Method and device for predicting convexity of medium-thickness plate based on data mining and multi-model fusion
Through the integration of data mining and multi-model, the problems of low plate convexity prediction accuracy and poor interpretation in medium-thick plate processing are solved, and higher-precision plate convexity prediction and process guidance are achieved, and the production capacity of steel plants is improved. and product quality.
Patent Information
- Application Number
- CN202411197152.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-11-29
AI Technical Summary
During the processing of medium and thick plates, traditional rolling mechanism models are difficult to effectively control the convexity of the plate, resulting in low prediction accuracy and poor interpretation, and cannot guide the actual process production.
Using a method based on data mining and multi-model fusion, the data is expanded through one-dimensional projection transformation and Bootstrap, and the ridge regression and BP neural network models are built to predict the convexity of the medium and thick plates, and the reverse derivation is analyzed. The influence of characteristic parameters on the convexity of the plate.
The accuracy and interpretability of the prediction of convexity of medium and thick plates can be improved, and the process production of steel plants can be guided more accurately, and the production capacity and product quality can be improved.
Smart Images

Figure CN119047322A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medium and thick plate rolling, and in particular to a method and device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion. Background Art
[0002] Medium and thick plates reflect the level of the steel industry to a certain extent. In recent years, improving the plate shape quality of plate products has gradually become an urgent problem to be solved by modern high-precision rolling mills. The plate shape problem has always been a pain point that restricts the quality of medium and thick plate products. In actual production, serious plate shape problems will cause the batch of products to directly become scrap, causing serious economic losses. As one of the main quality indicators of medium and thick plates, how to effectively control the plate convexity is one of the key issues studied in the industry. However, in the processing of medium and thick plates, the various rolling process parameters are coupled with each other and have strong real variability. The traditional rolling mechanism model is difficult to achieve the ideal control effect, making it difficult for process production personnel to accurately preset the plate convexity to meet product requirements in actual production.
[0003] Previous studies on the convexity of medium and thick plates were mostly based on a single influencing factor or multiple single influencing factors, ignoring the mutual influence between different parameters, resulting in unstable conclusions. With the development of artificial intelligence, the integration of mechanism models with industrial big data-driven models has gradually emerged. Scholars have applied big data models such as neural networks to plate convexity prediction. However, the production data of medium and thick plates covers all production processes from slabs to final forming. The data dimensions are not unified and present multi-source heterogeneity, resulting in defects such as difficulty in processing multi-source heterogeneous data in existing prediction models, resulting in low prediction accuracy; and the multi-layer complex neural network model has poor interpretability and can only achieve forward mapping from input parameters to output plate convexity, and cannot reversely deduce the influence of parameters on product convexity, making it difficult to guide and optimize actual process production.
[0004] To this end, a convexity prediction method for medium and thick plates with high precision and good interpretability is proposed to effectively guide process production personnel in designing process parameters, so that steel mills can achieve higher production capacity and product quality under safer operating conditions, which has strong practical significance. Summary of the invention
[0005] In view of this, the present invention provides a method and device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion to solve at least one of the problems mentioned above.
[0006] In order to achieve the above object, the present invention adopts the following scheme:
[0007] According to a first aspect of the present invention, a method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion is provided, the method comprising: obtaining historical industrial data of a medium and thick plate production line, the historical industrial data comprising slab performance parameters, rolling process parameters and rolling process parameters; performing a one-dimensional projection transformation on the rolling process parameters of each slab to obtain a one-dimensional projection value, and using the one-dimensional projection value as a principal component variable representing the slab rolling process; dividing the historical industrial data into three sub-sample sets according to a set convexity critical value, obtaining the optimal sample size ratio of the three sample subsets based on the logistic regression result, and expanding the three sample subsets according to the optimal imbalance ratio by using an improved SMOTE method based on Bootstrap; Historical industrial data are used to make the three sample sizes reach an optimal balance state, and the expanded historical industrial data are standardized; a ridge regression prediction model for predicting the convexity of medium and thick plates based on ridge regression is established; the error of the ridge regression prediction model is calculated, and the error is used as the dependent variable to establish an error prediction model based on a BP neural network; an integrated model is constructed with the predicted value of the ridge regression prediction model as the main value and the predicted value of the error prediction model as the deviation, and the integrated model is trained using the standardized historical industrial data; based on the training results of the integrated model, the weights and deviations of each layer of the ridge regression prediction model and the error prediction model are obtained, and the final prediction formula for predicting the convexity of medium and thick plates is obtained.
[0008] As an embodiment of the present invention, after obtaining the final prediction formula for predicting the convexity of medium and thick plates in the above method, the method also includes: quantitatively inferring the influence of each characteristic variable on the convexity value of medium and thick plates based on the weight coefficients of the ridge regression prediction model and the error prediction model to guide the actual production process.
[0009] As an embodiment of the present invention, in the above method, a one-dimensional projection transformation is performed on the rolling process parameters of each slab to obtain a one-dimensional projection value, and the one-dimensional projection value is used as the principal component variable representing the slab rolling process, including: standardizing the rolling process data corresponding to each slab; calculating the covariance matrix of the rolling process data set after standardization, and calculating the eigenvalues and corresponding eigenvectors of the covariance matrix, and sorting the eigenvalues in reverse order; projecting the standardized rolling process data set into the space spanned by the eigenvector corresponding to the first eigenvalue after sorting in reverse order, as the principal component variable representing the slab rolling process.
[0010] As an embodiment of the present invention, in the above method, the historical industrial data is divided into three sub-sample sets according to the set convexity critical value, and the optimal sample size ratio and of the three sample subsets are calculated based on the logistic regression result, and the improved SMOTE method based on Bootstrap is used to expand the historical industrial data according to the optimal imbalance ratio so that the three sample sizes reach the optimal balance state, and the expanded historical industrial data is standardized, including:
[0011] Step 1: Divide the original data set consisting of the historical industrial data into a benchmark training set T0 and a benchmark test set T1 according to an 8:2 ratio;
[0012] Step 2: Based on the distribution diagram of the convexity data of medium and thick plates, select the convexity critical values C1 and C2, where C1>C2, and divide the samples in the original data set into the maximum plate convexity value sample S1, the normal plate convexity value sample S0 and the minimum plate convexity value sample S-1, where the plate convexity value of S1 is greater than C1, the plate convexity value of S0 is between C1 and C2, and the plate convexity value of S-1 is less than C2;
[0013] Step 3: Set the unbalance ratio of S1 to S0 to r1, and the unbalance ratio of S-1 to S0 to r2;
[0014] Step 4: Calculate the sample center point x of S1 1,center , and calculate for each sample point x in S1 i Distance x 1,center The Euclidean distance d i , calculate {d i} i=1,2,…n The average value d average , if d i <d average , x i is divided into safe points, otherwise, x i Divided into noise points, the set of all safe points is recorded as S 1,safe ;
[0015] Step 5: Use Bootstrap to create S 1,safe Perform random weighted sampling and calculate the weighted average point c after resampling 1,weigh ted , according to the formula x new =c 1,weigh ted +a(c 1,weigh ted -x 1,center ) Generate new sample data x new , where a is a random number that follows a random distribution in the interval [0,1];
[0016] Step 6: According to the imbalance ratio r1, based on the safe point set S 1,safe ,generate New sample data;
[0017] Step 7: Repeat steps 4 to 6, based on the imbalance ratio r2 and the safe point set S 2,safe ,generate New sample data;
[0018] Step 8: Merge the new sample data with the benchmark training set T0 to form an expanded new data set T0';
[0019] Step 9: According to the convexity critical value selected in step 2, T0' is divided into abnormal convexity value samples and normal convexity value samples. The abnormal samples are marked as 1 and the normal samples are marked as 0. A plate convexity abnormal value classification model based on logistic regression is constructed. The classification effect of the logistic regression model is verified using the benchmark test set T1. The F value and accuracy are used as evaluation indicators. Starting from r1=0 and r2=0, the stepwise growth method is used to find the imbalance ratio r1 that makes the model work best. * and r2 * ;
[0020] Step 10: Optimal imbalance ratio r1 obtained from step 9 * and r2 * According to steps 4 to 8, we can obtain the data set T0' of the fused expanded data under the optimal imbalance ratio. * ;
[0021] Step 11: T0' according to the standardized formula * Perform standardization.
