A method and system for predicting the mechanical properties of hot-rolled strip steel

Abstract

This invention belongs to the field of material performance prediction technology, specifically a method and system for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and ensemble real-time learning. First, multiple feature quantification methods are used to obtain feature weights, and a time weight is introduced on top of these feature weights to construct a similarity metric that comprehensively considers input features and time factors to select suitable modeling samples, overcoming the limitations of modeling with a single similarity metric. Second, a cumulative similarity factor is introduced to ensure that the optimal modeling data sample set is selected for local modeling. Finally, an ensemble learning framework is used to integrate the output results of each local model, reducing the dependence of the prediction results on a single model, thereby significantly enhancing model performance and improving the accuracy and generalization of hot-rolled strip steel mechanical property prediction under complex production conditions.

Description

Technical Field

[0001] This invention relates to the field of material property prediction technology, specifically to a method and system for predicting the mechanical properties of hot-rolled strip steel, and more specifically, to a method and system for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning. Background Technology

[0002] Steel materials play a fundamental role in numerous industrial sectors due to their superior mechanical properties and advantages in economic and environmental sustainability. Mechanical properties, as a key indicator for evaluating the quality of hot-rolled strip steel, are crucial for accurately predicting performance, reducing sample testing frequency, ensuring product quality stability, optimizing steel composition, and promoting new product development.

[0003] Currently, mechanical property prediction methods are mainly divided into two categories: those based on traditional physical metallurgy models and those based on artificial intelligence models. The construction of traditional physical metallurgy models relies on laboratory test data, which is limited by professional knowledge, cost, time, and applicability to specific steel grades. While artificial intelligence-based models benefit from advancements in data acquisition and management technologies, they still fall short in terms of prediction accuracy and adaptability when faced with the complex, strongly coupled, and nonlinear relationships between the chemical composition, crystal phase structure, and mechanical properties of hot-rolled products.

[0004] A review of relevant literature reveals that data-driven prediction models for the mechanical properties of hot-rolled strip steel fall into two categories: global modeling and just-in-time (JITL), also known as local modeling. Global modeling relies on a model trained offline from historical datasets in a single instance, expecting it to provide services over a long period. While this approach offers advantages in operational efficiency, it has limitations in handling the time-varying nature of industrial data, making it difficult for the model to consistently and stably adapt to the dynamic changes in industrial processes. Furthermore, global modeling is insufficient in capturing process variable information, directly impacting the model's predictive ability. In contrast, JITL has garnered widespread attention in the industry due to its ability to address specific problems. JITL models have three core characteristics: first, they only begin building the model upon receiving a query request; second, they assess the similarity between the test sample and historical samples based on similarity, selecting relevant samples to build a local model based on this similarity; and finally, the model is discarded after prediction to improve adaptability and accuracy.

[0005] However, the predictive accuracy of JITL models is highly dependent on the modeling samples selected through similarity measurement methods. In sample analysis, multiple similarity dimensions are common, and these dimensions may have different impacts on the similarity relationships between different samples. Typically, a single similarity measurement method may not be sufficient to fully capture the true similarity structure between samples. Especially in the prediction of the mechanical properties of hot-rolled strip steel, the multidimensional similarity relationships between input and output features make it difficult for existing technologies to accurately select relevant samples, which is a prominent technical challenge in this field. Summary of the Invention

[0006] To address the problems existing in the prior art, the main objective of this invention is to propose a method and system for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning.

[0007] According to one aspect of the present invention, the present invention provides the following technical solution:

[0008] A method for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning includes the following steps:

[0009] S1. Collect and preprocess the hot-rolled strip steel production dataset, using the hot-rolling production process parameters X[X1, X2, …, X…]. n ] is the independent variable, and the mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel are used as the independent variable. n Using [] as the dependent variable, and taking the production data of each coil of hot-rolled strip steel as one data set, a historical dataset for the prediction model is constructed.

