Product production process modeling method, simulation method, device, equipment and medium

Abstract

The present application relates to the technical field of industrial modeling, and discloses a product production process modeling method, a simulation method, a device, equipment and a medium, the modeling method comprising: acquiring a plurality of process variables and a plurality of quality variables of product production; acquiring numerical values of the plurality of process variables and the plurality of quality variables; for each process variable, calculating a comprehensive index of the process variable under the plurality of quality variables according to the numerical value of the process variable and the numerical value of the quality variable; classifying the process variable based on the comprehensive index corresponding to the process variable and a set threshold value, into a quality high correlation variable, a quality low correlation variable and a quality irrelevant variable; constructing a deep learning model; and the deep learning model generating a corresponding quality variable prediction result based on the quality high correlation variable and the quality low correlation variable.The present application can quickly and accurately model the industrial product production process.

Description

Technical Field

[0001] This invention relates to the field of industrial modeling technology, specifically to a process modeling method, simulation method, device, equipment, and medium for product manufacturing. Background Technology

[0002] Timely monitoring and prediction of various state variables in industrial systems are prerequisites for maintaining stable operation of industrial equipment and improving production efficiency. Modern industry is characterized by system complexity, strong nonlinearity, and high coupling. Traditional mechanistic modeling is difficult to apply effectively to complex industrial systems.

[0003] For example, the production process of traditional Chinese medicine is complex and involves many process variables, such as the opening degree of various valves, container temperature, flow rate, and steam pressure. Therefore, modeling is difficult and inefficient. Summary of the Invention

[0004] In view of this, the present invention provides a process modeling method, simulation method, device, equipment and medium for product manufacturing, so as to solve the problems of difficult and inefficient modeling of industrial production processes.

[0005] In a first aspect, the present invention provides a process modeling method for product manufacturing, the method comprising:

[0006] Obtain multiple process variables and multiple quality variables for product manufacturing;

[0007] Acquire the values ​​of the multiple process variables and the values ​​of the multiple quality variables at the corresponding time points;

[0008] For each process variable, a comprehensive index of the process variable under multiple quality variables is calculated based on the value of the process variable and the value of the quality variable; wherein, the comprehensive index is used to indicate the correlation between the process variable and the quality variable;

[0009] Based on the comprehensive index corresponding to the process variable and the set threshold, the process variable is classified; wherein, the classified categories include high-quality related variables, low-quality related variables, and quality-irrelevant variables;

[0010] A deep learning model is constructed; the deep learning model includes two first network structures and one second network structure, the inputs of the two first network structures are the high-quality correlated variables and the low-quality correlated variables, respectively; the second network structure generates the corresponding quality variable prediction results based on the outputs of the two first network structures.

[0011] In one optional implementation, the step of calculating a comprehensive index of the process variable under multiple quality variables based on the value of the process variable and the value of the quality variable for each process variable includes:

[0012] For each process variable, the maximum information coefficient between the process variable and multiple quality variables is calculated based on the value of the process variable and the value of the quality variable.

[0013] Using the Bayesian fusion method, the maximum information coefficients of the process variable and multiple quality variables are fused to obtain the comprehensive index corresponding to the process variable.

[0014] In one optional implementation, the step of calculating the maximum information coefficient between the process variable and multiple quality variables for each process variable, based on the value of the process variable and the value of the quality variable, includes:

[0015] For the first process variable among the process variables and the first mass variable among the mass variables, the multiple values ​​of the first process variable and the multiple values ​​of the first mass variable are divided into a grid in a two-dimensional space;

[0016] By statistically analyzing the data within each grid, the joint probability density of the first process variable and the first quality variable falling into the grid is obtained;

[0017] Mutual information is calculated and normalized based on the joint probability density corresponding to each grid.

[0018] By iterating through various different grid partitioning methods, the corresponding normalized mutual information is obtained;

[0019] The values ​​of the maximum information coefficients for the first process variable and the first quality variable are determined based on the maximum mutual information after normalization.

[0020] In one optional implementation, the step of using a Bayesian fusion method to fuse the maximum information coefficients of the process variable and multiple quality variables to obtain the comprehensive index corresponding to the process variable includes:

[0021] For the second quality variable among the quality variables, calculate the mean of the maximum information coefficient of the second quality variable and each of the process variables respectively;

[0022] Based on the mean of the maximum information coefficients of the second quality variable and each of the process variables, calculate the prior conditional probability density of the second process variable and each of the quality variables;

[0023] Based on the prior conditional probability density, calculate the posterior probability density of the second process variable and each of the quality variables;

[0024] The comprehensive index of the second process variable is calculated based on the prior conditional probability density and the posterior probability density.

[0025] In one optional implementation, classifying the process variables based on the comprehensive index corresponding to the process variables and a set threshold includes:

[0026] For each of the process variables, the threshold is set using the cumulative variance percentage for the comprehensive index.

[0027] The process variables are classified according to the set threshold.

[0028] In one optional implementation, the first network structure is a Transformer network structure, including: a multi-head attention layer, a first residual connection normalization layer, a second residual connection normalization layer, a first feedforward neural network layer, a second feedforward neural network layer, a first discard layer, and a second discard layer.

[0029] In one optional implementation, the input of the multi-head attention layer is either the high-quality correlated variable or the low-quality correlated variable; the output of the multi-head attention layer is connected to the first discard layer; the output of the first discard layer is connected to the first residual connection normalization layer; the output of the first residual connection normalization layer is sequentially connected to the first feedforward neural network layer and the second feedforward neural network layer; the output of the second feedforward neural network layer is connected to the second discard layer; and the output of the second discard layer is connected to the second residual connection normalization layer.