[0022] As an embodiment of the present invention, the method for establishing a ridge regression prediction model for predicting the convexity of a medium and thick plate based on ridge regression includes: using a cross-validation method to find an optimal regularization parameter; based on the optimal regularization parameter, taking the convexity value of the medium and thick plate as output, constructing a ridge regression prediction model based on ridge regression:
[0023] y r =w r x+b r ;
[0024] In the above formula, y r is the plate convexity value predicted by the ridge regression prediction model, x is the matrix of independent variables, w r is the weight, b r is the deviation; based on the training results of the ridge regression prediction model, the weight w of the ridge regression model is obtained r and deviation b r .
[0025] As an embodiment of the present invention, the above method calculates the error of the ridge regression prediction model, takes the error as the dependent variable, and establishes an error prediction model based on the BP neural network, including: calculating the prediction error of the ridge regression model, taking the prediction error as the output, and constructing a ridge regression prediction error model based on the BP neural network; using the mean square error MSE as an evaluation index of the performance of the BP neural network model, starting from a single hidden layer, correcting the network parameters through a step-by-step layer-by-layer growth method, and finding the neuron combination that makes the MSE value the lowest.
[0026] As an embodiment of the present invention, the above method constructs an integrated model with the predicted value of the ridge regression prediction model as the main value and the predicted value of the error prediction model as the deviation, and uses the standardized historical industrial data to train the integrated model, including: constructing a ridge regression model prediction value y r is the main value, and the BP neural network model predicts the value e BP is the integrated model of deviation, and the predicted value of the integrated model is Use the standardized data set T0' obtained in step 11 * The integrated model is trained; and the weights and deviations between the neural network layers of the BP neural network model are obtained according to the model training results.
[0027] As an embodiment of the present invention, the final prediction formula for the convexity prediction of the medium and thick plate in the above method is as follows:
[0028]
[0029] In the above formula, w r is the ridge regression model weight, b r is the ridge regression model bias; The weights and biases transferred from the BP neural network input layer to the first hidden layer; is the weight and bias transferred from the first hidden layer to the second hidden layer of the BP neural network; w 3,j (j=1,2…,11) is the weight and bias transmitted from the second hidden layer of the BP neural network to the output layer; σ is the ReLu activation function.
[0030] According to a first aspect of the present invention, there is provided a medium and thick plate convexity prediction device based on data mining and multi-model fusion, the device comprising: a data acquisition unit, used to acquire historical industrial data of a medium and thick plate production line, the historical industrial data comprising slab performance parameters, rolling process parameters and rolling process parameters; a one-dimensional projection unit, used to perform a one-dimensional projection transformation on the rolling process parameters of each slab to obtain a one-dimensional projection value, and use the one-dimensional projection value as a principal component variable representing the slab rolling process; a data expansion unit, used to divide the historical industrial data into three sub-sample sets according to a set convexity critical value, obtain the optimal sample size ratio of the three sample subsets based on the logistic regression result, and use the improved SMOTE method based on Bootstrap to expand the historical industrial data according to the optimal imbalance ratio, so that the three The sample size reaches the optimal balance state, and the expanded historical industrial data is standardized; a ridge regression model building unit is used to establish a ridge regression prediction model for predicting the convexity of medium and thick plates based on ridge regression; an error prediction model building unit is used to calculate the error of the ridge regression prediction model, and use the error as the dependent variable to establish an error prediction model based on the BP neural network; an integrated model construction unit is used to construct an integrated model with the predicted value of the ridge regression prediction model as the main value and the predicted value of the error prediction model as the deviation, and the integrated model is trained using the standardized historical industrial data; a final prediction unit is used to obtain the weights and deviations of each layer of the ridge regression prediction model and the error prediction model based on the training results of the integrated model, and obtain the final prediction formula for predicting the convexity of medium and thick plates.
[0031] As an embodiment of the present invention, the above-mentioned device also includes: an influence back-estimation unit, which is used to quantitatively back-estimating the influence of each characteristic variable on the convexity value of medium and thick plates based on the weight coefficients of the ridge regression prediction model and the error prediction model, so as to guide the actual production process.
[0032] As an embodiment of the present invention, the above-mentioned one-dimensional projection unit includes: a standardization processing module, which is used to standardize the rolling process data corresponding to each slab; an eigenvalue sorting module, which is used to calculate the covariance matrix of the rolling process data set after standardization, and calculate the eigenvalues and corresponding eigenvectors of the covariance matrix, and sort the eigenvalues in reverse order; a projection module, which is used to project the standardized rolling process data set into the space spanned by the eigenvector corresponding to the first eigenvalue after reverse sorting, as the principal component variable representing the slab rolling process.
[0033] As an embodiment of the present invention, the data expansion unit is specifically used to perform the following steps:
[0034] Step 1: Divide the original data set consisting of the historical industrial data into a benchmark training set T0 and a benchmark test set T1 according to an 8:2 ratio;
[0035] Step 2: Based on the distribution diagram of the convexity data of medium and thick plates, select the convexity critical values C1 and C2, where C1>C2, and divide the samples in the original data set into the maximum plate convexity value sample S1, the normal plate convexity value sample S0 and the minimum plate convexity value sample S-1, where the plate convexity value of S1 is greater than C1, the plate convexity value of S0 is between C1 and C2, and the plate convexity value of S-1 is less than C2;
[0036] Step 3: Set the unbalance ratio of S1 to S0 to r1, and the unbalance ratio of S-1 to S0 to r2;
[0037] Step 4: Calculate the sample center point x of S1 1,center , and calculate for each sample point x in S1 i Distance x 1,center The Euclidean distance d i , calculate {d i} i=1,2,…n The average value d average , if d i <d average , x i is divided into safe points, otherwise, x i Divided into noise points, the set of all safe points is recorded as S 1,safe ;
[0038] Step 5: Use Bootstrap to create S 1,safe Perform random weighted sampling and calculate the weighted average point c after resampling 1,weigh ted , according to the formula x new =c 1,weigh ted +a(c 1,weigh ted -x 1,center ) Generate new sample data x new , where a is a random number that follows a random distribution in the interval [0,1];
[0039] Step 6: According to the imbalance ratio r1, based on the safe point set S 1,safe ,generate New sample data;
[0040] Step 7: Repeat steps 4 to 6, based on the imbalance ratio r2 and the safe point set S 2,safe ,generate New sample data;
[0041] Step 8: Merge the new sample data with the benchmark training set T0 to form an expanded new data set T0';
[0042] Step 9: According to the convexity critical value selected in step 2, T0' is divided into abnormal convexity value samples and normal convexity value samples. The abnormal samples are marked as 1 and the normal samples are marked as 0. A plate convexity abnormal value classification model based on logistic regression is constructed. The classification effect of the logistic regression model is verified using the benchmark test set T1. The F value and accuracy are used as evaluation indicators. Starting from r1=0 and r2=0, the stepwise growth method is used to find the imbalance ratio r1 that makes the model work best. * and r2 * ;
[0043] Step 10: Optimal imbalance ratio r1 obtained from step 9 * and r2 * According to steps 4 to 8, we can obtain the data set T0' of the fused expanded data under the optimal imbalance ratio. * ;
[0044] Step 11: T0' according to the standardized formula * Perform standardization.
[0045] As an embodiment of the present invention, the ridge regression model building unit includes: an optimal parameter acquisition module, which is used to find the optimal regularization parameter by using a cross-validation method; a ridge regression model building module, which is used to build a ridge regression prediction model based on ridge regression based on the optimal regularization parameter and taking the convexity value of the medium and thick plate as the output:
[0046] y r =w r x+b r ;
[0047] In the above formula, y r is the plate convexity value predicted by the ridge regression prediction model, x is the matrix of independent variables, w r is the weight, b r is a deviation; a weight and deviation acquisition module, which is used to obtain the weight w of the ridge regression model based on the training results of the ridge regression prediction model r and deviation b r .
[0048] As an embodiment of the present invention, the above-mentioned error prediction model establishment unit includes: an error model construction module, which is used to calculate the prediction error of the ridge regression model, take the prediction error as output, and construct a ridge regression prediction error model based on the BP neural network; a neural combination optimization module, which is used to use the mean square error MSE as an evaluation index of the performance of the BP neural network model, starting from a single hidden layer, and correcting the network parameters through a step-by-step layer-by-layer growth method to find the neuron combination that makes the MSE value the lowest.