[0010] S2. For the historical dataset, use feature quantization method A to calculate the hot rolling production process parameters X[X1, X2, …, X…]. n The mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel n The feature weights that influence the data are determined; at the same time, each data group is sorted by the corresponding hot-rolled strip steel production time and assigned a time weight.

[0011] S3. Based on feature weights and time weights, calculate the feature-time multidimensional similarity measure between the test sample and the samples in the historical dataset. Select multiple samples with high similarity to the test sample from the historical dataset by accumulating similarity factors, and use them as training datasets to build a local model based on instant learning. Output the prediction results of the mechanical properties of hot-rolled strip steel based on feature quantization method A.

[0012] S4. Use at least one other feature quantization method (feature quantization method B, C, D...) different from feature quantization method A, and repeat steps S2-S3;

[0013] S5. An ensemble learning strategy is adopted to weight and integrate the prediction results of multiple local models obtained under different feature quantization methods, and output the prediction results of the mechanical properties of hot-rolled strip steel based on multi-dimensional feature weighted similarity and ensemble real-time learning as the final output of the model.

[0014] As a preferred embodiment of the hot-rolled strip mechanical property prediction method based on multidimensional feature similarity and integrated real-time learning described in this invention, in step S1, data that meets one of the following three conditions during the dataset preprocessing process is considered abnormal data:

[0015] (1) The measured value is outside the predetermined range; (2) Data is missing; (3) The 3σ principle is not met.

[0016] As a preferred embodiment of the hot-rolled strip mechanical property prediction method based on multidimensional feature similarity and integrated real-time learning described in this invention, wherein: in step S1, the hot-rolling production process parameters X[X1, X2, …, X…] are… n Including but not limited to chemical element content, furnace exit temperature, roughing rolling start temperature, roughing rolling start thickness, roughing rolling finish thickness, finishing rolling start temperature, finishing rolling finish temperature, finishing rolling finish thickness, coiling temperature, etc.; mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel. n Including but not limited to tensile strength, yield strength, elongation, etc.

[0017] As a preferred embodiment of the hot-rolled strip mechanical property prediction method based on multidimensional feature similarity and integrated real-time learning described in this invention, in step S2, the feature quantization method includes, but is not limited to, Pearson correlation analysis (PPC), Spearman correlation analysis (SCC), mutual information correlation analysis (MICC), and mean decreasing precision (MDA).

[0018] As a preferred embodiment of the hot-rolled strip mechanical property prediction method based on multidimensional feature similarity and integrated real-time learning described in this invention, wherein: in step S3, the formula for calculating the feature-time multidimensional similarity metric between the sample to be tested and the samples in the historical dataset is:

[0019]

[0020] in: This represents the feature-time multidimensional similarity measure between the sample to be tested and historical samples. For the first j A historical sample, The sample to be tested. This is a measure of the similarity between the feature weights of the sample to be tested and historical samples. For historical samples, time weights This is an adjustable parameter.

[0021] As a preferred embodiment of the hot-rolled strip mechanical property prediction method based on multidimensional feature similarity and integrated real-time learning described in this invention, in step S3, the feature weight similarity metric is introduced using Pearson correlation analysis (PPC) as an example. Calculation method:

[0022] The formula for calculating feature weights using Pearson correlation analysis (PPC) is as follows:

[0023]

[0024] in: n Indicates the number of samples. and These represent sample dimensions as follows: d Input and output observations, and These represent the sample input mean and output mean, respectively.

[0025] Combining the feature weights calculated by Pearson correlation analysis (PCC) with traditional Euclidean similarity, the formula for calculating the feature weight similarity measure is as follows:

[0026]

[0027] in: For the j-th historical sample; For testing purposes; d Dimensions of the input data; Indicates the input feature weights; To test the similarity between the sample and historical samples.