[0030] In one alternative implementation, after constructing the deep learning model, the method further includes:

[0031] The collected values ​​of multiple process variables are used as training samples, and the collected values ​​of multiple quality variables at corresponding time points are used as training labels to perform supervised training on the deep learning model.

[0032] The trained deep learning model is evaluated using metrics including one or more of the following: mean absolute error, mean square error, root mean square error, and coefficient of determination.

[0033] In one optional implementation, acquiring the values ​​of the plurality of process variables and the values ​​of the plurality of quality variables at corresponding times includes:

[0034] Obtain the initial values ​​of the process variables and the initial values ​​of multiple quality variables at corresponding times;

[0035] The initial values ​​of the process variable and the initial values ​​of the quality variable are preprocessed to obtain the values ​​of the process variable and the quality variable; wherein the preprocessing includes one or more of the following: removing missing values, removing outliers, and standardization.

[0036] Secondly, the present invention provides a process simulation method for product manufacturing, the method comprising:

[0037] Obtain a deep learning model, wherein the deep learning model is a model constructed using the product manufacturing process modeling method described in the first aspect above or any corresponding embodiment;

[0038] Obtain the set values ​​of the variables with high quality correlation and the variables with low quality correlation in the process variables;

[0039] The set values ​​of the high-quality correlated variables and the low-quality correlated variables are input into the deep learning model, and the corresponding quality variable prediction results are output.

[0040] Thirdly, the present invention provides a process modeling apparatus for product manufacturing, the apparatus comprising:

[0041] The variable acquisition module is used to acquire multiple process variables and multiple quality variables in product manufacturing.

[0042] The numerical acquisition module is used to acquire the values ​​of multiple process variables and the values ​​of multiple mass variables at corresponding times;

[0043] The comprehensive index acquisition module is used to calculate a comprehensive index of each process variable under multiple quality variables based on the value of the process variable and the value of the quality variable; wherein the comprehensive index is used to indicate the correlation between the process variable and the quality variable.

[0044] The variable classification module is used to classify the process variables based on the comprehensive index corresponding to the process variables and the set threshold; wherein the classified categories include high-quality correlated variables, low-quality correlated variables, and quality-irrelevant variables;

[0045] A model building module is used to build a deep learning model; the deep learning model includes two first network structures and one second network structure, the inputs of the two first network structures are the high-quality correlated variables and the low-quality correlated variables, respectively; the second network structure generates the corresponding quality variable prediction results based on the outputs of the two first network structures.

[0046] Fourthly, the present invention provides a process simulation device for product manufacturing, the device comprising:

[0047] The model acquisition module is used to acquire a deep learning model, which is a model constructed using the product manufacturing process modeling method described in the first aspect above or any corresponding embodiment.

[0048] The setpoint acquisition module is used to acquire the setpoints of variables with high quality correlation and variables with low quality correlation in the process variables;

[0049] The prediction module is used to input the set values ​​of the high-quality correlated variables and the low-quality correlated variables into the deep learning model and output the corresponding quality variable prediction results.

[0050] Fifthly, the present invention provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the product manufacturing process modeling method of the first aspect or any corresponding embodiment thereof, or to perform the product manufacturing process simulation method of the second aspect or any corresponding embodiment thereof.

[0051] In a sixth aspect, the present invention provides a computer-readable storage medium storing computer instructions, the computer instructions being used to cause a computer to execute the product manufacturing process modeling method of the first aspect or any corresponding embodiment thereof, or to execute the product manufacturing process simulation method of the second aspect or any corresponding embodiment thereof.

[0052] In a seventh aspect, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the process modeling method for product manufacturing described in the first aspect or any corresponding embodiment thereof, or to execute the process simulation method for product manufacturing described in the second aspect or any corresponding embodiment thereof.

[0053] The product manufacturing process modeling method, simulation method, device, equipment, and medium proposed in this invention provide a classification method for process variables to ensure the stable operation of the product manufacturing process model. A comprehensive (evaluation) index is established to accurately measure the relationship between individual process variables and all quality indicators, thereby accurately distinguishing process variables (i.e., quality-related variables) that affect key performance indicators (i.e., quality variables). Furthermore, a data-driven quality-related process modeling method for key performance indicators is proposed, which improves modeling efficiency and accuracy.

[0054] In addition, in the embodiments of the present invention, for the modeling of quality variables and quality-related variables, a parallel input method is adopted for high-quality-related variables and low-quality-related variables to ensure the utilization of data, thereby ensuring the accuracy of the model and thus ensuring that the established model has high prediction accuracy. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the specific embodiments or related technologies of the present invention, the drawings used in the description of the specific embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0056] Figure 1 This is a flowchart illustrating a product manufacturing process modeling method according to an embodiment of the present invention;

[0057] Figure 2 This is a partial network structure diagram of a deep learning model according to an embodiment of the present invention;

[0058] Figure 3 This is a flowchart illustrating another product manufacturing process modeling method according to an embodiment of the present invention;

[0059] Figure 4 This is one of the schematic diagrams comparing the predicted output and the actual output of a deep learning model according to an embodiment of the present invention;

[0060] Figure 5 This is a second schematic diagram comparing the predicted output and the actual output of a deep learning model according to an embodiment of the present invention.

[0061] Figure 6 This is a structural block diagram of a product manufacturing process modeling apparatus according to an embodiment of the present invention;

[0062] Figure 7 This is a structural block diagram of a product manufacturing process simulation device according to an embodiment of the present invention;

[0063] Figure 8 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Mutual information and maximum information coefficient (MIC) methods in related technologies can only assess the correlation between two variables, and cannot measure the comprehensive correlation between a single process variable and multiple key performance indicators. Furthermore, multivariate statistical methods in related technologies, such as partial least squares (PLS) and principal component regression (PCR), can only construct linear relationships, and kernel-based PLS and PCR suffer from high computational costs. Therefore, none of these methods are suitable for modern industrial systems that are complex, highly nonlinear, and highly coupled.