[0049] As an embodiment of the present invention, the above-mentioned integrated model construction unit includes: an integrated model construction module for constructing a ridge regression model prediction value y r is the main value, and the BP neural network model predicts the value e BP is the integrated model of deviation, and the predicted value of the integrated model is Model training module, used to use the standardized data set T0' obtained in step 11 * The integrated model is trained; a weight and deviation acquisition module is used to obtain the weights and deviations between the neural network layers of the BP neural network model according to the model training results.
[0050] As an embodiment of the present invention, the final prediction formula for the above-mentioned convexity prediction of the medium and thick plate is as follows:
[0051]
[0052] In the above formula, w r is the ridge regression model weight, b r is the ridge regression model bias; The weights and biases transferred from the BP neural network input layer to the first hidden layer; is the weight and bias transferred from the first hidden layer to the second hidden layer of the BP neural network; w 3,j (j=1,2…,11) is the weight and bias transmitted from the second hidden layer of the BP neural network to the output layer; σ is the ReLu activation function.
[0053] According to a third aspect of the present invention, there is provided an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.
[0054] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.
[0055] It can be seen from the above technical scheme that the method and device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion provided by the present invention firstly adopts one-dimensional projection for the rolling process data, thereby weakening the multi-source heterogeneous phenomenon presented therein; secondly, the data set is expanded by using the improved SMOTE method integrating Bootstrap, thereby alleviating the data imbalance phenomenon; and then a plate convexity prediction model based on ridge regression is constructed, and then the difference between the predicted value and the actual value of the ridge regression model is used as the predicted value output by the BP neural network, thereby constructing an integrated model with ridge regression as the main value and BP neural network as the deviation. On the one hand, the present invention provides a new idea for processing the multi-source heterogeneity phenomenon of medium and thick plate production data and data mining, so that data mining technology can serve the actual process production; on the other hand, since the predicted value of ridge regression is close to the actual value in the same order of magnitude, the deviation between the predicted value of ridge regression and the actual value is used as the output of the BP neural network model, which can further reduce the range of the prediction error and be closer to the actual value, so that the prediction accuracy of the model is higher; finally, the prediction method proposed in the present invention can derive the influence of the change of each characteristic parameter on the final plate convexity through reverse deduction, so that the model has explanatory power, so that the data can reflect and effectively guide the actual process production; therefore, the present invention has a strong promotion ability, which can not only improve the prediction accuracy of the export plate convexity of the medium and thick plate, but also can quantitatively analyze the influence of the change of each characteristic parameter on the final plate convexity, so as to effectively guide the steel plant to control the convexity of the medium and thick plate products. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:
[0057] Figure 1 It is a flow chart of a method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion provided in an embodiment of the present application;
[0058] Figure 2 It is a schematic diagram of a process of performing one-dimensional projection transformation on rolling process parameters provided in an embodiment of the present application;
[0059] Figure 3 It is a schematic diagram of a process for establishing a ridge regression prediction model provided in an embodiment of the present application;
[0060] Figure 4 It is a schematic diagram of a process for establishing an error prediction model provided in an embodiment of the present application;
[0061] Figure 5It is a BP neural network structure diagram provided in an embodiment of the present application;
[0062] Figure 6 It is a schematic diagram of the process of establishing an integrated model provided in an embodiment of the present application;
[0063] Figure 7 is a comparison chart of the predicted value and the actual value of the integrated model provided in the embodiment of the present application;
[0064] Figure 8 It is a structural schematic diagram of a medium and thick plate convexity prediction device based on data mining and multi-model fusion provided in an embodiment of the present application;
[0065] Fig. 9 It is a structural schematic diagram of a medium and thick plate convexity prediction device based on data mining and multi-model fusion provided by another embodiment of the present application;
[0066] Fig.10 is a structural schematic diagram of a one-dimensional projection unit provided in an embodiment of the present application;
[0067] Fig.11 It is a structural schematic diagram of a ridge regression model establishment unit provided in an embodiment of the present application;
[0068] Fig.12 It is a structural schematic diagram of an error prediction model establishment unit provided in an embodiment of the present application;
[0069] Fig.13 is a structural schematic diagram of an integrated model building unit provided in an embodiment of the present application;
[0070] Fig.14 It is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0071] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.
[0072] like Figure 1 The figure shows a flow chart of a method for predicting the convexity of a medium and thick plate based on data mining and multi-model fusion provided in an embodiment of the present application. The method comprises the following steps:
[0073] Step S101: Acquire historical industrial data of a medium and thick plate production line, wherein the historical industrial data includes slab performance parameters, rolling process parameters and rolling process parameters.
[0074] The above historical industrial data can be obtained by collecting the actual values of the convexity of medium and thick plate products in actual steel mills, the measured data of each process flow of the production line, and the automated calculation data in the reversible rolling process.
[0075] When the amount of historical industrial data in actual production is insufficient, we can rely on digital twin technology to build a digital virtual entity of the medium and thick plate production process, simulate the medium and thick plate production process, and thus obtain the above-mentioned rolling industrial data; or by collecting historical data from the steel industry, build a big database, and then obtain the above-mentioned medium and thick plate production line data from the cloud platform.
[0076] Preferably, the above-mentioned slab performance parameters include but are not limited to: slab thickness, slab width, slab length, slab weight, slab temperature before charging into the furnace, finishing roll matching number, steel type, etc.
[0077] The rolling process parameters include but are not limited to: the number of intermediate cooling passes, whether controlled cooling technology is used, whether controlled rolling technology is used, final rolling temperature, time in the furnace, aspect ratio, number of small plates, upper working roll diameter, lower working roll diameter, final rolling thickness, final rolling width, final rolling length, number of rough rolling passes, number of finishing rolling passes, furnace exit temperature, maximum temperature after descaling, number of pre-straightening passes, total number of straightening passes, etc.
[0078] The rolling process parameters include but are not limited to: the rolling force of all rough rolling passes, the rolling force of all finishing rolling passes, and the roll shifting amount of all finishing rolling passes.
[0079] After obtaining the above historical industrial data and cleaning the data, the data can be stored in a relational database or a big data platform, and the data warehouse technology can be used to classify and manage the data for subsequent analysis and processing. At the same time, data management specifications are established and data is backed up regularly to ensure data security and confidentiality. Through the above steps, the historical industrial data obtained can be ensured to be comprehensive and accurate, providing a reliable data foundation for the subsequent medium and thick plate convexity prediction model.
[0080] Step S102: Perform one-dimensional projection transformation on the rolling process parameters of each slab to obtain a one-dimensional projection value, and use the one-dimensional projection value as the principal component variable representing the slab rolling process. That is, perform one-dimensional projection transformation on the rough rolling force, the finishing rolling force and the finishing roll shifting amount of each pass, and then use the one-dimensional projection value as the principal component variable representing the slab rolling process.
[0081] Preferably, Figure 2 As shown, this step may further include the following sub-steps:
[0082] Step S1021: Standardize the rolling process data corresponding to each slab. The standardization formula is as follows:
[0083]
[0084] In the above formula, x′ is the characteristic variable matrix after standardization; x is the original value of the characteristic variable; is the sample data mean, N is the sample size, This is the sample data standard.
[0085] Step S1022: Calculate the covariance matrix of the rolling process data set x′ after the standardization, and calculate the eigenvalues {λ i} i=1,2,...n and the corresponding eigenvectors, sorting the eigenvalues in reverse order.
[0086] Step S1023: Project the standardized rolling process data set x′ into the space spanned by the eigenvector corresponding to the first eigenvalue after being sorted in reverse order, as the principal component variable representing the slab rolling process.
[0087] Through the above steps, the multidimensional data is simplified into one-dimensional principal component variables, reducing the data dimension and improving the calculation efficiency. Moreover, the standardization and projection transformation processing make the data comparable and consistent, providing simplified and effective data input for the subsequent medium and thick plate crown prediction model.
[0088] Step S103: According to the set convexity critical value, the historical industrial data is divided into three sub-sample sets, and based on the logistic regression results, the optimal sample size ratio of the three sample subsets is obtained. The improved SMOTE method based on Bootstrap is used to expand the historical industrial data according to the optimal imbalance ratio so that the three sample sizes reach the optimal balance state, and the expanded historical industrial data is standardized.
