H-shaped steel rolling force prediction system based on machine learning technology

Abstract

The invention relates to an H-shaped steel rolling force prediction system based on a machine learning technology. The system comprises a parameter calculation module meeting actual production conditions, a data preprocessing module and a rolling force prediction module based on an optimized random forest. The parameter calculation module meeting the actual production conditions is used for obtaining multiple technological parameters of H-shaped steel of different specifications through programming according to actual rolling technological parameters of factory production, and input data are provided for follow-up rolling force prediction. The data preprocessing module is used for carrying out cleaning, feature selection and abnormal value detection on the obtained H-shaped steel rolling data; and the rolling force prediction module based on the optimized random forest is used for carrying out rolling force prediction on the preprocessed data and outputting a prediction result. According to the method, the rolling force of the H-shaped steel can be accurately predicted, the rolling process is optimized, the defect that a traditional method depends on artificial experience is overcome, meanwhile, the production efficiency is improved, and the rejection rate is reduced.

Description

Technical Field

[0001] The present invention relates to the field of H-beam rolling force prediction, and in particular to an H-beam rolling force prediction system based on machine learning technology. Background Art

[0002] H-beam is a commonly used profile in the field of construction and engineering. It is named because of its "H"-shaped cross section. It has the advantages of excellent strength-to-weight ratio, bending resistance, simple construction, cost saving, etc. The production of H-beam is mainly completed through hot rolling process. The accurate prediction of rolling force is crucial to control the thickness and shape of the web and flange of H-beam, which is directly related to the dimensional accuracy and surface quality of the finished product. However, the cross-sectional shape of H-beam is relatively complex, and continuous casting special-shaped billets are usually used as billets. During the rolling process, the thickness and width of the web and flange need to be accurately controlled. Since the rolling system is a multivariable, nonlinear, and strongly coupled system, and is affected by factors such as harsh working conditions, rolling mill vibration, and control system complexity, the traditional rolling force prediction model faces great challenges. Especially in the production process of H-beam, accurate prediction of rolling force to achieve efficient production is still an urgent problem to be solved.

[0003] Most of the existing rolling force models are based on traditional empirical formulas or theoretical models, such as the Sims formula and the Eklund formula. Although these models can provide rough guidance under certain conditions, they usually ignore many complex factors in actual production, such as changes in rolling process parameters and mill vibration, resulting in low prediction accuracy. Once new specifications or new process conditions are adopted, these traditional models often need to be recalculated in large quantities, and the accuracy of their prediction results cannot be fully guaranteed. In addition, the traditional method has a slow calculation speed, and when faced with a non-steady-state hot rolling environment, the prediction accuracy of the model often drops significantly, which cannot meet the actual needs of enterprises for high-precision rolling force prediction.

[0004] In recent years, with the rapid development of big data and artificial intelligence technology, machine learning algorithms have gradually been applied to many tasks in steel production, including rolling force prediction. In this field, deep learning, as an efficient feature learning method, has achieved remarkable results in image recognition, speech processing and other fields, and has gradually become an important prediction tool. Compared with traditional methods, rolling force prediction based on deep learning can automatically extract features by learning a large amount of historical data, avoiding the limitations of manual experience, and has strong adaptability and generalization capabilities.

[0005] Although researchers at home and abroad have made some progress in the application of machine learning to predict rolling force in the steel industry, most of the research focuses on rolling processes such as plates, strips, and bars, and there are relatively few prediction studies on H-beams. H-beams have fewer specifications, but their rolling processes are complex and have strong production adaptability. Existing models often cannot meet their needs for high precision and high real-time performance. Especially in the actual production process, due to the influence of temperature fluctuations, rolling mill vibrations and other environmental factors, traditional machine learning models show large prediction errors in the hot rolling process of H-beams, and fail to effectively solve the problems of high-precision prediction and real-time feedback.

[0006] Therefore, based on machine learning and big data analysis, a high-precision and adaptable H-beam rolling force prediction model is proposed. Summary of the invention

[0007] The purpose of the present invention is to provide an H-beam rolling force prediction system based on machine learning technology, which can accurately predict the rolling force of H-beam and optimize the rolling process, avoid the shortcomings of traditional methods that rely on manual experience, and at the same time improve production efficiency and reduce scrap rate.