[0066] This invention proposes a modeling scheme for industrial system production processes, arguing that only certain variables can affect key performance indicators (quality variables). Therefore, effectively distinguishing between quality-related and quality-independent variables is beneficial for improving modeling accuracy and efficiency.

[0067] According to an embodiment of the present invention, a process modeling method for product manufacturing is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of executable computer instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0068] This embodiment provides a process modeling method for product manufacturing, which can be used with various computer devices. Figure 1 This is a flowchart of a product manufacturing process modeling method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0069] Step S101: Obtain multiple process variables and multiple quality variables for product manufacturing.

[0070] Specifically, product production can be, for example, the extraction of traditional Chinese medicine. Process variables and quality variables in product production can be quantities in the production system that can be directly or indirectly adjusted and that are subject to change. Process variables may affect quality variables, which are key performance indicators (KPIs). In the context of traditional Chinese medicine extraction, quality indicator variables (i.e., quality variables) can be selected based on the internal mechanism of the extraction process, and process variables can be chosen based on their characteristics. Specifically, based on the internal mechanism of the extraction process in traditional Chinese medicine production, the boiling point inside the extraction tank directly affects the quality of the subsequent medicinal solution, and the extraction tank temperature and extraction steam pressure directly reflect the boiling point inside the tank. Therefore, in the modeling process of the traditional Chinese medicine extraction process, these two variables are selected as quality indicators. Additionally, process variables can include, for example, valve opening, purified water temperature, and insulation steam pressure.

[0071] Step S102: Obtain the values ​​of the multiple process variables and the values ​​of the multiple quality variables at the corresponding time points.

[0072] Specifically, the values ​​of multiple process variables and multiple quality variables can be collected at each moment within the data acquisition period. At the same moment the process variable values ​​are acquired, the corresponding quality variable value is also acquired. Data acquisition can be achieved using sensors.

[0073] In some optional implementations, step S102, namely acquiring the values ​​of the multiple process variables and the values ​​of the multiple quality variables at corresponding times, includes:

[0074] Step S1021: Obtain the initial values ​​of the process variables and the initial values ​​of the multiple quality variables at the corresponding time points.

[0075] Specifically, the initial values ​​of the multiple process variables and the initial values ​​of the multiple quality variables obtained can be expressed as follows:

[0076] X = [x1, x2, ..., x N ] T ∈R N×n

[0077] Y = [y1, y2, ..., y N ] T ∈R N×m

[0078] Where x1,x2,...,x N Let y1, y2, ..., yN represent the initial numerical vectors of multiple process variables collected at times 1 to N. NThese represent the initial numerical vectors of multiple quality variables collected at times 1 to N, where n and m are the number of process variables and quality variables, respectively.

[0079] Step S1022: Preprocess the initial values ​​of the process variable and the initial values ​​of the quality variable to obtain the values ​​of the process variable and the quality variable; wherein, the preprocessing includes one or more of the following: removing missing values, removing outliers, and standardization.

[0080] In this embodiment of the invention, the collected data is preprocessed, for example, including removing missing and outlier values, and then standardization is performed. Specifically, first, all collected data is traversed, and missing and outlier values ​​are deleted. If, at any given moment, a process variable or quality variable has a missing or outlier value, then all data collected at that moment must be deleted. Based on this, the processed data is then standardized, as follows:

[0081] First, the mean vector is obtained by taking the mean of each column in X. The calculation formula is:

[0082]

[0083] Then, calculate the standard deviation of each column of data in X. The formula is as follows:

[0084]

[0085] Where, σ i Let x be the standard deviation of the i-th column of data in X. i,k For the data in column i, x i The kth data in the dataset.

[0086] Finally, based on the mean vector and the standard deviation, the data after removing missing and outlier values ​​is standardized. Specifically, the standardized process variable data matrix is ​​as follows:

[0087] X n =[x n,1 ,x n,2 ,...,x n,N ] T ∈R N×n

[0088] in, x n,i The transpose of the i-th row of data after standardization of process variable X, x n,i ∈R n×1 .

[0089] The mass variable matrix Y undergoes the same standardization process to obtain the standardized mass variable data matrix:

[0090] Y n =[y n,1 ,y n,2 ,...,y n,N ] T ∈R N×m .

[0091] Step S103: For each process variable, calculate a comprehensive index of the process variable under multiple quality variables based on the value of the process variable and the value of the quality variable; wherein the comprehensive index is used to indicate the correlation between the process variable and the quality variable.

[0092] In some optional implementations, step S103, namely, calculating the comprehensive index of the process variable under multiple quality variables based on the value of the process variable and the value of the quality variable for each process variable, includes:

[0093] Step S1031: For each process variable, calculate the maximum information coefficient (MIC) of the process variable and multiple quality variables based on the value of the process variable and the value of the quality variable.

[0094] In some optional implementations, step S1031, namely, calculating the maximum information coefficient between the process variable and multiple quality variables for each process variable based on the value of the process variable and the value of the quality variable, includes:

[0095] Step S10311: For the first process variable among the process variables and the first mass variable among the mass variables, divide the multiple values ​​of the first process variable and the multiple values ​​of the first mass variable in a two-dimensional space using a grid.

[0096] For example, for the first process variable (i.e., the l1st process variable) and the first quality variable (i.e., the l2nd quality variable), each with a value range of X. n and Y n Columns l1 and l2. For and The data space in which it is located is divided into a grid of k1×k2.

