Intelligent correction method and system for heat treatment process parameter deviation and storage medium

Abstract

The invention discloses a heat treatment process parameter deviation intelligent correction method and system and a storage medium, and belongs to the technical field of production optimizing.The method comprises the steps that data expansion is conducted on original production data, a first prediction model is trained locally based on the expansion data, and a second prediction model is obtained, calculating the deviation of the network parameters at the same position of the different second prediction model of each factory, determining an adjustment value based on the deviation, correcting the network parameters based on the adjustment value, defining the corrected second prediction model as a third prediction model, and inputting the real-time production data into the third prediction model. A dynamic adjustment mechanism in the third prediction model adjusts the number of network layers according to the product type, and outputs a process adjustment amount and a corresponding confidence coefficient; and when the confidence coefficient is smaller than a second threshold value, the historical case library is called to correct the process adjustment amount, and real-time dynamic adjustment of the heat treatment process parameters is achieved through a cooperative correction mechanism of the local and cloud models.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of production optimization, and particularly relates to a heat treatment process parameter deviation intelligent correction method and system and a storage medium. BACKGROUND

[0002] Hot processing manufacturing is a typical long-process manufacturing industry, which generally includes smelting, casting, forging, welding, heat treatment and the like processes. There are many process factors affecting manufacturing quality, and the quantitative relationship between process data and product quality is difficult to determine. In addition, the traditional verification type process design optimization has a long cycle and high cost, and therefore there is a problem of difficulty in adjusting process parameters. To solve the technical problem, the prior art proposes the following method.

[0003] A control method, device and equipment for steel smelting and a storage medium are disclosed in a Chinese patent document with publication number CN117590810A. The method records the process parameters of each process such as coking, sintering, blast furnace, etc. and the corresponding quality indicators, compares the actual quality indicators with the theoretical estimated data, and if the deviation is within the allowable range, it is determined to be consistent. Only data that all processes are consistent is retained to form a quality indicator database, and 5-10 groups of parameters with the highest matching degree are selected from the database as the best process parameters.

[0004] However, the essence of this method is static optimization based on historical data, which cannot dynamically adjust process parameters according to real-time production conditions. When the raw material composition, equipment state and other production conditions change, the database needs to be manually updated, and therefore the adaptability is poor. SUMMARY

[0005] To solve the above problems, the application provides a heat treatment process parameter deviation intelligent correction method and system and a storage medium to solve the problems existing in the background art.

[0006] To achieve the above-mentioned purposes, the application provides a heat treatment process parameter deviation intelligent correction method, which comprises: Obtaining original production data of each process in each factory, performing data augmentation on the original production data to obtain augmented data; Establishing a first prediction model containing a dynamic adjustment mechanism, training the corresponding first prediction model locally based on the augmented data to obtain a second prediction model, and calculating the deviation of network parameters at the same position of the second prediction model in different factories; If the deviation exceeds a first threshold value, determining an adjustment value based on the deviation, correcting the network parameters of each second prediction model at the same position based on the adjustment value, and determining whether to retain the corrected network parameters based on the change in prediction accuracy of the second prediction model before and after correction; The second prediction model is defined as a third prediction model, real-time production data is input into the third prediction model, the third prediction model determines the product type of the target product of this production, the dynamic adjustment mechanism adjusts the network layer according to the product type, and outputs the process adjustment amount and the corresponding confidence; When the confidence is less than a second threshold value, the real-time production data is uploaded to a cloud adjustment model, and the adjustment model calls a historical case library to correct the process adjustment amount.

[0007] Further, the data augmentation of the original production data includes the following steps: A first quantity of the original production data is obtained, an interpolation quantity is set, a second quantity is determined based on the first quantity and the interpolation quantity, interpolation data of the second quantity and the same dimension as the original production data are generated under the constraint of the production process, and are defined as base data; The base data most similar to each of the original production data is screened out and defined as first data, a first correction amount is calculated based on the similarity of the original production data and the first data and the data sensitivity of each dimension data, the first data is corrected based on the first correction amount, second data is obtained, a plurality of interpolation data most similar to the second data are obtained and defined as third data, a second correction amount is determined based on the similarity of the second data and the third data, and fourth data is obtained after the third data is corrected based on the second correction amount. The similarity between the original production data and each of the second data and the fourth data is calculated and updated again, and the step is repeated until a preset number of times is reached. The original production data and a plurality of data most similar to the original production data in the base data are taken as the augmented data.

[0008] Further, the third prediction model outputs the confidence of the process adjustment amount, including the following steps: The output result of the third prediction model includes an intermediate quality parameter of each process and a final quality parameter after the product is completed production, a verification model is set, the verification model includes a plurality of calculation formulas, the calculation formulas calculate the intermediate quality parameter of the product after each process and the final quality parameter after the product is completed production according to the real-time production data, a plurality of unknown coefficients are included in the calculation formulas, the verification model is trained based on the original production data to determine the values of the unknown coefficients, the modified data is obtained after the real-time production data is corrected based on the process adjustment amount, the modified data is input into the verification model, the intermediate quality parameter after each process and the final quality parameter are calculated, and the intermediate quality parameter and the final quality parameter of the third prediction model and the verification model are compared to determine the confidence.

