Method, device and equipment for predicting magnetic performance of silicon steel and storage medium
By constructing a stacked basis model and a meta-model, and combining hyperparameter grid search and feature variable optimization, the prediction model for the magnetic properties of silicon steel was iteratively optimized, which solved the problem of low model accuracy, achieved efficient and accurate prediction of the magnetic properties of silicon steel, improved the yield rate and reduced the testing cost.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INST OF RES OF IRON & STEEL JIANGSU PROVINCE
- Filing Date
- 2025-12-18
- Publication Date
- 2026-05-01
AI Technical Summary
Existing models for predicting the magnetic properties of silicon steel suffer from underfitting or overfitting, resulting in low model accuracy. Furthermore, they fail to combine hyperparameter tuning and data feature tuning with a comprehensive grid search, leading to yield loss, increased testing costs, and unstable results.
By constructing stacked base models and meta-models, and combining hyperparameter grid search and feature variable optimization, the prediction model is iteratively tuned to optimize the hyperparameters and feature variables of the prediction model and improve the model accuracy.
It improves the accuracy of predicting the magnetic properties of silicon steel, reduces testing costs, reduces errors from manual testing, increases yield, and optimizes training efficiency and robustness.
Smart Images
Figure CN121963967A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of magnetic property prediction technology, and specifically to a method, apparatus, equipment, and storage medium for predicting the magnetic properties of silicon steel. Background Technology
[0002] Non-oriented silicon steel, as a core material for power equipment and drive motors in new energy vehicles, directly impacts equipment efficiency and energy consumption due to its magnetic properties (iron loss, magnetic induction, etc.). Traditional performance evaluation methods involve sampling from the beginning and end of annealed coils and sending them to a testing center to determine magnetic property indicators, which then serve as the basis for judging the magnetic properties of the silicon steel before it leaves the factory. This process not only results in yield loss and increased labor intensity but also incurs costs for sample processing and testing. Furthermore, differences in operator habits and sample processing locations can lead to performance instability and large fluctuations.
[0003] Common models for predicting the magnetic properties of silicon steel generally employ single model algorithms such as back propagation neural networks, random forests, and decision trees. These models suffer from underfitting or overfitting, which affects their accuracy. Furthermore, in terms of model optimization, they fail to combine hyperparameter tuning with data feature tuning to perform a comprehensive grid search, making it impossible to obtain the optimal training model. Summary of the Invention
[0004] In view of this, the present invention provides a method, apparatus, device and storage medium for predicting the magnetic properties of silicon steel, so as to solve the problem of low model prediction accuracy.
[0005] In a first aspect, the present invention provides a method for predicting the magnetic properties of silicon steel, the method comprising: Historical silicon steel composition data, historical process parameter data, and corresponding historical iron loss data and historical magnetic induction data are obtained, and training and testing sets are constructed after preprocessing each historical data set; The prediction model is initially trained using the training set, and the model accuracy is evaluated using the test set. If the model accuracy is not within the preset accuracy range, the prediction model is iteratively tuned using hyperparameter grid search and feature variable optimization until the model accuracy is within the preset accuracy range, thus obtaining an optimized prediction model. The feature variables include silicon steel composition and process parameters. Obtain silicon steel composition data and process parameter data from actual production, and use an optimized prediction model to predict iron loss data and magnetic induction data from actual production based on the silicon steel composition data and process parameter data from actual production.
[0006] The method for predicting the magnetic properties of silicon steel provided by this invention improves the yield of silicon steel by replacing manual inspection with a predictive model, while reducing inspection costs. It also eliminates the result errors caused by different operating methods of different personnel. By using hyperparameter grid search and feature variable optimization to iteratively fine-tune the predictive model, the optimal parameter combination is determined, which improves the prediction accuracy of the predictive model and ensures the accuracy of predicting the magnetic properties of silicon steel using the predictive model.
[0007] In one optional implementation, training and test sets are constructed after preprocessing the historical data, including: Historical data is integrated, cleaned, and noise-reduced to obtain effective data, and feature variables are selected based on the effective data; The effective data corresponding to the feature variables are divided into training set and test set according to a preset ratio, and the data in the training set is normalized.
[0008] The method for predicting the magnetic properties of silicon steel provided by this invention integrates, cleans, denoises, and filters feature variables from historical data, and divides the data into training and testing sets. This removes invalid, erroneous, and noisy data, ensures data quality, filters out key feature variables, reduces redundancy, and improves the efficiency and effectiveness of model training.
[0009] In one alternative implementation, the prediction model includes a stacked base model and a meta-model, wherein the stacked base model includes multiple decision trees and multiple random forests, and the meta-model includes a gradient boosting decision tree.
[0010] In one alternative implementation, the prediction model is initially trained using a training set, including: By simultaneously training a stacked base model using the training set, multiple prediction values can be obtained. The predicted values are used as input features to train the meta-model, resulting in a prediction model.