[0028] As a preferred embodiment of the hot-rolled strip mechanical property prediction method based on multidimensional feature similarity and integrated real-time learning described in this invention, wherein: in step S3, the cumulative similarity factor... L The calculation formula is:

[0029]

[0030] Wherein: the numerator represents the sum of the similarity contributions of the top n groups of samples with high similarity, and the denominator represents the sum of the similarity between the test sample and all historical samples.

[0031] As a preferred embodiment of the hot-rolled strip mechanical property prediction method based on multidimensional feature similarity and integrated real-time learning described in this invention, in step S3, the local model construction method based on real-time learning includes, but is not limited to, multiple linear regression (MLR), partial least squares regression (PLS), support vector machine (SVM), etc.

[0032] As a preferred embodiment of the hot-rolled strip mechanical property prediction method based on multidimensional feature similarity and integrated real-time learning described in this invention, wherein: in step S5, the weighted integration of prediction results from multiple local models depends on the performance of each local model on the validation dataset, the... r The mean absolute error of each model ( MAE The calculation formula is:

[0033]

[0034] in: and The first i Observed and predicted values ​​for each test sample. m This represents the number of samples in the test set.

[0035] As a preferred embodiment of the hot-rolled strip mechanical property prediction method based on multidimensional feature similarity and integrated real-time learning described in this invention, wherein: in step S5, through each model... MAE Definition of the first r The weights of the output results of each sub-model are:

[0036]

[0037] Where: the numerator term represents the first... r Sub-models e For the base, MAE The function value is the exponent, and the denominator represents all sub-models. e For the base, MAE It is the sum of the function values ​​of the exponent.

[0038] According to one aspect of the present invention, the present invention provides the following technical solution:

[0039] A system for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning, comprising:

[0040] Production Dataset Acquisition and Preprocessing Module: Acquires and preprocesses hot-rolled strip steel production data, using hot-rolling production process parameters X[X1, X2, …, X…]. n ] is the independent variable, and the mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel are used as the independent variable. n Using [] as the dependent variable, and taking the production data of each coil of hot-rolled strip steel as one data set, a historical dataset for the prediction model is constructed.

[0041] Weight determination module: For historical datasets, feature quantization method A is used to calculate hot rolling production process parameters X[X1, X2, …, X…]. nThe mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel n The feature weights that influence the data are determined; at the same time, each data group is sorted by the corresponding hot-rolled strip steel production time and assigned a time weight.

[0042] The multidimensional feature similarity calculation and local model construction module calculates the feature-time multidimensional similarity measure between the test sample and samples in the historical dataset based on feature weights and time weights. It then selects multiple samples with high similarity to the test sample from the historical dataset through accumulated similarity factors, using them as training datasets to construct a local model based on real-time learning. The module also outputs the predicted mechanical properties of hot-rolled strip steel based on feature quantization method A. Furthermore, it employs at least one other feature quantization method (feature quantization method B, C, D…) different from feature quantization method A, and outputs the predicted mechanical properties of hot-rolled strip steel based on that method.

[0043] The final output module of the ensemble learning: adopts the ensemble learning strategy to weight and integrate the prediction results of multiple local models obtained under different feature quantization methods, and outputs the prediction results of the mechanical properties of hot-rolled strip steel based on multi-dimensional feature weighted similarity and ensemble real-time learning as the final output of the model.

[0044] The beneficial effects of this invention are as follows:

[0045] This invention proposes a method and system for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and ensemble real-time learning. First, multiple feature quantification methods are used to obtain feature weights, and a time weight is introduced on top of these feature weights to construct a similarity metric that comprehensively considers input features and time factors to select suitable modeling samples, overcoming the limitations of modeling with a single similarity metric. Second, a cumulative similarity factor is introduced to ensure that the optimal modeling data sample set is selected for local modeling. Finally, an ensemble learning framework is used to integrate the output results of each local model, reducing the dependence of the prediction results on a single model, thereby significantly enhancing model performance and improving the accuracy and generalization of hot-rolled strip steel mechanical property prediction under complex production conditions. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of a framework for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature weighting and integrated real-time learning, provided by an embodiment of the present invention.

