Method for calculating process parameter

Abstract

To provide a method of applying optimization technique-based algorithms and trained predictive models to deduce the unconfirmed process parameters inversely.SOLUTION: A method constructs an optimization model by treating a trained predictive model as a function of process parameters to be determined. Subject to the conditions that the process parameters to be determined must satisfy, either individually or in relation to each other. Moreover, in order to solve the constrained optimization problem, the method with penalty function and barrier function can be used to convert the constrained optimization problem into an unconstrained optimization problem. For the optimization problem, a dataset of the trained predictive model can be used to find multiple samples that meet target results as anchor points to form a multi-dimensional feasible solution space, and set or make initial solutions in this feasible solution space, and then use different numerical analysis methods to find the direction and step size, and the optimal feasible solution is an input parameter that can be used.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a method for calculating process parameters, and more particularly to a method for inversely estimating unidentified process parameters by applying an optimization-based algorithm and a trained predictive model. [Background technology]

[0002] With the progress of medical engineering technology, the application of regenerative medicine in clinical treatment of diseases is becoming more and more diverse. Regenerative medicine is a medical technology that mainly uses the regenerative ability of cells to repair damaged tissues and organs, and has a very wide range of applications. In addition, by combining medical technologies such as tissue engineering and molecular biology, it is expected that diseases that were once difficult to treat, such as diabetes, neurological diseases, cardiovascular diseases, and cancer, can be improved and treated. Currently, regenerative medicine is mainly applied to organ repair, immune cell therapy, and stem cell therapy. In addition, research and applications of cell therapy in regenerative medicine are attracting more and more attention from various fields. Cell therapy involves the transplantation of human body cells into an individual's body after a procedure of culturing or processing the cells outside the body.

[0003] In recent years, many governments have gradually lifted the ban on the application of cell therapy, and many scholars at home and abroad have been entering the research of cell therapy one after another. As a result, cell therapy has made remarkable progress in the treatment of many diseases, such as the treatment of skin defects with autologous fibroblasts, the treatment of knee joint cartilage defects with autologous chondrocytes, and the treatment of spinal cord injury with autologous bone marrow mesenchymal stem cells. In addition, the quality of cell therapy products has a direct impact on the safety and efficacy of the treatment. Therefore, in the cell culture process, it is necessary to strictly control the cell growth state and monitor the culture parameters and environmental parameters of cell growth in real time to avoid cell contamination or deterioration of quality during the culture process. In addition, previous research has shown that the variability of cells is extremely high between different cases, so the optimal culture parameters and environmental parameters of cell preparations applied to different cases are not completely the same. Therefore, it is necessary to design and adjust each process parameter in the process for each cell preparation to achieve the desired result, and therefore each cell preparation cannot be produced with fixed process parameters. In addition, due to the complexity and interrelationship of the process, the prior art usually only allows optimization of a single parameter, while neglecting the overall mutual influences between each parameter in the entire process.

[0004] In the prior art, a prediction model is obtained by training on a large number of sample data sets using machine learning. This prediction model can generate predicted results of cell culture by inputting various process parameters, so that a user can simulate the effect of the cell culture process to be designed in advance. However, the above prediction model trained by machine learning cannot reverse-estimate the process parameters of the cell culture process from the user's expected results. In other words, when designing a cell culture process, a user needs to try a large number of different process parameters, so a large amount of effort and resources are spent on designing and improving the cell culture process. In addition, there is a problem that the input parameters cannot be reverse-estimated for prediction models obtained by machine learning in other fields besides cell processes.

[0005] In general, a prediction model obtained by training using machine learning can be expressed as a function (f(x)). In addition, a method of inversely estimating appropriate input parameters from the desired results of a prediction model can be considered as an optimization operation of the function. In this optimization operation, the derivative of the function is often used. In other words, the function is often differentiated. However, the objective function of a prediction model obtained by training using machine learning is usually non-symbolic, and in many cases cannot be differentiated, and may even be discontinuous. Therefore, it is difficult to calculate the derivative of this function using a numerical analysis method. In other words, it is difficult to directly inversely estimate input parameters using the objective function of the prediction model.

[0006] Therefore, there is a need to develop a more comprehensive and accurate method for identifying and optimizing multiple process parameters that can be used to inversely estimate process parameters that will result in a desired outcome. Summary of the Invention [Problem to be solved by the invention]

[0007] In view of the above, the present invention provides a method for calculating process parameters to solve the above-mentioned problems of the prior art. [Means for solving the problem]

[0008] The present invention provides a method for calculating process parameters used to confirm and optimize a plurality of process parameters in a process, the process parameters including a plurality of confirmed input parameters and a plurality of unconfirmed input parameters. The method for calculating process parameters includes the steps of: providing a trained predictive model, the trained predictive model being obtained by machine learning a dataset using a machine learning method, the dataset including a plurality of samples, each of the samples including a plurality of sample parameters, the trained predictive model being used to input the plurality of input parameters and generate a predicted result corresponding to the input parameters; setting a target result corresponding to the predicted result of the trained predictive model and providing the confirmed input parameters among the input parameters; and checking whether the predicted result generated by inputting each of the samples in the dataset into the trained predictive model matches the target result, and checking whether the predicted result matches the target result. forming a multidimensional subspace with the samples matching a target result as a plurality of anchor samples, and identifying an initial sample in the multidimensional subspace; obtaining a direction of the initial sample in the multidimensional subspace; increasing a step size along the direction for the sample parameters of the initial sample, inputting the confirmed input parameters and the sample parameters with the increased step size that do not correspond to the confirmed input parameters into the trained prediction model to confirm whether the generated prediction result matches the target result; and if the prediction result matches the target result, setting the sample parameters with the increased step size that do not correspond to the confirmed input parameters as the unconfirmed input parameters.

[0009] Identifying the initial sample in the multi-dimensional subspace includes determining the initial sample as a midpoint of a distance between at least two of the anchor samples in the multi-dimensional subspace.