[0011] The method for predicting the magnetic properties of silicon steel provided by this invention utilizes the complementary advantages of decision trees and random forests, secondary learning of meta-models, and the integration of the advantages of multiple models to improve model accuracy and enhance the model's adaptability and generalization ability to different scenarios and data. The method involves parallel training of base models and targeted learning of meta-models, which ensures the effectiveness while optimizing the training process and improving modeling efficiency, thus helping to build a better prediction model for the magnetic properties of silicon steel.
[0012] In one alternative implementation, the prediction model is iteratively tuned using hyperparameter grid search and feature variable optimization, including: Determine the optimization range and optimization step size for each hyperparameter in the prediction model; The hyperparameters in the prediction model are initialized, and the hyperparameters are iteratively tuned sequentially based on their optimization range and optimization step size until each hyperparameter exceeds its optimization range. Based on the total number of feature variables in historical data, determine the optimization range and optimization step size for the number of feature variables; The number of feature variables is initialized, and the number of feature variables is iteratively tuned based on the optimization range and optimization step size of the number of feature variables until the model accuracy is within the preset accuracy range or the number of feature variables exceeds the corresponding optimization range.
[0013] In one optional implementation, the hyperparameters in the prediction model are initialized, and the hyperparameters are iteratively tuned sequentially based on their optimization range and optimization step size, including: The hyperparameters of each decision tree, each random forest, and the meta-model are initialized. The hyperparameters of each decision tree are tuned in ascending order and with corresponding optimization step sizes until the hyperparameters of each decision tree reach the upper limit of the corresponding optimization range. The hyperparameters of each random forest are tuned in ascending order and with corresponding optimization step sizes until the hyperparameters of each random forest reach the upper limit of the corresponding optimization range. The hyperparameters of the meta-model are tuned in ascending order with corresponding optimization step sizes until the hyperparameters of the meta-model reach the upper limit of the corresponding optimization range.
[0014] The method for predicting the magnetic properties of silicon steel provided by this invention determines the optimization range and step size of hyperparameters and iterates step by step to make the internal logic of the model conform to the data pattern; it optimizes based on the total number of feature variables, simplifies redundant variables, and reduces the computational burden. The two work together to improve the model's accuracy and generalization ability, making the prediction results more reliable and adaptable to various working conditions, while also optimizing training efficiency and saving computing power costs. Through multiple rounds of iteration, the robustness of the model is enhanced, and it can output stably even in the face of noise and outliers. By using equal step size search, the possibility of missing the extreme value of the objective function is reduced, which can improve the model training accuracy and flexibly balance accuracy and cost.
[0015] In one optional implementation, the number of feature variables is initialized, and the number of feature variables is tuned based on the optimization range and optimization step size, including: The optimization range for the number of feature variables is determined based on the total number of feature variables in historical data. The upper limit of the optimization range is the total number of feature variables, and the lower limit of the optimization range is half of the total number of feature variables. Calculate the correlation between each characteristic variable and iron loss and magnetic induction, and sort them in descending order according to the magnitude of the correlation to obtain the characteristic variable correlation sequence; The total number of feature variables is used as the initial number, and the number of feature variables is optimized based on the correlation sequence of feature variables with the corresponding optimization step size.
[0016] The method for predicting the magnetic properties of silicon steel provided by this invention selects feature variables based on the magnitude of their correlation with magnetic properties, avoiding redundant feature variables from interfering with the predictive ability of important feature variables for magnetic properties, thereby improving model optimization efficiency and prediction performance.
[0017] Secondly, the present invention provides a device for predicting the magnetic properties of silicon steel, the device comprising: The data acquisition and preprocessing module is used to acquire historical silicon steel composition data, historical process parameter data, and corresponding historical iron loss data and historical magnetic induction data, and to construct training and test sets after preprocessing each historical data. The model training and optimization module is used to initially train the prediction model using the training set and evaluate the model accuracy using the test set. If the model accuracy is not within the preset accuracy range, the prediction model is iteratively optimized using hyperparameter grid search and feature variable optimization until the model accuracy is within the preset accuracy range, thus obtaining an optimized prediction model. The feature variables include silicon steel composition and process parameters. The magnetic performance prediction module is used to acquire silicon steel composition data and process parameter data in actual production, and based on the silicon steel composition data and process parameter data in actual production, use the optimized prediction model to predict iron loss data and magnetic induction data in actual production.
[0018] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method described in the first aspect or any corresponding embodiment thereof.
