Steel quality multi-index joint prediction method and device under small sample and medium

Abstract

The application discloses a small-sample steel quality multi-index joint prediction method and device and a medium, relates to the field of material performance prediction, and comprises the following steps: determining a standardized data set comprising steel chemical composition, process parameter characteristics and multi-dimensional mechanical performance index data; configuring a target order of the multi-dimensional mechanical performance index; constructing a regression chain model based on the target order; in the regression chain model, the prediction value of a previous mechanical performance index is used as an additional input feature of a corresponding regressor of a subsequent mechanical performance index; adopting a nested cross-validation strategy to perform hyperparameter search and optimization on the regression chain model, and determining a target hyperparameter combination; performing full-amount retraining on the regression chain model based on the target hyperparameter combination, and obtaining a trained steel performance joint prediction model; and inputting the chemical composition and production process parameters of the steel to be predicted into the steel performance joint prediction model, and obtaining a joint prediction result of each steel performance index corresponding to the steel to be predicted.

Description

Technical Field

[0001] This application relates to the field of material performance prediction, and in particular to a method, device and medium for joint prediction of multiple indicators of steel quality under small sample conditions. Background Technology

[0002] In the research and development and production of steel materials, accurate prediction of the mechanical properties of steel is a core link in achieving process optimization, cost control, and product quality improvement. The formation of these properties follows a strong coupling mechanism of "composition-process-structure-performance," requiring comprehensive prediction of multiple mechanical indicators such as yield strength, tensile strength, and elongation based on parameters such as chemical composition and heat treatment processes. With the implementation of intelligent manufacturing in the metallurgical industry, data-driven performance prediction methods, due to their efficiency and flexibility, are gradually becoming an important means to replace traditional experimental trial-and-error methods. However, in actual production, limitations such as testing costs, production rhythm, and safety compliance result in a limited sample size for multiple indicators. Furthermore, issues such as strong correlations among multiple performance indicators, large dimensional differences, and long-tailed data distribution pose serious challenges to the joint prediction of multiple steel properties under small sample conditions.

[0003] Existing technologies for predicting steel performance mainly include three implementation schemes: physical / empirical models, single-objective data-driven models, and multi-output regression models. Among them, physical / empirical models rely on a large number of experimentally calibrated parameters and have limited support for joint modeling of multiple indicators. Single-objective data-driven models construct regression models for each performance indicator independently, which easily ignores the conditional dependencies between indicators, leading to inconsistent prediction results. Although multi-output regression models can utilize the correlation between objectives to some extent, they mine the correlation between indicators through implicit shared representations and cannot directly reflect the conditional influence of previous objectives on subsequent objectives. Furthermore, it is difficult to set the multi-task loss weights under small sample conditions, which can easily lead to some objectives dominating the training. In addition, although regression chains, as a structured modeling strategy for multi-output prediction, can explicitly construct the conditional dependency links between objectives, when directly applied to steel performance prediction, they still do not solve the problems of large evaluation variance under small sample conditions, easy occurrence of accidental optimality in hyperparameter search, and lack of engineering specifications for chain order configuration, making it difficult to meet the requirements of robustness and reproducibility of prediction models in industrial scenarios. Summary of the Invention

[0004] In view of this, this application provides a method, device and medium for joint prediction of multiple indicators of steel quality under small sample conditions, which can solve the technical problems of low prediction accuracy and unstable model evaluation results caused by strong correlation and significant differences in dimensions of multiple performance indicators under small sample conditions.

[0005] According to the first aspect of this application, a method for joint prediction of multiple indicators of steel quality under small sample conditions is provided, including:

[0006] A standardized dataset for steel materials is determined. The standardized dataset includes steel chemical composition, process parameter characteristics, and multi-dimensional mechanical property index data. The multi-dimensional mechanical property index includes at least two of the following: yield strength, tensile strength, elongation, reduction of area, and impact energy.

[0007] The target order of the multi-dimensional mechanical performance indicators is configured based on preset rules or traversal methods, and a regression chain model is constructed based on the target order. In the regression chain model, the predicted value of the preceding mechanical performance indicator is used as an additional input feature of the regressor corresponding to the subsequent mechanical performance indicator.

[0008] A nested cross-validation strategy adapted to small samples is used to search and optimize the hyperparameters of the regression chain model to determine the target hyperparameter combination;

[0009] Based on the target hyperparameter combination, the regression chain model is fully retrained using the standardized dataset to obtain a trained joint prediction model for steel performance.

[0010] The chemical composition and production process parameters of the steel to be predicted are input into the joint prediction model of steel performance to obtain the joint prediction results of each steel performance index corresponding to the steel to be predicted.