[0089] Preferably, this step may further include the following sub-steps 1-11:
[0090] Step 1: Divide the original data set consisting of the historical industrial data into a benchmark training set T0 and a benchmark test set T1 according to an 8:2 ratio;
[0091] Step 2: Based on the distribution diagram of the convexity data of medium and thick plates, select the convexity critical values C1 and C2, where C1>C2, and divide the samples in the original data set into the maximum plate convexity value sample S1, the normal plate convexity value sample S0 and the minimum plate convexity value sample S-1, where the plate convexity value of S1 is greater than C1, the plate convexity value of S0 is between C1 and C2, and the plate convexity value of S-1 is less than C2;
[0092] Step 3: Set the unbalance ratio of S1 to S0 to r1, and the unbalance ratio of S-1 to S0 to r2;
[0093] Step 4: Calculate the sample center point x of S1 1,center , and calculate for each sample point x in S1 i Distance x 1,center The Euclidean distance d i , calculate {d i} i=1,2,…n The average value d average , if d i <d average , x i is divided into safe points, otherwise, x i Divided into noise points, the set of all safe points is recorded as S 1,safe ;
[0094] Step 5: Use Bootstrap to create S 1,safe Perform random weighted sampling and calculate the weighted average point c after resampling 1,weigh ted , according to the formula x new =c 1,weigh ted +a(c 1,weigh ted -x 1,center ) Generate new sample data x new , where a is a random number that follows a random distribution in the interval [0,1];
[0095] Step 6: According to the imbalance ratio r1, based on the safe point set S 1,safe ,generate New sample data;
[0096] Step 7: Repeat steps 4 to 6, based on the imbalance ratio r2 and the safe point set S 2,safe ,generate New sample data;
[0097] Step 8: Merge the new sample data with the benchmark training set T0 to form an expanded new data set T0';
[0098] Step 9: According to the convexity critical value selected in step 2, T0' is divided into abnormal convexity value samples and normal convexity value samples. The abnormal samples are marked as 1 and the normal samples are marked as 0. A plate convexity abnormal value classification model based on logistic regression is constructed. The classification effect of the logistic regression model is verified using the benchmark test set T1. The F value and accuracy are used as evaluation indicators. Starting from r1=0 and r2=0, the stepwise growth method is used to find the imbalance ratio r1 that makes the model work best. * and r2 * ;
[0099] Step 10: Optimal imbalance ratio r1 obtained from step 9* and r2 * According to steps 4 to 8, we can obtain the data set T0' of the fused expanded data under the optimal imbalance ratio. * ;
[0100] Step 11: T0' according to the standardized formula * Perform standardization.
[0101] From the above steps, it can be seen that this application weakens the data imbalance and improves the learning performance and generalization ability of the model through Bootstrap and improved SMOTE methods; secondly, through standardization, the dimensional differences are eliminated to make the data comparable, which is convenient for subsequent model training and prediction. Through reasonable data expansion and standardization, the prediction accuracy and stability of the model can be improved. Therefore, through the above steps, the imbalance problem in historical industrial data can be effectively handled, and the data can be standardized to provide balanced and consistent data input for the subsequent medium and thick plate convexity prediction model.
[0102] Step S104: Establishing a ridge regression prediction model for predicting the convexity of medium and thick plates based on ridge regression.
[0103] Preferably, Figure 3 As shown, this step may further include the following sub-steps:
[0104] Step S1041: Use the cross-validation method to find the optimal regularization parameter α. Cross-validation can effectively prevent overfitting and find the parameters that make the model most generalizable. Common cross-validation methods include K-fold cross-validation (such as 5-fold or 10-fold cross-validation). During the cross-validation process, try different regularization parameter values and select the value that makes the model perform best on the validation set as the optimal regularization parameter α.
[0105] Step S1042: Based on the optimal regularization parameter α, the convexity value of the medium and thick plate is used as the output to construct a ridge regression prediction model based on ridge regression:
[0106] y r =w r x+b r ;
[0107] In the above formula, y r is the plate convexity value predicted by the ridge regression prediction model, x is the matrix of independent variables, w r is the weight, b r For deviation.
[0108] Step S1043: Based on the training results of the ridge regression prediction model, obtain the weight w of the ridge regression model r and deviation b r .
[0109] Step S105: Calculate the error of the ridge regression prediction model, take the error as the dependent variable, and establish an error prediction model based on the BP neural network.
[0110] Preferably, Figure 4 As shown, this step may further include the following sub-steps:
[0111] Step S1051: Calculate the prediction error of the ridge regression model, take the prediction error as output, and construct a ridge regression prediction error model based on the BP neural network.
[0112] Prediction error err = y true -y ridge , where y true Represents the actual value of the convexity of the medium and thick plate, y ridge is the predicted value of ridge regression.
[0113] Step S1052: using mean square error (MSE) as the performance evaluation index of the BP neural network model, starting from a single hidden layer, modifying the network parameters by gradually increasing the layer by layer method, and finding the neuron combination that minimizes the MSE value.
[0114] In this embodiment, the mean square error MSE is calculated according to the following formula:
[0115]
[0116] In the above formula, e i , are the actual and predicted values of the plate convexity of sample i, respectively, and N is the number of samples.
[0117] The BP neural network model selected in this application includes: input layer, two hidden layers and output layer. The neural network is as follows: Figure 5 shown.
[0118] Step S106: constructing an integrated model with the predicted value of the ridge regression prediction model as the main value and the predicted value of the error prediction model as the deviation, and using the standardized historical industrial data to train the integrated model.
[0119] Preferably, Figure 6 As shown, this step may further include the following sub-steps:
[0120] Step S1061: Construct the predicted value y of the ridge regression model r is the main value, and the BP neural network model predicts the value e BP is the integrated model of deviation, and the predicted value of the integrated model is
[0121] Step S1062: Use the standardized data set T0' obtained in step 11 above * The integrated model is trained.
[0122] Step S1063: Obtain the weights and biases between the neural network layers of the BP neural network model according to the model training results.
[0123] Step S107: Based on the training results of the integrated model, the weights and deviations of each layer of the ridge regression prediction model and the error prediction model are obtained to obtain the final prediction formula for the convexity prediction of the medium and thick plate. The convexity of the medium and thick plate can be predicted using the final prediction formula.
[0124] The final prediction formula for the convexity prediction of medium and thick plates is as follows:
[0125]
[0126] In the above formula, w r is the ridge regression model weight, b r is the ridge regression model bias; The weights and biases transferred from the BP neural network input layer to the first hidden layer; is the weight and bias transferred from the first hidden layer to the second hidden layer of the BP neural network; w 3,j (j=1,2…,11), is the weight and bias transmitted from the second hidden layer of the BP neural network to the output layer; σ is the ReLu activation function. The weights w of each model * With deviation b * All can be obtained based on their respective training results.
[0127] Preferably, the medium and thick plate convexity prediction method of this embodiment may further include, after step S107: quantitatively inferring the influence of each characteristic variable on the convexity value of the medium and thick plate based on the weight coefficients of the ridge regression prediction model and the error prediction model, to guide the actual production process.
[0128] It can be seen from the above technical scheme that the medium and thick plate convexity prediction method based on data mining and multi-model fusion provided by the present invention first adopts one-dimensional projection for the rolling process data to solve the problem of multi-source heterogeneity presented therein; secondly, the data set is expanded by using the improved SMOTE method fused with Bootstrap, which weakens the data imbalance and deals with the problem of low model accuracy caused by data imbalance; then, a plate convexity prediction model based on ridge regression is constructed, and then the difference between the predicted value and the actual value of the ridge regression model is used as the predicted value output by the BP neural network, and an integrated model with ridge regression as the main value and BP neural network as the deviation is constructed. On the one hand, the present invention provides a new idea for processing the multi-source heterogeneity phenomenon of medium and thick plate production data and data mining, so that data mining technology can serve the actual process production; on the other hand, since the predicted value of ridge regression is close to the actual value in the same order of magnitude, the deviation between the predicted value of ridge regression and the actual value is used as the output of the BP neural network model, which can further reduce the range of the prediction error and be closer to the actual value, so that the prediction accuracy of the model is higher; finally, the prediction method proposed in the present invention can derive the influence of the change of each characteristic parameter on the final plate convexity through reverse deduction, so that the model has interpretability, so that the data can reflect and effectively guide the actual process production; therefore, the present invention has a strong promotion ability, which can not only improve the prediction accuracy of the export plate convexity of medium and thick plates, but also can quantitatively and effectively guide steel mills to control the convexity of medium and thick plate products.