[0009] Further, the adjustment model calls the historical case library to correct the process adjustment amount, including the following steps: The historical case library includes a plurality of optimal production cases, the adjustment model calculates the similarity of raw material composition, equipment state and environmental parameters between the real-time production data and each production case, if the similarity of the three parameters of a production case and the real-time production data is greater than a reliable threshold, the process adjustment amount is generated based on the process parameters in the production case and is issued to the factory, otherwise an artificial review process is triggered.

[0010] Further, the third prediction model outputs the process adjustment amount, including the following steps: The third prediction model is a graph neural network, after the real-time production data is input into the third prediction model, the third prediction model obtains the data sensitivity, divides the real-time production data into high-sensitivity data and low-sensitivity data based on the data sensitivity, and predicts the process adjustment amount based on the high-sensitivity data.

[0011] Further, calculating the deviation of network parameters at the same position of the second prediction model in different factories includes the following steps: After the number of the second prediction model reaches a first value, the correction model of the cloud obtains the network parameters of each second prediction model, aggregates the network parameters at the same position into a data group, selects one network parameter in the data group as a first data, and the remaining network parameters as second data, calculates the average value of the second data, and takes the difference between the first data and the average value as the deviation.

[0012] Further, determining the adjustment value based on the deviation includes the following steps: Obtaining the number of data including network parameters in the data group, if the number of data is not greater than a third threshold, setting an adjustment coefficient, reducing the deviation based on the adjustment coefficient to obtain a transition value, correcting the network parameters based on the transition value to obtain the adjustment value; If the number of data is greater than the third threshold, the data is density estimated based on a Gaussian kernel function, the mode of the network parameters in the data group is determined based on the density estimation result, the kurtosis coefficient of each mode is calculated, the correction coefficient is determined based on the kurtosis coefficient, the main mode corresponding to the network parameters is determined, and the distribution distance between the network parameters and the main mode is obtained, the network parameters are corrected based on the distribution distance and the correction coefficient to obtain the adjustment value, the difference between the information entropy before and after the correction of the data group is taken as the entropy change, and when the entropy change is less than a preset entropy tolerance, the correction of the data group is stopped, otherwise the correction of the network parameters in the data group is continued.

[0013] Further, the graph neural network divides the real-time production data into high-sensitive data and low-sensitive data based on the data sensitivity, including the following steps: If the data sensitivity is greater than a critical threshold, the real-time production data is divided into the high-sensitive data, otherwise, the real-time production data is divided into the low-sensitive data.

[0014] The application further provides a heat treatment process parameter deviation intelligent correction system, which is used for realizing the method, and the system comprises: An expansion module is configured to obtain original production data of each process in each factory, expand the original production data, and obtain expanded data. A training module is configured to establish a first prediction model comprising a dynamic adjustment mechanism, train the corresponding first prediction model based on the expanded data, obtain a second prediction model, calculate the deviation of network parameters at the same position of the second prediction model of different factories, determine an adjustment value based on the deviation if the deviation exceeds a first threshold, correct the network parameters of each second prediction model at the same position based on the adjustment value, and determine whether to retain the corrected network parameters based on the change of prediction accuracy of the second prediction model before and after correction. A prediction module defines the second prediction model after correction as a third prediction model, inputs real-time production data into the third prediction model, determines the product type of the target product of this production by the third prediction model, adjusts the network layer according to the product type by the dynamic adjustment mechanism, and outputs a process adjustment amount and a corresponding confidence. A correction module uploads the real-time production data to a cloud adjustment model when the confidence is less than a second threshold, and the adjustment model corrects the process adjustment amount by calling a historical case library.

[0015] The application further provides a computer readable storage medium, which stores instructions, and the instructions are executed by a processor to realize the method.

[0016] The application has the following advantages: The application firstly performs data augmentation on the original production data, improves the number of training samples, and ensures that the model can be completely trained; by establishing a prediction model containing a dynamic adjustment mechanism, the model can automatically adjust the network layer according to the product type, adapt to the process characteristic differences of different products, balance the prediction speed and accuracy; by calculating the deviation of network parameters at the same position in different factories and performing collaborative correction, the influence of single factory data deviation on the model can be reduced, and the generalization ability of the model in multiple scenarios can be improved; by calculating the confidence and combining the historical case library to correct the low confidence result, the frequency of manual re-calibration error threshold is reduced, and through the combination of local training and cloud collaborative correction, the cross-factory knowledge sharing is realized while protecting the factory data privacy, and the accuracy and real-time performance of the whole process parameter correction are improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flow chart of the steps of the heat treatment process parameter deviation intelligent correction method of the application; Figure 2 A structural schematic diagram of the heat treatment process parameter deviation intelligent correction system of the application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the application clearer and more apparent, the application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application, and are not used to limit the application.

