CN116108932A - Method for establishing fusion model of steel production process data and mechanism - Google Patents

Abstract

The invention discloses a method for establishing a fusion model of steel production process data and mechanism, which comprises the following steps: s1: obtaining and processing original data; s2: establishing a mechanism model; s3: establishing a data and mechanism fusion model; s4: bayesian optimization of the XGBoost model; s5: and solving a data and mechanism fusion model. According to the method for establishing the fusion model of the steel production process data and the mechanism, the depth perception of the production process is realized through the method of fusion modeling of the data and the mechanism, so that the accurate prediction of the quality target is realized, the on-site actual production process can be guided, and the stability of the product quality is effectively improved.

Description

A method for establishing a steel production process data and mechanism fusion model

Technical Field

The present invention relates to the technical field of steel rolling automatic control, and in particular to a method for establishing a steel production process data and mechanism fusion model.

Background Art

In the steel production process, there are complex physical and chemical changes, many external random interference factors, and different processes have different functions but are interrelated, mutually supportive, and mutually constrained. After being integrated in series, they form a complex production system. Under such a production organization model, there is a lot of uncertainty in the generation of product quality problems, which leads to product quality fluctuations. There are many key process parameters related to the quality of steel products, and the parameters within the process are nonlinearly coupled, and quality problems will also accumulate and inherit between processes. Improving the model's adaptability to complex dynamic conditions is the key to further improving product quality.

There are several forms of mathematical models used in steel production sites. The first is static or semi-dynamic process mechanism models, which are limited by factors such as simplified assumptions and changes in uncertain boundary conditions. They are difficult to support high-precision quality control under complex working conditions, and there are large deviations in model predictions when working conditions and operating states change. The second is data models, which can often achieve good results in complex model scenarios, but data-based modeling requires a lot of data support, takes a long time, and is not interpretable. In addition, there are some statistical models based on empirical knowledge, which generally have poor modeling effects and are often only used in special process situations.

How to give full play to the respective advantages of process mechanism, production data and experience knowledge, and accurately understand the complex relationship between key parameters such as process and quality, is a key issue that needs to be overcome in the construction of steel production process model. Therefore, a method for establishing a steel production process data and mechanism fusion model is urgently needed.

Summary of the invention

The purpose of the present invention is to provide a method for establishing a steel production process data and mechanism fusion model, which can achieve deep perception of the production process through the data and mechanism fusion modeling method, and then achieve accurate prediction of quality goals.

To achieve the above object, the present invention provides a method for establishing a steel production process data and mechanism fusion model, comprising the following steps:

S1: raw data acquisition and processing;

S2: Establishment of the mechanism model;

S3: Establishment of data and mechanism fusion model;

S4: Bayesian optimization of XGBoost model;

S5: Data and mechanism fusion model solution.

Preferably, in step S1, the method for acquiring the original data is: based on the process automation level of the plate and strip rolling computer control system, acquiring the field data, and exporting and storing the field data.

Preferably, in step S1, the original data processing method comprises the following steps:

S11. Data processing: Delete null values and missing values from the field data, and delete the text data to obtain production data;

S12. Establishment of original data set: Screen the production data, obtain the input features of the model, and establish the original data set.

Preferably, the field data are setting data of the production process and actual field detection data fed back by the basic automation level.

Preferably, in step S2, the establishment of the mechanism model takes the actual production process on site as input, and the mechanism model prediction data is calculated based on the rolling process mechanism formula.

Preferably, in step S3, the establishment of data and mechanism model specifically includes the following steps:

S31. Determination and processing of model data: Supplement the mechanism model prediction data as an input feature to the original data set and merge them into a new data set;

S32, dividing the new data set into a training set and a test set;

S33. Standardize the new data set.

Preferably, in step S4, the specific process of Bayesian optimization of XGBoost model is as follows:

S41: Establish a hyperparameter space, and randomly generate n initial hyperparameter combinations in the space, perform model training based on the initial hyperparameter combinations, calculate r ² of the predicted data, and obtain the initial Bayesian data set as:

D ₀ ={(X ₁ ,y ₁ ),(X ₂ ,y ₂ ),…,(X _n ,y _| )}

Where _Xn represents the hyperparameter combination, _yn is the corresponding ^r2 ;

S42: Substitute the initial Bayesian data set D ₀ into the Gaussian process, calculate its Gaussian distribution model, and use the acquisition function EI model to select the parameter combination with the largest r ² to obtain the next hyperparameter combination to be calculated;

S43: Calculate r ² of the next hyperparameter combination to be calculated based on the model, add it to the initial Bayesian data set D ₀ to form a new Bayesian data set D, update the Gaussian process regression, and repeat steps S42-S43 until the set maximum number of iterations is reached;

S44: After the model optimization stops, the hyperparameter combination with the largest r ² is selected as the optimal hyperparameter.