[0019] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description
[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0021] Figure 1This is a flowchart illustrating a method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the model structure of the prediction model in the method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating another method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the overall process of predicting model optimization in the method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention. Figure 5 This is a schematic diagram of the tuning process of the hyperparameter tuning module of the decision tree in the method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention. Figure 6 This is a schematic diagram of the tuning process of the hyperparameter tuning module of random forest in the method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention. Figure 7 This is a schematic diagram of the tuning process of the hyperparameter tuning module of the GBDT model in the method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention. Figure 8 This is a schematic diagram of the iterative optimization process of characteristic variables in the method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention; Figure 9 This is a schematic flowchart of a specific embodiment of the method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention; Figure 10 This is a scatter plot comparing the predicted and actual values of iron loss in the method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention. Figure 11 This is a scatter plot comparing the predicted magnetic induction value with the actual value in the method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention. Figure 12 This is a structural block diagram of a device for predicting the magnetic properties of silicon steel according to an embodiment of the present invention; Figure 13 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] This invention provides a method for predicting the magnetic properties of silicon steel. By constructing a stacked basis model and a meta-model, and iteratively optimizing the model hyperparameters and feature variables, the prediction accuracy of the magnetic properties of silicon steel can be improved.
[0024] According to an embodiment of the present invention, a method for predicting the magnetic properties of silicon steel is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.
[0025] This embodiment provides a method for predicting the magnetic properties of silicon steel, which can be used in the aforementioned computer system. Figure 1 This is a flowchart of a method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention, as shown below. Figure 1 As shown, the process includes the following steps: Step S101: Obtain historical silicon steel composition data, historical process parameter data, and corresponding historical iron loss data and historical magnetic induction data, and construct training set and test set after preprocessing each historical data.
[0026] Specifically, historical data on silicon steel composition, process parameters, iron loss, and magnetic induction are obtained from the production management system. After preprocessing, the historical data is divided into training and testing sets according to a certain ratio. The silicon steel composition includes, but is not limited to, the main components such as carbon (C), silicon (Si), manganese (Mn), phosphorus (P), sulfur (S), aluminum (Al), nitrogen (N), chromium (Cr), copper (Cu), and titanium (Ti). The process parameters include, but are not limited to, the hot rolling furnace temperature, roughing rolling temperature, finishing rolling temperature, coiling temperature, normalizing temperature, normalizing rate, annealing temperature, and annealing rate. The iron loss and magnetic induction data are derived from the comprehensive magnetic performance evaluation data. The iron loss data reflects the energy loss of silicon steel, affecting the energy efficiency of equipment such as motors and transformers. The magnetic induction data reflects the magnetic permeability of silicon steel, which is related to the magnetic circuit design and performance of electrical equipment.
[0027] Step S102: Initially train the prediction model using the training set, and evaluate the model accuracy using the test set. If the model accuracy is not within the preset accuracy range, iteratively optimize the prediction model using hyperparameter grid search and feature variable optimization until the model accuracy is within the preset accuracy range, thus obtaining an optimized prediction model. The feature variables include silicon steel composition and process parameters.
[0028] In some alternative implementations, the prediction model includes a stacking base model and a meta-model, wherein the stacking base model includes multiple decision trees and multiple random forests, and the meta-model includes a gradient boosting decision tree.
[0029] Specifically, a suitable stacked base model and meta-model are selected as the prediction model. The training set is substituted into the prediction model to complete the initial training. The model accuracy of the prediction model is evaluated using the test set. If the model accuracy is not within the preset accuracy range, the prediction model is iteratively tuned using hyperparameter grid search and feature variable optimization until the model accuracy is within the preset accuracy range, thus obtaining an optimized prediction model.
[0030] like Figure 2 The diagram shows the model structure of the prediction model, which includes a stacked model and a base model. The stacked base model includes multiple decision trees and multiple random forests. The base model contains random forest and decision tree algorithms. There are N decision trees, named decision trees 1 to N, and M random forests, named random forests 1 to M. The meta-model is a gradient boosting decision tree (GBDT).
[0031] Step S103: Obtain silicon steel composition data and process parameter data in actual production, and use an optimized prediction model to predict iron loss data and magnetic induction data in actual production based on the silicon steel composition data and process parameter data in actual production.
[0032] Specifically, the optimized prediction model, based on prior training and optimization, incorporates the correlation between silicon steel composition, process parameters, and iron loss and magnetic induction. In actual production, silicon steel composition data and process parameter data can be acquired according to a preset sampling period. This data is then input into the optimized prediction model to quickly calculate and derive iron loss and magnetic induction data from actual production. Real-time prediction of iron loss and magnetic induction allows production personnel to anticipate the product's magnetic performance. If the prediction results are unsatisfactory, process parameters can be adjusted promptly (by changing the annealing temperature) to avoid batches of defective products, ensuring the stability of production equipment and improving product quality and energy efficiency.
[0033] The method for predicting the magnetic properties of silicon steel provided in this embodiment improves the yield of silicon steel by replacing manual inspection with a predictive model, while reducing inspection costs. It also eliminates the result errors caused by different operating methods of different personnel. By using hyperparameter grid search and feature variable optimization to iteratively fine-tune the predictive model, the optimal parameter combination is determined, which improves the prediction accuracy of the predictive model and ensures the accuracy of predicting the magnetic properties of silicon steel using the predictive model.