[0129] The above method is further described below by a specific embodiment, which comprises the following steps:
[0130] Step a: Based on the specific production process of medium and thick plates, collect the following historical industrial data:
[0131] Slab performance parameters: slab thickness, slab width, slab length, slab weight, slab temperature before loading into the furnace, finishing roll number, steel type;
[0132] Rolling process parameters: number of intermediate cooling passes, whether controlled cooling technology is used, whether controlled rolling technology is used, final rolling temperature, furnace time, width ratio, number of small plates, upper working roll diameter, lower working roll diameter, final rolling thickness, final rolling width, final rolling length, number of rough rolling passes, number of finishing rolling passes, furnace exit temperature, maximum temperature after descaling, number of pre-straightening passes, and total number of straightening passes;
[0133] Rolling process parameters: rolling force in all rough rolling passes, rolling force in all finishing passes, and roll shifting in all finishing passes.
[0134] Step b: For the rolling process data of each slab, perform one-dimensional projection transformation on the rough rolling force, finishing rolling force and finishing roll shifting amount of each pass, and use the one-dimensional projection value as the principal component variable representing the slab rolling process.
[0135] Step c: According to the appropriate convexity critical value, the samples in the maximum and minimum ranges are divided into two sub-sample sets, and the normal plate convexity value is taken as a separate sample subset. Based on the logistic regression results, the optimal sample size ratio r of the three sample subsets is calculated. 板凸度极大值 and r 板凸度极小值 ,The improved SMOTE method based on Bootstrap is used to expand the original data according to the optimal imbalance ratio so that the three sample sizes reach the optimal balance state, and the expanded data set is standardized.
[0136] Specifically, the original data set is divided into a benchmark training set T0 and a benchmark test set T1 at an 8:2 ratio; based on the obtained distribution diagram of the convexity data of the medium and thick plates, an appropriate convexity critical value c1>c2 is selected. In this embodiment, c1 is selected as 0.5 mm, c2 is selected as -0.25 mm, and the samples are divided into a maximum plate convexity value sample S1, a normal plate convexity value sample S0, and a minimum plate convexity value sample S -1 ; Set the imbalance ratio of S1 to S0 to r1, S -1 The imbalance ratio for S0 is r2; based on the convexity data distribution diagram of medium and thick plates, samples with convexity values greater than 0.5mm are divided into maximum plate convexity value samples, samples with convexity values less than -0.25mm are divided into minimum plate convexity value samples, and samples with convexity values between -0.25mm and 0.5mm are divided into normal plate convexity value samples.
[0137] Calculate the sample center point x of S1 1,center , and calculate for each sample point x in S1 i Distance x 1,center The Euclidean distance d i , calculate {d i} i=1,2,... The average value d average , if d i <d average , x i is divided into safe points, otherwise, x i Divided into noise points, the set of all safe points is recorded as S 1,safe ; According to the imbalance ratios r1 and r2, based on S 1,safe and S 2,safe ,generate and new sample data; merge the new sample data with the benchmark training set T0 to form an expanded new data set T0′;
[0138] According to the selected convexity critical value, T0′ is divided into abnormal convexity value samples and normal convexity value samples. The abnormal samples are marked as 1 and the normal samples are marked as 0. A plate convexity abnormal value classification model based on logistic regression is constructed. The classification effect of the logistic regression model is verified using the benchmark test set T1. The F value and accuracy are used as evaluation indicators. Starting from r1=0 and r2=0, the stepwise growth method is used to find the imbalance ratio r1 that makes the model work best. * and r2 * .
[0139] The samples whose convexity values in T0′ fall outside the interval [-0.25mm, 0.5mm] are marked as 1, and the rest of the samples are marked as 0. The corresponding logistic regression classification model of plate convexity anomalies is trained and verified on the benchmark test set T1 containing only original sample points, starting from r1=0 and r2=0, and gradually increasing. After empirical research, the optimal imbalance ratio r1 * is 0.75, r2 * is 0.20.
[0140] Step d: Establishing the convexity prediction model y for medium and thick plates based on ridge regression ridge =W ridge X+b ridge , where y ridge represents the plate convexity value predicted by the ridge regression model, and X is the matrix of independent variables.
[0141] Step e: Calculate the ridge regression model error err = y true -y ridge , where y true Represents the actual value of the convexity of the medium and thick plate, y ridge The prediction value of ridge regression is taken, err is taken as the dependent variable, and an error prediction model based on BP neural network is established. The mean square error (MSE) is used to evaluate the prediction effect of the BP neural network model. Starting from a single hidden layer, the network parameters are corrected by the stepwise growth method to find the neural network combination with the lowest MSE value. Finally, the BP neural network model selected by the present invention includes: an input layer, two hidden layers and an output layer.
[0142] After experiments, the BP neural network model structure with the lowest MSE value contains four layers: input layer, the first hidden layer with 50 neurons, the second hidden layer with 11 neurons and the output layer. ReLu is selected as the activation function, and the ReLu function formula is σ(x)=max(0,x).
[0143] Step f: Construct an integrated model with the prediction value of the ridge regression model as the main value and the prediction value of the BP neural network model as the deviation. Use the data set of the standardized fused and expanded data in step 3 to train the integrated model. The trained integrated model can be used to predict the convexity of medium and thick plates. Use T0′ as the training set to train the integrated model, and use the benchmark test set T1 containing only the original sample points as the test set. The results Figure 7 As shown, the mean square error of the model is 0.004mm 2 .
[0144] Step g: Based on the model training results, the weights and deviations of each layer of the ridge regression model and the BP neural network model are obtained to obtain the final result of the convexity prediction of medium and thick plates. Based on the weight ratio of the network model, it is inferred how each characteristic variable affects the convexity value to guide the actual production process.
[0145] The prediction result obtained by the ridge regression model is recorded as y r =w r x+b r In the BP neural network, the output of the jth neuron in the first hidden layer is recorded as The output of the jth neuron in the second hidden layer is denoted as The final output of the BP neural network is recorded as Where σ is the activation function ReLu, and the weights w of each model * With deviation b * are obtained from the model training results, so the integrated model prediction value can be written as:
[0146]
[0147] Control other variables unchanged and let the characteristic variable Increasing by one unit, the predicted value becomes
[0148]
[0149] where k * Satisfy k * ∈{1,2,…,50} and j * Satisfy * ∈{1…,11} and
[0150] Therefore, when the variable When increasing by one unit, the plate convexity will change If ΔC is positive, increase the variable The convexity of the plate will increase. If ΔC is negative, increasing the variable The convexity of the plate will be reduced. Therefore, this model can be used to guide the process production in actual process production, effectively control the convexity of the final product, and improve the production capacity of the steel plant.
[0151] like Figure 8 The figure shows a schematic diagram of the structure of a medium and thick plate convexity prediction device based on data mining and multi-model fusion provided by an embodiment of the present application, the device comprises: a data acquisition unit 810, a one-dimensional projection unit 820, a data expansion unit 830, a ridge regression model establishment unit 840, an error prediction model establishment unit 850, an integrated model construction unit 860 and a final prediction unit 870, which are sequentially connected. Among them:
[0152] The data acquisition unit 810 is used to acquire the historical industrial data of the medium and thick plate production line, wherein the historical industrial data includes slab performance parameters, rolling process parameters and rolling process parameters.
[0153] The one-dimensional projection unit 820 is used to perform one-dimensional projection transformation on the rolling process parameters of each slab to obtain a one-dimensional projection value, and use the one-dimensional projection value as a principal component variable representing the slab rolling process.
[0154] The data expansion unit 830 is used to divide the historical industrial data into three sub-sample sets according to the set convexity critical value, obtain the optimal sample size ratio of the three sample subsets based on the logistic regression results, adopt the improved SMOTE method based on Bootstrap, expand the historical industrial data according to the optimal imbalance ratio, make the three sample sizes reach the optimal balance state, and standardize the expanded historical industrial data.
[0155] A ridge regression model building unit 840 is used to build a ridge regression prediction model for predicting the convexity of medium-thick plates based on ridge regression;
[0156] The error prediction model building unit 850 is used to calculate the error of the ridge regression prediction model, take the error as the dependent variable, and build an error prediction model based on the BP neural network.
[0157] The integrated model construction unit 860 is used to construct an integrated model with the predicted value of the ridge regression prediction model as the main value and the predicted value of the error prediction model as the deviation, and to train the integrated model using the standardized historical industrial data.
[0158] The final prediction unit 870 is used to obtain the weights and deviations of each layer of the ridge regression prediction model and the error prediction model based on the training results of the integrated model, and obtain the final prediction formula for the convexity prediction of medium and thick plates.