[0019] It can be understood that the terms "first", "second", etc. used in the present application can be used herein to describe various elements, but unless specifically stated, these elements are not limited by these terms. These terms are only used to distinguish the first element from another element. For example, without departing from the scope of the present application, the first xx script can be referred to as the second xx script, and similarly, the second xx script can be referred to as the first xx script.

[0020] As shown in Figure 1 A heat treatment process parameter deviation intelligent correction method, comprising: S1: obtaining original production data of each process of each factory, performing data augmentation on the original production data, and obtaining augmented data.

[0021] Taking the manufacture of steel as an example, the manufacturing process includes coking, sintering, smelting, steelmaking and rolling. In order to protect the privacy of the factory production data as much as possible, the embodiment only collects the data in the factory and trains the prediction model locally. In order to obtain the influence relationship between process parameters, a graph neural network is used to establish a first prediction model, and a BP neural network can also be used in other embodiments. The raw material characteristics, process parameters and quality indicators are used as nodes of the graph neural network, and the process adjustment amount is obtained through the fully connected layer set inside. Considering the update iteration of the steelmaking equipment and the change of the environmental data, only the production data in the past three years is used for training. Since only the data in the factory is used to train the model, the amount of data will be relatively small, and the time limit will further reduce the amount of available data. Therefore, the embodiment needs to expand the original production data before training the model, and the specific expansion method will be described in detail later.

[0022] S2: Establish a first prediction model containing a dynamic adjustment mechanism, train the corresponding first prediction model locally based on the expanded data, obtain a second prediction model, and calculate the deviation of the network parameters at the same position of the second prediction models of different factories.

[0023] The first prediction model is specifically the model before training, which has a pre-set dynamic adjustment mechanism. The dynamic adjustment mechanism dynamically adjusts the number of network layers according to the production target. For example, the production target is divided into low-carbon steel, medium-carbon steel and high-carbon steel. The rolling temperature range of low-carbon steel is wide, the deformation resistance is low, and the quality control is relatively simple. The rolling temperature control of high-carbon steel is strict, the deformation resistance is high, and the production process parameters need to be accurately controlled. When the production target is low-carbon steel, fewer network layers are used, such as two graph convolution layers, which can speed up the prediction of the model. When the production target is high-carbon steel, more network layers are used, such as four graph convolution layers. The more the number of graph convolution layers, the more the influence relationship between nodes far apart can be captured, that is, a more complex network structure is used, which can obtain higher prediction accuracy.

[0024] In the embodiment, calculating the deviation of the network parameters at the same position of the second prediction models of different factories includes the following steps: After the number of second prediction models reaches the first value, the correction model of the cloud obtains the network parameters of each second prediction model, aggregates the network parameters at the same position into a data group, selects one network parameter as the first data in the data group, and selects the remaining network parameters as the second data. Calculate the average value of the second data, and the difference between the first data and the average value is the deviation.

[0025] The equipment states and environmental conditions of different factories are different, and the local data may overfit the specific equipment, raw materials or operation habits, resulting in low generalization ability of the model. When the production equipment is replaced or the environmental conditions change, it may have a greater impact on the prediction accuracy of the model. To solve this problem, the network parameters of the second prediction model need to be detected and corrected. Specifically, by calculating the deviation of the same position parameters of different factory models, the network parameters are corrected according to the numerical value of the deviation, and the more the number of second prediction models participating in the correction, the more reliable the correction result. In this embodiment, the first value is set to 10. When the number of second prediction models does not reach the first value, that is, the number of factories where the second prediction model is deployed is less than 10, the factories where the second prediction model has been deployed use the second prediction model that has been trained to correct the production process parameters.

[0026] When the number of second prediction models reaches the first value, the adjustment model of the cloud obtains the network parameters of each second prediction model. For a graph neural network, the network parameter is the weight of the edge, and the network parameter at the same position is the edge with the same two end nodes selected from each second prediction model. The weight of this edge is the network parameter, for example, the edges connecting endpoint A and endpoint B are selected from each second prediction model, and the weights of these edges are combined into a data group. For a fully connected neural layer, the same position refers to the edge with the same two end neurons.

[0027] The first network parameter is extracted from the data group, which is used as the first value. The average value of the remaining network parameters in the data group is calculated as the second value. The difference between the first value and the second value is calculated as the deviation of the first network parameter. The larger the deviation, the greater the difference between the first data and the overall network parameter distribution value. Then the deviation of the second network parameter is calculated, and the process is repeated until the deviations of all network parameters in all data groups are calculated.

