Process parameter optimization interpretation method, device, equipment and storage medium

Abstract

The application provides an interpretation method, device and equipment of process parameter optimization and a storage medium. The method comprises the following steps: obtaining a first sample set according to an original process procedure, and obtaining a second sample set according to a new process procedure; performing primary interpretation on original features of each sample in the first sample set by a model interpreter, determining an adjustment feature according to changes between the new process procedure and the original process procedure, and generating a first correlation trend chart; determining a second target value group distribution in combination with the first correlation trend chart and a first target value group distribution; performing secondary interpretation on adjustment features corresponding to each sample in the second sample set by the model interpreter, and generating a second correlation trend chart; determining a third target value group distribution in combination with the second correlation trend chart and the second target value group distribution; and performing display on the first target value group distribution and the third target value group distribution, thereby completing the interpretation of process parameter optimization. The application not only provides an idea of process parameter optimization, but also enhances the credibility of process parameter optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of process parameter optimization, and in particular, to a process parameter optimization interpretation method and device, equipment and a storage medium. BACKGROUND

[0002] Process parameter optimization is the adjustment and optimization of various parameters in the production process, and its purpose is to achieve the best production effect and product quality in product production, which is a very important link in manufacturing industry, and can effectively improve product quality and production efficiency, thereby reducing production cost. In the process of process parameter optimization, the best parameter combination needs to be found to achieve the best product quality and production efficiency, however, due to the existence of a large number of variables and uncertainties in the manufacturing process, the process of optimizing parameters is often very complex.

[0003] Common process parameter optimization often uses experimental design method, response surface method, etc., but in actual production, the experimental design method has many problems such as high experimental cost, long experimental period, large experimental error and difficult experimental design, and the response surface method may also have errors in the modeling process, resulting in large deviation of the optimization result, therefore, for the process parameter optimization demand, the use of machine learning in the industrial field is increased. However, in the related technology, the decision rules and variable weights of the machine learning model and other factors cannot be explained, and the transparency of the model is low, which makes the correctness of the predicted optimization result of the model be questioned, thereby affecting the application of the model, and it is also not conducive to people to develop appropriate measures according to the optimization deviation. SUMMARY

[0004] To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. The summary is not an extensive review of the disclosure, nor is it intended to identify key / critical elements of the embodiments or to delineate the scope of the disclosure, but to present some aspects of the disclosure as a prelude to the detailed description.

[0005] In view of the above-mentioned disadvantages of the prior art, the present application discloses a process parameter optimization interpretation method, device, equipment and storage medium, which is used to solve the technical problems of low transparency of process parameter optimization model in related technology, resulting in the predicted result being questioned, and difficulty in verifying the process optimization idea.

[0006] In a first aspect, the present application provides a process parameter optimization interpretation method, the method comprising: performing a first screening on a historical data set according to an original process procedure to obtain a first sample set, and performing a second screening on the historical data set according to a new process procedure to obtain a second sample set; performing a first interpretation on original features corresponding to each sample in the first sample set by a model interpreter, and determining an adjustment feature according to changes between the new process procedure and the original process procedure to generate a first correlation trend chart between the adjustment feature and a first target value of product quality, wherein the original features corresponding to each sample are at least two, and the adjustment feature is at least one; determining a second target value group distribution in combination with the first correlation trend chart and a first target value group distribution, the first target value group distribution being composed of target values of the product quality corresponding to each sample in the first sample set; performing a second interpretation on the adjustment feature corresponding to each sample in the second sample set by the model interpreter to generate a second correlation trend chart between the adjustment feature and a second target value of the product quality; determining a third target value group distribution in combination with the second correlation trend chart and the second target value group distribution; and displaying the first target value group distribution and the third target value group distribution to complete the interpretation of the process parameter optimization.

[0007] In an embodiment of the present application, before the first sample screening on the historical data set according to the original process procedure, the method comprises: cleaning each data in the historical data set to obtain a training sample set; extracting features corresponding to each sample in the training sample set to obtain a training feature set; inputting each feature in the training feature set into an initial model for regression fitting and training to obtain a prediction model; and training the prediction model and the historical data set to obtain the model interpreter.

[0008] In an embodiment of the present application, after the prediction model and the historical data set are trained to obtain the model interpreter, the method comprises: detecting a first fitting degree of the prediction model and each sample in the training sample set and a second fitting degree of the prediction model and process priori knowledge based on working condition changes; comparing the first fitting degree with a first preset fitting threshold, and comparing the second fitting degree with a second preset fitting threshold; if the first fitting degree is less than the first preset fitting threshold, and / or the second fitting degree is less than the second preset fitting threshold, updating the prediction model and the model interpreter.

[0009] In one embodiment of the present invention, generating a first correlation trend graph between the original adjustment feature and the first target value of product quality includes: interpreting the relationship between the original feature of each sample in the second sample set and the first target value through individual condition expectation; keeping all features except one adjustment feature unchanged; replacing the adjustment feature value of the one adjustment feature with the value in the grid and creating a variant of each sample in the first sample set; using the prediction model to predict the variant of each sample in the first sample set to obtain an adjustment feature curve; and determining a centralized adjustment feature point through a centralized individual condition expectation graph to present the difference between the adjustment feature curve and the centralized adjustment feature point.

[0010] In one embodiment of the present invention, the expression for adjusting the characteristic curve is:

[0011]

[0012] In the formula, Indicates the first One sample; Indicates the first Adjusted characteristic curves for each sample; Indicates the first The first correlation trend curve for each sample; 1 represents a vector of appropriate size, and f is the prediction model. Indicates the anchor point; C Indicates features other than adjustment features; Indicates the first Other feature values ​​in each sample besides the adjusted features.