[0159] Preferably, Fig. 9As shown, the above-mentioned device also includes: an impact back-estimation unit 880, which is used to back-estimating the influence of each characteristic variable on the convexity value of medium and thick plates based on the weight ratio of the ridge regression prediction model and the error prediction model, so as to guide the actual production process.
[0160] Preferably, Fig.10 As shown, the one-dimensional projection unit 820 includes:
[0161] A standardization processing module 821 is used to perform standardization processing on the rolling process data corresponding to each slab;
[0162] An eigenvalue sorting module 822 is used to calculate the covariance matrix of the rolling process data set after the standardization process, calculate the eigenvalues and corresponding eigenvectors of the covariance matrix, and sort the eigenvalues in reverse order;
[0163] The projection module 823 is used to project the standardized rolling process data set into the space spanned by the eigenvector corresponding to the first eigenvalue after being sorted in reverse order, as the principal component variable representing the slab rolling process.
[0164] Preferably, the data expansion unit 830 is specifically used to perform the following steps:
[0165] Step 1: Divide the original data set consisting of the historical industrial data into a benchmark training set T0 and a benchmark test set T1 according to an 8:2 ratio;
[0166] Step 2: Based on the distribution diagram of the convexity data of medium and thick plates, select the convexity critical values C1 and C2, where C1>C2, and divide the samples in the original data set into the maximum plate convexity value sample S1, the normal plate convexity value sample S0 and the minimum plate convexity value sample S-1, where the plate convexity value of S1 is greater than C1, the plate convexity value of S0 is between C1 and C2, and the plate convexity value of S-1 is less than C2;
[0167] Step 3: Set the unbalance ratio of S1 to S0 to r1, and the unbalance ratio of S-1 to S0 to r2;
[0168] Step 4: Calculate the sample center point x of S1 1,center , and calculate for each sample point x in S1 i Distance x 1,center The Euclidean distance d i , calculate {d i} i=1,2,…n The average value d average , if d i <d average , x i is divided into safe points, otherwise, x i Divided into noise points, the set of all safe points is recorded as S1,safe ;
[0169] Step 5: Use Bootstrap to create S 1,safe Perform random weighted sampling and calculate the weighted average point c after resampling 1,weigh ted , according to the formula x new =c 1,weigh ted +a(c 1,weigh ted -x 1,center ) Generate new sample data x new , where a is a random number that follows a random distribution in the interval [0,1];
[0170] Step 6: According to the imbalance ratio r1, based on the safe point set S 1,safe ,generate New sample data;
[0171] Step 7: Repeat steps 4 to 6, based on the imbalance ratio r2 and the safe point set S 2,safe ,generate New sample data;
[0172] Step 8: Merge the new sample data with the benchmark training set T0 to form an expanded new data set T0';
[0173] Step 9: According to the convexity critical value selected in step 2, T0' is divided into abnormal convexity value samples and normal convexity value samples. The abnormal samples are marked as 1 and the normal samples are marked as 0. A plate convexity abnormal value classification model based on logistic regression is constructed. The classification effect of the logistic regression model is verified using the benchmark test set T1. The F value and accuracy are used as evaluation indicators. Starting from r1=0 and r2=0, the stepwise growth method is used to find the imbalance ratio r1 that makes the model work best. * and r2 * ;
[0174] Step 10: Optimal imbalance ratio r1 obtained from step 9 * and r2 * According to steps 4 to 8, we can obtain the data set T0' of the fused expanded data under the optimal imbalance ratio. * ;
[0175] Step 11: T0' according to the standardized formula * Perform standardization.
[0176] Preferably, Fig.11 As shown, the ridge regression model building unit 840 includes:
[0177] The optimal parameter acquisition module 841 is used to find the optimal regularization parameter by using a cross-validation method;
[0178] The ridge regression model building module 842 is used to build a ridge regression prediction model based on ridge regression based on the optimal regularization parameter and taking the convexity value of the medium and thick plate as the output:
[0179] y r =w r x+b r ;
[0180] In the above formula, y r is the plate convexity value predicted by the ridge regression prediction model, x is the matrix of independent variables, w r is the weight, b r for deviation;
[0181] The weight and bias acquisition module 843 is used to obtain the weight w of the ridge regression model based on the training results of the ridge regression prediction model. r and deviation b r .
[0182] Preferably, Fig.12 As shown, the above-mentioned error prediction model building unit 850 includes:
[0183] The error model construction module 851 is used to calculate the prediction error of the ridge regression model, take the prediction error as output, and construct a ridge regression prediction error model based on the BP neural network.
[0184] The neural combination optimization module 852 is used to use the mean square error MSE as the evaluation index of the BP neural network model performance, starting from a single hidden layer, and correcting the network parameters by a step-by-step layer-by-layer growth method to find the neural network combination with the lowest MSE value.
[0185] Preferably, Fig.13 As shown, the above-mentioned integrated model building unit 860 includes:
[0186] The integrated model building module 861 is used to build the predicted value y of the Yiling regression model r is the main value, and the BP neural network model predicts the value e BP is the integrated model of deviation, and the predicted value of the integrated model is
[0187] Model training module 862, used to use the standardized data set T0' obtained in step 11 * The integrated model is trained.
[0188] The weight and bias acquisition module 863 obtains the weights and biases between the neural network layers of the BP neural network model according to the model training results.
[0189] Preferably, the final prediction formula for the above-mentioned plate convexity prediction is as follows:
[0190]
[0191] In the above formula, w r is the ridge regression model weight, b r is the ridge regression model bias; The weights and biases transferred from the BP neural network input layer to the first hidden layer; is the weight and bias transferred from the first hidden layer to the second hidden layer of the BP neural network; w 3,j (j=1,2…,11) is the weight and bias transmitted from the second hidden layer of the BP neural network to the output layer; σ is the ReLu activation function.
[0192] For a detailed description of the above-mentioned units and models, please refer to the corresponding description in the aforementioned method embodiment, which will not be further elaborated here.
[0193] It can be seen from the above technical scheme that the medium and thick plate convexity prediction device based on data mining and multi-model fusion provided by the present invention first adopts one-dimensional projection for the rolling process data to solve the problem of multi-source heterogeneity phenomenon presented therein; secondly, the data set is expanded by using the improved SMOTE method fused with Bootstrap to deal with the problem of low model learning accuracy caused by data imbalance; then, a plate convexity prediction model based on ridge regression is constructed, and then the difference between the predicted value and the actual value of the ridge regression model is used as the predicted value output by the BP neural network, to construct an integrated model with ridge regression as the main value and BP neural network as the deviation. On the one hand, the present invention provides a new idea for processing the multi-source heterogeneity phenomenon of medium and thick plate production data and data mining, so that data mining technology can serve the actual process production; on the other hand, since the predicted value of ridge regression is close to the actual value in the same order of magnitude, the deviation between the predicted value of ridge regression and the actual value is used as the output of the BP neural network model, which can further reduce the range of the prediction error and be closer to the actual value, so that the prediction accuracy of the model is higher; finally, the prediction method proposed in the present invention can derive the influence of the change of each characteristic parameter on the final plate convexity through reverse deduction, so that the model has interpretability, so that the data can reflect and effectively guide the actual process production; therefore, the present invention has a strong promotion ability, which can not only improve the prediction accuracy of the export plate convexity of medium and thick plates, but also can quantitatively and effectively guide steel mills to control the convexity of medium and thick plate products.
[0194] Fig.14 is a schematic diagram of an electronic device provided by an embodiment of the present invention. Fig.14The electronic device shown is a general data processing device, which includes a general computer hardware structure, which at least includes a processor 801 and a memory 802. The processor 801 and the memory 802 are connected via a bus 803. The memory 802 is suitable for storing one or more instructions or programs executable by the processor 801. The one or more instructions or programs are executed by the processor 801 to implement the steps in the above-mentioned medium and thick plate convexity prediction method based on data mining and multi-model fusion.
[0195] The processor 801 may be an independent microprocessor or a collection of one or more microprocessors. Thus, the processor 801 executes the commands stored in the memory 802 to execute the method flow of the embodiment of the present invention as described above to realize the processing of data and the control of other devices. The bus 803 connects the above-mentioned multiple components together, and at the same time connects the above-mentioned components to the display controller 804 and the display device and the input / output (I / O) device 805. The input / output (I / O) device 805 may be a mouse, a keyboard, a modem, a network interface, a touch input device, a somatosensory input device, a printer, and other devices known in the art. Typically, the input / output (I / O) device 805 is connected to the system via an input / output (I / O) controller 806.