[0097] Step S10312: Statistically analyze the data within each grid to obtain the probability that the first process variable and the first quality variable fall into the grid, i.e., the joint probability density.

[0098] Step S10313: Calculate and normalize the mutual information based on the joint probability density corresponding to each grid. Specifically, the first process variable... (i.e., the l1st process variable) and the first mass variable y l2 Mutual information between (i.e., the l2th quality variable) The calculation formula is:

[0099]

[0100] in, This is the joint probability density (also known as the joint probability distribution); This represents a marginal probability distribution.

[0101] The formula for calculating mutual information normalization is:

[0102]

[0103] In this embodiment of the invention, normalizing the mutual information can avoid the influence of grid size on the results.

[0104] Step S10314: Iterate through various different grid partitioning methods to obtain the corresponding normalized mutual information. Specifically, the values ​​of k1 and k2 can be different in different grid partitioning methods, and there are multiple partitioning methods even when the values ​​of k1 and k2 are the same, and it is not necessary to partition according to equal width.

[0105] Step S10315: Based on the maximum mutual information after normalization, determine the value of the maximum information coefficient of the first process variable and the first quality variable.

[0106] Specifically, the value of the maximum mutual information after normalization can be taken as the value of the maximum information coefficient, that is... Here, B(N) is a constraint variable that limits the size of the grid division, and its size can be set to approximately 0.6 times the amount of data.

[0107] For different combinations of process variables and quality variables, the corresponding maximum information coefficient is calculated using the method described above.

[0108] Step S1032: Using the Bayesian Fusion (BF) method, the maximum information coefficients of the process variable and multiple quality variables are fused to obtain the comprehensive index corresponding to the process variable.

[0109] In some optional implementations, step S1032, namely, using the Bayesian fusion method to fuse the maximum information coefficients of the process variable and multiple quality variables to obtain the comprehensive index corresponding to the process variable, includes:

[0110] Step S10321: For the second quality variable among the quality variables, calculate the mean of the maximum information coefficient of the second quality variable and each of the process variables. The specific calculation formula is as follows:

[0111]

[0112] in, It is the mean of the maximum information coefficients of the l2th quality variable and all process variables (a total of n).

[0113] Step S10322: Calculate the prior conditional probability density of the second process variable and each of the process variables based on the mean of the maximum information coefficients of the second quality variable and each of the process variables.

[0114] For example, the prior conditional probability density of the l1st process variable and the l2nd quality variable. The calculation formula can be:

[0115]

[0116] Among them, R l2 This indicates that a certain process variable is related to the 12th quality indicator.

[0117] That is, the prior conditional probability density of the second process variable and the quality variable is determined based on the maximum information coefficient (value) of the second process variable and the quality variable, as well as the mean of the maximum information coefficients of the quality variable and each process variable.

[0118] Step S10323: Based on the prior conditional probability density, calculate the posterior probability density of the second process variable and each of the quality variables.

[0119] For example, the posterior probability density related to the l1st process variable and the l2nd quality variable The calculation formula can be:

[0120]

[0121] in, The calculation formula can be expressed as:

[0122]

[0123] in, This indicates that a certain process variable is unrelated to the l2th quality indicator. and This can be obtained through process variables. Exceeding and Not Exceeding The proportion of the number is determined. It can be calculated using the following formula:

[0124]

[0125] In other words, the posterior probability density of the second process variable and the quality variable is determined not only based on the prior conditional probability density of the second process variable and the quality variable, but also in combination with the probability density of the process variable. Exceeding and Not Exceeding The proportion of the number of processes, the maximum information coefficient (value) of the second process variable and the quality variable, and the mean of the maximum information coefficients of the quality variable and each process variable are used to determine the quality variable.

[0126] Step S10324: Based on the prior conditional probability density and the posterior probability density, calculate the comprehensive index of the second process variable. For example, the comprehensive index of the l1th process variable is:

[0127]

[0128] In this embodiment of the invention, a Bayesian fusion method is used to assign weights to the maximum information coefficients between a single process variable and different quality indicators.

[0129] In this embodiment of the invention, the Bayesian fusion method and the maximum information coefficient are combined to obtain a comprehensive (evaluation) index for each process variable under multiple quality indicators to quantitatively describe the correlation between process variables and quality variables.

[0130] In other embodiments, methods such as distance correlation coefficient can be used to determine the correlation between process variables and different quality variables.

[0131] Step S104: Based on the comprehensive index corresponding to the process variable and the set threshold, classify the process variable; wherein the classified categories include high-quality related variables, low-quality related variables, and quality-irrelevant variables.

[0132] In some optional implementations, step S104, namely classifying the process variables based on the comprehensive index corresponding to the process variables and the set threshold, includes:

[0133] Step S1041: For the comprehensive index corresponding to each process variable, the threshold is set using the Cumulative Percentage of Variance (CPV).

[0134] Step S1042: Classify the process variables according to the set threshold.

[0135] Specifically, the comprehensive indices corresponding to multiple process variables can be sorted in descending order, and then the CPV method can be used to select the optimal index. Process variables with a large proportion are considered high-quality relevant variables, and the selection criteria are as follows:

[0136]

[0137] It should be noted that the original order is shuffled after sorting in descending order, so the subscript of the comprehensive index BFMIC in the above formula is ω. i Instead of i, k3 represents the number of high-quality correlated variables selected. The remaining variables are then filtered out as low-quality correlated variables. Furthermore, if any composite index is 0, its corresponding variable is classified as a quality-irrelevant variable.