[0013] In one embodiment of the present invention, generating a second correlation trend graph between the adjustment feature and the second target value of the product quality includes: displaying the average marginal effect of the adjustment feature conforming to the second sample set on the prediction result of the prediction model through a partial dependency graph; and presenting the second correlation trend graph based on the average marginal effect.

[0014] In one embodiment of the present invention, the formula for calculating the average marginal effect is:

[0015]

[0016] In the formula, S Indicates adjustment features; This indicates adjustment of the eigenvalues; This represents the average marginal effect; n represents the number of samples in the second sample set; Indicates the first One sample; C Indicates features other than adjustment features; Indicates the first a value of a feature other than the adjustment feature in a sample.

[0017] In an embodiment of the present application, for the new process recipe whose occurrence times in the historical data are less than a second preset value, each mean value of the first target value group is taken as a second target value group.

[0018] In an embodiment of the present application, before the explanation of the process parameter optimization is completed, the adjustment feature and the second correlation trend graph are displayed as an auxiliary explanation of the process parameter optimization.

[0019] In a second aspect, the present application provides an explanation device of process parameter optimization, the device comprising: a data screening module, configured to perform a first screening on a historical data set according to an original process recipe to obtain a first sample set, and perform a second screening on the historical data set according to a new process recipe to obtain a second sample set; a first generation module, configured to perform a first explanation on original features corresponding to each sample in the first sample set by a model interpreter, and determine an adjustment feature according to changes between the new process recipe and the original process recipe, and generate a first correlation trend graph between the adjustment feature and a first target value of product quality, wherein the original features corresponding to each sample are at least two, and the adjustment feature is at least one; a first determination module, configured to determine a second target value group distribution in combination with the first correlation trend graph and a first target value group distribution, the first target value group distribution being composed of target values of the product quality corresponding to each sample in the first sample set; a second generation module, configured to perform a second explanation on the adjustment feature corresponding to each sample in the second sample set by the model interpreter, and generate a second correlation trend graph between the adjustment feature and a second target value of the product quality; a second determination module, configured to determine a third target value group distribution in combination with the second correlation trend graph and the second target value group distribution; and an explanation module, configured to display the first target value group distribution and the third target value group distribution to complete the explanation of the process parameter optimization.

[0020] In a third aspect, the present application provides an electronic device, comprising: one or more processors; a storage device, configured to store one or more programs, when the one or more programs are executed by the one or more processors, the electronic device implements the process parameter optimization explanation method described in the first aspect.

[0021] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, when the computer program is executed by a processor of a computer, the computer executes the process parameter optimization explanation method described in the first aspect.

[0022] As described above, the process parameter optimization explanation method, device, equipment and storage medium provided by the embodiment of the application have the following beneficial effects:

[0023] First, the historical data set is filtered according to the original process procedure and the new process procedure respectively to obtain a first sample set and a second sample set respectively, then the original features corresponding to each sample in the first sample set are explained by the model explainer, and the adjustment features are determined according to the changes of the new process procedure and the original process procedure to generate a first correlation trend graph between the adjustment features and the first target value of the product quality, the first correlation trend graph and the first target value group distribution are combined to determine the second target value group distribution, the adjustment features corresponding to each sample in the second sample set are explained again by the model explainer to generate a second correlation trend graph between the adjustment features and the second target value of the product quality, the second correlation trend graph and the second target value group distribution are further combined to determine the third target value group distribution, and finally the first target value group distribution and the third target value group distribution are displayed to complete the explanation of the process parameter optimization. The transparency of the process parameter optimization process is realized, the credibility of the process parameter optimization in production is enhanced, and the correctness of the prediction result of the prediction model is affirmed. At the same time, due to the high transparency of the process parameter optimization, it is beneficial for people to formulate appropriate optimization measures according to the deviation of the actual production quality from the deviation of the process parameter optimization, and the problems of difficult process optimization idea verification and high cost are solved, thereby improving the process parameter optimization efficiency.

[0024] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0025] The drawings incorporated into the specification and forming a part thereof, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the principles of the application. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. In the drawings:

[0026] Figure 1 is a schematic diagram of an implementation environment of the process parameter optimization explanation device according to an exemplary embodiment of the present application;

[0027] Figure 2 is a flowchart of the process parameter optimization explanation method according to an exemplary embodiment of the present application;

[0028] Figure 3 is a flowchart of obtaining a model explainer according to an exemplary embodiment of the present application;

[0029] Figure 4is a flowchart of an update model interpreter according to an example embodiment of the present application;

[0030] Figure 5 is a flowchart of generating a first correlation trend chart according to an example embodiment of the present application;

[0031] Figure 6 is a flowchart of generating a second correlation trend chart according to an example embodiment of the present application;

[0032] Figure 7 is a graph of the relationship between C-element and yield strength according to an example embodiment of the present application;

[0033] Figure 8 is a partial dependence plot of C-element and yield strength according to an example embodiment of the present application;

[0034] Figure 9 is a block diagram of an explanation device for process parameter optimization according to an example embodiment of the present application;

[0035] Figure 10 is a structural schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0036] Other advantages and novel features of the present application will be readily appreciated as the same becomes better understood by reference to the following detailed description when considered in connection with the accompanying drawings wherein:

[0037] It should be noted that the drawings included in the following embodiments are only schematic and are not drawn to scale. They are provided to aid in understanding the present application and are not provided to limit the present application. The specific instrumentalities and materials disclosed in the detailed description section merely provide example embodiments of the present application. Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the present application disclosed herein. Therefore, it is intended that the present application not be limited to the particular embodiments disclosed, but that the present application include all embodiments falling within the scope of the appended claims, and their equivalents.