Preferably, in step S5, a model is built based on the model parameters obtained by Bayesian optimization, training set data is imported for model training, and model performance is evaluated using test set data.

Preferably, the indicators for evaluating model performance include mean square error (MSE), maximum percentage error (MAPE), and mean absolute error (MAE).

Therefore, the present invention adopts the above-mentioned method for establishing a steel production process data and mechanism fusion model, and its technical effects are as follows: through the data and mechanism fusion modeling method, deep perception of the production process is achieved, and then accurate prediction of quality goals is achieved, which can provide guidance for the actual production process on site and effectively improve the stability of product quality.

The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of a hot rolling production line in a method for establishing a steel production process data and mechanism fusion model according to the present invention;

FIG2 is a calculation flow of a mechanism model of a method for establishing a steel production process data and mechanism fusion model according to the present invention;

FIG3 is a modeling step of a data and mechanism fusion model of a method for establishing a data and mechanism fusion model of a steel production process according to the present invention;

FIG4 is a Bayesian optimization iteration process of a data and mechanism fusion model of a method for establishing a data and mechanism fusion model of a steel production process according to the present invention;

FIG5 is a prediction result of a data and mechanism fusion model of a method for establishing a data and mechanism fusion model of a steel production process according to the present invention;

FIG6 is a prediction result of a pure data-driven model of a method for establishing a steel production process data and mechanism fusion model according to the present invention.

Reference numerals

1. Heating furnace; 2. Slab; 3. Roughing mill; 4. Sensor; 5. Insulation cover; 6. Finishing mill; 7. Laminar cooling; 8. Coiler; 9. Width gauge; 10. Hot metal detector; 11. Pyrometer; 12. Thickness gauge.

DETAILED DESCRIPTION

The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the present invention can be implemented in other specific forms without departing from the gist or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the present invention, and any reference numerals in the claims should not be considered as limiting the claims to which they relate.

In addition, it should be understood that although this specification is described according to the implementation modes, not every implementation mode includes only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation modes that can be understood by those skilled in the art. These other implementation modes are also covered within the protection scope of the present invention.

It should also be understood that the specific embodiments described above are only used to explain the present invention, and the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, can make equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, which should be covered by the protection scope of the present invention/invention.

Technologies, methods, and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

The disclosed contents of the prior art documents cited in the present specification are incorporated into the present invention by reference in their entirety and are therefore part of the disclosure of the present invention.

Embodiment 1

A method for establishing a steel production process data and mechanism fusion model. Figure 1 shows the layout of a hot rolling production line. The rolled piece is rolled and deformed in 6 stands in the finishing area to obtain the final product. A width gauge is arranged at the exit of the last stand to measure the width of the final product. The finishing width is the difference between the width of the final product and the width at the finishing entrance.

The specific implementation steps are as follows:

1. Raw data acquisition and processing

(1) Acquisition of original data

Based on the hot rolling production line, the actual measured value of the width of each strip steel is obtained, so as to obtain the rolling width of the slab. Each strip steel corresponds to a set of production processes, which serves as the source data for data-driven modeling. Based on the process automation level of the plate and strip rolling computer control system, the field data is obtained. The field data is the setting data of the production process and the actual field detection data fed back by the basic automation level, and the field data is exported and stored.

(2) Processing of raw data

In addition to the required process parameters, the data obtained on site also includes some text-related data, such as material number, production time, etc., which are features that are obviously irrelevant to the width expansion of the finishing rolling, and are directly deleted manually. In addition, for data null values caused by missed detection or data transmission packet loss, they are deleted based on their small proportion of the total data. A total of 8 missing data were removed, and the remaining data were production data, totaling 1,860.

(3) Establishment of original data set

The final data set that can be used for calculation contains 1860 production data. After removing unnecessary features, there are 30 input features left, including those obtained from on-site detection instruments and calculated setting values in the rolling model. The measured data include: intermediate billet length, F1 inlet temperature, F6 outlet temperature, F1 inlet slab width, F6 outlet slab width, F1 inlet slab thickness, F1~F6 rolling force, F1~F6 rolling speed; the set data calculated by the rolling model are: F1~F6 roll gap, F1~F6 strip tension. The above data are used as the original data set.

2. Establishment of the mechanism model

In the hot rolling finishing process, the factors affecting the change of slab width mainly include mill entrance width, thickness, slab temperature, roll radius, roll speed, rolling speed, and friction coefficient. Therefore, this embodiment uses the Bakhchinov model as the mechanism model. The calculation formula is as follows:

Where: Δb is the slab width, Δh is the reduction, _h0 is the initial slab width, R is the roller radius, and μ is the friction coefficient.