[0008] To achieve the above object, the present invention provides the following technical solutions: a rolling force prediction system for H-beam based on machine learning technology, the system comprising a parameter calculation module that meets actual production conditions, a data preprocessing module, and a rolling force prediction module based on optimized random forest;

[0009] The parameter calculation module that meets the actual production conditions is used to obtain multiple process parameters of H-beams of different specifications by programming according to the actual rolling process parameters of the factory production, so as to provide input data for the subsequent rolling force prediction;

[0010] The data preprocessing module is used to clean, select features and detect abnormal values ​​of the obtained H-beam rolling data;

[0011] The rolling force prediction module based on optimized random forest is used to predict the rolling force of the preprocessed data and output the prediction result.

[0012] Furthermore, the parameter calculation module that meets the actual production conditions includes a basic parameter setting unit, a rolling procedure design unit, and a parameter acquisition and storage unit;

[0013] The basic parameter setting unit: according to the specifications of H-beam production, input the leader size, finished product size and roll size, and design the rolling pass according to this information;

[0014] The rolling procedure design unit: according to the input basic parameters, the reduction rate is set, the reduction amount is inverted, and the rolling speed of each pass is set;

[0015] The parameter acquisition and storage unit stores the acquired rolling process parameters to provide an input basis for the subsequent rolling force prediction module.

[0016] Further, the rolling schedule design unit comprises:

[0017] Rolling pass setting: Users can set the number of passes through different rolling mills according to the actual layout of UR and UF rolling mills; each pass through the universal rolling mill is a PASS, and each pass through three tandem reversible rolling mills is a pass; the setting of the rolling pass N here is actually the number of passes, because passing through a universal rolling mill corresponds to a PASS, which corresponds to a rolling force;

[0018] Reduction rate setting and inversion: Users can set the web and reduction rate from the 2nd PASS to the Nth PASS based on known LEADER size, finished product size and rolling pass parameters. The system can inversely obtain the reduction rate and reduction amount from the first PASS to the last PASS.

[0019] Furthermore, the data preprocessing module is further as follows:

[0020] The obtained rolling data are processed by isolation forest algorithm to filter out outliers in the data;

[0021] Perform feature selection on the remaining data after screening to select the features that have a greater impact on rolling force prediction and reduce redundant input variables;

[0022] Standardize the parameters of different dimensions and normalize them to ensure the balance of parameter weights during model training and improve prediction accuracy;

[0023] The preprocessed data are divided into training set, validation set and test set for subsequent model training and verification.

[0024] Furthermore, the rolling force prediction module based on the optimized random forest is further:

[0025] The hyperparameters of the random forest model were optimized by using the Northern Goshawk Optimization Algorithm, and the model was trained with the training set data.

[0026] During the training process, the validation set is used to evaluate the performance of the model, and the predictive ability of the model is optimized through cross-validation.

[0027] Furthermore, the rolling force prediction module based on optimized random forest includes an isolated forest data outlier screening unit, a feature selection unit, a northern goshawk optimized random forest model unit and a result output unit;

[0028] Furthermore, the isolation forest data outlier screening unit is used to perform outlier detection on the rolling data using the isolation forest algorithm, and to remove abnormal data that does not meet the expected range, thereby ensuring the quality of the data set and the effectiveness of the model training.

[0029] Furthermore, the feature selection unit is used to screen the features in the data set using a tree model method, identify features that are highly relevant to rolling force prediction, remove redundant information, and improve the training efficiency of the model.

[0030] Furthermore, the Northern Goshawk Optimization Random Forest Model Unit is used to optimize the hyperparameters of the random forest model using the Northern Goshawk Optimization Algorithm. The Northern Goshawk Optimization Algorithm first initializes the population and determines the initial solution space; then obtains the candidate solution through the first stage optimization, and further refines the search for the optimal solution in the second stage; finally, it is determined whether the optimal parameter combination is achieved. If not, it is iterated repeatedly until the optimization conditions are met. The accuracy and robustness of the model in predicting the rolling force of horizontal rolls and vertical rolls under different rolling conditions are improved through parameter optimization.