[0034] This embodiment provides a method for predicting the magnetic properties of silicon steel, which can be used in the aforementioned computer system. Figure 3 This is a flowchart of a method for predicting the magnetic properties of silicon steel according to an embodiment of the present invention, as shown below. Figure 3 As shown, the process includes the following steps: Step S201: Obtain historical silicon steel composition data, historical process parameter data, and corresponding historical iron loss data and historical magnetic induction data, and construct training set and test set after preprocessing each historical data.
[0035] Specifically, step S201 includes: Step S2011: Integrate, clean, and denoise the historical data to obtain valid data, and select feature variables based on the valid data.
[0036] Specifically, the acquired raw historical data needs to undergo integration, cleaning, and noise reduction to eliminate interference from noisy data and retain high-quality data as valid data. A three-standard-deviation screening method is typically used; this process is a mature existing technology and will not be elaborated here. The processed valid data is then initially screened based on its correlation with magnetic properties. The initial iteration trains the model using all feature variables, and subsequently, the number of feature variables is gradually reduced based on their correlation with magnetic properties. In this embodiment, the silicon steel composition data and process parameter data affecting magnetic properties are collectively referred to as feature variables. A higher correlation with magnetic properties indicates greater importance to magnetic properties. The correlation calculation process can employ mature existing technologies and will not be elaborated here.
[0037] Step S2012: Divide the valid data corresponding to the feature variables into training set and test set according to a preset ratio, and normalize the data in the training set.
[0038] Specifically, the valid data corresponding to the feature variables (e.g., carbon content of 10%, silicon content of 3%, etc. as a set of valid data) are divided into training set and test set according to a preset ratio, such as 80% training set and 20% test set. This is just an example, but not a limitation.
[0039] After splitting the dataset, the training set is normalized to eliminate the influence of different feature variables. The order of dataset splitting and data normalization should not be reversed, otherwise it will affect the model's generalization ability.
[0040] The method for predicting the magnetic properties of silicon steel provided in this embodiment integrates, cleans, denoises, and filters feature variables from historical data, and divides the data into training and testing sets. This removes invalid, erroneous, and noisy data, ensures data quality, filters out key feature variables, reduces redundancy, and improves the efficiency and effectiveness of model training.
[0041] Step S202: Initially train the prediction model using the training set, and evaluate the model accuracy using the test set. If the model accuracy is not within the preset accuracy range, iteratively optimize the prediction model using hyperparameter grid search and feature variable optimization until the model accuracy is within the preset accuracy range, thus obtaining an optimized prediction model. The feature variables include silicon steel composition and process parameters.
[0042] Specifically, step S202 above, which involves initially training the prediction model using the training set, includes: Step S2021: Simultaneously train the stacked base model using the training set to obtain multiple predicted values.
[0043] Specifically, the core of the Stacking ensemble learning algorithm lies in simultaneously training several base models using a training set, and then using the predictions from multiple models as input features to perform secondary training on a meta-model to obtain the final prediction model. For example... Figure 2 As shown, when training a stacked model simultaneously using the training set, each decision tree and each random forest will obtain a predicted result.
[0044] Step S2022: Use the predicted values as input features of the meta-model to train the meta-model and obtain the prediction model.
[0045] Specifically, the predicted values are used as input features to train the GBDT model, resulting in an initially trained prediction model. The process of training the model using the training set is a mature existing technique and will not be elaborated upon here. After the initial (first) model training is completed, the accuracy of the current model is evaluated. If the accuracy meets the standard, the model is directly output. If the accuracy does not meet the standard, the hyperparameters are first fine-tuned using a grid search method. Once the hyperparameters reach a critical value, the number of feature variables is optimized. This process of hyperparameter tuning is repeated iteratively until the model accuracy meets the requirements. If the model iteration is completely finished (hyperparameters reach critical values and the number of feature variables reaches the optimization threshold), and the accuracy requirement is still not met, the highest-accurate model is output from the historical training model list.
[0046] The method for predicting the magnetic properties of silicon steel provided in this embodiment integrates, cleans, denoises, and filters feature variables from historical data, and divides the data into training and testing sets. This removes invalid, erroneous, and noisy data, ensures data quality, filters out key feature variables, reduces redundancy, and improves the efficiency and effectiveness of model training.
[0047] Specifically, step S202 above involves iteratively optimizing the prediction model using hyperparameter grid search and feature variable optimization, including: Step S2023: Determine the optimization range and optimization step size of each hyperparameter in the prediction model.
[0048] Specifically, such as Figure 4 The diagram shows the overall process of predictive model optimization, which involves iterative model optimization in the order of base model hyperparameter tuning, meta-model hyperparameter tuning, and data feature selection.