[0138] Step S105: Construct a deep learning model; the deep learning model includes two first network structures and one second network structure, the inputs of the two first network structures are the high-quality correlated variables and the low-quality correlated variables, respectively; the second network structure generates the corresponding quality variable prediction results based on the outputs of the two first network structures.

[0139] Some optional implementations, such as Figure 2 As shown, the first network structure is a Transformer network structure, including: a multi-head attention layer, a first residual connection normalization layer (Add&Norm), a second residual connection normalization layer, a first feedforward neural network layer, a second feedforward neural network layer, a first discard layer, and a second discard layer.

[0140] In this embodiment of the invention, the deep learning model used to predict quality variables is a parallel Transformer (i.e., PTransformer) network structure.

[0141] Transformer is a deep learning architecture based on a self-attention mechanism. Based on multi-head attention, Transformer can focus on the interactions between multiple variables in parallel, thus better uncovering multivariate relationships in industrial data.

[0142] In some optional implementations, the inputs of the multi-head attention layer are either the high-quality correlated variables or the low-quality correlated variables. Specifically, the input of a multi-head attention layer of one first network structure is the high-quality correlated variable, and the input of a multi-head attention layer of another first network structure is the low-quality correlated variable. The output of the multi-head attention layer is connected to the first discard layer to reduce overfitting and enhance the generalization ability of the model. The output of the first discard layer is connected to the first residual connection normalization layer. The output of the first residual connection normalization layer is sequentially connected to the first feedforward neural network layer and the second feedforward neural network layer. The output of the second feedforward neural network layer is connected to the second discard layer, and the output of the second discard layer is connected to the second residual connection normalization layer.

[0143] In addition, the second network structure may include a fully connected layer.

[0144] In some optional implementations, after constructing the deep learning model, the method further includes:

[0145] Step S106: Use the collected values ​​of multiple process variables as training samples and the collected values ​​of multiple quality variables at corresponding times as training labels to perform supervised training on the deep learning model.

[0146] Specifically, the numerical values ​​of process variables and quality variables used as training samples and training labels are not limited to the data used to classify process variables as described above.

[0147] Step S107: Evaluate the trained deep learning model. The evaluation metrics include Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Coefficient of Determination (R-squared). 2 One or more of the following.

[0148] Specifically, during training, Python (a high-level programming language) is used to simulate the model, that is, to build the model using Python and obtain the prediction results of the deep learning model. The deep learning model is evaluated using a test set (which can also be constructed using the collected values ​​of multiple process variables and multiple quality variables, but the data in the test set is different from the training samples).

[0149] The formula for calculating MAE is:

[0150]

[0151] Where ||·||1 represents the 1-norm, T represents the length of the test set, and y t,i and and represent the quality index and its predicted value at time i in the test set, respectively. The above formula calculates the average of the sum of the sample absolute errors.

[0152] The formula for calculating MSE is as follows:

[0153]

[0154] in This represents the square of the 2-norm. The above formula calculates the average of the sum of squared errors of the sample.

[0155] The RMSE calculation formula is as follows:

[0156]

[0157] The above formula calculates the square root of MSE.

[0158] R 2 The calculation formula is as follows:

[0159]

[0160] Where y t This represents the mean of the quality metrics in the test set.

[0161] In addition, the Adaptive Moment Estimation (ADAM) method is used for parameter optimization during model training, and the loss function used during training is Mean Squared Error (MSE).

[0162] In this embodiment, to ensure the stable operation of the product manufacturing process model, a classification method for process variables is proposed. A comprehensive (evaluation) index is established to accurately measure the relationship between individual process variables and all quality indicators, thereby accurately distinguishing process variables (i.e., quality-related variables) that affect key performance indicators (i.e., quality variables). Furthermore, a data-driven quality-related process modeling method for key performance indicators is proposed, which improves modeling efficiency and accuracy.

[0163] In addition, in the embodiments of the present invention, for the modeling of quality variables and quality-related variables, a parallel input method is adopted for high-quality-related variables and low-quality-related variables to ensure the utilization of data, thereby ensuring the accuracy of the model and thus ensuring that the established model has high prediction accuracy.

[0164] This invention proposes a novel method based on the Bayesian Fusion (BF) method and the Micro-Information Coefficient (MIC) index to accurately measure the correlation between a single process variable and all quality indicators, thereby accurately classifying the quality correlation of process variables. In other words, this invention provides a quality-related process modeling method based on the BFMIC-PTransformer (Bayesian Fusion Maximal Information Coefficient Parallel Transformer).

[0165] This embodiment provides a process modeling method for product manufacturing, specifically a process modeling method for the extraction section of traditional Chinese medicine, such as... Figure 3 As shown, the specific steps include the following:

[0166] Step 1: Determine the process variables and quality variables in the extraction process of traditional Chinese medicine;

[0167] Taking the honeysuckle extraction process in traditional Chinese medicine extraction as an example, data from 59 variables were collected by sensors installed throughout the process. The variable labels and names are shown in Table 1. Considering the internal relationship between the extraction process and subsequent traditional Chinese medicine production, the following issues exist: the boiling point in the extraction tank directly affects the quality of the effluent, which in turn directly affects the quality of the final traditional Chinese medicine product. Therefore, boiling point will be used as a quality variable in the extraction tank modeling process. The variables measured by the aforementioned sensors that directly reflect the boiling point in the tank are: extraction steam pressure and extraction tank temperature. In summary, the above variables are decomposed into process variables and quality variables required for modeling. Based on the above variable classification, the variables are divided according to a training set:test set ratio of 0.85:0.15.

[0168] Table 1. Extraction of process variable information

[0169]

[0170]

[0171] Step 2: Perform preprocessing operations on the collected data, including removing missing and outlier values, and standardizing the obtained data.