[0031] Furthermore, the result output unit is used to output the rolling force results predicted based on the optimized random forest model in an Excel table format for factory staff to perform process optimization and production decision-making.

[0032] The beneficial effects of the present invention are as follows: on the basis of the existing traditional rolling force prediction model, machine learning technology is used for the first time to improve the accuracy and efficiency of rolling force prediction. Specifically, the present invention introduces a parameter calculation module that meets the actual production conditions, and ensures the accuracy and reliability of the input data by accurately calculating various process parameters in the steel rolling production process; the isolation forest algorithm is used to screen outliers in the data, which significantly improves the quality of the data, and the tree model is used for feature selection to screen out the features most relevant to the rolling force prediction, thereby reducing redundant features and improving the training efficiency of the model; the Northern Goshawk Optimization Algorithm (NGO) is used to optimize the number of decision trees and the maximum depth hyperparameters of the tree in the random forest model, so that the model can better adapt to the complex rolling process of H-beams, thereby improving the prediction accuracy and robustness. At the same time, the present invention can simultaneously realize the accurate prediction of the rolling force of the horizontal rolls and vertical rolls of H-beams without relying on traditional trial rolling through machine learning technology, which significantly improves the prediction efficiency. Compared with traditional manual experience or simplified theoretical models, the present invention can quickly respond to changes in parameters during the production process, and export the prediction results in Excel format through the result output unit, which is convenient for factory staff to perform subsequent operations and analysis on the rolling force prediction results, helping the factory to quickly adjust the process and optimize the production process.

[0033] The prediction model has the ability to respond quickly, giving the predicted value of rolling force in a short time, and can adjust the process parameters in real time to optimize the production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the overall framework of the H-beam rolling force prediction system in the present invention;

[0035] Figure 2 It is a flow chart of a parameter calculation module in the present invention that meets actual production conditions;

[0036] Figure 3 This is a flow chart of the rolling force prediction module based on optimized random forest in the present invention. DETAILED DESCRIPTION

[0037] The present invention will be further described below in conjunction with the accompanying drawings.

[0038] See also Figures 1 to 3 , the present invention provides an embodiment: a rolling force prediction system for H-beam based on machine learning technology, the system comprising a parameter calculation module that meets actual production conditions, a data preprocessing module, and a rolling force prediction module based on optimized random forest;

[0039] The parameter calculation module that meets the actual production conditions is used to obtain multiple process parameters of H-beams of different specifications by programming according to the actual rolling process parameters of factory production, and provide input data for subsequent rolling force prediction; the parameter calculation module that meets the actual production conditions is used to generate key parameters required in the rolling process according to the parameters such as LEADER size, rolling procedure design, finished product size and roller size in H-beam production, including LEADER size, rolling procedure design, finished product size, roller size, rolled product size, equipment parameters, reduction, temperature, contact arc length, and width. The collected H-beam multiple production data (such as rolled product size, rolling mill parameters, reduction, etc.) are input into the module. According to the production conditions and process requirements, by using the traditional rolling force calculation formula and related parameters, combined with the equipment parameters in actual production (such as roller diameter, roller width) and the actual size parameters of the rolled product (such as flange thickness, flange width, web thickness, etc.), the calculation formula is algorithmized by Python programming to obtain the parameters required for the rolling force prediction of H-beams of different specifications.

[0040] The data preprocessing module is used to clean, select features and detect outliers for the acquired H-beam rolling data; specifically, the isolation forest algorithm is used to screen outliers in the data and remove data points that do not meet expectations; at the same time, the tree model is used for feature selection to extract key features that have a significant impact on rolling force prediction.

[0041] The rolling force prediction module based on optimized random forest is used to predict the rolling force of the preprocessed data and output the prediction results. The module uses the Northern Goshawk Optimization Algorithm (NGO) to optimize the hyperparameters of the random forest (RF) model to improve the prediction accuracy. The output of the module is the rolling force prediction result, which can be used for production decision-making and process optimization.