[0196] The memory 802 may store software components such as an operating system, a communication module, an interaction module, and an application program. Each of the modules and applications described above corresponds to a set of executable program instructions that implement one or more functions and methods described in the embodiments of the invention.
[0197] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion are implemented.
[0198] An embodiment of the present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion.
[0199] In summary, the method and device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion provided by the present invention firstly adopt one-dimensional projection for the rolling process data to solve the problem of multi-source heterogeneity phenomenon presented therein; secondly, the data set is expanded by adopting the improved SMOTE method integrating Bootstrap to deal with the problem of low model learning accuracy caused by data imbalance; then, a plate convexity prediction model based on ridge regression is constructed, and then the difference between the predicted value and the actual value of the ridge regression model is used as the predicted value output by the BP neural network, so as to construct an integrated model with ridge regression as the main value and BP neural network as the deviation. On the one hand, the present invention provides a new idea for processing the multi-source heterogeneity phenomenon of medium and thick plate production data and data mining, so that data mining technology can serve the actual process production; on the other hand, since the predicted value of ridge regression is close to the actual value in the same order of magnitude, the deviation between the predicted value of ridge regression and the actual value is used as the output of the BP neural network model, which can further reduce the range of the prediction error and be closer to the actual value, so that the prediction accuracy of the model is higher; finally, the prediction method proposed in the present invention can derive the influence of the change of each characteristic parameter on the final plate convexity through reverse deduction, so that the model has interpretability, so that the data can reflect and effectively guide the actual process production; therefore, the present invention has a strong promotion ability, which can not only improve the prediction accuracy of the export plate convexity of medium and thick plates, but also can quantitatively and effectively guide steel mills to control the convexity of medium and thick plate products.
[0200] The preferred embodiments of the present invention have been described above with reference to the accompanying drawings. Many features and advantages of these embodiments are clear from this detailed description, and therefore the claims are intended to cover all such features and advantages of these embodiments that fall within their true spirit and scope. In addition, since many modifications and changes are readily apparent to those skilled in the art, the embodiments of the present invention are not intended to be limited to the precise structure and operation illustrated and described, but all suitable modifications and equivalents that fall within their scope may be included.
[0201] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0202] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0203] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0204] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0205] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion, It is characterized in that The method comprises: Acquire historical industrial data of a medium and thick plate production line, wherein the historical industrial data includes slab performance parameters, rolling process parameters, and rolling process parameters; Performing a one-dimensional projection transformation on the rolling process parameters of each slab to obtain a one-dimensional projection value, and using the one-dimensional projection value as a principal component variable representing the slab rolling process; According to the set convexity critical value, the historical industrial data is divided into three sub-sample sets, and based on the logistic regression results, the optimal sample size ratio of the three sub-sample sets is obtained. The improved SMOTE method based on Bootstrap is used to expand the historical industrial data according to the optimal imbalance ratio, so that the three sample sizes reach the optimal balance state, and the expanded historical industrial data is standardized; A ridge regression prediction model based on ridge regression is established to predict the convexity of medium and thick plates; Calculating the error of the ridge regression prediction model, taking the error as the dependent variable, and establishing an error prediction model based on the BP neural network; Constructing an integrated model with the predicted value of the ridge regression prediction model as the main value and the predicted value of the error prediction model as the deviation, and using the standardized historical industrial data to train the integrated model; Based on the training results of the integrated model, the weights and deviations of each layer of the ridge regression prediction model and the error prediction model are obtained, and the final prediction formula for the convexity prediction of medium and thick plates is obtained.
2. The method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 1, It is characterized in that After obtaining the final prediction formula for predicting the convexity of medium and thick plates, the method further includes: quantitatively inferring the influence of each characteristic variable on the convexity value of medium and thick plates based on the weight coefficients of the ridge regression prediction model and the error prediction model to guide the actual production process.
3. The method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 1, It is characterized in that The one-dimensional projection transformation is performed on the rolling process parameters of each slab to obtain a one-dimensional projection value, and the one-dimensional projection value is used as a principal component variable representing the slab rolling process, including: Standardize the rolling process data corresponding to each slab; Calculating the covariance matrix of the rolling process data set after the standardization, calculating the eigenvalues and corresponding eigenvectors of the covariance matrix, and sorting the eigenvalues in reverse order; The standardized rolling process data set is projected into the space spanned by the eigenvector corresponding to the first eigenvalue after being sorted in reverse order as the principal component variable representing the slab rolling process.
4. The method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 1, It is characterized in that According to the set convexity critical value, the historical industrial data is divided into three sub-sample sets, based on the logistic regression results, the optimal sample size ratio of the three sample subsets is obtained, the improved SMOTE method based on Bootstrap is used to expand the historical industrial data according to the optimal imbalance ratio, so that the three sample sizes reach the optimal balance state, and the expanded historical industrial data is standardized, including: Step 1: Divide the original data set consisting of the historical industrial data into a benchmark training set T according to an 8:2 ratio 0 and the benchmark set T 1 ; Step 2: Based on the convexity data distribution diagram of medium and thick plates, select the convexity critical value C 1 and C 2 , where C 1 >C 2 , divide the samples in the original data set into the maximum plate convexity value samples S 1 , Normal plate convexity value sample S 0 And the minimum plate convexity value sample S- 1 , where S 1 The plate convexity value is greater than C 1 , S 0 The plate convexity value is between C 1 , C 2 Between, S- 1 The plate convexity value is less than C 2 ; Step 3: Set S 1 For S 0 The imbalance ratio is r 1 , S- 1 For S 0 The imbalance ratio is r 2 ; Step 4: Calculate S 1 The sample center point x 1,center , and calculate S 1 For each sample point x i Distance x 1,center The Euclidean distance d i , calculate {d i } i=1,2,…n The average value d average , if d i <d average , x i is divided into safe points, otherwise, x i Divided into noise points, the set of all safe points is recorded as S 1,safe ; Step 5: Use Bootstrap to create S 1,safe Perform random weighted sampling and calculate the weighted average point c after resampling 1,weighted , according to the formula x new =c 1,weighted +a(c 1,weighted -x 1,center ) Generate new sample data x new , where a is a random number that follows a random distribution in the interval [0,1]; Step 6: According to the imbalance ratio r 1 , based on the security point set S 1,safe ,generate New sample data; Step 7: Repeat steps 4 to 6, according to the imbalance ratio r 2 , based on the security point set S 2,safe ,generate New sample data; Step 8: Compare the new sample data with the benchmark training set T 0 Fusion, forming an expanded new data set T 0 '; Step 9: According to the convexity critical value selected in step 2, T 0 'Divided into abnormal convexity value samples and normal convexity value samples, abnormal samples are marked as 1, and normal samples are marked as 0. A plate convexity abnormal value classification model based on logistic regression is constructed, using the benchmark test set T 1 Verify the classification effect of the logistic regression model, using F value and accuracy as evaluation indicators, from r 1 = 0 and r 2 = 0, and use the stepwise growth method to find the imbalance ratio r that makes the model work best. 1 * and r 2 * ; Step 10: Optimal imbalance ratio r obtained from step 9 1 * and r 2 * According to steps 4 to 8, we can obtain the dataset T of fused and expanded data with the optimal imbalance ratio. 0 ' * ; Step 11: Use the standardized formula to calculate T 0 ' * Perform standardization.
5. The method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 4, It is characterized in that The method of establishing a ridge regression prediction model for predicting the convexity of medium and thick plates based on ridge regression includes: Use the cross-validation method to find the optimal regularization parameter; Based on the optimal regularization parameter, the convexity value of the medium and thick plate is taken as the output, and a ridge regression prediction model based on ridge regression is constructed: y r =w r x+b r ; In the above formula, y r is the plate convexity value predicted by the ridge regression prediction model, x is the matrix of independent variables, w r is the weight, b r for deviation; Based on the training results of the ridge regression prediction model, the weight w of the ridge regression model is obtained r and deviation b r .
6. The method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 5, It is characterized in that The method of calculating the error of the ridge regression prediction model, taking the error as a dependent variable, and establishing an error prediction model based on a BP neural network includes: Calculate the prediction error of the ridge regression model, take the prediction error as output, and construct a ridge regression prediction error model based on the BP neural network; The mean square error (MSE) is used as the performance evaluation index of the BP neural network model. Starting from a single hidden layer, the network parameters are corrected by gradually increasing the layer by layer method to find the neuron combination with the lowest MSE value.