[0049] First, determine the types and number of base models, including N decision trees and M random forests. The hyperparameters of the decision tree models include, but are not limited to: decision tree depth n1 (corresponding to an optimization step size of n). 1_step The minimum number of samples for downward splitting is n2 (corresponding to an optimization step size of n). 2_step The maximum number of leaf nodes is n3 (corresponding to an optimization step size of n). 3_step The hyperparameters of the random forest model include, but are not limited to: the number of trees n4 (corresponding to an optimization step size of n). 4_step The tree depth is n5 (corresponding to an optimization step size of n). 5_step ), the number of splitting features n6 (corresponding to an optimization step size of n) 6_step The GBDT algorithm is chosen as the metamodel. The hyperparameters of the metamodel include, but are not limited to: the number of trees n7 (corresponding to an optimization step size of n). 7_step The learning rate is n / 8 (corresponding to an optimization step size of n). 8_step The tree depth is n9 (corresponding to an optimization step size of n). 9_step ).
[0050] Step S2024: Initialize the hyperparameters in the prediction model, and iteratively tune the hyperparameters sequentially based on their optimization range and optimization step size until each hyperparameter exceeds its optimization range.
[0051] In some optional implementations, step S2024 above includes: Step a1: Initialize the hyperparameters of each decision tree, each random forest, and the meta-model.
[0052] Specifically, let n1 = n 1_min n2=n 2_min n3=n 3_min n4=n 4_min n5=n 5_min n6=n 6_min n7=n 7_min n8=n 8_min n9=n 9_min This allows for the initialization of various hyperparameters. 。
[0053] Step a2: The hyperparameters of each decision tree are tuned in ascending order and with the corresponding optimization step size until the hyperparameters of each decision tree reach the upper limit of the corresponding optimization range.
[0054] Specifically, each decision tree corresponds to a hyperparameter tuning module, such as Figure 5 The diagram shows the tuning process of the hyperparameter tuning module for a decision tree. The optimization range and step size for each hyperparameter are set as follows: the critical value of the decision tree depth n1 is defined as n 1_min=10, n 1_max =100, optimal step size n 1_step =10; the critical value of the minimum number of samples for downward splitting, n², is defined as n... 2_min =1, n 2_max =50, optimal step size n 2_step =5; the critical value for the maximum number of leaf nodes n3 is defined as n 3_min =1, n 3_max =100, optimal step size n 3_step =10.
[0055] Step a3: Optimize the hyperparameters of each random forest in ascending order with the corresponding optimization step size until the hyperparameters of each random forest reach the upper limit of the corresponding optimization range.
[0056] Specifically, each random forest corresponds to a hyperparameter tuning module, such as Figure 6 The diagram shows the tuning process of the hyperparameter tuning module for random forests. The optimization range and step size for each hyperparameter are set as follows: the number of trees n4 and the critical value are defined as n 4_min =10, n 4_max =100, optimal step size n 4_step =10; the critical value of tree depth n5 is defined as n 5_min =10, n 5_max =100, optimal step size n 5_step =10; the critical value of the splitting characteristic number n6 is defined as n 6_min =1, n 6_max =20, optimal step size n 6_step =2.
[0057] Step a4: The hyperparameters of the meta-model are tuned in ascending order with corresponding optimization step sizes until the hyperparameters of the meta-model reach the upper limit of the corresponding optimization range.
[0058] Specifically, after all the base models have completed one round of hyperparameter tuning, the metamodel hyperparameters are then tuned. For example... Figure 7 The diagram shows the tuning process of the hyperparameter tuning module for the GBDT model. The optimization range and step size for each hyperparameter are set as follows: the number of trees n7 and the critical value are defined as n7. 7_min =10, n 7_max =200, optimal step size n 7_step =20; the critical value of the learning rate n8 is defined as n 8_min =0.01, n 8_max =1, optimal step size n 8_step =10; the critical value of tree depth n6 is defined as n 9_min =1, n 9_max =20, optimal step size n 9_step=2.
[0059] Step S2025: Determine the optimization range and optimization step size of the number of feature variables based on the total number of feature variables in the historical data.
[0060] Specifically, after the base model and meta-model complete one round of hyperparameter tuning, the data feature selection module is entered to iteratively fine-tune the feature variables of the data source. The total number of feature variables n 10 The sum of the number of silicon steel components and the number of production parameters in historical data, with an initial value of n. 10 =n 10_max The upper limit of the optimization range for the number of feature variables is n. 10_max The lower limit can be set to n. 10_min =0.5*n 10_max The optimization step size can be set to n. 10_step =1, is just an example, but is not limited to this.
[0061] Step S2026: Initialize the number of feature variables, and iteratively optimize the number of feature variables based on the optimization range and optimization step size until the model accuracy is within the preset accuracy range or the number of feature variables exceeds the corresponding optimization range.
[0062] The method for predicting the magnetic properties of silicon steel provided in this embodiment determines the optimization range and step size of hyperparameters and iterates step by step to make the internal logic of the model conform to the data pattern. Based on the total number of feature variables, the optimization is carried out to simplify redundant variables and reduce the computational burden. The two work together to improve the model's accuracy and generalization ability, making the prediction results more reliable and adaptable to various working conditions. At the same time, the training efficiency is optimized and the computing power cost is saved. Through multiple rounds of iteration, the robustness of the model is enhanced, and it can output stably in the face of noise and outliers. By using equal step size search, the possibility of missing the extreme value of the objective function is reduced, which can improve the model training accuracy and flexibly balance accuracy and cost.