[0042] Please continue reading Figures 1 to 3 As shown, in one embodiment of the present invention, the parameter calculation module that meets the actual production conditions includes a basic parameter setting unit, a rolling procedure design unit, and a parameter acquisition and storage unit;

[0043] The basic parameter setting unit: according to the specifications of H-beam production, input the leader size, finished product size and roll size, and design the rolling pass according to this information;

[0044] The rolling procedure design unit: according to the input basic parameters, the reduction rate is set, the reduction amount is inverted, and the rolling speed of each pass is set; the rolling speed is set manually.

[0045] The parameter acquisition and storage unit: saves the obtained rolling process parameters to provide input basis for the subsequent rolling force prediction module. This module can accurately generate rolling parameters based on actual production conditions, avoiding rolling force prediction errors caused by parameter setting deviations. The rolling procedure design unit is the core part of this module; its workflow can be found in Figure 2 ,

[0046] Please continue reading Figures 1 to 3 As shown, in one embodiment of the present invention, the rolling schedule design unit includes:

[0047] Rolling pass setting: Users can set the number of passes through different rolling mills according to the actual layout of UR and UF rolling mills; each pass through the universal rolling mill is a PASS, and each pass through three tandem reversible rolling mills is a pass; the setting of the rolling pass N here is actually the number of passes, because passing through a universal rolling mill corresponds to a PASS, which corresponds to a rolling force;

[0048] Reduction rate setting and inversion: Users can set the web and reduction rate from the 2nd PASS to the Nth PASS based on the known LEADER size, finished product size and rolling pass parameters. The system can inversely obtain the reduction rate and reduction amount from the first PASS to the last PASS. The reduction rate can be set in the following two ways: Experience setting: Based on production experience and combined with the historical data of conventional H-beams, the operator can directly set the reduction rate of the web and flange. Auxiliary design table setting: Using the factory TDM reduction procedure auxiliary design table (Excel), according to a certain specification of LEADER size and finished product size, pre-set the reduction procedure, and then set the reduction rate of each PASS according to the table content. Since the thickness of the LEADER web and flange, and the finished web and flange thickness are known, they can be calculated one by one from the finished product size forward. For example, the thickness of the web after rolling is (also the finished web thickness), the thickness of the web before rolling is (also the thickness of the web before rolling in the previous PASS), and the reduction rate of this PASS is set to, then it can be calculated, and so on, it can be calculated from the last Nth PASS to the second PASS. When calculating to the second PASS, no further calculation is required, because the LEADER size is known at this time, and there is no need to set the reduction rate and then calculate it in reverse. At the same time, the system will automatically generate the TDM reduction auxiliary design curve based on the inverted reduction rate data. Specifically, the system selects two columns of reduction rate data (web and flange reduction rate) through Python programming and automatically draws them, and draws two curves in the same area. It is used to verify the rationality of the reduction setting. Reasonable curves usually present a parabolic shape from small to large and then to small.

[0049] Please continue reading Figure 1 to Figure 2As shown, in one embodiment of the present invention, the data preprocessing module is further:

[0050] The obtained rolling data are processed by isolation forest algorithm to filter out outliers in the data;

[0051] Perform feature selection on the remaining data after screening (combining the tree model to score the importance of each parameter), screen out the features that have a greater impact on rolling force prediction, and reduce redundant input variables;

[0052] Standardize the parameters of different dimensions and normalize them to ensure the balance of parameter weights during model training and improve prediction accuracy;

[0053] The preprocessed data is divided into training set, validation set and test set for subsequent model training and verification. For example, the training set accounts for 80%, the validation set accounts for 10%, and the test set accounts for 10%. Through the data preprocessing module, the training efficiency and prediction accuracy of the model can be significantly improved, while reducing the impact of redundant data on the system computing performance.