[0010] Identifying the initial sample in the multidimensional subspace includes determining the initial sample as a linear combination of at least two of the anchor samples in the multidimensional subspace.

[0011] The step of obtaining the orientation of the initial sample within the multidimensional subspace includes a step of calculating, by the numerical analysis method, a second directional derivative of the initial sample along each of the anchor samples as the orientation of the initial sample.

[0012] The step of obtaining the orientation of the initial sample within the multidimensional subspace includes a step of setting the initial sample to an origin and calculating, by the numerical analysis method, a third directional derivative of the origin toward each of the anchor samples as the orientation of the initial sample.

[0013] The trained prediction model is represented by an objective function, and obtaining the orientation of the initial sample in the multi-dimensional subspace includes obtaining a secant directional derivative of the initial sample toward each of the anchor samples by calculating a change in a function value of the objective function based on a change in a distance of the initial sample toward each of the anchor samples through the numerical analysis method.

[0014] The trained predictive model is represented by an objective function. and obtaining the direction of the initial sample in the multidimensional subspace includes: increasing and adjusting the step size of the initial sample towards a directional derivative of each of the anchor samples in sequence; checking in sequence whether the function value of the objective function is closer to the target result after increasing and adjusting the step size of the initial sample towards the directional derivative of each of the anchor samples; if it is determined that the function value of the objective function is closer to the target result after increasing the step size of the initial sample towards the directional derivative of any of the anchor samples, increasing the step size of the initial sample in the directional derivative; if it is determined that the function value of the objective function is not closer to the target result after increasing the step size of the initial sample towards the directional derivative of any of the anchor samples, increasing the step size of the initial sample towards the directional derivative of another of the anchor samples and checking again whether the function value of the objective function is closer to the target result.

[0015] The step of increasing the step size along the direction for the sample parameters of the initial sample further includes the steps of: setting an initial size of the step size; increasing the step size along the direction for the sample parameters of the initial sample by the initial size, and then inputting the sample parameters to the trained prediction model, and checking whether the prediction result output from the trained prediction model approaches the target result; if the prediction result output from the trained prediction model approaches the target result, determining the step size to the initial size; if the prediction result output from the trained prediction model does not approach the target result, continuing to adjust the size of the step size until the prediction result approaches the target result, and increasing the step size along the direction for the sample parameters of the initial sample by the size, and then inputting the sample parameters to the trained prediction model, and determining the step size to the final size.

[0016] The trained prediction model includes a function of a process parameter waiting to be identified, an objective function, and at least one constraint. The objective function minimizes a difference between an output value of the function of the process parameter waiting to be identified and the target result, and a constrained optimization problem is modeled based on the function of the process parameter waiting to be identified, the objective function, the at least one constraint, and the prediction result. The method of calculating the process parameters obtains the unconfirmed input parameters by solving the optimization problem.

[0017] The process parameter calculation method of the present invention further includes a step of adding a barrier function to the objective function so as to constrain the process parameters to satisfy an inequality constraint condition.

[0018] The process parameter calculation method of the present invention further includes a step of adding a penalty function to the objective function so as to constrain the process parameters to satisfy an equality constraint condition.

[0019] The data set further includes a process data set, the process data set including a plurality of process samples, and the sample parameters of each of the process samples include a source parameter and an incubation parameter.

[0020] The source parameters of each of the process samples further include attribute data of the source of each of the process samples. The incubation parameters of each of the process samples further include human, equipment, material, method and environment parameters of each of the process samples.

[0021] The trained predictive model is obtained by machine learning the dataset using any of Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), and Recurrent Neural Networks (RNNs).

[0022] The process parameter calculation method of the present invention further includes the steps of: if the prediction result does not match the target result, checking the confirmed input parameters and the sample parameters of the sample again to select a second sample in the sample; calculating the gradient of the second sample in the objective function of the trained prediction model by the numerical analysis method; increasing the step size of the sample parameters of the second sample along the opposite direction of the gradient, and inputting all the sample parameters with the increased step size into the trained prediction model to check whether the generated prediction result matches the target result; and if the prediction result matches the target result, setting the sample parameters with the increased step size that do not correspond to the confirmed input parameters as unconfirmed input parameters.

[0023] In another aspect, the present invention provides a method for calculating process parameters used to confirm and optimize a plurality of process parameters in a process, the process parameters including a plurality of confirmed input parameters and a plurality of unconfirmed input parameters. The method for calculating process parameters includes the steps of: providing a plurality of trained prediction models, the trained prediction models being obtained by machine learning a data set using a machine learning method, the data set including a plurality of samples, each of the samples including a plurality of sample parameters, the trained models corresponding to a plurality of objective functions, the trained prediction models being used to input the plurality of input parameters to generate a plurality of predicted results corresponding to the input parameters, the predicted results respectively belonging to different target result types; combining the objective functions of the trained prediction models to form a composite objective function, and inputting the input parameters into the composite objective function to generate the predicted results; setting a plurality of target results corresponding to the predicted results of the trained prediction models and providing the confirmed input parameters among the input parameters; the sample parameters of the initial sample are increased along the direction, and the confirmed input parameters and the sample parameters with the increased step size that do not correspond to the confirmed input parameters are input to the composite objective function to confirm whether the generated prediction result matches the target result; and if the prediction result matches the target result, the sample parameters with the increased step size that do not correspond to the confirmed input parameters are set as unconfirmed input parameters.

[0024] The method for calculating process parameters in the present invention further includes the step of adding a barrier function to the composite objective function so as to constrain the process parameters to satisfy an inequality constraint condition.