[0011] According to a second aspect of this application, a device for joint prediction of multiple indicators of steel quality under small sample conditions is provided, comprising:

[0012] The determination module is used to determine a standardized dataset for steel materials. The standardized dataset includes steel chemical composition, process parameter characteristics, and multi-dimensional mechanical property index data. The multi-dimensional mechanical property index includes at least two of the following: yield strength, tensile strength, elongation, reduction of area, and impact energy.

[0013] The construction module is used to configure the target order of the multi-dimensional mechanical performance indicators based on preset rules or traversal methods, and to construct a regression chain model based on the target order. In the regression chain model, the predicted value of the preceding mechanical performance indicator is used as an additional input feature of the regressor corresponding to the subsequent mechanical performance indicator.

[0014] The validation module is used to perform hyperparameter search and optimization on the regression chain model using a nested cross-validation strategy adapted to small samples, and to determine the target hyperparameter combination.

[0015] The training module is used to fully retrain the regression chain model based on the target hyperparameter combination and the standardized dataset to obtain the trained joint prediction model for steel performance.

[0016] The prediction module is used to input the chemical composition and production process parameters of the steel to be predicted into the joint prediction model of steel performance, and obtain the joint prediction results of each steel performance index corresponding to the steel to be predicted.

[0017] According to a third aspect of this application, a storage medium is provided on which a computer program is stored, which, when executed by a processor, implements the above-described method for joint prediction of multiple indicators of steel quality under small sample conditions.

[0018] According to a fourth aspect of this application, an electronic device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described method for joint prediction of multiple indicators of steel quality under small sample conditions.

[0019] By employing the above technical solutions, the method, apparatus, and medium for joint prediction of multiple steel quality indicators under small sample conditions provided in this application, through determining a standardized dataset containing steel chemical composition, process parameter characteristics, and multiple mechanical property indicators, configuring the target order of indicators by combining preset rules or traversal methods, and constructing a regression chain model in which the predicted values ​​of the preceding indicators are additional inputs to the subsequent regressor, can specifically solve the problems of insufficient joint modeling capability of physical / empirical models for multiple indicators, inconsistent prediction logic caused by single-objective data-driven models ignoring the conditional dependencies between indicators, and the problems of implicitly mining indicator correlations in multi-output regression models and easy training imbalance under small sample conditions; at the same time, it adopts a nested approach adapted to small samples. Cross-validation strategy is used to perform hyperparameter search and optimization. By combining the target hyperparameter combination, the regression chain model is fully retrained to obtain a joint prediction model and output the steel performance indicators to be predicted. This can effectively solve the pain points of large evaluation variance under small samples, easy occurrence of accidental optimality in hyperparameter search, and lack of engineering specifications for chain order configuration when the regression chain is directly applied to steel performance prediction. It can significantly improve the robustness and prediction accuracy of the steel performance prediction model. Moreover, through the engineering chain order configuration, hyperparameter optimization and model retraining process, the model can meet the core requirements of reproducibility and practicality in industrial scenarios, and achieve accurate joint prediction of multi-dimensional mechanical performance indicators of steel.

[0020] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0022] Figure 1 The diagram illustrates a flowchart of a method for jointly predicting multiple indicators of steel quality under a small sample size, as provided in an embodiment of this application.

[0023] Figure 2 A flowchart illustrating a method for joint prediction of multiple indicators of steel quality under small sample conditions, provided in another embodiment of this application, is shown.

[0024] Figure 3 The diagram shows a structural schematic of a multi-index joint prediction device for steel quality under a small sample size, provided in an embodiment of this application. Detailed Implementation

[0025] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0026] Existing technologies for predicting steel performance mainly include three implementation schemes: physical / empirical models, single-objective data-driven models, and multi-output regression models. Among them, physical / empirical models rely on a large number of experimentally calibrated parameters and have limited support for joint modeling of multiple indicators. Single-objective data-driven models construct regression models for each performance indicator independently, which easily ignores the conditional dependencies between indicators, leading to inconsistent prediction results. Although multi-output regression models can utilize the correlation between objectives to some extent, they mine the correlation between indicators through implicit shared representations and cannot directly reflect the conditional influence of previous objectives on subsequent objectives. Furthermore, it is difficult to set the multi-task loss weights under small sample conditions, which can easily lead to some objectives dominating the training. In addition, although regression chains, as a structured modeling strategy for multi-output prediction, can explicitly construct the conditional dependency links between objectives, when directly applied to steel performance prediction, they still do not solve the problems of large evaluation variance under small sample conditions, easy occurrence of accidental optimality in hyperparameter search, and lack of engineering specifications for chain order configuration, making it difficult to meet the requirements of robustness and reproducibility of prediction models in industrial scenarios.