[0028] S3: If the deviation exceeds the first threshold value, determine the adjustment value based on the deviation, correct the network parameters of each second prediction model at the same position based on the adjustment value, and determine whether to retain the corrected network parameters based on the change of the prediction accuracy of the second prediction model before and after correction.

[0029] The adjustment value is a value indicating how much the network parameter needs to be modified, and the network parameter is modified by adding the adjustment value to the network parameter. Each time the network parameter of the second prediction model needs to be modified, the second prediction model after the network parameter is modified is defined as a test model. The prediction effect of the test model is verified. Specifically, the verification data is divided from the original production data in advance, and the verification data is other data different from the training data. The accuracy of the output results of the second prediction model and the test model is obtained by inputting the verification data into the second prediction model and the test model respectively. If the accuracy of the test model is greater than that of the second prediction model, the adjustment of the network parameter at this place is retained. If the accuracy of the test model is much less than that of the second prediction model, the network parameter at this place is not modified. Through this step, the prediction accuracy of the model is avoided to be reduced after the network parameter is modified.

[0030] S4: defining the second prediction model after the modification as a third prediction model, inputting the real-time production data into the third prediction model, and the third prediction model determining the product type of the target product of this production, the dynamic adjustment mechanism adjusting the network layer according to the product type, and outputting the process adjustment amount and the corresponding confidence.

[0031] The modified second prediction model is defined as a third prediction model, and real-time production data is input into the third prediction model. The real-time production data includes the current set process parameters, the target quality parameters of the product, the actual quality parameters of the produced product, the environmental parameters and the raw material characteristic parameters, etc. The third prediction model outputs the process adjustment amount and the corresponding confidence according to the difference between the actual quality parameters and the target quality parameters, in combination with the current production process parameters. The confidence represents the reliability of the process adjustment amount. The higher the confidence, the higher the reliability of the process adjustment amount. Generally, a Softmax function can be added to the graph neural network to output the confidence. The present application obtains the confidence by additionally setting a verification model, which will be introduced later.

[0032] S5: when the confidence is less than a second threshold, uploading the real-time production data to an adjustment model in the cloud, and the adjustment model calling a historical case library to modify the process adjustment amount.

[0033] When the confidence is low, that is, less than the second threshold, it indicates that the reliability of the process adjustment amount output by the second prediction model is low. At this time, the real-time production data is uploaded to the cloud, and the cloud is provided with an adjustment model. The adjustment model selects the best process adjustment amount from the historical case library through similarity comparison. If the best process adjustment amount is close to the process adjustment amount output by the third prediction model, the process adjustment amount based on the third prediction model is adjusted. Otherwise, the adjustment is made in an artificial calculation manner. In particular, the prediction result of the third prediction model is not always low in confidence, so the production data is only occasionally uploaded to the cloud, and the privacy of the factory data is still greatly protected.

[0034] In this embodiment, the data augmentation of the original production data includes the following steps: A first quantity of the original production data is obtained, an interpolation quantity is set, a second quantity is determined based on the first quantity and the interpolation quantity, interpolation data of the second quantity is generated under the constraints of the production process and has the same dimension as the original production data, and is defined as the basic data.

[0035] There are various methods for determining the second quantity, such as determining the second quantity by a first formula, and the first formula is specifically N = M * (n + 1), where N is the second quantity, M is the first quantity, and n is the interpolation quantity. When M is 1000 and n is 1, the second quantity is 2000, and the interpolation data generated has the same data dimension as the original production data. For example, the original production data is 40-dimensional data, including smelting temperature, furnace pressure, and cooling speed, etc. The generated interpolation data is also 40-dimensional.

[0036] The interpolation data is generated in this embodiment by the following method. First, according to the upper and lower limit ranges of the data, it is equally divided into K subspaces, and the value of K can be set according to actual needs. For example, the smelting temperature is between 800-1200℃, which is divided into [800, 840), [840, 880) intervals, and then a sample point is randomly selected in each interval. Finally, the random permutation method is used to combine the sample points of each dimension to avoid correlation between dimensions. After generating the interpolation data, the interpolation data is also verified using the process rule library. For example, there is a rule in the process rule library that the quenching temperature must be at least 150℃ higher than the tempering temperature. If the generated interpolation data does not meet this rule, it will be regenerated. The finally retained interpolation data is defined as the basic data.

[0037] The most similar basic data to each original production data is selected and defined as the first data. The first correction quantity is calculated based on the similarity of the original production data and the first data and the data sensitivity of each dimension. The first data is corrected based on the first correction quantity to obtain the second data. The most similar interpolation data to the second data is obtained and defined as the third data. The second correction quantity is determined based on the similarity of the second data and the third data. The fourth data is obtained after the third data is corrected based on the second correction quantity. The similarity between the original production data and each second data and fourth data is calculated and updated again. This step is repeated until a preset number of times. The original production data and the most similar data in the basic data to the original production data are used as the augmented data.

