may be abbreviated as policy trend. The CPU 11 may read a policy trend previously stored in the memory 12, may acquire a policy trend stored in the memory of another computer via communication, or may acquire a policy trend stored in a portable recording medium. The CPU 11 may also cause the display device 15 to display an input screen, and acquire a policy trend based on data input by the user via the input screen.

[0027] The control policy trend information is a collection of optimization target indexes for each culture time point, and may be formed of one optimization target index for each culture time point, may be formed of multiple optimization target indexes for each culture time point, or may be formed of any number of optimization target indexes for each culture time point. The optimization target index is an index that is optimized in determining the optimal value in the optimal condition determination step (S32), and at least one of a plurality of culture state indexes that are output information of the AI ​​model is specified. Note that, in this embodiment, an example is given in which the policy trend is formed by one optimization target index for each culture time point. For example, when the AI ​​model is constructed to output the production rate of the target product, the production rate of the by-product, the production rate of carbon dioxide, and the consumption rate of the substrate as culture state indicators, and when the production rate of the target product, the production rate of the by-product, the production rate of carbon dioxide, and the consumption rate of the substrate are represented by identifiers "0", "1", "2", and "3", the policy trend is formed as follows: Control policy trend information = [22223000011···]

[0028] When the policy trend is acquired, the CPU 11 reads out the optimization target indexes from the policy trend in the order of the culture time points, and executes the optimal condition determination process (S32) and the culture behavior prediction process (S33) (S34) based on the optimization target indexes for each culture time point (loop L1(S) to loop L1(E)).

[0029] The culture behavior prediction step is a step of repeating a prediction cycle including a first prediction step (S33) and a second prediction step (S34) for each culture time point within a predetermined culture time, to predict the culture behavior in the culture vessel. In the first prediction step (S33), the CPU 11 inputs culture information at a certain culture time point into the AI ​​model to obtain a plurality of culture state indices. Although not shown in FIG. 3, the culture information at the initial culture time point (0) includes initial values ​​of specific culture conditions and is obtained from the memory 12. In the second prediction step (S34), the CPU 11 acquires culture information at the next culture time point. In this manner, the second prediction step (S34) is a preparation step for the next prediction cycle. There are various methods for acquiring the culture information in the second prediction step (S34). In one embodiment, a part or all of the culture information at the next culture time point is acquired from a culture information list in which culture information predetermined for each culture time point is stored. In another embodiment, a part or all of the culture information at the next culture time point is calculated using a part or all of the multiple culture state indices acquired in the immediately preceding first prediction step.

[0030] Here, an embodiment is exemplified in which the AI ​​model is constructed so that the culture information includes the concentration of the target substance in addition to the specific culture conditions, and also includes the production rate of the target substance as one of the multiple culture state indexes. In this embodiment, in a first prediction step (S33), the CPU 11 inputs culture information including the specific culture conditions and the concentration of the target substance at a certain culture time point into the AI ​​model to obtain multiple culture state indexes including the production rate of the target substance. In a second prediction step (S34), the CPU 11 calculates the concentration of the target substance at the next culture time point using at least the production rate of the target substance included in the multiple culture state indexes obtained in the first prediction step (S33). A calculation formula such as a material balance formula described later is used to calculate the concentration of the target substance.

[0031] The specific culture conditions include, for example, the temperature in the culture vessel, the hydrogen ion exponent of the culture solution in the culture vessel, and the oxygen supply rate or oxygen consumption rate, and optimal values ​​of such specific culture conditions are determined for each culture time point. Note that candidate values ​​for deriving the initial values ​​and optimal values ​​of the specific culture conditions can be obtained from the culture information list stored in the memory 12. Oxygen Consumption Rate (OCR) is the rate at which microorganisms in a culture vessel consume oxygen per unit volume of culture medium (mol / m 3 / h). The oxygen transfer rate (OTR) is the rate at which oxygen supplied to the culture vessel moves into the culture solution and becomes dissolved oxygen (mol / m 3 / h). Regarding the oxygen consumption rate and oxygen supply rate, either one may be used equally in the first and second support methods. This is because the rate at which oxygen dissolves in an aqueous solution (culture solution) is slow, and oxygen supply becomes a bottleneck during cultivation. For example, the oxygen concentration and air flow rate can be measured at the air supply side and exhaust side of the cultivation system, and the unit culture solution volume (m 3 ) or the oxygen supply rate OTR (mol / m 3 On the other hand, since the amount of oxygen that can be dissolved in the culture solution is very small compared to the oxygen supply (consumption) rate per unit time, the oxygen supply rate OTR and the oxygen consumption rate OCR can be treated equally by regarding the dissolved oxygen concentration as approximately 0 according to the following balance equation for oxygen in the culture solution. OTR (mol / m 3 / h) = OCR (mol / m 3 / h) + (dissolved oxygen concentration (g / m 3 ) / amount of oxygen (g / mol) / Δt(h) Although gaseous oxygen is also present in the culture solution, the bacteria can only utilize the oxygen present in the solution and oxygen dissolves slowly in the aqueous solution (culture solution), so the gaseous oxygen in the solution is ignored in the above balance equation. Furthermore, the temperature inside the culture vessel may be the temperature of the culture solution inside the culture vessel, the temperature of the air inside the culture vessel, or both.