[0025] The process parameter calculation method of the present invention further includes the step of adding a penalty function to the composite objective function so as to constrain the process parameters to satisfy an equality constraint condition. Effect of the Invention

[0026] In summary, the process parameter calculation method of the present invention regards the trained prediction model as a function of the process parameters to be identified and constructs an optimization model. The objective function is to minimize the difference between the function value and the target result, and is subject to constraints of the constraints to be satisfied for each of the process parameters to be identified or between them. In addition, in order to find a solution to the above-mentioned constrained optimization problem, a method including a penalty function and a barrier function is used to convert the constrained optimization problem into an unconstrained optimization problem, making it easier to find a solution. For this optimization problem, first, from the data set when the trained prediction model is trained, a number of samples that match the target result are found and used as anchors to form a multidimensional feasible solution space. In addition, an initial solution is set or created in this feasible solution space, and then a better feasible solution is found by finding the direction and step size using a different numerical analysis method. Then, this better feasible solution becomes a usable input parameter. In addition, the process effect prediction and optimization method of the present invention can be applied to the cell process field, as well as to some other trained machine learning model, and can be used for reverse estimation of process parameters and unconfirmed input parameters. [Brief description of the drawings]

[0027] [Figure 1] FIG. 1 shows a flow chart of the steps of a method for calculating process parameters according to an embodiment of the present invention. [Diagram 2]FIG. 2 is a flow chart of further steps of a method for calculating process parameters according to some illustrative embodiments of the present invention. [Diagram 3] FIG. 3 is a flow chart of further steps of a method for calculating process parameters according to some illustrative embodiments of the present invention. [Figure 4] FIG. 4 shows a flow chart of a direction determination step in a process parameter calculation method according to some illustrative embodiments of the present invention. [Diagram 5] FIG. 5 shows a flow chart of a direction determination step in a process parameter calculation method according to some illustrative embodiments of the present invention. [Figure 6] FIG. 6 shows a flow chart of a direction determination step in a process parameter calculation method according to another specific embodiment of the present invention. [Figure 7] FIG. 7 shows a flow chart of a direction determination step in a process parameter calculation method according to another specific embodiment of the present invention. [Figure 8] FIG. 8 shows a flow chart of steps of a method for calculating process parameters according to another embodiment of the present invention. [Figure 9] FIG. 9 shows a flow chart of steps of a method for calculating process parameters according to another embodiment of the present invention. [Figure 10] FIG. 10 shows a flow chart of steps of a method for calculating process parameters according to another embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0028] In order to make the advantages, spirits and features of the present invention easier and clearer to understand, the present invention will be described and discussed in detail with reference to the drawings, using specific examples. It should be noted that these specific examples are merely representative specific examples of the present invention, and the specific methods, devices, conditions, materials, etc. illustrated do not limit the present invention or the corresponding specific examples. In addition, each component in the drawings is used only to represent their relative positions, and is not described based on actual ratios. In addition, the step numbers of the present invention are only used to distinguish different steps, and do not represent the order of steps. The above will be explained in advance.

[0029] Please refer to Fig. 1. Fig. 1 shows a flow chart of steps of a process parameter calculation method according to a specific embodiment of the present invention. The process parameter calculation method in this specific embodiment is used to confirm and optimize multiple process parameters in a process. The process parameters include multiple confirmed input parameters and multiple unconfirmed input parameters. The confirmed input parameters are input parameters that have been confirmed by a user during operation or have been actually implemented. And the unconfirmed input parameters are input parameters that have not been confirmed or implemented by a user. The process parameter calculation method includes:

[0030] Step S10: Provide a trained prediction model. The trained prediction model is obtained by machine learning a dataset through a machine learning method. The dataset includes a plurality of samples. Each sample includes a plurality of sample parameters. The trained prediction model is used to input the plurality of input parameters and generate a prediction result corresponding to the input parameters.

[0031] Step S11: Set a target result corresponding to the predicted result of the trained prediction model, and provide confirmed input parameters among the input parameters.

[0032] Step S12: Check whether the predicted result generated by inputting each sample in the dataset into the trained prediction model matches the target result, and form a multidimensional subspace with the samples whose predicted results match the target result as multiple anchor samples, and identify an initial sample in the multidimensional subspace.

[0033] Step S13: Obtain the direction of the initial sample in the multi-dimensional subspace.

[0034] Step S14: Increase the step size along the direction for the sample parameters of the initial sample, and then input the confirmed input parameters and other sample parameters with increased step size that do not correspond to the confirmed input parameters into the trained prediction model to check whether the generated prediction result is consistent with the target result.

[0035] Step S15: If the predicted result coincides with the target result, the sample parameters with increased step size that do not correspond to the confirmed input parameters are set as unconfirmed input parameters.

[0036] The trained prediction model in step S10 in this specific embodiment may be any model that has completed machine learning training and is available on an open platform, or may be a prediction model that has been trained by a user himself. The data set in this specific embodiment may be a set formed of any data that can be used for training, testing, and validation of machine learning. In practical applications, the process parameter calculation method in this specific embodiment may be applied to any other trained machine learning model for inverse estimation of unconfirmed input parameters. The trained prediction model in this specific embodiment is obtained by machine learning the data set using Artificial Neural Networks (ANN), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), or any other machine learning or neural network algorithm. The machine learning or neural network algorithm is selected according to the needs of the user.

[0037] In practice, a cell culture process may have many input parameters at the same time, including parameters that have been identified or implemented. The system and method of the present invention can be used to inversely estimate parameters that have not been implemented and execute the process to obtain a desired target result. For example, a user may be executing a cell culture process for 21 days, and the cell preparation to be produced must have a cell viability of 95%, and the user may have already executed the process for 7 days (i.e., already has confirmed input parameters), but may not have confirmed how to execute the process for the subsequent 8th to 21st days or adjust the process parameters (i.e., unconfirmed input parameters) to obtain the desired target result on the pre-set 21st day. Therefore, step S11 is a step of setting the inverse estimation conditions for the input parameters.