[0048] Figure 2 This is a schematic diagram of dataset partitioning during a simulated hot rolling process provided in one embodiment of the present invention.

[0049] Figure 3 This is a schematic diagram of the MDA principle provided in one embodiment of the present invention.

[0050] Figure 4 This is a feature importance graph under different feature analysis methods provided in one embodiment of the present invention.

[0051] Figure 5 This is a local model prediction error map provided by an embodiment of the present invention under different cumulative similarity factors.

[0052] Figure 6 This is a comparison chart of prediction results from different models provided in one embodiment of the present invention.

[0053] Figure 7 This is a scatter plot comparison of prediction results from different models provided in one embodiment of the present invention.

[0054] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0055] The technical solutions described below in conjunction with the embodiments will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the 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.

[0056] This invention proposes a method and system for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning. The technical solution of this invention will be further explained below with reference to specific embodiments.

[0057] Example

[0058] Figure 1 This is a schematic diagram of the framework for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature weighting and integrated real-time learning in this embodiment, including the following steps:

[0059] S1. Collect and preprocess the hot-rolled strip steel production dataset, using the hot-rolling production process parameters X[X1, X2, …, X…]. n ] is the independent variable, and the mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel are used as the independent variable. n Using [] as the dependent variable, and taking the production data of each coil of hot-rolled strip steel as one data set, a historical dataset for the prediction model is constructed.

[0060] Hot rolling production process parameters X[X1, X2, …, X n This includes the content of chemical elements (C, Si, Mn, P, S, Cu, Ni, Cr, V, Al, N, Nb, Ti), furnace exit temperature (TT), roughing rolling start temperature (RRST), roughing rolling start thickness (TRR), roughing rolling finish thickness (FTRR), finishing rolling start temperature (FRST), finishing rolling finish temperature (FDT), finishing rolling finish thickness (THK), coiling temperature (CT), and the mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel. n This includes tensile strength (TS), yield strength (YS), and elongation (EL); the data preprocessing method considers measurement data that meets one of the following three conditions as outlier data: (1) the measurement value is outside the predetermined range; (2) the data is missing; (3) the 3σ principle is not met; in order to more accurately simulate and reflect the actual production process, a dataset partitioning method is adopted, such as Figure 2 As shown, the original data is divided into three parts according to the production time series: (1) Historical dataset: consisting of the earliest production data; (2) Intermediate dataset: lagging samples caused by objective factors in the actual production process; (3) The remaining part is the test sample set.

[0061] After data preprocessing, 10,000 samples were retained. The statistical distribution of each parameter in the dataset is shown in Table 1.

[0062] Table 1 Statistical distribution of parameters in the sample

[0063]

[0064] The samples were sorted according to production time sequence and divided into historical datasets, intermediate datasets, and test sets, with 5970, 30, and 4000 samples respectively. To eliminate the influence of differences in scale, the Min-Max normalization method was used to normalize the original data.

[0065]

[0066] in: and Let represent the maximum and minimum values ​​of the d-th feature, respectively. Let be the original value of the d-th feature of the i-th sample. This represents the normalized value of the d-th feature of the i-th sample.

[0067] S2. For the historical dataset, Pearson correlation analysis (PCC) is used to calculate the feature weights of the influence of hot rolling production process parameters on the mechanical properties of hot-rolled strip steel; at the same time, each data group is sorted according to the corresponding hot-rolled strip steel production time and assigned a time weight.

[0068] S3. Based on feature weights and time weights, calculate the feature-time multidimensional similarity measure between the test sample and the samples in the historical dataset. Adaptively select multiple samples with high similarity to the test sample from the historical dataset as training datasets to build a local model based on instant learning, and output the prediction results of the mechanical properties of hot-rolled strip steel based on Pearson correlation analysis (PCC).