[0032] In a first prediction step (S33), the CPU 11 inputs culture information including the concentration of the target substance, temperature, hydrogen ion exponent, and oxygen supply rate or oxygen consumption rate into an AI model to obtain multiple culture state indexes including the production rate of the target substance. In a second prediction step (S34), the CPU 11 determines the optimal values ​​of the specific culture conditions (temperature, hydrogen ion exponent, and oxygen supply rate or oxygen consumption rate) determined in the optimal condition determination step (S32) as the specific culture conditions at the next culture time point, and in addition, calculates the concentration of the target substance at the next culture time point using at least the production rate of the target substance obtained in the first prediction step (S33).

[0033] In this manner, by adding the concentration of the target substance in the culture medium in the culture vessel as input information and the production rate of the target substance as output information, it is possible to construct an AI model that can estimate culture behavior with high accuracy. Furthermore, by including such specific culture conditions (particularly oxygen supply rate or oxygen consumption rate) in addition to the target substance concentration in the culture information input to the AI ​​model, the estimation accuracy of the AI ​​model can be improved, and ultimately the prediction accuracy of the culture behavior can be improved. However, the specific culture conditions may further include information other than the temperature, pH, oxygen supply rate, and oxygen consumption rate, and the culture information may further include the concentration of a specific component in the culture vessel, such as a substrate concentration, in addition to the target substance concentration.

[0034] The AI ​​model may be constructed so that the culture information further includes bacterial activity information indicating the bacterial activity state of the microorganism using a categorical variable. In this case, in the first prediction step (S33), the CPU 11 inputs culture information including the target substance concentration and specific culture conditions at a certain culture time point as well as the bacterial cell activity information into the AI ​​model to obtain multiple types of culture state indicators including the target substance production rate. The microbial activity information indicates, for example, the microbial activity state of the microorganisms in the culture vessel by a categorical variable (0 or 1), whether the microbial activity state is in the proliferation phase or the production phase of the target product. In this way, by adding bacterial activity information to the culture information input into the AI ​​model, it is possible to simulate the bacterial activity response in the AI ​​model, thereby further improving the accuracy of predicting culture behavior.

[0035] When the culture information includes bacterial activity information, it is preferable that the specific culture conditions included in the culture information include an oxygen supply rate or an oxygen consumption rate. This is because the accuracy of estimating the production rate of the target substance in the AI ​​model can be improved by using the oxygen supply rate or the oxygen consumption rate as input information. Furthermore, since the bacterial activity state can be estimated using the oxygen supply rate or the oxygen consumption rate, the oxygen supply rate or the oxygen consumption rate can also be used to calculate bacterial activity information at the next culture time point.

[0036] The AI ​​model may be constructed so that the culture information further includes by-product state information indicating the increase or decrease state of the by-product in the culture vessel using a categorical variable. In this case, in the first prediction step (S33), the CPU 11 acquires the production rate of the target substance and the production rate of the by-product by inputting culture information including the target substance concentration and specific culture conditions at a certain culture time point as well as by-product state information into the AI ​​model. The by-product state information is information indicating the increase or decrease state of a by-product accompanying the production of the by-product by a microorganism, and is, for example, information indicating an increase or decrease state by a categorical variable (0 or 1). In this case, the by-product is a substance corresponding to an intermediate in the metabolic pathway in which the bacterial cell produces the target product. Therefore, the increase state of the by-product indicates a metabolic state in which the metabolism after the intermediate is stagnant in the metabolic pathway, and the decrease state of the by-product indicates a metabolic state in which the metabolism after the intermediate is promoted. Furthermore, the production rate of the by-product indicates a positive value in a metabolic state in which the metabolism after the intermediate is stagnant, and indicates a negative value in a metabolic state in which the metabolism after the intermediate is promoted. In this case, the specific component indicates the target product or substrate. In this way, by adding by-product state information to the culture information input into the AI ​​model, it is possible to simulate the bacterial activity response in the AI ​​model, thereby further improving the accuracy of predicting culture behavior.