[0038] In step S12, a certain number of samples from the data set used to train the trained prediction model match the target result set by the user (for example, the above-mentioned cell preparation must have a cell viability of 95%). Therefore, these samples have considerable reference value in the inverse estimation of the input parameters, and can be used to form a multidimensional subspace as anchor samples. When points (representing vectors of input parameters) in this multidimensional subspace are input to the objective function of the trained prediction model, function values ​​that match the target result are obtained. Therefore, this multidimensional subspace can be a feasible solution space for the function of the trained prediction model. However, the function values ​​of the points in the feasible solution space merely match the target result, and are not optimal function values. Therefore, in step S12, it is necessary to further identify points in the feasible solution space that will be the initial points for finding a feasible solution. That is, it is necessary to identify an initial sample.

[0039] As described in steps S13 and S14, after identifying the initial sample, it is necessary to proceed from the initial sample in at least one direction and at least one step size toward a feasible solution. After determining the direction in step S13, proceed along the direction by at least one step size as described in step S14, and input the incremented sample parameters that do not correspond to the confirmed input parameters together into the trained prediction model to check whether the output function value matches the target result. It should be noted that the input parameters input to the trained prediction model described above are the confirmed input parameters and some parameters of the points in the feasible solution space. That is, the combined input parameters.

[0040] If the function value obtained by inputting the combined input parameters into the trained predictive model matches the target result, the point in the feasible solution space is a feasible solution. Thus, as described in step S15, some parameters of the point (i.e., other parameters that do not correspond to the confirmed input parameters) can be unconfirmed input parameters. For example, if the first 7 days of a 21-day cell culture process have been identified (confirmed input parameters), the unconfirmed input parameters identified in step S15 are the input parameters of the point from days 8 to 21.

[0041] In the above-mentioned specific embodiment, in step S12, an initial sample is identified in the multi-dimensional subspace, but the method of the present invention further includes several methods in the step of identifying the initial sample. Here, please refer to FIG. 2 and FIG. 3. FIG. 2 and FIG. 3 are flowcharts of further steps of the method for calculating process parameters according to several specific embodiments of the present invention. As shown in FIG. 2, this specific embodiment is different from the above-mentioned specific embodiment in that step S12 further includes step S120 and sub-step S122. Step S120 is a step of finding several anchor samples from the data set in the above-mentioned step S12 to form a multi-dimensional subspace. In addition, in step S122, the midpoint of the distance between at least two anchor samples in the multi-dimensional subspace is taken as the initial sample. In addition, as shown in FIG. 3, this specific embodiment is different in that step S12 further includes step S120 and step S124. In step S124, a linear combination between at least two anchor samples in the multi-dimensional subspace is taken as the initial sample. The above-mentioned specific embodiments are only two of several methods for identifying an initial point in a feasible solution space. In practice, the present invention can use any method that can identify the initial point or initial sample. In addition, in the specific embodiment of Figure 2 and Figure 3, the steps corresponding to the specific embodiment described above are the same, so they will not be described in detail again here.

[0042] After the initial point is identified, it is necessary to determine the direction of travel from the initial point. Now, please refer to Fig. 4 and Fig. 5. Fig. 4 and Fig. 5 show a flow chart of the direction determination step in the process parameter calculation method according to several specific embodiments of the present invention. As shown in Fig. 4, this specific embodiment is different from the above-mentioned specific embodiment in the following points. That is, the process parameter calculation method further includes:

[0043] Step S132: Using a numerical analysis method, calculate the second directional derivative of the initial sample along each anchor sample as the direction of the initial sample.

[0044] 5, this specific embodiment is different from the above specific embodiment in the following respects: That is, the method for calculating the process parameters further includes:

[0045] Step S134: Set the initial sample as the origin, and use a numerical analysis method to calculate the third directional derivative of the origin toward each anchor sample as the direction of the initial sample.

[0046] Alternatively, in the specific embodiment of FIG. 4, the tangent direction of the initial sample toward the anchor sample may be selected as the direction of travel. That is, the second directional derivative may be calculated by a numerical analysis method. It should be noted that since there are multiple anchor samples, the initial sample also has multiple second directional derivatives for these anchor samples. In practice, one second directional derivative may be selected from the initial sample and advanced by one step size, and some input parameters of the advanced point may be combined with the confirmed input parameters and then input into the trained prediction model to check whether the output function value is better (i.e., whether the output function value is closer to the target result). Then, if the function value is better, this second directional derivative is set as the direction of the initial sample. On the other hand, if the function value is not better, the next second directional derivative may be selected, and similar steps may be performed until a second directional derivative that can provide a better function value is found. Also, in the specific embodiment of FIG. 5, the initial sample may first be set to the origin in the multidimensional subspace, and then the third directional derivative of the origin toward each anchor sample may be calculated by the numerical analysis method. Then, similar to the embodiment of Fig. 4, the third directional derivative that may give a better function value is selected as the direction of the initial sample. In the embodiment of Fig. 5, after moving one step from the initial sample, the new point reached in the feasible solution space may be used as the new starting point. That is, the current solution may be set as the origin again, and step S134 of Fig. 5 may be repeated to move one step again.

[0047] In the specific embodiment of FIG. 4, the tangential direction is the travel direction, but as shown in the specific embodiment of FIG. 6, the secant direction from the initial sample to the anchor sample may be the travel direction. FIG. 6 shows a flow chart of the direction determination step in the method for calculating process parameters according to another specific embodiment of the present invention. As shown in FIG. 6, this specific embodiment is different from the above-mentioned specific embodiment in the following points. That is, the trained prediction model in this specific embodiment can be expressed by an objective function. In addition, it further includes the following.

[0048] Step S136: Using a numerical analysis method, calculate the change in function value of the objective function based on the change in distance of the initial sample toward each anchor sample to obtain the secant direction derivative of the initial sample toward each anchor sample.