7. The method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 6, It is characterized in that The construction of the integrated model using the predicted value of the ridge regression prediction model as the main value and the predicted value of the error prediction model as the deviation, and using the standardized historical industrial data to train the integrated model includes: Construct the predicted value y using the ridge regression model r is the main value, and the BP neural network model predicts the value e BP is the integrated model of deviation, and the predicted value of the integrated model is Use the standardized data set T obtained in step 11 0 ' * training the integrated model; The weights and deviations between the neural network layers of the BP neural network model are obtained according to the model training results.
8. The method for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 7, It is characterized in that The final prediction formula for the convexity prediction of medium and thick plates is as follows: In the above formula, w r is the ridge regression model weight, b r is the ridge regression model bias; (k=1,2…,50) is the weight and bias transmitted from the BP neural network input layer to the first hidden layer; (k=1,2…,50,j=1,2…,11) are the weights and biases transferred from the first hidden layer to the second hidden layer of the BP neural network; w 3,j (j=1,2…,11) is the weight and bias transmitted from the second hidden layer of the BP neural network to the output layer; σ is the ReLu activation function.
9. A device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion, It is characterized in that The device comprises: A data acquisition unit, used to acquire historical industrial data of a medium and thick plate production line, wherein the historical industrial data includes slab performance parameters, rolling process parameters and rolling process parameters; A one-dimensional projection unit, used for performing one-dimensional projection transformation on the rolling process parameters of each slab to obtain a one-dimensional projection value, and using the one-dimensional projection value as a principal component variable representing the slab rolling process; A data expansion unit, used for dividing the historical industrial data into three sub-sample sets according to a set convexity critical value, obtaining the optimal sample size ratio of the three sub-sample sets based on the logistic regression result, using an improved SMOTE method based on Bootstrap to expand the historical industrial data according to the optimal imbalance ratio so that the three sample sizes reach an optimal balance state, and performing standardization on the expanded historical industrial data; A ridge regression model building unit is used to build a ridge regression prediction model for predicting the convexity of medium and thick plates based on ridge regression; An error prediction model building unit is used to calculate the error of the ridge regression prediction model, take the error as a dependent variable, and build an error prediction model based on a BP neural network; An integrated model construction unit, used to construct an integrated model with the predicted value of the ridge regression prediction model as the main value and the predicted value of the error prediction model as the deviation, and to train the integrated model using the standardized historical industrial data; The final prediction unit is used to obtain the weights and deviations of each layer of the ridge regression prediction model and the error prediction model based on the training results of the integrated model, and obtain the final prediction formula for the convexity prediction of medium and thick plates.
10. The device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 9, It is characterized in that The device also includes: an influence back-estimation unit, which is used to quantitatively back-estimating the influence of each characteristic variable on the convexity value of the medium and thick plate based on the weight coefficients of the ridge regression prediction model and the error prediction model, so as to guide the actual production process.
11. The device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 9, It is characterized in that The one-dimensional projection unit comprises: A standardization processing module is used to perform standardization processing on the rolling process data corresponding to each slab; An eigenvalue sorting module is used to calculate the covariance matrix of the rolling process data set after standardization, calculate the eigenvalues and corresponding eigenvectors of the covariance matrix, and sort the eigenvalues in reverse order; The projection module is used to project the standardized rolling process data set into the space spanned by the eigenvector corresponding to the first eigenvalue after being sorted in reverse order, as the principal component variable representing the slab rolling process.
12. The device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 9, It is characterized in that The data expansion unit is specifically used to perform the following steps: Step 1: Divide the original data set consisting of the historical industrial data into a benchmark training set T according to an 8:2 ratio 0 and the benchmark set T 1 ; Step 2: Based on the convexity data distribution diagram of medium and thick plates, select the convexity critical value C 1 and C 2 , where C 1 >C 2 , divide the samples in the original data set into the maximum plate convexity value samples S 1 , Normal plate convexity value sample S 0 And the minimum plate convexity value sample S- 1 , where S 1 The plate convexity value is greater than C 1 , S 0 The plate convexity value is between C 1 , C 2 Between, S- 1 The plate convexity value is less than C 2 ; Step 3: Set S 1 For S 0 The imbalance ratio is r 1 , S- 1 For S 0 The imbalance ratio is r 2 ; Step 4: Calculate S 1 The sample center point x 1,center , and calculate S 1 For each sample point x i Distance x 1,center The Euclidean distance d i , calculate {d i } i=1,2,…n The average value d average , if d i <d average , x i is divided into safe points, otherwise, x i Divided into noise points, the set of all safe points is recorded as S 1,safe ; Step 5: Use Bootstrap to create S 1,safe Perform random weighted sampling and calculate the weighted average point c after resampling 1,weighted , according to the formula x new =c 1,weighted +a(c 1,weighted -x 1,center ) Generate new sample data x new , where a is a random number that follows a random distribution in the interval [0,1]; Step 6: According to the imbalance ratio r 1 , based on the security point set S 1,safe ,generate New sample data; Step 7: Repeat steps 4 to 6, according to the imbalance ratio r 2 , based on the security point set S 2,safe ,generate New sample data; Step 8: Compare the new sample data with the benchmark training set T 0 Fusion, forming an expanded new data set T 0 '; Step 9: According to the convexity critical value selected in step 2, T 0 'Divided into abnormal convexity value samples and normal convexity value samples, abnormal samples are marked as 1, and normal samples are marked as 0. A plate convexity abnormal value classification model based on logistic regression is constructed, using the benchmark test set T 1 Verify the classification effect of the logistic regression model, using F value and accuracy as evaluation indicators, from r 1 = 0 and r 2 = 0, and use the stepwise growth method to find the imbalance ratio r that makes the model work best. 1 * and r 2 * ; Step 10: Optimal imbalance ratio r obtained from step 9 1 * and r 2 * According to steps 4 to 8, we can obtain the dataset T of fused and expanded data with the optimal imbalance ratio. 0 ' * ; Step 11: Use the standardized formula to calculate T 0 ' * Perform standardization.
13. The device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 12, It is characterized in that The ridge regression model building unit comprises: The optimal parameter acquisition module is used to find the optimal regularization parameter by using the cross-validation method; The ridge regression model building module is used to build a ridge regression prediction model based on ridge regression based on the optimal regularization parameter and taking the convexity value of the medium and thick plate as the output: y r =w r x+b r ; In the above formula, y r is the plate convexity value predicted by the ridge regression prediction model, x is the matrix of independent variables, w r is the weight, b r for deviation; The weight and deviation acquisition module is used to obtain the weight w of the ridge regression model based on the training results of the ridge regression prediction model. r and deviation b r .
14. The device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 13, It is characterized in that The error prediction model building unit comprises: An error model building module is used to calculate the prediction error of the ridge regression model, take the prediction error as output, and build a ridge regression prediction error model based on the BP neural network; The neural combination optimization module is used to use the mean square error MSE as the evaluation index of the BP neural network model performance. Starting from a single hidden layer, the network parameters are corrected by the step-by-step layer-by-layer growth method to find the neuron combination that makes the MSE value the lowest.
15. The device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 14, It is characterized in that The integrated model building unit comprises: Integrated model building module for building the predicted value y of the Yiling regression model r is the main value, and the BP neural network model predicts the value e BP is the integrated model of deviation, and the predicted value of the integrated model is Model training module, used to use the standardized data set T obtained in step 11 0 ' * training the integrated model; The weight and bias acquisition module obtains the weights and biases between the neural network layers of the BP neural network model according to the model training results.
16. The device for predicting the convexity of medium and thick plates based on data mining and multi-model fusion according to claim 15, It is characterized in that The final prediction formula for the convexity prediction of medium and thick plates is as follows: In the above formula, w r is the ridge regression model weight, b r is the ridge regression model bias; (k=1,2…,50) is the weight and bias transmitted from the BP neural network input layer to the first hidden layer; (k=1,2…,50,j=1,2…,11) are the weights and biases transferred from the first hidden layer to the second hidden layer of the BP neural network; w 3,j (j=1,2…,11) is the weight and bias transmitted from the second hidden layer of the BP neural network to the output layer; σ is the ReLu activation function.
17. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.
18. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.
Citation Information
Cited By
Electric quantity prediction method based on small sample data
CN120031264A
Method for evaluating reliability of convexity setting calculation result
CN122346662A
A convexity setting calculation result reliability evaluation method
CN122346662B