[0172] First, any outlier data in the collected data is checked and removed. Then, the mean and standard deviation of the training data are calculated, and these two metrics are used to standardize the training data. This standardization is then used as a basis for further standardizing the test data.

[0173] Step 3: Traditional MIC only quantifies the correlation between two single variables. However, the extraction process considers two quality indicator variables. Therefore, to consider the comprehensive correlation of each process variable with the two quality indicators, a BFMIC method is proposed, which integrates the comprehensive correlation of the two quality indicators based on BF.

[0174] The Business Factor (BF) method is applied to integrate the MICs (Minimum Indicators) of the obtained process variables with those of each quality indicator variable. First, the MIC between each process variable and each quality indicator variable is calculated. Based on this, for each process variable, a comprehensive index is derived using BF as the evaluation criterion for the current process variable, taking into account the two MICs obtained.

[0175] Step 4: For the two sets of process variables obtained in Step 3—the high-quality-related variables and the low-quality-related variables—a parallel input method is used to construct Transformer structures separately, in order to comprehensively consider the final impact of the two sets of variables on the quality index. Based on the above structures, the outputs obtained from the two Transformers are connected and integrated using fully connected layers to establish the final BFMIC-PTransformer prediction model.

[0176] The trained model is used to predict quality variables from the collected process variables. The predicted quality variables are then compared with the actual collected quality variable values. Figure 4 and Figure 5 As shown.

[0177] To verify the feasibility of the models proposed in the embodiments of this invention, Transformer, AE-Transformer, AE-LSTM, AE-BiLSTM, and BFMIC-PTransformer models were established respectively. Simulations were performed on each established model using the Python toolbox, and the prediction accuracy of each model was calculated. MAE, MSE, RMSE, and RMS were used as the metrics. 2 The specific evaluation indicators are shown in Tables 2, 3, and 4 below.

[0178] Table 2 shows the MAE, MSE, RMSE, and R of the extracted vapor pressure. 2 index

[0179]

[0180] Table 3 shows the MAE, MSE, RMSE, and R values ​​for the temperature inside the extraction vessel. 2 index

[0181]

[0182] Table 4. MAE, MSE, RMSE, and R under the comprehensive quality indicators 2 index

[0183]

[0184] In comparison, the BFMIC-PTransformer model established in this embodiment of the invention has the best accuracy.

[0185] In this embodiment, the extraction process of traditional Chinese medicine is taken as the research object. Based on the working mechanism, the extraction steam pressure and the temperature inside the extraction tank are selected as quality index variables, and a parallel Transformer prediction model based on BFMIC is established. Simulation results show that the BFMIC-PTransformer model of the present invention has good accuracy and stability in predicting quality indicators.

[0186] In addition, embodiments of the present invention also provide a process simulation method for product manufacturing, the method comprising:

[0187] Step 1: Obtain a deep learning model, which is a model constructed using any of the product manufacturing process modeling methods in the above embodiments;

[0188] Step 2: Obtain the set values ​​of the variables with high quality correlation and the variables with low quality correlation in the process variables;

[0189] Step 3: Input the set values ​​of the high-quality correlated variables and the low-quality correlated variables into the deep learning model, and output the corresponding quality variable prediction results.

[0190] The product manufacturing process simulation method provided in this invention can not only be applied to product manufacturing processes where quality variables are difficult to measure, for estimating quality variables, but also used to determine setpoints or adjustment values ​​for process variables to ensure that quality variables meet requirements.

[0191] This embodiment also provides a process modeling apparatus for product manufacturing, which is used to implement the above-described process modeling method embodiments and preferred embodiments for product manufacturing; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0192] This embodiment provides a process modeling device for product manufacturing, such as... Figure 6 As shown, it includes:

[0193] The variable acquisition module 501 is used to acquire multiple process variables and multiple quality variables in product manufacturing.

[0194] The numerical acquisition module 502 is used to acquire the numerical values ​​of the multiple process variables and the numerical values ​​of the multiple mass variables at the corresponding time.

[0195] The comprehensive index acquisition module 503 is used to calculate a comprehensive index of each process variable under multiple quality variables based on the value of the process variable and the value of the quality variable; wherein the comprehensive index is used to indicate the correlation between the process variable and the quality variable.

[0196] The variable classification module 504 is used to classify the process variables based on the comprehensive index corresponding to the process variables and the set threshold; wherein the classified categories include high-quality correlated variables, low-quality correlated variables and quality-irrelevant variables;

[0197] The model building module 505 is used to build a deep learning model; the deep learning model includes two first network structures and one second network structure, the inputs of the two first network structures are the high-quality correlated variables and the low-quality correlated variables, respectively; the second network structure generates the corresponding quality variable prediction results based on the outputs of the two first network structures.

[0198] In some optional implementations, the comprehensive index acquisition module 503 includes:

[0199] The MIC calculation unit is used to calculate, for each process variable, the maximum information coefficient under multiple quality variables based on the value of the process variable and the value of the quality variable.

[0200] The fusion unit is used to fuse the process variable with the maximum information coefficients of multiple quality variables using a Bayesian fusion method to obtain the comprehensive index corresponding to the process variable.

[0201] In some optional implementations, the MIC computing unit includes:

[0202] The sub-unit is used to divide the first process variable and the first mass variable in the process variables and the first mass variable in the mass variables into a grid in a two-dimensional space.

[0203] The statistical sub-unit is used to statistically analyze the data within each grid to obtain the joint probability density of the first process variable and the first quality variable falling into the grid.

[0204] The mutual information calculation subunit is used to calculate and normalize the mutual information based on the joint probability density corresponding to each grid.