[0049] In the above-mentioned direction determination step, if the initial sample proceeds in the opposite direction to the feasible solution, the function value may become worse or not match the expectation. Therefore, according to the method of the present invention, the direction in which the function value may become better can be searched in sequence. Now, refer to FIG. 7. FIG. 7 shows a flow chart of the direction determination step in the method for calculating process parameters according to another specific embodiment of the present invention. As shown in FIG. 7, this specific embodiment is different from the above-mentioned specific embodiment in the following points. That is, the trained prediction model in this specific embodiment can be expressed by an objective function. In addition, it further includes the following steps.

[0050] Step S1380: Increase and adjust the step size of the initial sample in sequence towards the directional derivative of each anchor sample.

[0051] Step S1382: After increasing and adjusting the step size of the initial sample towards the directional derivative of each anchor sample, check in sequence whether the function value of the objective function is better (i.e., whether the function value of the objective function is closer to the target result).

[0052] Step S1384: If it is confirmed that the function value of the objective function becomes better after increasing the step size of the initial sample toward the directional derivative of any of the anchor samples, then increase the step size of the initial sample.

[0053] Step S1386: If it is determined that the function value of the objective function is not better after increasing the step size of the initial sample toward the directional derivative of one of the anchor samples, then it is checked again whether the function value of the objective function is better after increasing the step size of the initial sample toward the directional derivative of another anchor sample.

[0054] Therefore, in this specific embodiment, first, the step size is repeatedly adjusted from the initial sample toward the first anchor sample, and it is confirmed whether the function value becomes better after the step size is increased. Then, if an appropriate step size is found, the step size may be advanced along that direction. However, if an appropriate step size for which the function value becomes better cannot be found even after adjusting the step size a certain number of times, the above adjustment is repeated again for the initial sample toward the next anchor sample. Note that, for the above step size adjustment, an initial size may be set for the step size first, and it may be confirmed whether the prediction result output after inputting the trained prediction model approaches the target result after the progress of the initial sample has this initial size. Then, if the target result is further approached, it is possible to proceed with the step size of this initial size. On the other hand, if the target result is not approached, the size of the step size may be repeatedly adjusted until the prediction result output from the trained prediction model approaches the target result, and the initial sample may be advanced with the step size approaching the target result. In practice, if we cannot find a suitable step size that gives a better function value in all anchor sample directions (e.g., after adjusting the step size 10 times), we stop the process parameter calculation method. The above steps are cyclical search steps. When the initial point has progressed to the next point in the feasible solution space, we may set the new point as the current solution and perform the above steps again.

[0055] When the process parameter calculation method in this specific embodiment is applied to the field of cell culture process, the data set in this specific embodiment may further include a cell data set. The cell data set includes a plurality of cell samples. And the sample parameters of each cell sample include source parameters and culture parameters. In practice, the cell data set may be any data obtained from an open platform, or may be data collected by the user himself. In addition, the types of cell samples may be immune cells (e.g., dendritic cells (DC cells), cytokine-induced killer cells (CIK), tumor-infiltrating lymphocytes (TIL), natural killer cells (NK cells) and CAR-T cells), stem cells (e.g., peripheral blood stem cells, adipose stem cells, bone marrow mesenchymal stem cells), chondrocytes, fibroblasts, etc., but are not limited thereto in actual applications and may be determined according to the type of cell culture that the user wants to perform. In addition, in this specific embodiment, the source parameters of each cell sample may further include attribute data of the source of each cell sample. Each cell sample in the cell data set may include a cell source corresponding to the cell and attribute data of the source. The attribute data may be physiological data of the source or other data related to the source, such as the source's gender, age, medical history, living environment, residential area, etc. However, in practical applications, the source parameters of the cell sample may further include other parameters related to the cell source that may affect cellular processes.

[0056] In addition, in this specific embodiment, the culture parameters of each cell sample further include human, equipment, material, method, and environment parameters for each cell sample. The cell culture process includes many steps, and each step is related to many culture parameters. The culture parameters include human-related parameters such as the gender and age of the cell source and the experience and stability of the cell culture operator, equipment-related parameters such as the type and grade of the cell operation platform, and the stability and accuracy of the temperature control and humidity control of the cell culture device, material-related parameters such as the material of the cell culture dish, the components and ratios and mixtures of the cell culture medium, method-related parameters such as the technique of the cell culture operator and the method of the cell culture process, and environment-related parameters such as the environmental temperature, humidity, and carbon dioxide concentration of the cell culture environment.

[0057] In the process, each process parameter is not arbitrarily set, but may have various constraints according to the actual situation. The trained prediction model includes a function of the waiting-to-be-identified process parameters, an objective function, and at least one constraint. The objective function is for minimizing the difference between the output value of the function of the waiting-to-be-identified process parameters and the target result, and a constrained optimization problem is modeled based on the function of the waiting-to-be-identified process parameters, the objective function, the at least one constraint, and the prediction result. The process parameter calculation method obtains the unconfirmed input parameters by solving the optimization problem. Here, please refer to FIG. 8. FIG. 8 shows a flowchart of steps of the process parameter calculation method according to another specific embodiment of the present invention. As shown in FIG. 8, this specific embodiment is different from the above-mentioned specific embodiment in the following points. That is, the process parameter calculation method in this specific embodiment further includes the following:

[0058] Step S100: Provide a trained predictive model. The trained predictive model is obtained by machine learning the dataset using a machine learning method. The trained predictive model includes a function of a to-be-identified process parameter, an objective function, and at least one constraint. The objective function includes a barrier function. The dataset includes a plurality of samples. Also, each sample includes a plurality of sample parameters. The trained predictive model is used to input a plurality of input parameters and generate a predictive result corresponding to the input parameters.

[0059] In this specific embodiment, the barrier function is set to constrain the process parameters to satisfy the inequality constraint. The barrier function includes a barrier region and a barrier function value. When the process parameters satisfy the inequality constraint, the process parameters are outside the barrier region, and the corresponding barrier function value is 0. That is, the prediction result of the trained prediction model is not affected. On the other hand, when the process parameters do not satisfy the inequality constraint, the process parameters are within the barrier region, and the corresponding barrier function value is large, which greatly affects the prediction result of the trained prediction model. It should be noted that the other steps in this specific embodiment are the same as the corresponding steps in the above-mentioned specific embodiment, and therefore will not be described in detail here.