[0027] To address the aforementioned technical problems, embodiments of the present invention provide a method for joint prediction of multiple indicators of steel quality under small sample conditions, such as... Figure 1 As shown, the method includes:

[0028] Step 110: Determine the standardized dataset for steel materials. The standardized dataset includes steel chemical composition, process parameter characteristics, and multi-dimensional mechanical property index data. The multi-dimensional mechanical property index includes at least two of the following: yield strength, tensile strength, elongation, reduction of area, and impact energy.

[0029] The standardized dataset is obtained by standardizing and handling missing and outlier values ​​in the original data related to steel materials; the chemical composition of steel refers to the proportion of various chemical elements that constitute steel materials, which is the core characteristic that determines the basic properties of steel materials; the process parameter characteristics are various process control parameters in the production and processing of steel materials, which are key processing characteristics that affect the formation of steel material properties; the multi-dimensional mechanical performance indicators are multiple core indicators for measuring the mechanical properties of steel materials. In this scheme, at least two of the following can be included: yield strength, tensile strength, elongation, reduction of area, and impact energy, which are the predictive indicators for steel performance. The core objectives of the test are as follows: Yield strength is the critical stress at which steel undergoes plastic deformation, reflecting the material's ability to resist plastic deformation; Tensile strength is the maximum stress that steel can withstand before fracture, reflecting the material's ultimate tensile strength; Elongation is the ratio of the elongation of the gauge length after fracture to the original gauge length, reflecting the material's plastic deformation capacity; Reduction of area is the ratio of the reduction in cross-sectional area at the necking point after fracture to the original cross-sectional area, reflecting the degree of plastic deformation; Impact energy is the ability of steel to absorb deformation and fracture energy under impact load, reflecting the material's impact resistance.

[0030] In this embodiment of the disclosure, when determining the standardized dataset for steel materials, raw data covering steel chemical composition, process parameter characteristics, and corresponding multi-dimensional mechanical performance indicators can be obtained first. Then, the raw data is preprocessed using appropriate data processing methods to effectively handle missing or abnormal parts of the data. At the same time, standardization operations are performed on the processed data to eliminate modeling biases caused by differences in dimensions and numerical scales of different features and indicators. Finally, a standardized dataset containing steel chemical composition, process parameter characteristics, and multi-dimensional mechanical performance indicators is formed, laying a unified, standardized, and high-quality data foundation for the subsequent joint prediction of steel performance models.

[0031] Establishing a standardized dataset that includes steel chemical composition, process parameter characteristics, and multi-dimensional mechanical performance indicators can eliminate the adverse effects of dimensional differences and data noise on subsequent modeling from the data source. This provides a unified, high-quality foundation of data for the construction and optimization of regression chain models, effectively avoiding model search and evaluation bias problems caused by inconsistent data scales. Furthermore, it fully incorporates the core influencing features and core prediction objectives of steel performance formation, allowing subsequent models to fully learn the intrinsic relationships between chemical composition, process parameters, and multi-dimensional mechanical performance indicators. This lays a solid data foundation for accurate joint prediction of multiple steel performance indicators and provides data assurance for robust training and optimization of models in small-sample scenarios, thereby improving the scientific rigor and reliability of the overall prediction method.

[0032] Step 120: Configure the target order of multi-dimensional mechanical performance indicators based on preset rules or traversal methods, and construct a regression chain model based on the target order. In the regression chain model, the predicted value of the preceding mechanical performance indicator is used as an additional input feature of the regressor corresponding to the subsequent mechanical performance indicator.

[0033] Among them, the preset rules are configuration rules formulated by combining the industrial mechanism, engineering priors, or statistical exploration results of steel performance, and are the basis for determining the ranking of mechanical performance indicators; the traversal method is a configuration method that verifies all permutations and combinations of multi-dimensional mechanical performance indicators one by one, which can uncover the index ranking that fits the model; the target order is the order in which multi-dimensional mechanical performance indicators are predicted in the regression chain model, and is the core basis for constructing the regression chain model; the regression chain model is a structured prediction model that decomposes multi-objective prediction into ordered links, with each node in the link corresponding to the regression prediction of a single mechanical performance indicator, and is the core model for realizing the joint prediction of multiple steel properties; the regressor is the learner in each node of the regression chain model used to realize the prediction of a single mechanical performance indicator, and is the basic unit constituting the regression chain model; the additional input features are the input features introduced into the regressor of the subsequent nodes of the regression chain model, in addition to the characteristics of steel chemical composition and process parameters, and in this scheme, they are the predicted values ​​of the preceding mechanical performance indicators.