The friction coefficient can be calculated by the following formula:

μ＝0.8(1.05-0.0005θ)(2)

Wherein: θ is the rolled piece temperature, and 700℃<θ<1200℃.

Figure 2 is a flow chart for calculating the width expansion of multiple passes of finishing rolling. In the hot rolling finishing process, except for the first stand, the slab inlet thickness of each stand cannot be measured and can only be calculated cumulatively through the rolling process. The inlet thickness of each stand is calculated, and the remaining data are calculated using actual process data and imported into the mechanism formula to obtain the slab width expansion after rolling each stand.

After 6-stand rolling, the final slab width is:

3. Establishment of data and mechanism fusion model

(1) Determination and processing of model data

The prediction data of the mechanism model is added to the original data set as an input feature of the finishing width expansion to generate a new data set for predicting the finishing width expansion of hot rolling. The original 30 features are supplemented and the prediction value of the mechanism model width expansion is added as the 31st feature to form a new data set.

The new data set is divided into a training set and a test set in a ratio of 8:2. The training set contains 1488 data items and the test set contains 372 data items.

In view of the different types of features in the new data set and the large differences in unit dimensions, Z-Score standardization is used, that is, data of different dimensions are converted into Z-Score scores of unified measurement for representation. The formula is:

为原始数据的标准差，N为数据数目。Among them, μ is the mean of the original data,

is the standard deviation of the original data, and N is the number of data.

(2) Establishing a data and mechanism fusion model

As an integrated learning method, XGBoost uses the boosted tree CART algorithm as the base model decision tree, and reduces model complexity and overfitting through regularization terms, and has strong generalization ability. A preliminary XGBoost model is established based on python, with the model parameters using default values, and the model is trained using the training set data to preliminarily establish a data and mechanism fusion model.

4. Bayesian Optimization of XGBoost Model

The model parameters are optimized based on the Bayesian optimization method to improve the prediction performance of the model. Figure 3 shows the process of modeling and model tuning of the data and mechanism fusion model.

The Bayesian optimization process is as follows:

(1) Establish a hyperparameter space and randomly generate five initial hyperparameter combinations in the space. Perform model training based on the initial hyperparameter combinations and calculate r ² of the predicted data. The initial Bayesian data set is: D ₀ =({X ₁ ,y ₁ ),(X ₂ ,y ₂ ),…,(X _n ,y _n )}.

Where _Xn represents the nth hyperparameter combination, and _yn is the corresponding ^r2 . The calculation formula of ^r2 is as follows:

为第i组工艺下的宽展预测值，n为原数据数目，

为原始数据的平均值。Where _yi is the measured value of the finishing width under the i-th group of processes,

is the predicted value of the width expansion under the i-th group of processes, n is the number of original data,

is the average value of the original data.

(2) Substitute D ₀ into the Gaussian process, calculate its Gaussian distribution model, and use the acquisition function EI model to select the hyperparameter combination with the largest r ² to obtain the next hyperparameter combination to be calculated;

(3) Substitute the next hyperparameter combination to be calculated into the model for calculation, add it to the initial Bayesian data set _D0 to form a new Bayesian data set D, and update the Gaussian process regression.

Repeat steps (2) and (3) until the maximum number of iterations is reached.

(4) After the model optimization stops, the hyperparameter combination with the largest r ² is taken as the optimal hyperparameter.

Table 1 shows the model hyperparameters, parameter space and optimal values of hyperparameters that need to be optimized. The maximum number of model iterations is 200, and the parameter optimization iteration process is shown in Figure 4. All model parameters except those in the table use default parameters, and finally the optimal hyperparameter combination is obtained through Bayesian optimization.

Table 1 XGBoost parameters to be optimized

5. Data and mechanism fusion model solution

Based on the optimal hyperparameter combination obtained by Bayesian optimization, a data and mechanism fusion model is established for training. In order to verify the accuracy of the established model, it is necessary to evaluate its performance. The mean square error MSE, maximum percentage error MAPE, and mean absolute error MAE are used to evaluate the prediction ability of the model. The formulas for each evaluation indicator are as follows:

为第i组工艺下的宽展预测值，n为原数据数目，

为原始数据的平均值。Where _yi is the measured value of the finishing width under the i-th group of processes,

is the predicted value of the width expansion under the i-th group of processes, n is the number of original data,

is the average value of the original data.

Table 2 shows the performance measurement results of the data and mechanism fusion model. The average absolute error of the test model is 1.0809 mm, and the model determination coefficient is 0.9598. The model has strong generalization ability and is competent for the prediction task of the actual production process.