[0060] Also refer to Figure 9. Figure 9 shows a flow chart of steps of a method for calculating process parameters according to another specific embodiment of the present invention. As shown in Figure 9, this specific embodiment is different from the above-mentioned specific embodiment in the following points. That is, the method for calculating process parameters in this specific embodiment further includes:

[0061] Step S102: Provide a trained prediction model. The trained prediction model is obtained by machine learning the dataset through a machine learning method. The trained prediction model has an objective function and at least one constraint, where the objective function includes a penalty function. The dataset includes a plurality of samples. Also, each sample includes a plurality of sample parameters. The trained prediction model is used to input a plurality of input parameters and generate a prediction result corresponding to the input parameters.

[0062] In this specific embodiment, the penalty function is set to constrain the process parameters to satisfy the constraint condition of the equality. The penalty function includes a penalty region and a penalty function value. If the process parameters satisfy the constraint condition of the equality, the process parameters are in the penalty region, and the corresponding penalty function value is 0. On the other hand, if the process parameters do not satisfy the constraint condition of the equality, the process parameters are outside the penalty region, and the corresponding penalty function value is large, which greatly affects the prediction result of the trained prediction model. Thus, when a sample parameter is input to a trained prediction model having a barrier function and / or a penalty function, not only can it be confirmed whether the generated prediction result matches the target result, but also whether the input parameter satisfies the actual constraint condition. If the input process parameter does not satisfy the condition of the barrier function and / or the penalty function, the prediction result will be significantly deviated, indicating that the process parameter cannot be used because it does not satisfy the condition. It should be noted that the other steps in this specific embodiment are the same as the corresponding steps in the specific embodiment described above, and will not be described in detail again here.

[0063] The practical constraints mentioned above may be, for example, that in a cell process, the combined antibiotic concentration of Streptomycin and Amphotericin B added to the cell culture medium on day 10 must be 200 μg / mL or less, and the sum of the serum concentration values ​​in the cell culture medium on days 8 and 10 must be 20% or less.

[0064] In practice, penalty functions and barrier functions are commonly used in machine learning to restrict the parameters of a machine learning model to a reasonable range so as to match the actual process situation in natural environmental conditions. For example, they are used to restrict the conditions under which the cell culture temperature and relative humidity are not negative and the component concentrations of the medium do not damage cells. By setting the barrier function and the penalty function to restrict the range of the parameters, the relationship between the parameters is comprehensively considered to find a combination of process parameters that is closest to the desired target result. In addition, in practical applications, the barrier function may be applied to the constraint of unconfirmed input parameters, and the penalty function may be applied to the constraint of confirmed input parameters. These can be used interchangeably according to the needs of the user. In this specific embodiment, the objective function minimizes the difference value between the function value and the target result, and is subject to the constraints of each of the process parameters waiting to be identified or between them. In addition, in order to find a solution to the above constrained optimization problem, a method including the penalty function and the barrier function is used to convert the constrained optimization problem into an unconstrained optimization problem, making it easier to find a solution.

[0065] Also, in practice, multiple different trained predictive models can be used to predict different outcomes of input parameters, such as cell viability, number of proliferated cells, and culture time. Now, refer to FIG. 10. FIG. 10 shows a flow chart of steps of a process parameter calculation method according to another specific embodiment of the present invention. The process parameter calculation method in this specific embodiment is used to confirm and optimize multiple process parameters in a process. The process parameters include multiple confirmed input parameters and multiple unconfirmed input parameters. The process parameter calculation method includes:

[0066] Step S10': provide a plurality of trained prediction models. The trained prediction models are obtained by machine learning the datasets respectively through a machine learning method. The datasets include a plurality of samples. And each sample includes a plurality of sample parameters. The trained models respectively correspond to a plurality of objective functions. The trained prediction models are used to input a plurality of input parameters to generate a plurality of prediction results corresponding to the input parameters. The prediction results respectively belong to different target result types.

[0067] Step S11': The objective functions of the trained prediction models are combined to form a composite objective function, and the prediction result is generated by inputting the input parameters into the composite objective function.

[0068] Step S12': Setting a plurality of target results corresponding to the predicted results of the trained prediction model, and providing confirmed input parameters among the input parameters.

[0069] Step S13': input each sample in the data set into the trained prediction model to check whether the generated prediction results match the target results, and form a multidimensional subspace using the samples whose prediction results match the target results as multiple anchor samples, and identify initial samples in the multidimensional subspace.

[0070] Step S14': Obtain the direction of the initial sample in the multi-dimensional subspace.

[0071] Step S15': Increase the step size along the direction for the sample parameters of the initial sample, and input the confirmed input parameters and the sample parameters with the increased step size that do not correspond to the confirmed input parameters into the composite objective function to check whether the generated prediction result is consistent with the target result.

[0072] Step S16': If the predicted result coincides with the target result, the sample parameters with increased step size that do not correspond to the confirmed input parameters are set as unconfirmed input parameters.

[0073] According to the process parameter calculation method in this specific embodiment, each trained prediction model can back-estimate a group of input parameters to simultaneously meet different desired results (e.g., the cell viability of the cell product after culture is higher than 90% and the cell number is 1×10 7 (The number of input parameters is more than 1 and the culture time is 7 days.) In this specific embodiment, the objective functions of the trained prediction models can be combined in any manner, such as by directly adding them to each other or forming a new function. Then, by performing inverse estimation on this new function by the various steps described in the present invention, a group of input parameters that can simultaneously satisfy the above conditions can be obtained. In another specific embodiment, the trained prediction models may also add a barrier function and / or a penalty function to the composite objective function of the trained prediction models, each of which may use the above method. By adding a barrier function or a penalty function to the objective function, relatively complex constraints are transferred to the objective function, making it easier to find a solution.