[0063] In some optional implementations, step S2026 above includes: Step b1: Determine the optimization range of the number of feature variables based on the total number of feature variables in the historical data. The upper limit of the optimization range is the total number of feature variables, and the lower limit of the optimization range is half of the total number of feature variables.
[0064] Specifically, such as Figure 8 The diagram illustrates the iterative optimization process of feature variables. The optimization range for the number of feature variables is determined based on the total number of feature variables in historical data. For example, if the number of silicon steel components is 12 and the number of process parameters is 28 in historical data, then the total number of feature variables is 41, and the optimization range for the number of feature variables can be n. 10_max =41,n 10_min=20 or 21, are just examples, but are not limited to.
[0065] Step b2: Calculate the correlation between each characteristic variable and iron loss and magnetic induction, and sort them in descending order according to the magnitude of the correlation to obtain the characteristic variable correlation sequence.
[0066] Specifically, the correlation between each characteristic variable and iron loss and magnetic induction is calculated. The calculation process can use mature existing technology, which will not be elaborated here. The variables are then sorted in descending order according to the magnitude of the correlation to obtain the correlation sequence of the characteristic variables.
[0067] Step b3: Using the total number of feature variables as the initial number, the number of feature variables is optimized based on the correlation sequence of feature variables and the corresponding optimization step size.
[0068] Specifically, the upper limit of the optimization range of the number of feature variables is used as the initial number. Based on the correlation sequence of feature variables, the number of feature variables is decreased sequentially with the corresponding optimization step size, reducing one feature variable each time. The model accuracy of the prediction model is calculated each time, and it is determined whether the model accuracy meets the requirements. If the requirements are met, the iterative optimization stops. If the model accuracy does not meet the requirements, iterative optimization continues until the number of feature variables reaches the lower limit of the optimization range. The hyperparameters and the model corresponding to the feature variables when the model accuracy is the highest are taken as the final optimized prediction model.
[0069] The method for predicting the magnetic properties of silicon steel provided in this embodiment filters feature variables based on their correlation with magnetic properties, avoiding redundant feature variables from interfering with the predictive ability of important feature variables for magnetic properties, thereby improving model optimization efficiency and prediction performance.
[0070] Step S203: Obtain silicon steel composition data and process parameter data from actual production, and based on this data, use an optimized prediction model to predict iron loss data and magnetic induction data from actual production. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.
[0071] In one specific embodiment, historical production data of a certain silicon steel is first obtained from the production management system of the Manufacturing Execution System (MES). The data contains 20,000 records, including 12 characteristic variables of silicon steel composition (carbon (C), silicon (Si), manganese (Mn), phosphorus (P), sulfur (S), aluminum (Al), nitrogen (N), chromium (Cr), copper (Cu), titanium (Ti), tin (Sn), nickel (Ni), and molybdenum (Mo)). The process parameters include hot rolling, normalizing, pickling, and annealing processes, with a total of 28 characteristic variables (heating furnace soaking zone temperature, normalizing process speed, pickling temperature, annealing temperature, etc.). The performance parameters include two magnetic properties: process speed, furnace time, NOF3 in normalizing furnace, pickling speed, RTF1 section furnace temperature, roughing mill exit temperature, NOF5 in normalizing furnace, tension leveling, RTF2 section furnace temperature, final rolling temperature, SF3 in normalizing furnace, process belt speed, RTF3 section furnace temperature, coiling temperature, SF5 in normalizing furnace, RTF4 section furnace temperature, SF6 section furnace temperature, SF7 section furnace temperature, SF8 section furnace temperature, SF9 section furnace temperature, SF10 section furnace temperature, SF11 section furnace temperature, hydrogen content, dew point, and furnace tension.
[0072] like Figure 9 As shown, this is a schematic diagram of the overall process of this embodiment, following the... Figure 10 The process shown is used to train and optimize the prediction model. Due to the numerous interfering factors and large fluctuations in industrial big data, it is necessary to re-clean and integrate 20,000 historical silicon steel production data points. A three-standard-deviation denoising method is used to remove abnormal production data. The core idea is to consider data exceeding the mean ± three standard deviations as outliers and remove them. The standard deviation calculation formula is as follows:
[0073] in Represents the average value of the characteristic variable. n Indicates the number of samples. This represents a single sample. The threshold range is set to... ±3 Sample data that exceeds the threshold range will be removed.
[0074] After the original data was processed by data denoising, there were about 18,000 data points left. The next step was to select the feature variables. All 41 feature variables were retained in the initial training without any selection.
[0075] The data was divided into training and test sets in an 8:2 ratio. The training set contained 14,400 data points, which were used to train the Stacking model, while the test set contained 3,600 data points, which were used to verify and evaluate the model's accuracy.