[0037] In a form in which the production rate of the target substance is used to calculate the concentration of the target substance at the next time point of culture, for example, the following mass balance equation can be used. In the following example of the balance equation (Equation 1), n TGT indicates the amount of the target substance (mol), and r TGT indicates the production rate of the target substance (mol / g / h), and B(t) indicates the bacterial mass (g) of a specific bacterial cell of a microorganism at the time t of culture. In the example of (Equation 2), MW TGT indicates the molecular weight of the target substance (g / mol), V indicates the volume of culture medium (L) at the time of culture (t+1), and C TGT indicates the concentration of the target substance (g / L) in the culture medium at the time point of culture (t+1). In this case, in the second prediction step (S34), the CPU 11 calculates the amount of substance n of the target substance at the next culture time point (t+1) by solving the differential equation (Formula 1). TGT By calculating the amount of substance and substituting it into (Equation 2), the target substance concentration C in the culture solution at the next culture time point (t+1) is obtained. TGT can be calculated.

number

[0038] B(t) and V in the above (Equation 1) and (Equation 2) may be extracted from a list of values ​​determined in advance for each culture time point, or may be calculated using a material balance equation or the like. For example, the CPU 11 can calculate a specific bacterial mass of a microorganism at each culture time point based on the elapsed culture time and the oxygen supply rate or oxygen consumption rate at each culture time point, as exemplified by the following (Equation 3). The specific bacterial mass calculated here means the wet weight or dry weight of a specific type of bacterial mass of a microorganism cultured in a culture vessel, and may be one type of bacterial mass or two or more types of bacterial mass. In formula 3, min{} is a function that takes the smaller of the values ​​on the left and right separated by a comma. The left side indicates a state in which there is an excess of dissolved oxygen in the culture solution and the specific bacteria are growing according to time t, while the right side indicates a state in which there is a shortage of dissolved oxygen and the oxygen supply rate or oxygen consumption rate is at the upper limit. In addition, in (Formula 3), BCI indicates the initial cell concentration of a specific microbial cell in a culture vessel, and k1, k2, and k3 are fixed values. The value of BCI is obtained as an initial value in the first step (S31), for example. k1, k2, and k3 are values ​​obtained by experiments by the present inventors, and are, for example, k1=0.18, k2=17.5, and k3=0.334. However, k1, k2, and k3 may be appropriately changed depending on the value of BCI or the type of microorganism or the medium. In addition, in equation (3), the oxygen supply rate OTR may be replaced with the oxygen consumption rate OCR.

number

[0039] In the above example of (Equation 3), the culture solution volume V at the next culture time point (t+1) may be extracted from a predetermined value list, or may be calculated using a material balance equation or the like. The CPU 11 can calculate the concentration of the target substance at the next culture time point by using at least the specific bacterial mass calculated in this way and the production rate of the target substance obtained in the first prediction step (S33). By using the above (Equation 1), (Equation 2), and (Equation 3), the bacterial mass B(t) of the specific bacterial mass of the microorganism at the culture time point t is calculated by (Equation 3), and the calculated B(t) and the production rate r of the target substance obtained in the first prediction step (S33) are used to calculate the concentration of the target substance at the next culture time point. TGT By solving the differential equation (Equation 1) using TGT is calculated, and the target substance concentration at the next culture time point (t+1) can be calculated using (Equation 2).

[0040] When the culture information input to the AI ​​model includes bacterial cell activity information, in the second prediction step (S34), the CPU 11 acquires bacterial cell activity information at the next culture time point. The next bacterial activity information may be obtained from a list of values ​​that are predetermined for each culture time point, or may be calculated using the oxygen supply rate or oxygen consumption rate included in the specific culture conditions. In the latter case, for example, when the left side of the min{} function in the above (Equation 3) is selected (when it becomes smaller), the bacterial activity information is set to a value (0) indicating that the bacterial activity state is in the bacterial proliferation phase, and when the right side is selected (when it becomes smaller), the information is set to a value (1) indicating that the bacterial activity state is in the production phase of the target product.

[0041] In addition, when the culture information input to the AI ​​model includes by-product state information and the by-product production rate is acquired from the AI ​​model as one of the change rates of the specific component, in the second prediction step (S34), the CPU 11 can acquire by-product state information at the next culture time point using the acquired by-product production rate. For example, when the by-product production rate is equal to or greater than 0, the by-product status information is set to a value (0) indicating an increasing state of the by-product, and when the by-product production rate is less than 0, the by-product status information is set to a value (1) indicating a decreasing state of the by-product.