Table 2 Performance metrics of data and mechanism fusion models

Metrics MSE MAPE MAE Training set 0.2152 1.0964% 0.3386 Test Set 2.1909 3.3901% 1.0809

Comparative Example 1

In order to verify the prediction effect of the established data and mechanism fusion model, a pure data-driven model based on Bayesian optimization was established and compared with the data and mechanism fusion model.

Table 3 shows the performance measurement results of the pure data model. Compared with the data and mechanism models, MSE, MAPE, and MAE are relatively large, that is, the performance of the pure data model is poor.

Table 3. Performance metrics of pure data models

Metrics MSE MAPE MAE Training set 2.1533 0.3420% 1.0807 Test Set 3.2826 3.8288% 1.2995

Figures 5 and 6 show the prediction results of the data and mechanism fusion model and the pure data model, respectively. The prediction results of the data and mechanism fusion model are closer to the actual values and have higher prediction accuracy. After adding the prediction data of the mechanism model, it has a certain guiding effect on the training of the model, the prediction results have a higher credibility range and smaller errors, and the prediction performance of the model is greatly improved.

Therefore, the present invention adopts the above-mentioned steel production process data and mechanism fusion model establishment method, by adding the mechanism model prediction data to the steel production data, performing dimension increase processing, merging into a new data set, guiding the model training through the mechanism data, and using the Bayesian optimization algorithm to select the optimal model parameter combination, to achieve accurate prediction of quality targets. It can provide guidance for the actual production process on site and effectively improve the stability of product quality.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims (9)

1. A method for establishing a steel production process data and mechanism fusion model, characterized in that it includes the following steps: S1: raw data acquisition and processing; S2: Establishment of the mechanism model; S3: Establishment of data and mechanism fusion model; S4: Bayesian optimization of XGBoost model; S5: Data and mechanism fusion model solution. 2. A method for establishing a steel production process data and mechanism fusion model according to claim 1, characterized in that: in step S1, the method for obtaining original data is: based on the process automation level of the plate and strip rolling computer control system, field data is obtained, and the field data is exported and stored. 3. The method for establishing a steel production process data and mechanism fusion model according to claim 1, characterized in that: in step S1, the original data processing method includes the following steps: S11. Data processing: Delete null values and missing values from the field data, and delete the text data to obtain production data; S12. Establishment of original data set: Screen the production data, obtain the input features of the model, and establish the original data set. 4. A method for establishing a steel production process data and mechanism fusion model according to claim 2, characterized in that the field data is the setting data of the production process and the actual field detection data fed back by the basic automation level. 5. A method for establishing a steel production process data and mechanism fusion model according to claim 1, characterized in that: in step S2, the establishment of the mechanism model takes the actual production process on site as input, and obtains the mechanism model prediction data based on the rolling process mechanism formula. 6. The method for establishing a steel production process data and mechanism fusion model according to claim 1, characterized in that: in step S3, the establishment of the data and mechanism model specifically includes the following steps: S31, determination and processing of model data, adding the mechanism model prediction data as an input feature to the original data set and merging them into a new data set; S32, dividing the new data set into a training set and a test set; S33. Standardize the new data set. 7. The method for establishing a steel production process data and mechanism fusion model according to claim 1, characterized in that: in step S4, the specific process of the Bayesian optimization XGBoost model is as follows: S41: Establish a hyperparameter space, and randomly generate n initial hyperparameter combinations in the space, perform model training based on the initial hyperparameter combinations, calculate r ² of the predicted data, and obtain the initial Bayesian data set as: D ₀ ={(X ₁ ,y ₁ ),(X ₂ ,y ₂ ),…,(X _n ,y _n )} Where _Xn represents the nth hyperparameter combination, and _yn is the corresponding ^r2 ; S42: Substitute the initial Bayesian data set D ₀ into the Gaussian process, calculate its Gaussian distribution model, and use the acquisition function EI model to select the parameter combination with the largest r ² to obtain the next hyperparameter combination to be calculated; S43: Calculate r ² of the next hyperparameter combination to be calculated based on the model, add it to the initial Bayesian data set D ₀ to form a new Bayesian data set D, update the Gaussian process regression, and repeat steps S42-S43 until the set maximum number of iterations is reached; S44: After the model optimization stops, the hyperparameter combination with the largest r ² is selected as the optimal hyperparameter. 8. A method for establishing a steel production process data and mechanism fusion model according to claim 1, characterized in that: in step S5, a model is built based on the optimal hyperparameter combination obtained by Bayesian optimization, training set data is imported for model training, and model performance is evaluated using test set data. 9. A method for establishing a steel production process data and mechanism fusion model according to claim 8, characterized in that the indicators for model performance evaluation include mean square error (MSE), maximum percentage error (MAPE), and mean absolute error (MAE).

Images