[0076] Because different feature variables have different dimensions and significantly different orders of magnitude, they can interfere with model training. Therefore, data normalization is needed to convert all feature variables to a dimensionless format and scale the data fluctuation range to 0~1. The data normalization formula is as follows:
[0077] The above formula is used to uniformly normalize all feature variables in the training set. Additionally, during the data normalization operation for the test set, the formula... , The maximum and minimum values in the training set are used as the standard.
[0078] After data preprocessing, the model selection module was entered. Decision trees (1-10) and random forests (1-10) were selected as the base models, for a total of 20. The GBDT gradient boosting algorithm was selected as the meta-model. The optimization range and preset values of all model hyperparameters and feature variables are shown in Table 1.
[0079] Table 1. Optimization Range and Preset Values for Model Hyperparameters and Feature Variables
[0080] After all model hyperparameters are preset, the model training phase begins. After each training iteration, the model accuracy is evaluated using a test set, and the coefficient of determination R is output. 2 (R) 2 The closer to 1, the higher the model accuracy. 2 If the value is greater than 0.9, output the training model for this round. If the value is not met, perform gridded search optimization in the order of decision tree hyperparameters - random forest hyperparameters - GBDT hyperparameters - number of feature variables until the model accuracy meets the requirements.
[0081] In this example, after approximately 10,000 iterations of training, the model accuracy met the requirements, and the optimal model hyperparameters were obtained as shown in Table 2.
[0082] Table 2. Optimal Model Hyperparameter Table
[0083] The final number of characteristic variables is 26: C, Si, Mn, P, S, Ti, N, Al, heating furnace soaking zone temperature, furnace time, roughing mill exit temperature, final rolling temperature, coiling temperature, normalizing process speed, normalizing furnace SF3, normalizing furnace SF5, pickling temperature, pickling speed, process belt speed, annealing process speed, SF9 section furnace temperature, SF10 section furnace temperature, SF11 section furnace temperature, hydrogen content, dew point, and furnace tension.
[0084] The accuracy evaluation of the silicon steel performance prediction model is shown in Table 3.
[0085] Table 3 Prediction Model Accuracy Evaluation Table
[0086] A scatter plot comparing the predicted and actual values of silicon steel performance obtained based on the optimized prediction model is shown below. Figure 10 and Figure 11 As shown, where Figure 10 This is a scatter plot comparing the predicted and actual iron loss values. Figure 11 This is a scatter plot comparing the predicted and actual magnetic field values. Figure 10 and Figure 11 The scattered points are roughly distributed near a straight line passing through the origin with a slope of 1, indicating that the predicted values of the two magnetic properties are quite close to the actual values. The prediction accuracy of the optimized prediction model provided in this embodiment is high.
[0087] This embodiment also provides a device for predicting the magnetic properties of silicon steel, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.
[0088] This embodiment provides a device for predicting the magnetic properties of silicon steel, such as... Figure 12 As shown, it includes: The data acquisition and preprocessing module 1201 is used to acquire historical silicon steel composition data, historical process parameter data, and corresponding historical iron loss data and historical magnetic induction data, and to construct training sets and test sets after preprocessing each historical data.
[0089] The model training and optimization module 1202 is used to initially train the prediction model using the training set and evaluate the model accuracy using the test set. If the model accuracy is not within the preset accuracy range, the prediction model is iteratively optimized using hyperparameter grid search and feature variable optimization until the model accuracy is within the preset accuracy range, thus obtaining an optimized prediction model. The feature variables include silicon steel composition and process parameters.
[0090] The magnetic performance prediction module 1203 is used to acquire silicon steel composition data and process parameter data in actual production, and to predict iron loss data and magnetic induction data in actual production using the optimized prediction model based on the silicon steel composition data and process parameter data in actual production.
[0091] In some optional implementations, the data acquisition and preprocessing module 1201 includes: The data processing unit is used to integrate, clean, and reduce noise in historical data to obtain effective data, and to filter feature variables based on the effective data.
[0092] The data partitioning unit is used to divide the valid data corresponding to the feature variables into training set and test set according to a preset ratio.
[0093] In some alternative implementations, the model training and tuning module 1202 includes: The base model training unit is used to train stacked base models simultaneously using the training set to obtain multiple prediction values.
[0094] The meta-model training unit is used to train the meta-model by using the predicted values as input features, thereby obtaining the prediction model.
[0095] The hyperparameter optimization condition determination unit is used to determine the optimization range and optimization step size of each hyperparameter in the prediction model.
[0096] The hyperparameter iterative tuning unit is used to initialize the hyperparameters in the prediction model and iteratively tune the hyperparameters sequentially based on their optimization range and optimization step size until each hyperparameter exceeds its optimization range.
[0097] The feature variable optimization condition determination unit is used to determine the optimization range and optimization step size of the number of feature variables based on the total number of feature variables in historical data.
[0098] The feature variable iterative optimization unit is used to initialize the number of feature variables and iteratively optimize the number of feature variables based on the optimization range and optimization step size until the model accuracy is within the preset accuracy range or the number of feature variables exceeds the corresponding optimization range.
[0099] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.