[0038] In the following description, numerous specific details are discussed to provide a thorough understanding of embodiments of the present application. However, one of ordinary skill in the art will recognize that the application can be practiced without one or more of the specific details. In other instances, well-known structures and devices are shown in block diagram form to avoid obscuring the underlying principles of the application.

[0039] First of all, it needs to be pointed out that process parameter optimization is to adjust and optimize various parameters in the production process, the purpose is to make the product in the production manufacturing to achieve the best production effect and product quality, is a very important link in manufacturing industry, can effectively improve product quality and production efficiency, thereby reducing production cost. In the process of process parameter optimization, the best parameter combination needs to be found to achieve the best product quality and production efficiency, however, due to the existence of a large number of variables and uncertainties in the manufacturing process, the process of optimizing parameters is often very complex.

[0040] Common process parameter optimization often uses experimental design method, response surface method, etc., but in actual production, experimental design method has many problems such as high experimental cost, long experimental period, large experimental error and difficult experimental design, and response surface method may also have errors in modeling process, resulting in large deviation of optimization result, therefore, for the demand of process parameter optimization, the use of machine learning in industrial field increases. Through the research of the inventors of the present application, it is found that in the related technology, the decision rules and variable weights of the machine learning model and other factors cannot be explained, and the transparency of the model is low, which makes the correctness of the prediction optimization result of the model be questioned, thereby affecting the application of the model, and it is also not conducive for people to formulate appropriate measures according to the optimization deviation.

[0041] Therefore, please see Figure 1 , Figure 1 is an implementation environment schematic diagram of the process parameter optimization explanation device shown in an exemplary embodiment of the present application. As Figure 1 shown, the implementation environment can include an explanation device 101 and a computer device 102, the explanation device 101 can be arranged in the computer device 102, and is used for the explanation of process parameter optimization. Wherein, the computer device 102 can be at least one of desktop graphic processing unit (GPU) computer, GPU computing cluster, neural network computer, etc. Through the explanation device 101, the process parameter optimization process is explained, the transparency of the process parameter optimization process is realized, the credibility of the process parameter optimization in production is enhanced, and the correctness of the prediction result of the prediction model is affirmed. At the same time, because the transparency of process parameter optimization is high, it is also more conducive for people to formulate appropriate optimization measures according to the deviation of actual production quality and the deviation of process parameter optimization, solve the problem of difficult process optimization idea verification and high cost, thereby improve the process parameter optimization efficiency.

[0042] Please see Figure 2 , Figure 2 is a flow chart of the process parameter optimization explanation method shown in an exemplary embodiment of the present application. The method can be applied to Figure 1The implementation environment shown is specifically executed by the interpretation system within that implementation environment. It should be understood that this method can also be applied to other exemplary implementation environments and executed by devices in other implementation environments; this embodiment does not limit the implementation environment to which the method is applicable.

[0043] like Figure 2 As shown, in an exemplary embodiment, the method for interpreting process parameter optimization includes at least steps S210 to S260, which are described in detail below:

[0044] Step S210: The historical dataset is filtered once according to the original process specification to obtain the first sample set, and the historical dataset is filtered a second time according to the new process specification to obtain the second sample set.

[0045] It should be noted that the actual production of the product is carried out according to the process parameters. It can be assumed that there is a corresponding relationship between the process parameters and the product quality, and the process specification consists of a series of process parameters.

[0046] In this embodiment, the original process procedure refers to the process procedure for product production before process parameter optimization, corresponding to the unoptimized process parameters; the new process procedure refers to the process procedure for product production after process parameter optimization, corresponding to the optimized process parameters; the historical dataset refers to the collection of production data from the product's historical production, including product quality data. Specifically, the first filtering of the historical dataset based on the original process procedure involves selecting production data obtained from the historical dataset based on the original process procedure and using it as the first sample set; the second filtering of the historical dataset based on the new process procedure involves selecting production data obtained from the historical dataset based on the new process procedure and using it as the second sample set.

[0047] In one embodiment, please refer to Figure 3 , Figure 3 This is a flowchart illustrating the process of obtaining a model interpreter, as shown in an exemplary embodiment of this application. Figure 3 As shown, before performing a sample screening of the historical dataset according to the original process specifications, at least steps S310 to S340 are included, as follows: Step S310, clean each data in the historical dataset to obtain a training sample set; Step S320, extract the features corresponding to each sample in the training sample set to obtain a training feature set; Step S330, perform regression fitting on each feature in the training feature set and input it into the initial model for training to obtain a prediction model; Step S340, train the model interpreter based on the prediction model and the historical dataset.

[0048] It should be noted that cleaning each data in the historical data set includes checking data consistency, processing invalid values and missing values, etc., and each feature in the training feature set is subjected to regression fitting, which can use a random forest algorithm to obtain a prediction model. Here, the regression fitting algorithm is not limited, and any algorithm model is applicable to this embodiment.