[0038] After obtaining the basic data, the similarity between the original production data and the basic data is calculated. For example, there are 1000 pieces of original production data and 2000 pieces of basic data, and the similarity between each piece of data in the 1000 pieces of original production data and each piece of data in the 2000 pieces of basic data is calculated. In this embodiment, cosine similarity is used for calculation. Before calculating the similarity, the original production data and the basic data need to be standardized to eliminate the influence of dimension. After calculating the similarity between a piece of original production data and each basic data, the basic data most similar to the original production data is selected as the first data.

[0039] Then, the first correction amount is calculated by combining the similarity and the data sensitivity. Specifically, first, the difference between each dimension data of the original production data and the first data is obtained, and then the correction weight of each dimension data is determined through the data sensitivity. The greater the data sensitivity, the greater the influence of the data on the product quality, and thus the smaller the correction weight, so as to avoid large correction of the data value. Before the training of the graph neural network is completed, the data sensitivity can be obtained through statistical analysis of historical data, such as using the Pearson correlation coefficient between the data as the data sensitivity. After the training of the graph neural network is completed, when data augmentation is needed again, more accurate data sensitivity can be determined by parameter analysis of the graph neural network. The specific analysis method is prior art, which will not be introduced here.

[0040] In this embodiment, the first correction amount is calculated by the following second formula: Si=Ci*(1-X)*Wi, where Si is the first correction amount of the i-th dimension data in the first data, C is the difference between the i-th dimension data in the original production data and the i-th dimension data in the first data, X is the similarity between the original production data and the first data, and Wi is the correction weight of the i-th dimension data. In the second formula, the higher the similarity or the greater the data sensitivity, the smaller the first correction amount obtained, that is, the greater the similarity, the smaller the correction value of the first data, and the greater the data sensitivity, the smaller the correction value of the dimension data, so as to avoid large value correction of sensitive data. The first data after final correction is defined as the second data.

[0041] After obtaining the second data, a plurality of data most similar to the second data is obtained from the basic data, such as 10 or 20 data most similar to the second data, which is defined as the third data. The second data is used to correct the third data based on the method of correcting the second data from the first data, and the fourth data is obtained. Through this way, the value of the interpolation data is constantly updated to approach the original production data, and when a preset number of times is reached, the updating is stopped. Finally, a plurality of data most similar to each original production data is selected from the updated basic data as target data, and the original production data and the target data are used as augmented data.

[0042] In this embodiment, the third prediction model outputs the confidence of the process adjustment amount, including the following steps: The output result of the third prediction model includes the intermediate quality parameters of each process and the final quality parameters of the product after completing production. A verification model is set, and the verification model includes a plurality of calculation formulas. The calculation formulas calculate the intermediate quality parameters of the product after each process and the final quality parameters after completing production according to real-time production data. The calculation formulas include a plurality of unknown coefficients. The verification model is trained based on the original production data to determine the values of the unknown coefficients. The modified data is obtained by correcting the real-time production data based on the process adjustment amount. The modified data is input into the verification model to calculate the intermediate quality parameters and the final quality parameters after each process. The intermediate quality parameters and the final quality parameters of the third prediction model and the verification model are compared to determine the confidence.

[0043] As previously introduced, the production process of steelmaking includes multiple processes. Therefore, the raw material is processed into a semi-finished product after each process. The quality parameters of the semi-finished product are the intermediate quality parameters. For hot processing production, there is a clear boundary between processes. Therefore, the intermediate quality parameters can be easily obtained. The final product of steel is obtained after completing production. The quality parameters of the final product of steel are the final quality parameters. In this embodiment, the intermediate quality parameters and the final quality parameters are used as the prediction results of the graph neural network when constructing the first prediction model.

[0044] In order to verify the process adjustment amount, a verification model is also set in this embodiment. The verification model is specifically a formula library. The formula library includes a plurality of physical formulas and empirical formulas for calculating various parameters in the production process, such as the converter charging amount formula, the oxygen flow formula, and the addition amount formula of various substances. There are unknown coefficients that need to be calibrated in these formulas, which need to be obtained by analyzing historical production data. The intermediate quality parameters and the final quality parameters can be obtained by substituting the currently used process parameters into the calculation formulas. The results obtained by the verification model have high interpretability. Therefore, the modified data is obtained by correcting the real-time production data based on the process adjustment amount. The modified data is input into the verification model for calculation to obtain the intermediate quality parameters and the final quality parameters. If the output result of the third prediction model about the quality parameters is similar to the output result of the verification model, it is determined that the output result of the third prediction model has high confidence. Specifically, the confidence can be obtained in the following manner. First, the first similarity of the two intermediate quality parameters and the second similarity of the two final quality parameters are calculated. The average of the first similarity and the second similarity is used as the confidence. The larger the average is, the greater the confidence is.