[0034] In this embodiment of the disclosure, for multi-dimensional mechanical performance indicators of steel, the target order of each mechanical performance indicator in the model can be determined first based on the relevant mechanism of steel performance formation or engineering experience, or by traversing through all permutations and combinations of indicators one by one. Then, a structured regression chain model is built according to the target order, and a corresponding regressor is matched for each mechanical performance indicator in the model. At the same time, the information transmission logic of the model is set, and the mechanical performance indicator results that have been predicted in the preceding node of the regression chain model are used as the input features of the regressor of the mechanical performance indicator corresponding to the following node. Together with the characteristics of steel chemical composition and process parameters, they participate in the indicator prediction of the following node, thereby constructing a regression chain model that can realize the correlation prediction of multiple indicators.

[0035] By configuring the target order of multi-dimensional mechanical performance indicators through preset rules or traversal methods, the construction of the regression chain model has an engineering-standard basis, which can fully explore the inherent correlation between indicators. The regression chain model, which uses the predicted values ​​of the preceding indicators as additional input features of the subsequent regressor based on the target order, can explicitly model the conditional dependencies between multi-dimensional mechanical performance indicators of steel, effectively utilize the relevant information between indicators, and solve the problems of ignoring the correlation between indicators in single-objective modeling and implicitly mining the correlation in multi-output regression. This makes the model prediction results more consistent with the actual logic of steel performance formation. At the same time, the structured regression chain model structure can flexibly configure the node regressor and indicator order, improve the model's adaptability to steel performance prediction scenarios, and lay a reasonable model structure foundation for subsequent model optimization and accurate prediction in small sample scenarios.

[0036] Step 130: Use a nested cross-validation strategy adapted to small samples to search and optimize the hyperparameters of the regression chain model, and determine the target hyperparameter combination.

[0037] Among them, the nested cross-validation strategy adapted to small samples is a two-layer cross-validation method designed for the characteristics of small sample data. It improves the stability of model evaluation through multi-layer data partitioning and validation, and adapts to the model hyperparameter optimization needs under small sample conditions. Hyperparameter search and optimization is the process of finding hyperparameter combinations that fit the model and data within a set hyperparameter space through specific sampling and validation methods to improve the model's prediction performance. The target hyperparameter combination is the hyperparameter combination that fits the regression chain model and can achieve the best prediction effect after hyperparameter search and optimization. Hyperparameters are parameters that need to be set in advance before training the regression chain model. They cannot be obtained through data training and are key parameters that affect the model training effect and prediction performance.

[0038] In this embodiment of the disclosure, a customized nested cross-validation strategy can be adopted for the constructed regression chain model, taking into account the characteristics of small sample data. The standardized dataset is divided and validated in multiple levels, and a validation framework that takes into account both the evaluation of model generalization ability and the selection of hyperparameters is established. Under this framework, a reasonable hyperparameter search space is defined based on data characteristics and model structure. Hyperparameter search is carried out in the space through scientific sampling methods. At the same time, the model performance under different hyperparameter configurations is verified and selected using evaluation criteria that are suitable for multi-index prediction. This completes the hyperparameter optimization of the regression chain model and finally determines the target hyperparameter combination that is suitable for small sample scenarios and allows the regression chain model to achieve the best prediction performance.

[0039] By employing a nested cross-validation strategy adapted to small samples to search and optimize hyperparameters in the regression chain model and determine the target hyperparameter combination, this approach effectively addresses the issues of large model evaluation variance and the tendency for accidental optima to occur during hyperparameter search in small sample scenarios. Multi-level cross-validation makes the hyperparameter selection process more scientific and robust. The selected target hyperparameter combination ensures that the regression chain model's performance is highly adapted to small sample data, significantly improving the model's generalization ability and prediction accuracy in small sample scenarios. Simultaneously, the nested cross-validation logic makes the hyperparameter optimization process more standardized, avoiding evaluation bias caused by single data partitioning. This provides optimal parameter support for subsequent full retraining of the regression chain model and accurate prediction of steel performance, ensuring the model's robustness in small sample scenarios from a parameter perspective.

[0040] Step 140: Based on the target hyperparameter combination, the regression chain model is fully retrained using a standardized dataset to obtain the trained joint prediction model for steel performance.