[0049] It should also be noted that the model interpreter is obtained by training the prediction model and the historical data set, i.e., the historical data set is input into the prediction model, the key features are extracted, and the linear function of the explanation feature group formed according to the key features is obtained. The model interpreter can increase the transparency and interpretability of the prediction model, i.e., by explaining the key features in the sample data, the prediction of the prediction model is promoted, and according to the prediction result of the prediction model, combined with the process prior knowledge, a judgment can be made on whether to believe the prediction result.

[0050] In an embodiment, please refer to Figure 4 , Figure 4 is a flowchart for updating the model interpreter according to an exemplary embodiment of the present application. As shown in Figure 4 , after obtaining the model interpreter by training the prediction model and the historical data set, it further includes steps S410 to S430, as follows: step S410, detecting the first fitting degree of the prediction model and each sample in the training sample set based on the working condition change, and the second fitting degree of the prediction model and the process prior knowledge; step S420, comparing the first fitting degree with the first preset fitting threshold, and comparing the second fitting degree with the second preset fitting threshold; step S430, if the first fitting degree is less than the first preset fitting threshold, and / or the second fitting degree is less than the second preset fitting threshold, then updating the prediction model and the model interpreter.

[0051] In this embodiment, the change of working condition can be understood as a change of product production line or a change of product produced. In view of the change of product production line or the change of product produced, the model interpreter may no longer be applicable, and the model interpreter needs to be updated. In view of the dynamic change of working condition, the corresponding updating module in the device detects the first fitting degree of the prediction model and each sample in the training sample set, and detects the second fitting degree of each sample in the training sample set and the process priori knowledge. The process priori knowledge refers to the process knowledge obtained based on experience through a large amount of production data. In actual application, in addition to the process knowledge obtained by machine learning, the process priori knowledge is often used to deal with complex problems. If the first fitting degree of the prediction model and each sample in the training sample set is less than the corresponding standard value, i.e., the first preset fitting threshold, and / or the second fitting degree of the prediction model and the process priori knowledge is less than the corresponding standard value, i.e., the second preset fitting threshold, it indicates that the prediction model is no longer applicable, i.e., the model interpreter is no longer applicable, and the prediction model needs to be updated, so as to further update the model interpreter, thereby ensuring that the interpretation result of the model interpreter is more accurate. It should be understood that the update of the prediction model and the update of the model interpreter are the same as the above-mentioned implementation manner of obtaining the prediction model and the model interpreter, which will not be described here.

[0052] In step S220, the model interpreter is used to interpret the original features corresponding to each sample in the first sample set, and a first correlation trend graph between the adjusted features and the first target value of product quality is generated according to the change of the new process procedure and the original process procedure, wherein the original features corresponding to each sample in the first sample set are at least two, and the adjusted features are at least one.

[0053] It should be noted that the first sample set is obtained by screening the historical data set according to the original process procedure, and the features in each sample in the first sample set correspond to the original features. The model interpreter interprets the original features corresponding to each sample in the first sample set, which means that the original features corresponding to each sample in the first sample set are analyzed, and the adjusted features are determined according to the change of the new process procedure and the original process procedure, so that the first correlation trend graph between the adjusted features and the first target value of product quality can be obtained. The original features are the features of the samples screened according to the original process procedure, and the first target value of product quality is the quality value obtained by producing products according to the original process procedure. The adjusted features change based on the original features and have a certain influence on the first target value, so there is a certain correlation influence trend between the adjusted features and the first target value of product quality.

[0054] In addition, because the process parameter optimization process involves adjustment of features in the sample, accordingly, there are adjusted features and unchanged features in the sample, so the original features of each sample in the first sample set are at least two, and since there is adjustment of features, the adjusted features are at least one.

[0055] Step S230, determine the second target value group distribution in combination with the first correlation trend graph and the first target value group distribution, and the first target value group distribution is composed of target values of product quality corresponding to each sample in the first sample set.

[0056] The first correlation trend graph represents the relationship between the adjusted features in the original features of each sample in the first sample set and the first target value of product quality, and there is a first correlation trend graph between each adjusted feature in each sample and the first target value of product quality. The first target value group distribution is composed of target values of the product quality corresponding to each sample in the first sample set. According to the first correlation trend graph and the first target value group distribution, the second target value group distribution represents the combination of quality values of product quality after eliminating the influence of adjusted features for each original sample, i.e., each sample in the first sample set.

[0057] Step S240, second interpretation of the adjusted features corresponding to each sample in the second sample set by the model interpreter to generate the second correlation trend graph between the adjusted features and the second target value of product quality.

[0058] It should be noted that the interpretation of the features corresponding to each sample in the second sample set by the model interpreter is understood as the analysis of the features corresponding to each sample in the second sample set. The second sample set is obtained by secondary screening of the historical data set according to the new process procedure, and the new process procedure is considered to have certain changes relative to the original process procedure, involving adjusted features, therefore, the adjusted features will affect the product quality, so there is a certain corresponding relationship between the adjusted features and the second target value of product quality, i.e., the second correlation trend graph is obtained.

[0059] Specifically, please refer to Figure 5 , Figure 5 is a flowchart for generating the first correlation trend graph according to an exemplary embodiment of the present application. As Figure 5As shown, generating the first correlation trend chart between the adjustment feature and the first target value of the product quality at least includes steps S510 to S550, as follows: step S510, explaining the relationship between the original feature of each sample in the first sample set and the first target value by individual conditional expectation; step S520, keeping other features except for one adjustment feature unchanged; step S530, replacing the adjustment feature value of the adjustment feature with the value in the grid, and creating a variant of each sample in the first sample set; step S540, using the prediction model to predict the variant of each sample in the first sample set, obtaining an adjustment feature curve; and step S550, determining a centralized adjustment feature point by a centralized individual conditional expectation chart, and presenting the difference between the adjustment feature curve and the centralized adjustment feature point.