[0045] In this embodiment, the adjustment model corrects the process adjustment amount by calling the historical case library, including the following steps: The historical case library includes multiple best production cases, the adjustment model calculates the similarity of raw material composition, equipment state and environmental parameters between real-time production data and each production case, if the similarity of the production case and the real-time production data of the three parameters is greater than the reliable threshold, the process adjustment amount is generated based on the process parameters in the production case and is sent to the factory, otherwise the manual review process is triggered.

[0046] Through multi-dimensional similarity comparison, the reliability of the comparison result is ensured, for example, only when the three similarities are greater than 0.85 can the process parameters in the production case be used, avoiding decision bias caused by single-dimensional matching.

[0047] In this embodiment, the third prediction model outputs the process adjustment amount, including the following steps: The third prediction model is a graph neural network, after the real-time production data is input into the third prediction model, the third prediction model obtains the data sensitivity, divides the real-time production data into high-sensitivity data and low-sensitivity data based on the data sensitivity, and predicts the process adjustment amount based on the high-sensitivity data.

[0048] The graph neural network divides the real-time production data into high-sensitivity data and low-sensitivity data based on the data sensitivity, including the following steps: If the data sensitivity is greater than the critical threshold, the real-time production data is divided into high-sensitivity data, otherwise it is divided into low-sensitivity data.

[0049] The third prediction model includes a fully connected layer, after the real-time production data is input into the third prediction model, the real-time production data is divided into high-sensitivity data and low-sensitivity data according to the network parameters in the graph neural network, for example, if the data sensitivity is between 0-1, the critical threshold can be set to 0.6, the data sensitivity greater than 0.6 is divided into high-sensitivity data, and the rest is divided into low-sensitivity data, the critical threshold can be set according to actual needs. High-sensitivity data has a greater impact on the generated quality, and the fully connected layer predicts and calculates the process adjustment amount according to the numerical value of the high-sensitivity data and the data sensitivity. Through this step, only high-sensitivity data is input into the model for processing and prediction, which can reduce the computational complexity and increase the data processing efficiency.

[0050] In this embodiment, the adjustment value is determined based on the bias, including the following steps: After the number of the second prediction model reaches the first value, the correction model of the cloud obtains the network parameters of each second prediction model, aggregates the network parameters at the same position into a data group, selects one network parameter as the first data in the data group, and the remaining network parameters as the second data, calculates the average value of the second data, and the difference between the first data and the average value is the bias.

[0051] The third threshold value is set to 20 for example, when the number of network parameters included in the data set is less than 20, the network parameters are adjusted based on the following method, first, set the adjustment coefficient, the adjustment coefficient can be set according to experience, for example, set the adjustment coefficient to 0.5, then get the deviation, multiply the deviation by the adjustment coefficient, that is, reduce the deviation by 1 / 2, to obtain the transition value, and finally add the network parameters to the transition value, to obtain the adjustment value.

[0052] When the number of deployed second prediction models exceeds 20, the diversity of network parameter distribution increases, and using only a simple correction method may cause the corrected network parameters to deviate far from the original values. Therefore, the embodiment also proposes the following technical solutions.

[0053] If the number of data is greater than the third threshold value, the data is density estimated based on the Gaussian kernel function, the mode of the network parameters in the data set is determined based on the density estimation result, the kurtosis coefficient of each mode is calculated, the correction coefficient is determined based on the kurtosis coefficient, the main mode corresponding to the network parameters is determined, and the distribution distance between the network parameters and the main mode is obtained. The network parameters are corrected based on the distribution distance and the correction coefficient to obtain the adjustment value. The difference between the information entropy before and after the correction of the data set is taken as the entropy change, when the entropy change is less than the preset entropy tolerance, the correction of the data set is stopped, otherwise the correction of the network parameters in the data set is continued.

[0054] The above steps are described below. First, the Gaussian kernel function is used for density estimation, the mode distribution of the data is determined through density estimation, and then it is determined whether the data is unimodal distribution or multimodal distribution. If it is unimodal distribution, it means that the network parameters only have one value tendency, if it is multimodal distribution, it means that the network parameters only have multiple value tendencies. For example, through density estimation, it is determined that the network parameters in the data set are multimodal distribution, and their values are 0.8 and 0.3 respectively, indicating that part of the network parameters in the data set tend to 0.8 and part of them tend to 0.3. Here, the two distributions are defined as the main peak value and the secondary peak value. Then, a plurality of network parameters are selected on the left and right sides of the main peak value and the secondary peak value. The purpose is to calculate the kurtosis coefficient of the main peak value and the secondary peak value. The kurtosis coefficient is used to measure the steepness of the data distribution. When the kurtosis value is larger, the probability distribution graph is higher and sharper, representing that the value distribution of the network parameters is more concentrated.