[0069] S4. Spearman correlation analysis (SCC), mutual information correlation analysis (MICC), and mean decreasing precision analysis (MDA) are used respectively, and their schematic diagrams are shown below. Figure 3 The feature quantization method (as shown) is used to repeat steps S2-S3.

[0070] Given that the choice of fixed data length may be limited by the model's adaptability to different industrial environments, this invention employs an adaptive method based on a cumulative similarity factor to dynamically determine the number of similar samples, making the selected samples more reasonable; the formula for calculating the cumulative similarity factor L is:

[0071]

[0072] Where: the denominator represents the sum of similarities between the test sample and all historical samples, and the numerator represents the terms with the highest similarity. n The sum of the similarity contributions of the group samples.

[0073] To evaluate model performance, root mean square error (RMSE) was used. RMSE Mean absolute error ( MAE ) and mean absolute percentage error ( MAPE Its expression is as follows:

[0074]

[0075]

[0076]

[0077] in: m For the number of test samples, For the first i The true value of each sample for i The predicted value for each sample.

[0078] Figure 4Taking tensile strength (TS) as the output feature, this paper ranks the absolute values ​​of the feature weights assigned by feature quantification methods such as Pearson correlation analysis (PPC), Spearman correlation analysis (SCC), mutual information correlation analysis (MICC), and mean decreasing precision (MDA), and only displays the values ​​of the first six features; from Figure 4 It can be seen that different feature quantification methods assign different weight values ​​to the same feature.

[0079] Figure 5 Showing different L The predictive performance of each local model for different mechanical properties at different values ​​( Figure 5 In the middle (a), tensile strength is represented. Figure 5 In the middle (b), the yield strength is represented. Figure 5 (c) represents the elongation rate); from Figure 5 It can be seen from this that MAE along with L The change shows a trend of first decreasing and then increasing; specifically, when L When the value is small, due to insufficient training samples, the model is susceptible to overfitting, resulting in a large prediction error; as the value increases... L With the increase of [a certain number of training samples], the model can utilize more training samples, thereby improving prediction accuracy; however, when [the number of training samples] increases, [the accuracy of predictions may decrease]. L As the value continues to increase, more irrelevant samples are introduced, making the nonlinear relationships in the dataset more complex, which in turn affects the predictive performance of the linear model.

[0080] S5. An ensemble learning strategy is adopted to weight and integrate the prediction results of multiple local models, and output the prediction results of the mechanical properties of hot-rolled strip steel based on multi-dimensional feature weighted similarity and ensemble real-time learning as the final output of the model.

[0081] To verify the superiority of the Multidimensional Feature Weighted and Ensemble Instant Learning Model (MDFWS-IJITL), MDFWS-IJITL was compared with JITL models constructed with different feature weighted similarity measures (MDA-JITL, MICC-JITL, PCC-JITL, and SCC-JITL) and the Global Linear Regression Model (G-LR). The prediction results of each model for different mechanical properties are as follows: Figure 6 As shown ( Figure 6 In the middle (a), the root mean square error is represented by the mean square error. Figure 6 In the middle (b), the mean absolute error is represented. Figure 6 (c)DB mean absolute percentage error). From Figure 6As can be seen, G-LR has the highest prediction errors for all three mechanical properties. This is because G-LR only performs modeling once and uses the established model for an extended period. This modeling method ignores the time-varying characteristics of hot-rolled strip steel production and cannot better reflect the dynamic changes in industrial data, thus leading to its large prediction errors. Compared with G-LR, the JITL model, which introduces feature weights and time weights, shows a reduction in error. This is because JITL selects relevant samples for remodeling for each prediction sample, resulting in a more targeted model. Furthermore, the constructed JITL model considers feature weights and time weights, further improving the correlation between the selected modeling samples and the test samples.