[0205] Traverse sub-cells to iterate through various different grid division methods and obtain the corresponding normalized mutual information;

[0206] The MIC determining subunit is used to determine the value of the maximum information coefficient of the first process variable and the first quality variable based on the maximum normalized mutual information.

[0207] In some optional embodiments, the fusion unit includes:

[0208] The mean calculation subunit is used to calculate the mean of the maximum information coefficient of the second quality variable and each of the process variables for the second quality variable among the quality variables.

[0209] A prior condition probability density calculation subunit is used to calculate the prior condition probability density of the second process variable and each of the process variables based on the mean of the maximum information coefficients of the second mass variable and each of the process variables.

[0210] The posterior probability density calculation subunit is used to calculate the posterior probability density of the second process variable and each of the quality variables based on the prior conditional probability density.

[0211] The comprehensive index calculation subunit is used to calculate the comprehensive index of the second process variable based on the prior conditional probability density and the posterior probability density.

[0212] In some optional implementations, the variable classification module 504 includes:

[0213] A threshold setting unit is used to set the threshold for the comprehensive index corresponding to each process variable using the cumulative variance percentage.

[0214] A classification unit is used to classify the process variables according to the set threshold.

[0215] In some optional implementations, the first network structure is a Transformer network structure, including: a multi-head attention layer, a first residual connection normalization layer, a second residual connection normalization layer, a first feedforward neural network layer, a second feedforward neural network layer, a first discard layer, and a second discard layer.

[0216] In some optional implementations, the input to the multi-head attention layer is either the high-quality correlated variable or the low-quality correlated variable; the output of the multi-head attention layer is connected to the first discard layer; the output of the first discard layer is connected to the first residual connection normalization layer; the output of the first residual connection normalization layer is sequentially connected to the first feedforward neural network layer and the second feedforward neural network layer; the output of the second feedforward neural network layer is connected to the second discard layer; and the output of the second discard layer is connected to the second residual connection normalization layer.

[0217] In some optional embodiments, the product manufacturing process modeling apparatus further includes:

[0218] The training sample construction module is used to use the collected values ​​of multiple process variables as training samples and the collected values ​​of multiple quality variables at corresponding times as training labels to perform supervised training on the deep learning model.

[0219] The model evaluation module is used to evaluate the trained deep learning model. The evaluation metrics include one or more of the following: mean absolute error, mean square error, root mean square error, and coefficient of determination.

[0220] In some optional embodiments, the value acquisition module 502 includes:

[0221] An initial value acquisition unit is used to acquire the initial values ​​of the process variables and the initial values ​​of multiple mass variables at corresponding times.

[0222] A preprocessing unit is used to preprocess the initial values ​​of the process variable and the initial values ​​of the quality variable to obtain the values ​​of the process variable and the quality variable; wherein the preprocessing includes one or more of the following: removing missing values, removing outliers, and standardization.

[0223] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0224] In this embodiment, the product manufacturing process modeling device is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0225] This embodiment also provides a product manufacturing process simulation device, which is used to implement the above-described product manufacturing process modeling method embodiments and preferred embodiments, and will not be repeated hereafter. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0226] This embodiment provides a product manufacturing process simulation device, such as... Figure 7 As shown, it includes:

[0227] The model acquisition module 601 is used to acquire a deep learning model, which is a model constructed using any of the above-mentioned product manufacturing process modeling methods;

[0228] The setpoint acquisition module 602 is used to acquire the setpoints of the high-quality-related variables and the low-quality-related variables in the process variables;

[0229] The prediction module 603 is used to input the set values ​​of the high-quality correlated variables and the low-quality correlated variables into the deep learning model and output the corresponding quality variable prediction results.

[0230] Further functional descriptions of the above modules are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0231] In this embodiment, the product manufacturing process modeling device is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0232] This invention also provides a computer device having the above-described features. Figure 6 The product manufacturing process modeling device shown, or having Figure 7 The product manufacturing process simulation device shown is shown.

[0233] Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 8As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 8 Take a processor 10 as an example.

[0234] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GPA), or any combination thereof.

[0235] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0236] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0237] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0238] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 8 Taking the example of a connection between China and Israel via a bus.

[0239] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0240] The computer device also includes a communication interface for communicating with other devices or communication networks.

[0241] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0242] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0243] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A process modeling method for product manufacturing, characterized in that, The method includes: Obtain multiple process variables and multiple quality variables for product manufacturing; Acquire the values ​​of the multiple process variables and the values ​​of the multiple quality variables at the corresponding time points; For each process variable, the maximum information coefficient between the process variable and multiple quality variables is calculated based on the value of the process variable and the value of the quality variable. Using a Bayesian fusion method, the maximum information coefficients of the process variable and multiple quality variables are fused to obtain a comprehensive index corresponding to the process variable. This includes: for the second quality variable among the quality variables, calculating the mean of the maximum information coefficients of the second quality variable and each of the process variables; calculating the prior conditional probability density of the second process variable and each of the quality variables based on the mean of the maximum information coefficients of the second quality variable and each of the process variables; calculating the posterior probability density of the second process variable and each of the quality variables based on the prior conditional probability density; and calculating the comprehensive index of the second process variable based on the prior conditional probability density and the posterior probability density. The comprehensive index is used to indicate the correlation between the process variable and the quality variables. Based on the comprehensive index corresponding to the process variable and the set threshold, the process variable is classified; wherein the classified categories include high-quality related variables, low-quality related variables, and quality-irrelevant variables. A deep learning model is constructed; the deep learning model includes two first network structures and one second network structure, the inputs of the two first network structures are the high-quality correlated variables and the low-quality correlated variables, respectively; the second network structure generates the corresponding quality variable prediction results based on the outputs of the two first network structures.