[0054] Please continue reading Figure 1 to Figure 2 As shown, in one embodiment of the present invention, the rolling force prediction module based on the optimized random forest is further:

[0055] The hyperparameters of the random forest (RF) model were optimized by using the Northern Goshawk Optimization Algorithm (NGO), and the model was trained with the training set data; the purpose was to improve the accuracy of the model in predicting rolling force;

[0056] During the training process, the validation set is used to evaluate the performance of the model, and the prediction ability of the model is optimized through cross-validation. The model is trained using the training set data, and the accuracy and robustness of the model are verified through the validation set. After the training is completed, the model is finally evaluated using the test set to test the prediction performance and generalization ability of the model on unseen data to ensure that it meets actual production needs; the output of the model is an accurate rolling force prediction value, which is used by the factory to guide production and process adjustments. The rolling force prediction module based on optimized random forest is the core part of the system, which is used to predict rolling force based on preprocessed data. For its specific workflow, see Figure 3 .

[0057] Please continue reading Figure 1 to Figure 2 As shown, in one embodiment of the present invention, the rolling force prediction module based on optimized random forest includes an isolated forest data outlier screening unit, a feature selection unit, a northern goshawk optimized random forest model unit and a result output unit;

[0058] Please continue reading Figure 1 to Figure 2As shown, in one embodiment of the present invention, the isolation forest data outlier screening unit is used to perform outlier detection on rolling data using an isolation forest algorithm, remove abnormal data that does not meet the expected range, and ensure the quality of the data set and the effectiveness of model training.

[0059] Please continue reading Figure 1 to Figure 2 As shown, in one embodiment of the present invention, the feature selection unit is used to screen the features in the data set using a tree model method, identify features that are highly relevant to rolling force prediction, remove redundant information, and improve the training efficiency of the model.

[0060] Please continue reading Figure 1 to Figure 2 As shown, in one embodiment of the present invention, the Northern Goshawk Optimized Random Forest Model Unit is used to optimize the hyperparameters of the Random Forest (RF) model using the Northern Goshawk Optimization Algorithm. The Northern Goshawk Optimization Algorithm (NGO) first initializes the population and determines the initial solution space; then obtains the candidate solution through the first stage optimization, and further refines the search for the optimal solution in the second stage; finally, determines whether the optimal parameter combination is achieved. If not, it iterates repeatedly until the optimization conditions are met. The accuracy and robustness of the model in predicting the rolling force of horizontal rolls and vertical rolls under different rolling conditions are improved through parameter optimization. The entire process effectively combines the global search capability of the NGO algorithm and the prediction performance of the random forest to ensure the improvement of model accuracy.

[0061] Please continue reading Figure 1 to Figure 2 As shown, in one embodiment of the present invention, the result output unit is used to output the rolling force results predicted based on the optimized random forest model in Excel format. The output results are provided to factory staff for process optimization and production decision-making, so that factory staff can quickly adjust process parameters according to actual production needs and improve production efficiency.

[0062] In summary, the present invention introduces a parameter calculation module that meets actual production conditions, and ensures the accuracy and reliability of input data by accurately calculating various process parameters in the steel rolling production process; the isolation forest algorithm is used to screen data outliers, which significantly improves the quality of the data, and the tree model is used for feature selection to screen out the features most relevant to the rolling force prediction, reduce redundant features, and improve the training efficiency of the model; the Northern Goshawk Optimization Algorithm (NGO) is used to optimize the number of decision trees and the maximum depth of the tree in the random forest model, so that the model can better adapt to the complex rolling process of H-beam, thereby improving the prediction accuracy and robustness. At the same time, the present invention can simultaneously achieve accurate prediction of the rolling force of the horizontal roll and the vertical roll of the H-beam without relying on traditional trial rolling through machine learning technology, significantly improving the prediction efficiency. Compared with traditional manual experience or simplified theoretical models, the present invention can quickly respond to changes in parameters during the production process, and export the prediction results in Excel format through the result output unit, which is convenient for factory staff to perform subsequent operations and analysis on the rolling force prediction results, helping the factory to quickly adjust the process and optimize the production process.