[0082] For predicting different mechanical properties, various local models exhibited varying predictive performance. For example, MDA-JITL showed the lowest MAE and RMSE in tensile strength prediction compared to other JITL models, demonstrating the best prediction effect. Conversely, PCC-JITL performed best in elongation prediction compared to other JITL models. This is because different JITL models assigned different weights to different features when selecting relevant modeling samples. The magnitude of the feature weights reflects the correlation between input and output features. If the assigned feature weights are closer to the true correlation between input and output, the selection of relevant samples during model construction will be more accurate, thus ensuring the effectiveness of the constructed model. While the introduction of feature weights can improve the accuracy of relevant samples, a single JITL model can only consider one aspect of correlation, while the input and output features of hot-rolled strip steel samples often have multiple similarity dimensions. Figure 6 As can be seen, the MDFWS-IJITL model outperforms all JITL models in predicting different mechanical properties. This is because, as mentioned earlier, each JITL model only analyzes the correlation between input and output data from one side, making it difficult to consider all the information from the original data in its modeling. MDFWS-IJITL analyzes the correlation between input and output data in multiple ways and integrates them into a single framework. This sample selection method based on multi-dimensional feature-weighted similarity reflects a more realistic correlation between samples, improving the accuracy of relevant sample selection. Furthermore, the ensemble strategy automatically assigns appropriate weights to each JITL model, improving the robustness of the regression model. These characteristics make the MDFWS-IJITL method superior to other methods.

[0083] Figure 7This section compares the prediction accuracy of different prediction models for various mechanical properties. (a-1)~(a-6), (b-1)~(b-6), and (c-1)~(c-6) show the detailed comparisons between the predicted and actual values ​​of tensile strength (TS), yield strength (YS), and elongation (EL) under different models. The black lines in (a-1)~(a-6) represent the 5% relative error boundary, while the black lines in (b-1)~(b-6) and (c-1)~(c-6) represent the 10% relative error boundary. Figure 7 It can be seen that G-LR, due to its one-time modeling, cannot achieve high accuracy. In contrast, the accuracy of each JITL model is improved compared to global modeling due to the introduction of both feature and temporal weights. Compared with other JITL models, the MDFWS-IJITL model improves the accuracy of relevant sample selection by introducing multi-dimensional feature weights and temporal weights. This results in better fitting performance in different mechanical property predictions, with the prediction results showing stronger clustering on the diagonal and generally falling within the defined error region. A comparison of the scatter plots of the prediction results from each model further demonstrates the superiority of MDFWS-IJITL in predicting the mechanical properties of tropical steel.

[0084] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning, characterized in that, Includes the following steps: S1. Collect and preprocess the hot-rolled strip steel production dataset, using the hot-rolling production process parameters X[X1, X2, …, X…]. n ] is the independent variable, and the mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel are used as the independent variable. n Using [] as the dependent variable, and taking the production data of each coil of hot-rolled strip steel as one data set, a historical dataset for the prediction model is constructed. S2. For the historical dataset, use feature quantization method A to calculate the hot rolling production process parameters X[X1, X2, …, X…]. n The mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel n The feature weights that influence the data are determined; at the same time, each data group is sorted by the corresponding hot-rolled strip steel production time and assigned a time weight. S3. Based on feature weights and time weights, calculate the feature-time multidimensional similarity measure between the test sample and the samples in the historical dataset. Select multiple samples with high similarity to the test sample from the historical dataset by accumulating similarity factors, and use them as training datasets to build a local model based on instant learning. Output the prediction results of the mechanical properties of hot-rolled strip steel based on feature quantization method A. S4. Use at least one other feature quantization method different from feature quantization method A, and repeat steps S2-S3; S5. An ensemble learning strategy is adopted to weight and integrate the prediction results of multiple local models obtained under different feature quantization methods, and output the prediction results of the mechanical properties of hot-rolled strip steel based on multi-dimensional feature weighted similarity and ensemble real-time learning as the final output of the model.

2. The method for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning according to claim 1, characterized in that, In step S1, data that meets one of the following three conditions during the dataset preprocessing is considered abnormal data: (1) The measured value is outside the predetermined range; (2) Data is missing; (3) The 3σ principle is not met.