2. The method according to claim 1, characterized in that, For each process variable, the maximum information coefficient between the process variable and multiple quality variables is calculated based on the value of the process variable and the value of the quality variable, including: For the first process variable among the process variables and the first mass variable among the mass variables, the multiple values ​​of the first process variable and the multiple values ​​of the first mass variable are divided into a grid in a two-dimensional space. By statistically analyzing the data within each grid, the joint probability density of the first process variable and the first quality variable falling into the grid is obtained; Mutual information is calculated and normalized based on the joint probability density corresponding to each grid. By traversing various different grid partitioning methods, the corresponding normalized mutual information is obtained; The values ​​of the maximum information coefficients for the first process variable and the first quality variable are determined based on the maximum mutual information after normalization.

3. The method according to claim 1, characterized in that, The process variables are classified based on the comprehensive index corresponding to the process variables and the set threshold, including: For each of the process variables, the threshold is set using the cumulative variance percentage for the comprehensive index. The process variables are classified according to the set threshold.

4. The method according to claim 1, characterized in that, The first network structure is a Transformer network structure, including: a multi-head attention layer, a first residual connection normalization layer, a second residual connection normalization layer, a first feedforward neural network layer, a second feedforward neural network layer, a first discard layer, and a second discard layer.

5. The method according to claim 4, characterized in that, The input to the multi-head attention layer is either the high-quality correlated variable or the low-quality correlated variable; the output of the multi-head attention layer is connected to the first discard layer; the output of the first discard layer is connected to the first residual connection normalization layer; the output of the first residual connection normalization layer is sequentially connected to the first feedforward neural network layer and the second feedforward neural network layer; the output of the second feedforward neural network layer is connected to the second discard layer; and the output of the second discard layer is connected to the second residual connection normalization layer.

6. The method according to claim 1, characterized in that, After constructing the deep learning model, the following is also included: The collected values ​​of multiple process variables are used as training samples, and the collected values ​​of multiple quality variables at corresponding time points are used as training labels to perform supervised training on the deep learning model. The trained deep learning model is evaluated using metrics including one or more of the following: mean absolute error, mean square error, root mean square error, and coefficient of determination.

7. The method according to claim 1, characterized in that, The acquisition of the values ​​of the multiple process variables and the values ​​of the multiple quality variables at corresponding times includes: Obtain the initial values ​​of the process variables and the initial values ​​of multiple quality variables at corresponding times; The initial values ​​of the process variable and the initial values ​​of the quality variable are preprocessed to obtain the values ​​of the process variable and the quality variable; wherein the preprocessing includes one or more of the following: removing missing values, removing outliers, and standardization.

8. A method for simulating the production process of a product, characterized in that, The method includes: Obtain a deep learning model, wherein the deep learning model is a model constructed using the product manufacturing process modeling method according to any one of claims 1-7; Obtain the set values ​​of the variables with high quality correlation and the variables with low quality correlation in the process variables; The set values ​​of the high-quality correlated variables and the low-quality correlated variables are input into the deep learning model, and the corresponding quality variable prediction results are output.

9. A process modeling device for product manufacturing, characterized in that, The device includes: The variable acquisition module is used to acquire multiple process variables and multiple quality variables in product manufacturing. The numerical acquisition module is used to acquire the values ​​of multiple process variables and the values ​​of multiple mass variables at corresponding times; The comprehensive index acquisition module is used to calculate a comprehensive index of each process variable under multiple quality variables based on the value of the process variable and the value of the quality variable; wherein the comprehensive index is used to indicate the correlation between the process variable and the quality variable. The variable classification module is used to classify the process variables based on the comprehensive index corresponding to the process variables and the set threshold; wherein the classified categories include high-quality correlated variables, low-quality correlated variables, and quality-irrelevant variables; A model building module is used to build a deep learning model; the deep learning model includes two first network structures and one second network structure, the inputs of the two first network structures are the high-quality correlated variables and the low-quality correlated variables, respectively; the second network structure generates corresponding quality variable prediction results based on the outputs of the two first network structures; The comprehensive indicator acquisition module includes: The MIC calculation unit is used to calculate, for each process variable, the maximum information coefficient under multiple quality variables based on the value of the process variable and the value of the quality variable. The fusion unit is used to fuse the process variable with the maximum information coefficients of multiple quality variables using the Bayesian fusion method to obtain the comprehensive index corresponding to the process variable. The fusion unit includes: The mean calculation subunit is used to calculate the mean of the maximum information coefficient of the second quality variable and each of the process variables for the second quality variable among the quality variables. A prior condition probability density calculation subunit is used to calculate the prior condition probability density of the second process variable and each of the process variables based on the mean of the maximum information coefficients of the second mass variable and each of the process variables. The posterior probability density calculation subunit is used to calculate the posterior probability density of the second process variable and each of the quality variables based on the prior conditional probability density. The comprehensive index calculation subunit is used to calculate the comprehensive index of the second process variable based on the prior conditional probability density and the posterior probability density.

10. A product manufacturing process simulation device, characterized in that, The device includes: The model acquisition module is used to acquire a deep learning model, wherein the deep learning model is a model constructed using the product manufacturing process modeling method according to any one of claims 1-7; The setpoint acquisition module is used to acquire the setpoints of variables with high quality correlation and variables with low quality correlation in the process variables; The prediction module is used to input the set values ​​of the high-quality correlated variables and the low-quality correlated variables into the deep learning model and output the corresponding quality variable prediction results.

11. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 8.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the method of any one of claims 1 to 8.

13. A computer program product, characterized in that, Includes computer instructions for causing a computer to perform the method of any one of claims 1 to 8.

Images