[0060] It should be noted that the model interpreter includes an individual conditional expectation chart, a centralized individual conditional expectation chart, and a partial dependence chart. The individual conditional expectation chart displays a line for each sample, showing how the prediction changes according to the sample when the feature changes. The centralized individual conditional expectation chart means that the expected result is adjusted by adjusting the adjustment. The partial dependence chart is used to observe how a certain feature affects the prediction result. The feature importance shows which variable has the greatest impact on the prediction. When other features remain unchanged, changing one or more features (i.e., adjustment features) will result in a dependence chart according to the average result predicted by the prediction model.

[0061] In this embodiment, the variant of each sample in the first sample set means the sample formed after replacing the adjustment feature value of each sample in the first sample set with the value in the grid. The difference between the finally presented adjustment feature curve and the centralized adjustment feature point is the first correlation trend chart.

[0062] Specifically, the expression of the adjustment feature curve is: , wherein, represents the i-th sample; represents the i-th sample; represents the i-th sample; represents the i-th sample; represents the i-th sample; represents the i-th sample; represents the anchor point; C represents other features except for the adjustment feature; represents the i-th sample; represents the i-th sample.

[0063] It should be noted that the method for calculating the Individual Conditional Expectation Map (ICM) is to keep all features except for the adjustment features the same. Variants of the sample are created by replacing the feature values ​​with values ​​from the grid, and the model is used to predict these newly created instances. The result is a set of points with feature values ​​from the grid and corresponding predictions. In the ICM, for... In each instance, the curve It is about At this time Fixed and unchanging.

[0064] In one embodiment, for a new process procedure that appears less than a second preset value in historical data, the average values ​​of the first target value group are used as the second target value group.

[0065] In this embodiment, if the number of historical data conforming to the new process specification is less than the second preset value, that is, the number of samples in the second sample set does not meet the requirements, in order to ensure the accuracy of the process parameter optimization interpretation, the mean values ​​of the first target value group are used as the second target value group, thereby forming the second target value group distribution.

[0066] Step S250: Determine the distribution of the third target value group by combining the second correlation trend chart and the distribution of the second target value group.

[0067] It should be noted that the second correlation trend graph represents the relationship between the adjustment characteristics of each sample in the second sample set and the second target value of product quality. The second target value group distribution represents the combination of product quality values ​​after eliminating the influence of the adjustment characteristics for each original sample, i.e., each sample in the first sample set. Based on the second target value group, the influence of the adjustment characteristics of the new process specification is added, i.e., the second correlation trend graph, to obtain the third target value group distribution, which represents the quality value obtained by each sample in the second sample set when the product is manufactured according to the new specification.

[0068] Specifically, please see Figure 6 , Figure 6 This is a flowchart illustrating the generation of a second correlation trend graph, as shown in an exemplary embodiment of this application. Figure 6 As shown, generating a second correlation trend graph between the adjustment features and the second target value of product quality includes at least steps S610 to S620, as follows: Step S610, displaying the average marginal effect of the adjustment features conforming to the second sample set on the prediction results of the prediction model through a partial dependency graph; Step S620, presenting the second correlation trend graph based on the average marginal effect.

[0069] In this embodiment, the partial dependence function is calculated by marginalizing other features to obtain a result that depends only on the adjustment feature and its interaction with other features using a Monte Carlo method, i.e., the average value in the training data is calculated to estimate the average marginal effect of a given value of the adjustment feature on the prediction, which can be calculated by the formula

[0070] Specifically, the formula for calculating the average marginal effect is: , wherein, S represents the adjustment feature; represents the value of the adjustment feature; represents the average marginal effect; n represents the number of samples in the second sample set; represents the th sample; C represents other features except the adjustment feature; represents the value of other features except the adjustment feature in the th sample.

[0071] In step S260, the first target value group distribution and the third target value group distribution are displayed to complete the explanation of the process parameter optimization.

[0072] The first target value group distribution is composed of the target values of the product quality corresponding to each sample in the first sample set. The third target value group distribution represents the quality values obtained according to the new procedure in the production of the product by each sample in the second sample set. By displaying the first target value group distribution and the third target value group distribution through the display interface, the process of the process parameter optimization can be displayed.

[0073] In an embodiment, before completing the explanation of the process parameter optimization, the adjustment feature and the second related trend chart are also displayed as auxiliary explanation of the process parameter optimization.

[0074] In this embodiment, the adjustment feature and the second related trend chart are displayed at the same time as the first target value group distribution and the third target value group distribution, which can enhance the explainability of the prediction model and make the transparency of the process parameter optimization higher.

[0075] The process parameter optimization explanation method provided by the above embodiment first filters the historical data set according to the original process procedure and the new process procedure respectively, obtains a first sample set and a second sample set respectively, then explains the original features corresponding to each sample in the first sample set through the model explainer, determines the adjustment features according to the changes of the new process procedure and the original process procedure, generates a first correlation trend graph between the adjustment features and the first target value of the product quality, combines the first correlation trend graph and the first target value group distribution to determine the second target value group distribution, again explains the adjustment features corresponding to each sample in the second sample set through the model explainer, generates a second correlation trend graph between the adjustment features and the second target value of the product quality, further combines the second correlation trend graph and the second target value group distribution to determine the third target value group distribution, and finally displays the first target value group distribution and the third target value group distribution to complete the explanation of the process parameter optimization. The transparency of the process parameter optimization process is realized, the credibility of the process parameter optimization in production is enhanced, and the correctness of the prediction result of the prediction model is affirmed. At the same time, due to the high transparency of the process parameter optimization, it is beneficial for people to formulate appropriate optimization measures according to the deviation of the actual production quality from the deviation of the process parameter optimization, thereby solving the problems of difficult process optimization thought verification and high cost, and improving the process parameter optimization efficiency.