[0055] Then a control table of kurtosis coefficient and correction coefficient is established, the larger the kurtosis coefficient is, the smaller the correction coefficient is, for example, when the kurtosis coefficient is between 1.5 and 2.5, the correction coefficient is 0.5, the larger the correction coefficient is, the larger the correction proportion of the network parameter is. If the data set is a multimodal distribution, the distance between the network parameter and each mode is obtained, the distance is the absolute value of the difference between the network parameter and the mode, the mode closest to the network parameter is selected according to the distance, and the mode is used as the main mode of the network parameter, for example, the main mode of the network parameter is 0.8, and the network parameter is corrected in the direction of 0.8.

[0056] The network parameter can be corrected by using the following formula: x'=x+D*F, wherein: x' is the corrected network parameter, x is the network parameter before correction, D is the distribution distance between the network parameter and the main mode, and F is the correction coefficient.

[0057] After the correction of all network parameters in the data set is completed, the change of information entropy before and after the correction is calculated, with the continuous correction of the network parameter, the value of the information entropy is smaller and smaller, because the network parameter is more and more concentrated, when the entropy change is less than the preset entropy tolerance, when the information entropy no longer changes greatly, it is indicated that the distribution of the network parameter tends to be stable, otherwise the network parameter is continuously corrected.

[0058] In the embodiment, if the main mode does not exist in the data set, the network parameter corresponding to the data set is not corrected.

[0059] The absence of the main mode represents that the distribution of the data is relatively average, which indicates that the network parameter has no distribution rule, so as to avoid reducing the prediction accuracy of the prediction model, the network parameter at this position is not modified.

[0060] As shown in Figure 2 The present application also provides a heat treatment process parameter deviation intelligent correction system, which is used for realizing the above method, and the system comprises: An expansion module is configured to obtain original production data of each process of each factory, expand the original production data, and obtain expanded data; A training module is configured to establish a first prediction model comprising a dynamic adjustment mechanism, train a corresponding second prediction model based on the expanded data, obtain the second prediction model, calculate the deviation of network parameters at the same position of the second prediction models of different factories, determine an adjustment value based on the deviation if the deviation exceeds a first threshold value, correct the network parameters at the same position of each second prediction model based on the adjustment value, and determine whether to retain the corrected network parameters based on the change of prediction accuracy of the second prediction models before and after the correction. The prediction module defines the second prediction model after the correction as a third prediction model, inputs the real-time production data into the third prediction model, the third prediction model determines the product type of the target product in this production, the dynamic adjustment mechanism adjusts the network layer according to the product type, and outputs the process adjustment amount and the corresponding confidence; The correction module uploads the real-time production data to the adjustment model in the cloud when the confidence is less than the second threshold value, and the adjustment model calls the historical case library to correct the process adjustment amount.

[0061] The application further provides a computer readable storage medium, and instructions are stored on the computer readable storage medium.

[0062] The technical features of the above embodiments can be combined arbitrarily, and to make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, it should be considered that they are within the scope of the present application.

[0063] The above embodiments only express several embodiments of the present application, the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

[0064] The above is only the preferred embodiment of the present application, and does not limit the present application, any modification, equivalent replacement and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for intelligent correction of deviations in heat treatment process parameters, characterized in that, include: Obtain the original production data of each process in each factory, and augment the original production data to obtain augmented data; A first prediction model with a dynamic adjustment mechanism is established. Based on the expanded data, the corresponding first prediction model is trained locally to obtain a second prediction model. The deviation of the network parameters at the same position of the second prediction model in different factories is calculated. If the deviation exceeds the first threshold, an adjustment value is determined based on the deviation, the network parameters of each of the second prediction models at the same position are corrected based on the adjustment value, and it is determined whether to retain the corrected network parameters based on the change in the prediction accuracy of the second prediction models before and after the correction. The revised second prediction model is defined as the third prediction model. Real-time production data is input into the third prediction model. The third prediction model determines the product type of the target product for this production. The dynamic adjustment mechanism adjusts the number of network layers according to the product type and outputs the process adjustment amount and the corresponding confidence level. When the confidence level is less than the second threshold, real-time production data is uploaded to the adjustment model in the cloud, and the adjustment model calls the historical case library to correct the process adjustment amount.

2. The method according to claim 1, characterized in that, Data augmentation of the original production data includes the following steps: Obtain a first quantity of the original production data, set an interpolation quantity, determine a second quantity based on the first quantity and the interpolation quantity, and generate interpolated data of the second quantity with the same dimension as the original production data under the constraints of the production process, which is defined as the basic data; The basic data most similar to each of the original production data is selected and defined as the first data. A first correction amount is calculated based on the similarity between the original production data and the first data and the data sensitivity of each dimension of the data. The first data is corrected based on the first correction amount to obtain the second data. Multiple interpolated data most similar to the second data are obtained and defined as the third data. A second correction amount is determined based on the similarity between the second data and the third data. The third data is corrected based on the second correction amount to obtain the fourth data. The similarity between the original production data and each of the second data and the fourth data is calculated again and updated. This step is repeated until a preset number of times is reached. The original production data and multiple data most similar to the original production data in the basic data are used as the expanded data.