[0042] Furthermore, if the AI ​​model is constructed so that the culture status index, which is the output information, includes the production rate of the target substance and the substrate consumption rate, the substrate concentration in the culture solution can be calculated using a material balance equation such as the following. In the following example of the balance equation (Equation 4), n SBS indicates the amount of substrate (mol), and r SBS indicates the consumption rate of the substrate (mol / g / h), B(t) indicates the bacterial mass (g) of a specific microorganism at the time t of the culture, and C Feed indicates the substrate concentration in the substrate input solution (g / L), and F SBS (t) is the substrate input rate (L / h) at time t of the culture, and MW SBSindicates the molecular weight of the substrate (g / mol). In the example of (Equation 5), V indicates the volume of the culture medium (L) at the time point of culture (t+1). In this case, the CPU 11 calculates C Feed and F SBS (t), and the substrate consumption rate r obtained from the AI ​​model in the first prediction step (S33) and the substrate input information is used. SBS By solving the differential equation (4) using SBS By calculating the amount of substance and substituting it into (Equation 5), the substrate concentration C in the culture medium at the next culture time point (t+1) is obtained. SBS The method for obtaining the bacterial mass B(t) at the culture time point t is as described above.

number

[0043] In the optimal condition determination step (S32), the CPU 11 determines optimal values ​​of the specific culture conditions at each culture time point so that the culture state index designated as the optimization target index at each culture time point in the policy trend acquired in step (S31) among the multiple culture state indexes acquired from the AI ​​model in the first prediction step (S33) is optimized. For example, if an AI model is constructed to output the production rate of the target product, the production rate of a by-product, the production rate of carbon dioxide, and the consumption rate of the substrate as culture state indices, and the identifier "2" for the carbon dioxide production rate is specified as the optimization target index at the culture time (t=2) in the policy trend, the value of the specific culture condition that optimizes the carbon dioxide production rate (the fastest or slowest, or a predetermined desired rate) among the four culture state indices obtained from the AI ​​model by inputting the culture information at the culture time (t=2) is determined as the optimal value. In the optimum condition determination step (S32), various optimization methods can be used. For example, the calculation cost can be reduced by using Bayesian optimization.

[0044] The CPU 11 reads out the optimization target indexes in the order of the culture time points from the policy trend, and executes an optimal condition determination step (S32) and a culture behavior prediction step (S33) (S34) based on the optimization target indexes for each culture time point, thereby obtaining optimal values ​​for specific culture conditions at each culture time point within a specified culture time. That is, according to the first support method, it is possible to obtain appropriate conditions in line with the policy trend as the culture conditions for each culture time point in the fermentation production process. Furthermore, by using multiple types of culture state indices output from the AI ​​model based on the optimal values ​​of the specific culture conditions at each culture time point, it is possible to obtain culture result information such as the production amount of the target product, by-products, etc., and the consumption amount of the substrate.

[0045] [Second embodiment] Next, details of the second support method will be described with reference to Fig. 4. Fig. 4 is a flowchart showing the fermentation production process support method according to the second embodiment. The second support method includes all the steps constituting the first support method described above, and in the following description, the details related to the first support method will be omitted as appropriate. In Fig. 4, the part marked "FIG. 3 FLOW" includes the steps of the first support method shown in Fig. 3. The second support method includes, in addition to the steps of the first support method, a goal obtaining step (S40), a policy generating step (S41), a result obtaining step (S43), and a policy selecting step (S45). Each step will now be described in detail.

[0046] In the target obtaining step (S40), the CPU 11 obtains target quality information. The "target quality information" is information indicating the final target quality in a fermentation production process, and is set, for example, to maximize the yield of the target product per input substrate, maximize the production amount of the target product, minimize the amount of remaining by-products, minimize the amount of remaining substrate, etc. The CPU 11 may read target quality information prestored in the memory 12, may acquire target quality information stored in the memory of another computer via communication, or may acquire target quality information stored in a portable recording medium. The CPU 11 may also cause the display device 15 to display an input screen and acquire target quality information based on data input by the user via the input screen.

[0047] In the policy generating step (S41), the CPU 11 generates a plurality of pieces of control policy trend information. The control policy trend information is the same as that used in the first support method, and is also referred to as policy trends here. The CPU 11 generates a plurality of policy trends such that the optimization target index at at least one culture time point is different from each other.

[0048] For example, the CPU 11 generates multiple policy trends using a genetic algorithm based on one or more policy trends selected in a policy selection step (S45) described later. Specifically, in the policy selection step (S45), one or more excellent policy trends that match the target quality information are selected, so that multiple policy trends are generated by crossing (recombining) the optimization target indexes for each culture time point indicated by the selected excellent policy trends or mutating (changing) a predetermined ratio (several percent) of the optimization target indexes. By using a genetic algorithm to generate a policy trend in this way, the policy trend can be optimized efficiently, and the calculation cost can be reduced. However, the optimization algorithm used in the method for generating the policy trend in step (S41) is not limited.

[0049] After executing the policy generating step (S41), the CPU 11 selects policy trends one by one, and executes the first support method step and the result obtaining step (S43) based on each policy trend (loop L2(S) to loop L2(E)).