[0076] The following takes the yield strength of the 2250 hot rolling plant as an example to specifically explain the explanation of the process parameter optimization.

[0077] The yield strength of the 2250 hot rolling plant is selected as the quality target, and the real sample set procedure data such as the composition, the heating furnace temperature and the rolling line data are associated to obtain a historical data set of the process parameters related to the yield strength as the target and the procedure data of the process procedure related to the yield strength as the target.

[0078] Specifically, the WL510 steel grade is selected, and after data cleaning and post-processing, 1201 samples are obtained, including 26 original features and performance targets corresponding to the features. A random forest is used to regress and fit the original features to obtain a yield strength prediction model. In particular, the algorithm of the regression model is not specified here, and any model is suitable for this method. The 26 original features specifically include C carbon element performance value, MN manganese element performance value, SI silicon element performance value, P phosphorus element performance value, S sulfur element performance value, ALS acid-soluble aluminum performance value, AL aluminum element performance value, CU copper element performance value, CR chromium element performance value, NI nickel element performance value, MO molybdenum element performance value, NB niobium element performance value, V vanadium element performance value, TI titanium element performance value, H hydrogen element performance value, O oxygen element performance value, B boron element performance value, N nitrogen element performance value, CA calcium element performance value, RE rare earth element performance value, SN tin element performance value, W tungsten element performance value, discharge temperature, intermediate rolling temperature, coiling temperature and billet thickness.

[0079] According to the original process specification limit of the WL510 steel grade, a first sample set that meets the specification is screened out, which can be understood as an original sample set. The individual conditional expectation graph is used to explain the relationship between a single feature and the yield strength sample by sample. Taking the C element as an example, in the individual conditional expectation graph, each curve represents the relationship between the C element and the yield strength of a sample. If the values of the C element are adjusted by replacing the values in the grid while keeping other features unchanged, a variant of the sample is created, and the prediction model is used to predict these newly created variant samples. The curve is centered at a certain point in the adjusted feature, and only the prediction difference to this point is displayed. Please refer to Figure 7 , Figure 7 is a curve graph of the relationship between the C element and the yield strength according to an example embodiment of the present application, wherein the abscissa represents the actual value of the C carbon element, and the ordinate represents the yield strength, as shown in Figure 7 As shown in the figure, the C element and the yield strength are approximately positively correlated, which is consistent with the process prior knowledge, but for a few samples, the C element and the yield strength will show a negative correlation, wherein the thick curve in the figure is the average trend of the original sample set.

[0080] After obtaining the correlation trend graph of the C element and the yield strength of each sample in the original sample set, the C element is subtracted for each sample, that is, the influence of the adjusted feature, to obtain the part that does not contain the adjusted feature, that is, the part that removes the influence of the C element on the yield strength.

[0081] The process specification is adjusted for the C element adjustment to form a new process specification. The sample set that meets the new process specification in the historical data set, that is, the second sample set, is screened out. For the second sample set, its partial dependence graph is calculated as the correlation trend graph of the C element and the yield strength in the new process specification. Please refer to Figure 8 , Figure 8 is a partial dependence graph of the C element and the yield strength according to an example embodiment of the present application, wherein the abscissa represents the actual value of the C carbon element, and the ordinate represents the yield strength, as shown in Figure 8 By marginalizing other features except the C element, the result that only depends on the C element and the interaction with other features is obtained. The Monte Carlo method is used for calculation, that is, the average value in the training data is calculated, and the average marginal effect on the prediction is calculated when the value of the feature C element is given, that is, the thick curve in the figure.

[0082] The transparency of the process parameter optimization process is realized by explaining the process parameter optimization, the credibility of the process parameter optimization in production is enhanced, and the correctness of the prediction result of the prediction model is affirmed. Meanwhile, due to the high transparency of the process parameter optimization, people can formulate appropriate optimization measures according to the deviation of the actual production quality from the deviation of the process parameter optimization, solve the problems of difficult verification and high cost of process optimization ideas, and thus improve the process parameter optimization efficiency.

[0083] Please refer to Figure 9 , Figure 9 is a block diagram of an explanation device for process parameter optimization according to an example embodiment of the present application. The device can be applied to Figure 1 the implementation environment shown in the figure, and it should be understood that the device can also be applied to other example implementation environments, and the present embodiment does not limit the implementation environment to which the device is applied.

[0084] As shown in Figure 9 , in an example embodiment, the process parameter optimization explanation device 900 at least includes a data screening module 910, a first generation module 920, a first determination module 930, a second generation module 940, a second determination module 950, and an explanation module 960, which are described in detail as follows:

[0085] The data screening module 910 is configured to perform a first screening on a historical data set according to an original process procedure to obtain a first sample set, and perform a second screening on the historical data set according to a new process procedure to obtain a second sample set.

[0086] The first generation module 920 is configured to perform a first explanation on original features corresponding to each sample in the first sample set by a model explainer, and determine an adjustment feature according to a change between the new process procedure and the original process procedure to generate a first correlation trend graph between the adjustment feature and a first target value of product quality, wherein the original features corresponding to each sample are at least two, and the adjustment feature is at least one.