[0074] In the above specific embodiment, the objective functions of each trained predictive model are directly added and combined to form a new function. That is, the weight value of the objective function of each trained predictive model is set to 1. However, in practical applications, the weight value of the objective function of each trained predictive model may be adjusted according to the priority or importance order of the desired results of each process. In addition, the barrier function and the penalty function included in each objective function of each trained predictive model may also be set and adjusted according to the priority of the desired results of the process. In addition, the construction of an optimization problem for multiple trained models can be achieved in the following two ways.

[0075] First method: When each trained predictive model holds a constrained optimization problem, first, normalize the objective function of each trained predictive model (e.g., convert the objective function of each trained predictive model into the form of a minimization objective function), and assign weight values ​​to each to form a composite objective function. Then, consider and integrate the constraints on the composite objective function (i.e., obtain a combination of constraints included in each trained predictive model), and convert the constrained optimization problem into an unconstrained optimization problem. For example, a new function is formed by processing equality constraints with a penalty function and processing inequality constraints with a barrier function. Note that the constraints included in the objective function of each trained predictive model have different weight values ​​according to the priority of the desired result of the process. Each constraint is integrated and converted into processing by a barrier function and processing by a penalty function, and weight values ​​of the barrier function and penalty function are derived by considering the objective function of each trained predictive model.

[0076] Second method: When the optimization problem of each trained prediction model is converted into an unconstrained optimization problem (i.e., when all the constraints included in each trained prediction model are processed by a barrier function and / or a penalty function and added to the objective function), there is no need to adjust the constraints, and the objective function of each trained prediction model is normalized and the corresponding weight value is adjusted to form a composite objective function. Then, the barrier function and / or the penalty function included in the objective function of each trained prediction model are added to form a new function. Note that the objective function of each trained prediction model may include each barrier function and penalty function. In addition, the barrier function and the penalty function can be assigned a corresponding weight value according to the priority order of the objective function. It should be noted that the other steps in this specific embodiment are the same as the corresponding steps in the specific embodiment described above, and therefore will not be described in detail here.

[0077] In summary, the process parameter calculation method of the present invention regards the trained prediction model as a function of the process parameters to be identified and constructs an optimization model. The objective function is to minimize the difference between the function value and the target result, and is subject to constraints of the constraints to be satisfied for each of the process parameters to be identified or between them. In addition, in order to find a solution to the above-mentioned constrained optimization problem, a method including a penalty function and a barrier function is used to convert the constrained optimization problem into an unconstrained optimization problem, making it easier to find a solution. For this optimization problem, first, from the data set when the trained prediction model is trained, a number of samples that match the target result are found and used as anchors to form a multidimensional feasible solution space. In addition, an initial solution is set or created in this feasible solution space, and then a better feasible solution is found by finding the direction and step size using a different numerical analysis method. Then, this better feasible solution becomes a usable input parameter. In addition, the process effect prediction and optimization method of the present invention can be applied to the cell process field, as well as to some other trained machine learning model, and can be used for reverse estimation of process parameters and unconfirmed input parameters.

[0078] The detailed description of the preferred specific embodiments above is intended to more clearly describe the features and spirit of the present invention, and is not intended to limit the scope of the present invention to the preferred specific embodiments disclosed above. Rather, it is intended that various modifications and equivalent configurations be included within the scope of the claims that are to be claimed in the present invention. Therefore, the scope of the claims that are to be claimed in the present invention should be interpreted in the broadest possible manner based on the above description so as to include all possible modifications and equivalent configurations. [Explanation of symbols]

[0079] S10~S15, S120~S124, S132~S136 Steps S10'~S16', S1380~S1386 Steps

Claims

1. 1. A method of process parameter calculation for use in identifying and optimizing a plurality of process parameters in a process, the process parameters including a plurality of identified input parameters and a plurality of unidentified input parameters, the method of process parameter calculation comprising: providing a trained prediction model, the trained prediction model being obtained by machine learning a dataset by a machine learning method, the dataset including a plurality of samples, and each of the samples including a plurality of sample parameters, the trained prediction model being used to input a plurality of input parameters and generate a prediction result corresponding to the input parameters; setting a target outcome corresponding to the predicted outcome of the trained predictive model and providing the verified input parameters among the input parameters; A step of inputting each of the samples in the dataset into the trained prediction model to check whether the predicted result generated matches the target result, forming a multidimensional subspace using the samples whose predicted result matches the target result as multiple anchor samples, and identifying an initial sample in the multidimensional subspace; obtaining an orientation of the initial sample within the multi-dimensional subspace; increasing the step size along the direction for the sample parameters of the initial sample, inputting the confirmed input parameters and the sample parameters with the increased step size that do not correspond to the confirmed input parameters into the trained prediction model, and verifying whether the generated prediction result is consistent with the target result; if the predicted result matches the target result, then determining that the sample parameter with the increased step size that does not correspond to the confirmed input parameter is the unconfirmed input parameter; The method includes:

2. The step of identifying the initial sample within the multi-dimensional subspace further comprises: The method of claim 1 , further comprising the step of determining, as the initial sample, a midpoint of a distance between at least two of the anchor samples in the multi-dimensional subspace.

3. The step of identifying the initial sample within the multi-dimensional subspace further comprises: The method of claim 1 , further comprising the step of: determining, as the initial sample, a linear combination of at least two of the anchor samples in the multi-dimensional subspace.

4. The step of obtaining the orientation of the initial sample within the multi-dimensional subspace further comprises:

2. The method of claim 1, further comprising the step of: calculating, by a numerical analysis method, a second directional derivative of the initial sample along each of the anchor samples as the direction of the initial sample.