[0050] When the first support method is executed based on each policy trend, optimal values ​​of specific culture conditions at each culture time point within a specified culture time are obtained in line with each policy trend, and a culture behavior prediction process for each culture time point is executed using the optimal values ​​of the specific culture conditions. As a result, in the result acquisition step (S43), the CPU 11 can acquire culture result indicators over a specified culture time for each policy trend as a result of predicting the culture behavior at each culture time point using specific culture conditions optimized along each policy trend (executing the first prediction step (S33) and the second prediction step (S34)). As a result of predicting the culture behavior at each time point within a specified culture time, information such as the production amount of specific components (target product, by-product, carbon dioxide, etc.) in the culture vessel over the specified culture time, the amount of substrate input, consumption, and remaining amount can be obtained. CPU 11 processes (calculates) such information so as to conform to the quality indicated by the target quality information acquired in step (S40), thereby acquiring a culture result index conforming to the target quality information. For example, when the target quality information indicates maximization of the yield of the target product per input substrate, CPU 11 acquires the yield of the target product as the culture result index by dividing the production amount of the target product by the input amount of substrate. When the target quality information indicates maximization of the production amount of the target product, the production amount of the target product is acquired as the culture result index, when the target quality information indicates minimization of the remaining amount of the by-product, the production amount of the by-product is acquired as the culture result index, and when the target quality information indicates minimization of the remaining amount of the substrate, the remaining amount of the substrate is acquired as the culture result index.

[0051] When the first support method step and the result acquisition step (S43) have been executed for all policy trends generated in the step (S41), the process leaves the loop L2 and a policy selection step (S45) is executed.

[0052] In the policy selection step (S45), the CPU 11 selects one or more policy trends that match the target quality information from among the multiple policy trends based on the culture result index acquired for each policy trend. Here, one policy trend corresponding to the culture result index with the highest degree of match with the target quality indicated by the target quality information may be selected, or multiple policy trends may be selected in descending order of the degree of match of the culture result index with the target quality. Note that, when a genetic algorithm is used in step (S41), for example, two policy trends are selected in descending order of the degree of match of the culture result index with the target quality.

[0053] Also, for example, when the target quality information indicates maximization of the yield of the target product per input substrate and the yield of the target product per input substrate is acquired as the culture result index, one policy trend corresponding to the maximum culture result index (yield) or a predetermined number of policy trends corresponding to a predetermined number of culture result indexes in descending order of the culture result index (yield) are selected. Also, when the target quality information indicates minimization of the remaining amount of substrate and the remaining amount of substrate is acquired as the culture result index, one policy trend corresponding to the minimum culture result index (remaining amount of substrate) or a predetermined number of policy trends corresponding to a predetermined number of culture result indexes in ascending order of the culture result index (remaining amount of substrate) are selected.

[0054] In step (S47), the CPU 11 judges the end condition of the optimization of the policy trend. If the end condition is satisfied (S47; YES), the CPU 11 ends the process, and if the end condition is not satisfied (S47; NO), the CPU 11 executes the policy generation step (S41) and subsequent steps again. The termination condition determined in step (S47) may be any condition that allows the selection of the policy trend that is most compatible with the target quality, and the specific condition is not limited. For example, in a form in which a genetic algorithm is used in the policy generation step (S41), the number of generations of the algorithm is set as the termination condition. In addition, the termination condition may be that the current culture result index is lower in compatibility with the target quality than the previous one, by comparing the culture result index of the policy trend selected in the previous policy selection step (S45) with the culture result index of the policy trend selected in the current policy selection step (S45).

[0055] In this way, according to the second support method, it is possible to obtain an optimal policy trend that conforms to the target quality, and also to obtain optimal values ​​for specific culture conditions at each culture point in line with this optimal policy trend. This enables automatic design of fermentation production process control by data-driven fermentation simulation. As a result, fermentation production technology development can shift from the conventional know-how-based development that relied on the experience of developers to data-driven development, realizing more efficient and faster fermentation production technology development.

[0056] Some or all of the above-described embodiments and modified examples may be specified as follows. However, the above-described embodiments and modified examples are not limited to the following descriptions.

[0057] <1> A method for supporting a fermentation production process in which a target substance is produced in a culture vessel in which a microorganism is cultured, comprising the steps of: One or more processors that can use a trained model that inputs culture information including at least a specific culture condition and outputs a plurality of culture state indicators, a culture behavior prediction step of repeating a prediction cycle for each culture time point within a predetermined culture time, the prediction cycle including a first prediction step of acquiring the plurality of culture state indexes by inputting the culture information at a certain culture time point into the trained model and a second prediction step of acquiring the culture information at a next culture time point; a trend acquisition step of acquiring control policy trend information, the control policy trend information being a set of optimization target indexes for each culture time point, in which at least one of the plurality of culture state indexes is designated as the optimization target index for each culture time point; An optimal condition determination step of determining optimal values ​​of the specific culture conditions at each culture time point so that a culture state index designated as an optimization target index at each culture time point in the acquired control policy trend information among the plurality of culture state indexes acquired in the first prediction step is optimized; A fermentation production process support method for carrying out the above steps.