[0063] The above description is only a preferred embodiment of the present invention and should not be construed as limiting the present application. All equivalent changes and modifications made within the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. An H-beam rolling force prediction system based on machine learning technology, characterized in that: The system includes a parameter calculation module that meets actual production conditions, a data preprocessing module, and a rolling force prediction module based on optimized random forest; The parameter calculation module that meets the actual production conditions is used to obtain multiple process parameters of H-beams of different specifications by programming according to the actual rolling process parameters of the factory production, so as to provide input data for the subsequent rolling force prediction; The data preprocessing module is used to clean, select features and detect abnormal values ​​of the obtained H-beam rolling data; The rolling force prediction module based on optimized random forest is used to predict the rolling force of the preprocessed data and output the prediction result; The parameter calculation module that meets the actual production conditions includes a basic parameter setting unit, a rolling procedure design unit, and a parameter acquisition and storage unit; The basic parameter setting unit: according to the specifications of H-beam production, input the leader size, finished product size and roll size, and design the rolling pass according to this information; The rolling procedure design unit: according to the input basic parameters, the reduction rate is set, the reduction amount is inverted, and the rolling speed of each pass is set; The parameter acquisition and storage unit: stores the obtained rolling process parameters to provide input basis for the subsequent rolling force prediction module; Rolling pass setting: Users can set the number of passes through different rolling mills according to the actual layout of UR and UF rolling mills; each pass through the universal rolling mill is a PASS, and each pass through three tandem reversible rolling mills is a pass; the setting of the rolling pass N here is actually the number of passes, because passing through a universal rolling mill corresponds to a PASS, which corresponds to a rolling force; Reduction rate setting and inversion: Users can set the web and reduction rate from the 2nd PASS to the Nth PASS based on the known leader size, finished product size and rolling pass parameters. The system can invert the reduction rate and reduction amount from the first PASS to the last PASS. The obtained rolling data are processed by isolation forest algorithm to filter out outliers in the data; Perform feature selection on the remaining data after screening to select the features that have a greater impact on rolling force prediction and reduce redundant input variables; Standardize the parameters of different dimensions and normalize them to ensure the balance of parameter weights during model training and improve prediction accuracy; The preprocessed data are divided into training set, validation set and test set for subsequent model training and verification.

2. The H-beam rolling force prediction system based on machine learning technology according to claim 1 is characterized in that: The rolling force prediction module based on optimized random forest is further: The hyperparameters of the random forest model were optimized by using the Northern Goshawk Optimization Algorithm, and the model was trained with the training set data. During the training process, the validation set is used to evaluate the performance of the model, and the predictive ability of the model is optimized through cross-validation.

3. The H-beam rolling force prediction system based on machine learning technology according to claim 1 is characterized in that: The rolling force prediction module based on optimized random forest includes an isolated forest data outlier screening unit, a feature selection unit, a northern goshawk optimized random forest model unit and a result output unit.

4. The H-beam rolling force prediction system based on machine learning technology according to claim 1, characterized in that: The isolation forest data outlier screening unit is used to perform outlier detection on rolling data using the isolation forest algorithm, remove abnormal data that does not meet the expected range, and ensure the quality of the data set and the effectiveness of model training.

5. The H-beam rolling force prediction system based on machine learning technology according to claim 1, characterized in that: The feature selection unit is used to screen the features in the data set using a tree model method, identify features that are highly relevant to rolling force prediction, remove redundant information, and improve the training efficiency of the model.

6. The H-beam rolling force prediction system based on machine learning technology according to claim 1, characterized in that: The northern goshawk optimization random forest model unit is used to optimize the hyperparameters of the random forest model using the northern goshawk optimization algorithm, and the northern goshawk optimization algorithm first initializes the population and determines the initial solution space; Then, the candidate solution is obtained through the first stage optimization, and the optimal solution is further refined in the second stage; finally, it is determined whether the optimal parameter combination is achieved. If not, it is iterated repeatedly until the optimization conditions are met. The accuracy and robustness of the model in predicting the rolling force of horizontal rolls and vertical rolls under different rolling conditions are improved through parameter optimization.

7. The H-beam rolling force prediction system based on machine learning technology according to claim 1, characterized in that: The result output unit is used to output the rolling force results predicted based on the optimized random forest model in an Excel table format for factory staff to perform process optimization and production decision-making.

Images