[0087] The first determination module 930 is configured to determine a second target value group distribution in combination with the first correlation trend graph and a first target value group distribution, the first target value group distribution being composed of target values of product quality corresponding to each sample in the first sample set.

[0088] The second generation module 940 is configured to perform a second explanation on the adjustment feature corresponding to each sample in the second sample set by the model explainer to generate a second correlation trend graph between the adjustment feature and a second target value of product quality.

[0089] The second determination module 950 is configured to determine a third target value group distribution in combination with the second correlation trend graph and the second target value group distribution.

[0090] The explanation module 960 is configured to display the first target value group distribution and the third target value group distribution to complete explanation of the process parameter optimization.

[0091] It should be noted that the process parameter optimization explanation apparatus provided in the above embodiment and the process parameter optimization explanation method provided in the above embodiment belong to the same concept, and the content of the operation of each module has been described in detail in the method embodiment, which will not be described here.

[0092] Please refer to Figure 10 , Figure 10 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. Figure 10 A structural schematic diagram of a computer system of the electronic device suitable for implementing the embodiments of the present application is shown. It should be noted that Figure 10 The computer system 1000 of the electronic device shown is only an example, and should not limit the functions and use range of the embodiments of the present application.

[0093] As Figure 10 shown, the computer system 1000 includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 1002 or programs loaded from a storage portion 1008 to a random access memory (RAM) 1003, such as performing the method in the above embodiment. In the RAM 1003, various programs and data required for system operation are also stored. The CPU 1001, the ROM 1002, and the RAM 1003 are connected to each other through a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0094] The following components are connected to the I / O interface 1005: an input part 1006 including a keyboard, a mouse, etc.; an output part 1007 including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), etc., and a speaker, etc.; a storage part 1008 including a hard disk, etc.; and a communication part 1009 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication part 1009 performs communication processing via a network such as the Internet. A drive 1013 is also connected to the I / O interface 1005 as necessary. A removable medium 1011 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 1013 as necessary, so that a computer program read out therefrom is installed in the storage part 1008 as necessary.

[0095] In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing a computer program for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication part 1009, and / or installed from the removable medium 1011. When the computer program is executed by the central processing unit (CPU) 1001, various functions defined in the system of the present application are executed.

[0096] It should be noted that the computer-readable medium in the embodiments of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination thereof. The computer-readable storage medium may, for example, be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination thereof. More specific examples of the computer-readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (Compact Disc Read-Only Memory, CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this application, the computer-readable signal medium can include a data signal propagated in a baseband or as a carrier wave in a propagated data signal, in which the computer-readable computer program is carried. Such a propagated data signal can take on various forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate, or transmit the program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wired, or the like, or any suitable combination thereof.

[0097] The flowcharts and block diagrams in the drawings illustrate the possible implementation architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present application. In the flowcharts or block diagrams, each block can represent a module, a program segment, or a part of code, which contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different orders than those noted in the drawings. For example, two blocks represented in succession can actually be executed substantially in parallel, and sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams or flowcharts, and the combination of blocks in the block diagrams or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0098] The units described in the embodiments of the present application can be implemented in the form of software, or can be implemented in the form of hardware, or can be implemented in the form of a combination of software and hardware. The units described can also be arranged in a processor. In some cases, the names of the units do not constitute a limitation on the units themselves.

[0099] The present application also provides a computer readable storage medium, which stores a computer program. When the computer program is executed by a processor of a computer, the computer is caused to execute the process parameter optimization interpretation method described above. The computer readable storage medium can be included in the electronic device described in the above embodiments, or can exist separately and not be assembled into the electronic device.

[0100] The above embodiments only illustrate the principles and effects of the present application, but are not used to limit the present application. Any person skilled in the art can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by those skilled in the art without departing from the spirit and technical ideas disclosed by the present application should be covered by the claims of the present application.

Claims

1. A method of interpreting process parameter optimization, characterized in that, The method comprises: According to the original process procedure, a first sample set is obtained by screening the historical data set once, and a second sample set is obtained by screening the historical data set again according to the new process procedure; The original features corresponding to each sample in the first sample set are explained once by a model interpreter, and an adjustment feature is determined according to the changes between the new process procedure and the original process procedure, to generate a first correlation trend chart between the adjustment feature and a first target value of product quality, wherein the original features corresponding to each sample are at least two, and the adjustment feature is at least one, and the model interpreter is trained according to a pre-trained prediction model and the historical data set; The first correlation trend chart and a first target value group distribution are combined to determine a second target value group distribution, and the first target value group distribution is composed of target values of product quality corresponding to each sample in the first sample set; The adjustment features corresponding to each sample in the second sample set are explained again by the model interpreter to generate a second correlation trend chart between the adjustment features and a second target value of product quality; The second correlation trend chart and the second target value group distribution are combined to determine a third target value group distribution; The first target value group distribution and the third target value group distribution are displayed to complete the explanation of the process parameter optimization; The first correlation trend chart is generated by: explaining the relationship between the original features of each sample in the first sample set and the first target value by individual conditional expectation; keeping other features unchanged except one adjustment feature; replacing the value of the adjustment feature in the grid with the adjustment feature value of the one adjustment feature, and creating a variant of each sample in the first sample set; using the prediction model to predict the variant of each sample in the first sample set to obtain an adjustment feature curve; determining a centralized adjustment feature point by a centralized individual conditional expectation chart to present the difference between the adjustment feature curve and the centralized adjustment feature point; The second correlation trend chart is generated by: displaying the average marginal effect of the adjustment feature on the prediction model prediction result in the second sample set by a partial dependence chart; and presenting the second correlation trend chart based on the average marginal effect; The average marginal effect is determined by: calculating the average value of the prediction results of the prediction model on all samples in the second sample set under the condition that the adjustment feature in the second sample set takes an adjustment feature value and other features in the second sample set take other feature values, to obtain the average marginal effect.