[0058] <2> the one or more processors: a target acquisition step of acquiring target quality information; a policy generating step of generating a plurality of pieces of control policy trend information; a result acquisition step of acquiring a culture result index corresponding to the target quality information over the predetermined culture time by executing the culture behavior prediction step using the optimal value of the specific culture condition at each culture time point determined in the optimal condition determination step for each of the generated control policy trend information; a policy selection step of selecting one or more pieces of control policy trend information that match the target quality information from among the plurality of pieces of control policy trend information based on the culture result index acquired for each of the control policy trend information; Run the following again <1> The fermentation production process support method according to claim 1. <3> In the policy generating step, the one or more processors generate a plurality of pieces of control policy trend information using a genetic algorithm based on the one or more pieces of control policy trend information selected in the policy selecting step. <2> The fermentation production process support method according to claim 1. <4> In the optimal condition determination step, the one or more processors determine optimal values ​​of the specific culture conditions at each of the culture time points using Bayesian optimization. <1> from <3> 13. A fermentation production process supporting method according to claim 12. <5> The one or more processors: In the first prediction step, the culture information including the specific culture conditions and the concentration of the target substance is input to the trained model to obtain the plurality of culture state indices including the production rate of the target substance; In the second prediction step, a concentration of the target substance at the next culture time point is calculated using at least a production rate of the target substance included in the acquired multiple culture state indicators. <1> from <4> 13. A fermentation production process supporting method according to claim 12. <6> The specific culture conditions include a temperature in the culture vessel, a hydrogen ion exponent of the culture solution in the culture vessel, and an oxygen supply rate or an oxygen consumption rate; The one or more processors: In the first prediction step, the culture information including the concentration of the target substance, the temperature, the hydrogen ion exponent, and the oxygen supply rate or the oxygen consumption rate is input to the trained model to obtain the multiple culture state indices including the production rate of the target substance; In the second prediction step, the temperature, the hydrogen ion exponent, and the oxygen supply rate or the oxygen consumption rate at the next culture time point are obtained. <5> The fermentation production process support method according to claim 1. <7> In the culture vessel, at least the target product, a by-product, and carbon dioxide are produced and at least the substrate is consumed, the multiple culture state indicators include a production rate of the target product, a production rate of the by-product, a production rate of the carbon dioxide, and a consumption rate of the substrate, The optimization target index for each culture time point indicated by the control policy trend information includes a designation of the production rate of the target product, the production rate of the by-product, the production rate of the carbon dioxide, or the consumption rate of the substrate. <1> from <4> 13. A fermentation production process supporting method according to claim 12. <8> The trained model receives only the culture information as input and outputs only the multiple culture state indexes, The culture information and the plurality of culture state indexes are composed of intensive variables or dimensionless numbers, or both. <1> from <7> 13. A fermentation production process supporting method according to claim 12. <9> A fermentation production process support apparatus comprising at least a memory and the one or more processors, <1> from <8> A fermentation production process assistance apparatus capable of executing any one of the fermentation production process assistance methods described above.

[0059] The above content will be described in more detail below with reference to examples, but the description of the following examples does not limit the above content in any way. EXAMPLES

[0060] In this example, it was confirmed that the fermentation production process support method according to the second embodiment described above can acquire an optimal policy trend and can acquire optimal values ​​of specific culture conditions at each culture time point in accordance with the acquired policy trend, and that the practicality of the method can be confirmed from the processing time of the method. A cultivation experiment was conducted to support a fermentation production process that uses aerobic bacteria to produce organic acids (target product). In the cultivation experiment, microorganisms (aerobic bacteria) were actually cultivated using a specified cultivation tank (aeration stirring tank), and various cultivation data was obtained by sampling at intervals of several hours during the cultivation experiment. The following AI model was trained using the obtained cultivation data. As will be described later, glucose was used as the substrate.