3. The method according to claim 1, characterized in that, The third prediction model outputs the confidence level of the process adjustment amount by including the following steps: The output of the third prediction model includes intermediate quality parameters for each process and final quality parameters after the product is completed. A validation model is set up, which includes multiple calculation formulas. These formulas calculate the intermediate quality parameters after each process and the final quality parameters after production based on the real-time production data. Each calculation formula includes multiple unknown coefficients. The validation model is trained based on the original production data to determine the values ​​of the unknown coefficients. Corrected data is obtained by adjusting the real-time production data based on the process adjustment amount. The corrected data is input into the validation model to calculate the intermediate quality parameters and the final quality parameters after each process. The intermediate quality parameters and the final quality parameters of the third prediction model and the validation model are compared to determine the confidence level.

4. The method according to claim 3, characterized in that, The adjustment model calls upon the historical case library to correct the process adjustment amount, including the following steps: The historical case library includes multiple best production cases. The adjustment model calculates the similarity of raw material composition, equipment status, and environmental parameters between the real-time production data and each production case. If the similarity of a production case with the real-time production data in all three parameters is greater than the reliable threshold, the process adjustment amount is generated based on the process parameters in that production case and sent to the factory; otherwise, a manual review process is triggered.

5. The method according to claim 2, characterized in that, The third prediction model outputs the process adjustment amount by including the following steps: The third prediction model is a graph neural network. After the real-time production data is input into the third prediction model, the third prediction model obtains the data sensitivity, divides the real-time production data into high-sensitivity data and low-sensitivity data based on the data sensitivity, and predicts the process adjustment amount based on the high-sensitivity data.

6. The method according to claim 5, characterized in that, Calculating the deviation of the network parameters at the same location in the second prediction model for different factories includes the following steps: After the number of the second prediction models reaches a first value, the cloud-based correction model obtains the network parameters of each of the second prediction models, aggregates the network parameters at the same location into a data group, selects one network parameter from the data group as the first data, and uses the remaining network parameters as the second data, calculates the average value of the second data, and uses the difference between the first data and the average value as the deviation.

7. The method according to claim 6, characterized in that, Determining the adjustment value based on the deviation includes the following steps: The number of data including network parameters in the data group is obtained. If the number of data is not greater than a third threshold, an adjustment coefficient is set. The deviation is reduced based on the adjustment coefficient to obtain a transition value. The network parameters are corrected based on the transition value to obtain the adjustment value. If the amount of data exceeds the third threshold, density estimation is performed on the data based on the Gaussian kernel function. Based on the density estimation result, the mode of the network parameters in the data group is determined, the kurtosis coefficient of each mode is calculated, a correction coefficient is determined based on the kurtosis coefficient, the dominant mode corresponding to the network parameter is determined, and the distribution distance between the network parameter and the dominant mode is obtained. The network parameter is corrected based on the distribution distance and the correction coefficient to obtain the adjusted value. The difference in information entropy before and after the correction of the data group is taken as the entropy change. When the entropy change is less than the preset entropy tolerance, the correction of the data group is stopped; otherwise, the correction of the network parameters in the data group continues.

8. The method according to claim 7, characterized in that, The graph neural network divides the real-time production data into high-sensitivity data and low-sensitivity data based on the data sensitivity, including the following steps: If the data sensitivity is greater than the critical threshold, the real-time production data is classified as high-sensitivity data; otherwise, it is classified as low-sensitivity data.

9. A heat treatment process parameter deviation intelligent correction system, used to implement the method as described in any one of claims 1-8, characterized in that, The system includes: An expansion module is used to acquire the original production data of each process in each factory, and to expand the original production data to obtain expanded data. The training module is used to establish a first prediction model containing a dynamic adjustment mechanism, train the corresponding first prediction model locally based on the expanded data to obtain a second prediction model, calculate the deviation of network parameters at the same position of the second prediction model in different factories, if the deviation exceeds a first threshold, determine an adjustment value based on the deviation, correct the network parameters of each second prediction model at the same position based on the adjustment value, and determine whether to retain the corrected network parameters based on the change in prediction accuracy of the second prediction model before and after correction. The prediction module defines the corrected second prediction model as the third prediction model. The prediction module inputs real-time production data into the third prediction model. The third prediction model determines the product type of the target product for this production. The dynamic adjustment mechanism adjusts the number of network layers according to the product type and outputs the process adjustment amount and the corresponding confidence level. The correction module uploads real-time production data to the adjustment model in the cloud when the confidence level is less than the second threshold. The adjustment model then calls the historical case library to correct the process adjustment amount.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the method as described in any one of claims 1-8.

Images