2. The process parameter optimization interpretation method of claim 1, wherein, Before the historical data set is screened once according to the original process procedure, the method comprises: Cleaning each data in the historical data set to obtain a training sample set; Extracting features corresponding to each sample in the training sample set to obtain a training feature set; Each feature in the training feature set is regressed and fitted into an initial model for training to obtain a prediction model; The model interpreter is obtained according to the prediction model and the historical data set.

3. The process parameter optimization interpretation method of claim 2, wherein, After the model interpreter is obtained by training according to the prediction model and the historical data set, the method comprises: detecting a first fitting degree of the prediction model and each sample in the training sample set and a second fitting degree of the prediction model and prior knowledge of the process based on a change in the working condition; comparing the first fitting degree with a first preset fitting threshold and comparing the second fitting degree with a second preset fitting threshold; updating the prediction model and the model interpreter if the first fitting degree is less than the first preset fitting threshold and / or the second fitting degree is less than the second preset fitting threshold.

4. The process parameter optimization interpretation method of claim 2, wherein, The expression of the adjusted characteristic curve is: In the formula, Indicates the first One sample; Indicates the first Adjusted characteristic curves for each sample; Indicates the first The first correlation trend curve for each sample; 1 represents a vector of appropriate size, and f is the prediction model. Indicates the anchor point; C Indicates features other than adjustment features; Indicates the first Other feature values ​​in each sample besides the adjusted features.

5. The process parameter optimization interpretation method of claim 1, wherein, The calculation formula of the average marginal effect is: wherein, S denotes the adjustment feature; denotes the adjustment feature value; denotes the average marginal effect; n denotes the number of samples in the second sample set; denotes the first sample; denotes the first sample; C denotes the other feature than the adjustment feature; denotes the other feature value than the adjustment feature value in the first sample; denotes the other feature value than the adjustment feature value in the first sample.

6. The process parameter optimization interpretation method of claim 1, wherein, For the new process recipe that appears less than a second preset value in the historical data, the average of each target value group of the first target value group is taken as a second target value group.

7. The process parameter optimization interpretation method according to any one of claims 1 to 6, characterized in that, Before the explanation of the optimization of the process parameters is completed, the method further comprises: displaying the adjusted characteristic and the second correlation trend chart as auxiliary explanation of the optimization of the process parameters.

8. An interpretation device of process parameter optimization, characterized in that, The device comprises: a data screening module configured to perform a first screening on a historical data set according to an original process recipe to obtain a first sample set and perform a second screening on the historical data set according to a new process recipe to obtain a second sample set; a first generation module configured to perform a first explanation on original characteristics corresponding to each sample in the first sample set by a model interpreter and determine an adjusted characteristic according to a change between the new process recipe and the original process recipe to generate a first correlation trend chart between the adjusted characteristic and a first target value of product quality, wherein the original characteristics corresponding to each sample are at least two, the adjusted characteristic is at least one, and the model interpreter is obtained by training according to a prediction model and the historical data set; a first determination module configured to determine a second target value group distribution in combination with the first correlation trend chart and a first target value group distribution, the first target value group distribution being composed of target values of the product quality corresponding to each sample in the first sample set; a second generation module configured to perform a second explanation on the adjusted characteristic corresponding to each sample in the second sample set by the model interpreter to generate a second correlation trend chart between the adjusted characteristic and a second target value of the product quality; a second determination module configured to determine a third target value group distribution in combination with the second correlation trend chart and the second target value group distribution; an explanation module configured to display the first target value group distribution and the third target value group distribution to complete the explanation of the optimization of the process parameters; and a display module configured to display the first correlation trend chart, the second correlation trend chart, the first target value group distribution, the second target value group distribution and the third target value group distribution. The first generation module is specifically configured to explain the relationship between the original features of each sample in the first sample set and the first target value through individual conditional expectation, keep other features unchanged except for one adjustment feature, replace the adjustment feature value of the one adjustment feature with the value in the grid, and create a variant of each sample in the first sample set; use the prediction model to predict the variant of each sample in the first sample set to obtain an adjustment feature curve; determine a centralized adjustment feature point through a centralized individual conditional expectation diagram, and present the difference between the adjustment feature curve and the centralized adjustment feature point. The second generation module is specifically configured to show the average marginal effect of the adjustment feature on the prediction result of the prediction model in the second sample set through a partial dependence diagram, present the second correlation trend diagram based on the average marginal effect, and determine the average marginal effect, including: calculating the average value of the prediction result of the prediction model on all samples in the second sample set under the condition that the adjustment feature takes an adjustment feature value and other features except for the adjustment feature take other feature values, and obtaining the average marginal effect.

9. An electronic device, comprising: Comprise: One or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, cause the electronic device to implement the process parameter optimization explanation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer readable instruction is stored thereon, which, when executed by a processor of a computer, causes the computer to execute the process parameter optimization explanation method according to any one of claims 1 to 7.

Images