[0061] FIG. 5 is a diagram showing a data flow for predicting culture behavior in an example. In this embodiment, an AI model consisting of a combination of five function models F0 to F4 was used. Each of the function models F0 to F4 is a function that approximately expresses the bacterial activity response, and is constructed using supervised learning regression with culture information as an input variable and a type of culture state index as an output variable. The input variables and output variables of each function model are as follows: <output variable> Function model F0 = target product production rate per unit cell weight Function model F1 = By-product production rate per unit cell weight Function model F2 = glucose consumption rate per unit cell weight Function model F3 = bacterial growth rate per unit bacterial weight Function model F4 = Carbon dioxide production rate per unit cell weight <Input variables> Function model F0 = temperature, pH, oxygen supply rate (OTR), glucose concentration, target concentration, oxygen supply rate per unit cell weight, carbon dioxide production rate per unit cell weight, cell growth rate per unit cell weight, glucose consumption rate per unit cell weight Function model F1 = temperature, pH, oxygen supply rate (OTR), glucose concentration, target concentration, target production rate per unit cell weight, oxygen supply rate per unit cell weight, carbon dioxide production rate per unit cell weight, cell growth rate per unit cell weight, glucose consumption rate per unit cell weight Function model F2 = temperature, pH, oxygen supply rate (OTR), glucose concentration, target product production rate per unit cell weight, target product concentration, oxygen supply rate per unit cell weight, carbon dioxide production rate per unit cell weight, cell growth rate per unit cell weight Function model F3 = temperature, pH, oxygen supply rate (OTR), glucose concentration, target product production rate per unit cell weight, target product concentration, oxygen supply rate per unit cell weight, cell growth rate per unit cell weight, glucose consumption rate per unit cell weight Function model F4 = temperature, pH, oxygen supply rate (OTR), glucose concentration, culture time, target product production rate per unit cell weight, oxygen supply rate per unit cell weight, carbon dioxide production rate per unit cell weight, glucose consumption rate per unit cell weight

[0062] In this embodiment, the output variables of the five function models are set as five types of culture state indexes as output information of the AI ​​model. In this embodiment, the control policy trend information (policy trend) is in a format in which one of these five types of culture state indexes is designated as an optimization target index for each culture time point. That is, the target product production rate per unit cell weight is indicated by identifier "0", the by-product production rate per unit cell weight is indicated by identifier "1", the glucose consumption rate per unit cell weight is indicated by identifier "2", the cell growth rate per unit cell weight is indicated by identifier "3", and the carbon dioxide production rate per unit cell weight is indicated by identifier "4". For example, the policy trend can be generated as follows: Policy Trend = [44422103·····] In this embodiment, the incubation time was set to 72 hours, and the optimization interval was set to 5 minutes. Therefore, in the policy trend, 864 optimization target indexes were specified every 5 minutes.

[0063] In addition, in this embodiment, each of the five function models has at least a portion of the culture information as input information for the AI ​​model as input variables, and the specific culture conditions included in the culture information are temperature, pH, oxygen supply rate (OTR), and glucose concentration. Then, in the optimal condition determination step (S32), the optimal values ​​of such specific culture conditions were determined. In this embodiment, Bayesian optimization was adopted for optimizing the specific culture conditions.

[0064] In the first prediction step (S33) of the culture behavior prediction step in this embodiment, culture information at a certain culture time point was input into the input variables of the five function models, and the above-mentioned five culture state indices (rate parameters) were obtained as the values ​​of the output variables of the five function models. In the second prediction step (S34) of the culture behavior prediction step, culture information at the next culture time point was obtained based on the five culture state indices obtained in the first prediction step (S33). The target product production rate obtained from the function model F0, the glucose consumption rate obtained from the function model F2, the carbon dioxide production rate obtained from the function model F3, and the bacterial growth rate obtained from the function model F4 were used as the culture information at the next culture time point. Other culture information at the next culture time point was derived using the material balance equations such as the above-mentioned formulas (1) to (5) and other calculation formulas, the five culture state indices obtained in the first prediction step (S33), etc.

[0065] In this embodiment, the target quality information was set to "maximize the production amount of the target product at the end of the culture" (target acquisition step (S40)), and a genetic algorithm was used to optimize the policy trend (50 individuals, 50 generations).

[0066] In this embodiment, the fermentation production process support method was executed on a computer having the following specifications: CPU = Intel(R) Xeon(R) Gold 6258R 28 cores 56 threads x 2 OS: Windows 10 Pro RAM=128GB As a result, the optimal control policy trend information and the optimal values ​​of the specific culture conditions at each culture point in accordance with the control policy trend information were obtained after a processing time of approximately 18.6 hours.

[0067] Furthermore, in this embodiment, a demonstration experiment was conducted under the same conditions as the above-mentioned culture experiment (aerobic bacterial species, culture tank scale, etc.) using the optimal control policy trend information obtained in this manner and the optimal values ​​of specific culture conditions at each culture point in accordance with the control policy trend information. In this demonstration experiment, the culture was actually performed for 72 hours, and the production amount (g) of the target product was measured at each of the following timings: 7 hours, 24 hours, 32 hours, 48 ​​hours, and 72 hours. Meanwhile, in the culture behavior prediction step of this embodiment, the predicted value (g) of the target product production amount at each timing was derived, and the measured value and predicted value of the target product production amount at each timing were normalized using the predicted value of the target product production amount after 72 hours (g / g-MAX).

[0068] FIG. 6 is a table showing the results of the demonstration experiment. In FIG. 6, the time column indicates the elapsed time of culture, and the predicted and measured values ​​columns indicate the predicted and measured values ​​of the amount of target product production normalized as described above. As a result, the prediction error of the target production volume after 72 hours was suppressed to 16%, and the predicted target production trend roughly matched the experimental results. Furthermore, in the culture under the optimum values ​​of the specific culture conditions obtained in this example, the final production amount of the target product was improved by 5% compared to the conventional value before the study of this example.