[0100] In this embodiment, the device for predicting the magnetic properties of silicon steel is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above-mentioned functions.
[0101] This invention also provides a computer device having the above-described features. Figure 12 The device shown is for predicting the magnetic properties of silicon steel.
[0102] Please see Figure 13 , Figure 13This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 13 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 13 Take a processor 10 as an example.
[0103] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.
[0104] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.
[0105] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0106] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.
[0107] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.
[0108] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.
[0109] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.
Claims
1. A method for predicting the magnetic properties of silicon steel, characterized in that, The method includes: Historical silicon steel composition data, historical process parameter data, and corresponding historical iron loss data and historical magnetic induction data are obtained, and training and testing sets are constructed after preprocessing each historical data set; The prediction model is initially trained using the training set, and the model accuracy is evaluated using the test set. If the model accuracy is not within the preset accuracy range, the prediction model is iteratively tuned using hyperparameter grid search and feature variable optimization until the model accuracy is within the preset accuracy range, thus obtaining an optimized prediction model. The feature variables include silicon steel composition and process parameters. Obtain silicon steel composition data and process parameter data from actual production, and use the optimized prediction model to predict iron loss data and magnetic induction data from actual production based on the silicon steel composition data and process parameter data from actual production.
2. The method according to claim 1, characterized in that, After preprocessing the historical data, training and test sets are constructed, including: The historical data is integrated, cleaned, and noise-reduced to obtain effective data, and feature variables are selected based on the effective data; The valid data corresponding to the feature variables are divided into training set and test set according to a preset ratio, and the data in the training set is normalized.
3. The method according to claim 1, characterized in that, The prediction model includes a stacked base model and a meta-model. The stacked base model includes multiple decision trees and multiple random forests. The meta-model includes a gradient boosting decision tree.
4. The method according to claim 3, characterized in that, The prediction model is initially trained using the training set, including: The stacked base model is trained simultaneously using the training set to obtain multiple predicted values; The predicted values are used as input features of the meta-model to train the meta-model and obtain the prediction model.
5. The method according to claim 3, characterized in that, The prediction model is iteratively tuned using hyperparameter grid search and feature variable optimization, including: Determine the optimization range and optimization step size for each hyperparameter in the prediction model; The hyperparameters in the prediction model are initialized, and the hyperparameters are iteratively tuned sequentially based on their optimization range and optimization step size until each hyperparameter exceeds its optimization range. Based on the total number of feature variables in historical data, determine the optimization range and optimization step size for the number of feature variables; The number of feature variables is initialized, and the number of feature variables is iteratively optimized based on the optimization range and optimization step size of the number of feature variables until the model accuracy is within the preset accuracy range or the number of feature variables exceeds the corresponding optimization range.
6. The method according to claim 5, characterized in that, The hyperparameters in the prediction model are initialized, and then iteratively tuned based on the optimization range and step size of each hyperparameter, including: The hyperparameters of each decision tree, each random forest, and the meta-model are initialized. The hyperparameters of each decision tree are tuned in ascending order and with corresponding optimization step sizes until the hyperparameters of each decision tree reach the upper limit of the corresponding optimization range. The hyperparameters of each random forest are tuned in ascending order and with corresponding optimization step sizes until the hyperparameters of each random forest reach the upper limit of the corresponding optimization range. The hyperparameters of the meta-model are tuned in ascending order with corresponding optimization step sizes until the hyperparameters of the meta-model reach the upper limit of the corresponding optimization range.
7. The method according to claim 5, characterized in that, The number of feature variables is initialized, and the number of feature variables is tuned based on the optimization range and optimization step size, including: The optimization range for the number of feature variables is determined based on the total number of feature variables in historical data. The upper limit of the optimization range is the total number of feature variables, and the lower limit of the optimization range is half of the total number of feature variables. Calculate the correlation between each characteristic variable and iron loss and magnetic induction, and sort them in descending order according to the magnitude of the correlation to obtain the characteristic variable correlation sequence; The total number of feature variables is used as the initial number, and the number of feature variables is optimized based on the correlation sequence of feature variables with the corresponding optimization step size.
8. A device for predicting the magnetic properties of silicon steel, characterized in that, The device includes: The data acquisition and preprocessing module is used to acquire historical silicon steel composition data, historical process parameter data, and corresponding historical iron loss data and historical magnetic induction data, and to construct training and test sets after preprocessing each historical data. The model training and optimization module is used to initially train the prediction model using the training set and evaluate the model accuracy using the test set. If the model accuracy is not within the preset accuracy range, the prediction model is iteratively optimized using hyperparameter grid search and feature variable optimization until the model accuracy is within the preset accuracy range, thus obtaining an optimized prediction model. The feature variables include silicon steel composition and process parameters. The magnetic performance prediction module is used to acquire silicon steel composition data and process parameter data in actual production, and based on the silicon steel composition data and process parameter data in actual production, use the optimized prediction model to predict iron loss data and magnetic induction data in actual production.
9. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the method of any one of claims 1 to 7.