3. The method for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning according to claim 1, characterized in that, In step S1, the hot rolling production process parameters X[X1, X2, …, X…] n This includes chemical element content, furnace exit temperature, roughing rolling start temperature, roughing rolling start thickness, roughing rolling finish thickness, finishing rolling start temperature, finishing rolling finish temperature, finishing rolling finish thickness, and coiling temperature; mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel. n This includes tensile strength, yield strength, and elongation.

4. The method for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning according to claim 1, characterized in that, In step S2, the feature quantification methods include Pearson correlation analysis, Spearman correlation analysis, mutual information correlation analysis, and average decreasing precision.

5. The method for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning according to claim 1, characterized in that, In step S3, the formula for calculating the feature-time multidimensional similarity measure between the sample to be tested and the samples in the historical dataset is as follows: in: This represents the feature-time multidimensional similarity measure between the sample to be tested and historical samples. For the first j A historical sample, The sample to be tested. This is a measure of the similarity between the feature weights of the sample to be tested and historical samples. For historical samples, time weights This is an adjustable parameter.

6. The method for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning according to claim 5, characterized in that, In step S3, the cumulative similarity factor L The calculation formula is: Wherein: the numerator represents the sum of the similarity contributions of the top n groups of samples with high similarity, and the denominator represents the sum of the similarity between the test sample and all historical samples.

7. The method for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning according to claim 1, characterized in that, In step S3, the local model construction method based on real-time learning includes multiple linear regression, partial least squares regression, and support vector machine.

8. The method for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning according to claim 1, characterized in that, In step S5, the weighted integration of predictions from multiple local models depends on the performance of each local model on the validation dataset. r The mean absolute error of each model ( MAE The calculation formula is: in: and The first i Observed and predicted values ​​for each test sample. m This represents the number of samples in the test set.

9. The method for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning according to claim 8, characterized in that, In step S5, through each model MAE Definition of the first r The weights of the output results of each sub-model are: Where: the numerator term represents the first... r Sub-models e For the base, MAE The function value is the exponent, and the denominator represents all sub-models. e For the base, MAE It is the sum of the function values ​​of the exponent.

10. A system for predicting the mechanical properties of hot-rolled strip steel based on multidimensional feature similarity and integrated real-time learning, characterized in that, include: Production Dataset Acquisition and Preprocessing Module: Acquires and preprocesses hot-rolled strip steel production data, using hot-rolling production process parameters X[X1, X2, …, X…]. n ] is the independent variable, and the mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel are used as the independent variable. n Using [] as the dependent variable, and taking the production data of each coil of hot-rolled strip steel as one data set, a historical dataset for the prediction model is constructed. Multidimensional feature similarity calculation and local model construction module: For historical datasets, feature quantization method A is used to calculate hot rolling production process parameters X[X1, X2, …, X…]. n The mechanical properties Y[Y1, Y2, …, Y] of hot-rolled strip steel n The feature weights that influence the data are determined; at the same time, each data group is sorted by the corresponding hot-rolled strip steel production time and assigned a time weight. Based on feature weights and time weights, the feature-time multidimensional similarity measure between the test sample and samples in the historical dataset is calculated. By accumulating similarity factors, multiple samples with high similarity to the test sample are selected from the historical dataset as training datasets to construct a local model based on instant learning, and the prediction results of the mechanical properties of hot-rolled strip steel based on feature quantization method A are output. At least one other feature quantization method different from feature quantization method A is used to output the prediction results of the mechanical properties of hot-rolled strip steel based on it. The final output module of the ensemble learning: adopts the ensemble learning strategy to weight and integrate the prediction results of multiple local models obtained under different feature quantization methods, and outputs the prediction results of the mechanical properties of hot-rolled strip steel based on multi-dimensional feature weighted similarity and ensemble real-time learning as the final output of the model.

Images