[0069] As described above, according to this embodiment, it was demonstrated that the fermentation production process support method according to the second embodiment can acquire an optimal policy trend and can acquire optimal values ​​of specific culture conditions at each culture time point in accordance with the acquired policy trend, as well as the practicality of the method in terms of processing time. [Explanation of symbols]

[0070] 10 Information processing device (fermentation production process support device) 11 CPU 12. Memory 13 Input / Output Interface 14 Communication unit 15 Display device 16 Input Devices

Claims

1. A method for supporting a fermentation production process in which a target substance is produced in a culture vessel for culturing microorganisms, comprising: one or more processors that can use a trained model that receives input of culture information including at least specific culture conditions and outputs a plurality of culture state indicators; a culture behavior prediction step of repeating a prediction cycle for each culture time point within a predetermined culture period, the prediction cycle including a first prediction step of acquiring the plurality of culture state indices by inputting the culture information at a certain culture time point into the trained model and a second prediction step of acquiring the culture information at the next culture time point; a trend acquisition step of acquiring control policy trend information, the control policy trend information being a set of optimization target indexes for each culture time point, in which at least one of the plurality of culture state indexes is designated as the optimization target index for each culture time point; an optimal condition determination step of determining optimal values ​​of the specific culture conditions at each culture time point so that a culture state index designated as an optimization target index at each culture time point in the acquired control policy trend information among the plurality of culture state indexes acquired in the first prediction step is optimized; A fermentation production process support method for performing the above steps.

2. the one or more processors: a target acquisition step of acquiring target quality information; a policy generating step of generating a plurality of pieces of control policy trend information; a result acquisition step of acquiring a culture result index corresponding to the target quality information over the predetermined culture time by executing the culture behavior prediction step using the optimal values ​​of the specific culture conditions at each culture time point determined in the optimal condition determination step for each of the generated control policy trend information; a policy selection step of selecting one or more pieces of control policy trend information that match the target quality information from among the plurality of pieces of control policy trend information based on the culture result index acquired for each piece of control policy trend information; The method for supporting a fermentation production process according to claim 1, further comprising:

3. In the policy generating step, the one or more processors generate a plurality of pieces of control policy trend information using a genetic algorithm based on the one or more pieces of control policy trend information selected in the policy selecting step. The method for supporting a fermentation production process according to claim 2.

4. In the optimal condition determination step, the one or more processors determine optimal values ​​of the specific culture conditions at each of the culture time points using Bayesian optimization. The method for supporting a fermentation production process according to claim 1.

5. The one or more processors: In the first prediction step, the culture information including the specific culture conditions and the concentration of the target substance is input to the trained model to obtain the plurality of culture state indices including the production rate of the target substance; In the second prediction step, the concentration of the target substance at the next culture time point is calculated using at least the production rate of the target substance included in the acquired plurality of culture state indicators. The method for supporting a fermentation production process according to claim 1.

6. the specific culture conditions include a temperature in the culture vessel, a hydrogen ion exponent of the culture solution in the culture vessel, and an oxygen supply rate or an oxygen consumption rate; The one or more processors: In the first prediction step, the culture information including the concentration of the target substance, the temperature, the hydrogen ion exponent, and the oxygen supply rate or the oxygen consumption rate is input to the trained model to obtain the plurality of culture state indices including the production rate of the target substance; In the second prediction step, the temperature, the hydrogen ion exponent, and the oxygen supply rate or the oxygen consumption rate at the next culture time point are acquired. The fermentation production process support method according to claim 5.

7. In the culture vessel, at least the target product, a by-product, and carbon dioxide are produced and at least the substrate is consumed, the plurality of culture state indicators include a production rate of the target product, a production rate of the by-product, a production rate of the carbon dioxide, and a consumption rate of the substrate; The optimization target index for each culture time point indicated in the control policy trend information includes a designation of the production rate of the target product, the production rate of the by-product, the production rate of the carbon dioxide, or the consumption rate of the substrate. The method for supporting a fermentation production process according to claim 1.

8. the trained model receives only the culture information as input and outputs only the plurality of culture state indices; The culture information and the plurality of culture state indicators are composed of only intensive variables or dimensionless numbers, or both. The method for supporting a fermentation production process according to claim 1.

9. A fermentation production process support apparatus comprising at least a memory and the one or more processors, A fermentation production process assistance device capable of executing the fermentation production process assistance method according to claim 1.

10. A computer program for causing a computer including a memory and one or more processors to execute the fermentation production process support method described in claim 1.

11. A computer-readable recording medium that records a computer program for causing a computer including a memory and one or more processors to execute the fermentation production process support method described in claim 1.

Images