5. The step of obtaining the orientation of the initial sample within the multi-dimensional subspace further comprises:

2. The method of claim 1, further comprising the steps of: setting the initial sample at an origin; and calculating, by a numerical analysis method, a third directional derivative of the origin toward each of the anchor samples as the direction of the initial sample.

6. The trained prediction model is expressed by an objective function, and obtaining the orientation of the initial sample in the multi-dimensional subspace further comprises:

2. The method of claim 1, further comprising: calculating a change in a function value of the objective function based on a change in a distance of the initial sample toward each of the anchor samples by a numerical analysis method to obtain a secant derivative of the initial sample toward each of the anchor samples.

7. The trained prediction model is expressed by an objective function, and obtaining the orientation of the initial sample in the multi-dimensional subspace further comprises: Incrementing and adjusting the step size of the initial sample towards the directional derivative of each of the anchor samples in turn; after increasing and adjusting the step size of the initial sample towards the directional derivative of each of the anchor samples, checking in turn whether the function value of the objective function gets closer to the target result; increasing the step size of the initial sample in the directional derivative if it is determined that the function value of the objective function approaches the target result even more after increasing the step size of the initial sample towards the directional derivative of any of the anchor samples; if it is determined that the function value of the objective function does not get any closer to the target result after increasing the step size of the initial sample towards the directional derivative of one of the anchor samples, then checking again whether the function value of the objective function gets any closer to the target result after increasing the step size of the initial sample towards the directional derivative of another of the anchor samples; 2. The method of claim 1, comprising:

8. The step of increasing the step size along the direction for the sample parameter of the initial sample further comprises: setting an initial size of the step size, increasing the step size by the initial size along the direction for the sample parameters of the initial samples, and then inputting the sample parameters to the trained prediction model, and checking whether the prediction result output from the trained prediction model approaches the target result; If the prediction result output from the trained prediction model approaches the target result, determining the step size to be the initial size; if the prediction result output from the trained prediction model does not approach the target result, continuously adjust the size of the step size until the prediction result approaches the target result, and increase the step size by the size along the direction for the sample parameters of the initial sample, and then input them into the trained prediction model to determine the step size to the final size; 2. The method of claim 1, comprising:

9. 2. The method of claim 1, wherein the trained prediction model includes a function of a process parameter waiting to be identified, an objective function, and at least one constraint condition, the objective function minimizing a difference value between an output value of the function of the process parameter waiting to be identified and the target result, a constrained optimization problem is modeled based on the function of the process parameter waiting to be identified, the objective function, the at least one constraint condition, and the prediction result, and the method of calculating the process parameters comprises obtaining a solution to the optimization problem to obtain the unconfirmed input parameters.

10. 10. The method of claim 9, further comprising the step of adding a barrier function to the objective function to constrain a relationship between the process parameters to satisfy an inequality constraint.

11. 10. The method of claim 9, further comprising the step of adding a penalty function to the objective function to constrain the process parameters to satisfy an equality constraint.

12. 2. The method of claim 1, wherein the data set further comprises a process data set, the process data set comprising a plurality of process samples, and the sample parameters of each of the process samples comprise a source parameter and an incubation parameter.

13. The method for calculating process parameters according to claim 12, wherein the source parameters of each of the process samples further include attribute data of the source of each of the process samples, and the culture parameters of each of the process samples further include parameters of people, equipment, materials, methods, and environments in each of the process samples.

14. 2. The method of claim 1, wherein the trained predictive model is obtained by machine learning the data set using one of Artificial Neural Networks (ANN), Convolutional Neural Networks (CNNs), and Recurrent Neural Networks (RNNs).

15. Furthermore, if the predicted outcome does not match the target outcome, then again matching the verified input parameters and the sample parameters of the sample to select a second sample within the sample; calculating, by a numerical analysis method, a gradient of the second sample in an objective function of the trained predictive model; increasing the step size along the opposite direction of the gradient for the sample parameters of the second sample, inputting all the sample parameters with the increased step size into the trained prediction model to check whether the generated prediction results match the target results; if the predicted result matches the target result, then determining that the sample parameter with the increased step size that does not correspond to the confirmed input parameter is the unconfirmed input parameter; 2. The method of claim 1, comprising:

16. 1. A method of process parameter calculation used to identify and optimize a plurality of process parameters in a process, the process parameters including a plurality of identified input parameters and a plurality of unconfirmed input parameters, the method of process parameter calculation comprising: providing a plurality of trained prediction models, each of which is obtained by machine learning a dataset using a machine learning method, the dataset including a plurality of samples, each of which includes a plurality of sample parameters, the trained models corresponding to a plurality of objective functions, the trained prediction models being used to input a plurality of input parameters to generate a plurality of predicted results corresponding to the input parameters, the predicted results belonging to different target result types, respectively; combining the objective functions of the trained predictive models to form a composite objective function, and inputting the input parameters into the composite objective function to generate the prediction result; setting a plurality of target outcomes corresponding to the predicted outcomes of the trained predictive model and providing the verified input parameters among the input parameters; A step of inputting each of the samples in the dataset into the trained prediction model to check whether the prediction results generated match the target results, forming a multidimensional subspace using the samples whose prediction results match the target results as multiple anchor samples, and identifying an initial sample in the multidimensional subspace; obtaining an orientation of the initial sample within the multi-dimensional subspace; increasing the step size along the direction for the sample parameters of the initial sample, inputting the confirmed input parameters and the sample parameters with the increased step size that do not correspond to the confirmed input parameters into the composite objective function to determine whether the predicted results generated match the target results; if the predicted result matches the target result, then determining that the sample parameter with the increased step size that does not correspond to the confirmed input parameter is the unconfirmed input parameter; The method includes:

17. 17. The method of claim 16, further comprising adding a barrier function to the composite objective function to constrain the process parameters to satisfy an inequality constraint.

18. 17. The method of claim 16, further comprising the step of adding a penalty function to the composite objective function to constrain relationships between the process parameters to satisfy an equality constraint.

Images