[0041] Among them, full retraining refers to the complete training process of the model using the complete standardized dataset, rather than the partitioned partial dataset, which can make full use of all data information to complete the model training; the steel performance joint prediction model is a regression chain model that can accurately predict multi-dimensional mechanical performance indicators based on the characteristics of steel chemical composition and process parameters after full training, and is the model carrier for steel performance prediction; training completion refers to the state in which the internal parameters of the model reach a stable state after training with full data, and can output reliable mechanical performance indicator prediction results based on input features.

[0042] In this embodiment of the disclosure, the target hyperparameter combination obtained through hyperparameter search and optimization can be used to configure the parameters of the regression chain model. A complete standardized dataset containing steel chemical composition, process parameter characteristics, and multi-dimensional mechanical performance indicators is used as training data and input into the regression chain model with completed parameter configuration to carry out full training. This allows the model to fully learn the inherent correlation between the core characteristics of steel and the multi-dimensional mechanical performance indicators, complete the iterative optimization and convergence of the model's internal parameters, and finally obtain a steel performance joint prediction model with stable parameters that can achieve joint prediction of multi-dimensional mechanical performance indicators of steel.

[0043] A joint prediction model for steel performance is obtained by combining target hyperparameters and fully retraining the regression chain model using a standardized dataset. This allows the model to fully learn the correlation information between features and performance indicators in all data under optimal parameter configuration, maximizing the discovery of effective patterns in the data. This solves the problems of insufficient training data and inadequate parameter optimization in small sample scenarios, effectively improving the model's fitting ability and prediction accuracy. At the same time, the full retraining process adapts the model parameters to complete steel data features, ensuring the stability and reliability of the model in actual prediction scenarios. The final joint prediction model for steel performance can achieve coordinated and accurate prediction of multi-dimensional mechanical performance indicators, meeting the actual needs of joint prediction of steel performance in industrial scenarios.

[0044] Step 150: Input the chemical composition and production process parameters of the steel to be predicted into the joint prediction model of steel performance to obtain the joint prediction results of each steel performance index corresponding to the steel to be predicted.

[0045] Among them, the steel to be predicted is a steel material whose mechanical property index testing has not yet been completed and whose relevant properties need to be predicted by the model; the production process parameters are the various process control parameters actually used in the production and processing of the steel to be predicted, which is one of the core input features for predicting its performance; the steel performance joint prediction model is a model that can simultaneously predict multiple mechanical property indexes based on the steel chemical composition and process parameters after hyperparameter optimization and full retraining; the joint prediction result is the set of predicted values ​​of multiple mechanical property indexes output by the model for the steel to be predicted, which can fully reflect the comprehensive mechanical properties of the steel material.

[0046] In this embodiment of the present disclosure, data related to the chemical composition and production process parameters of the steel to be predicted can be obtained, and after being standardized according to the same processing standard as the model training data, it is input into the steel performance joint prediction model that has been trained. The model relies on the inherent correlation between the steel chemical composition, production process parameters and multi-dimensional mechanical performance indicators learned during the training process to complete the prediction calculation of each mechanical performance indicator in sequence, and finally outputs the predicted values ​​of all mechanical performance indicators corresponding to the steel to be predicted, forming a complete steel performance joint prediction result.

[0047] By inputting the chemical composition and production process parameters of the steel to be predicted into the joint prediction model for steel performance, joint prediction results can be obtained. Based on the correlation between the features and performance indicators learned by the model, rapid and accurate joint prediction of multi-dimensional mechanical performance indicators of steel can be achieved. There is no need to obtain performance data through a large number of experiments, which can significantly reduce the time and cost of steel performance testing. At the same time, the model is based on standardized feature inputs and structured regression chain design, which can ensure the logic and consistency of the joint prediction results and fully reflect the comprehensive mechanical properties of the steel to be predicted. This provides an efficient and reliable performance reference for the optimization of steel production processes and quality control, and meets the actual needs of rapid prediction of steel performance in industrial scenarios.

[0048] In summary, the method for joint prediction of multiple steel quality indicators under small sample conditions provided in this application, by determining a standardized dataset containing steel chemical composition, process parameter characteristics, and multiple mechanical property indicators, and configuring the target order of indicators according to preset rules or traversal methods, and constructing a regression chain model in which the predicted values ​​of the preceding indicators are added as additional inputs to the subsequent regressor, can specifically solve the problems of insufficient joint modeling capability of physical / empirical models for multiple indicators, inconsistent prediction logic caused by single-objective data-driven models ignoring the conditional dependencies between indicators, and the problems of implicitly mining indicator correlations in multi-output regression models and easy training imbalance under small sample conditions. At the same time, a nested cross-validation strategy adapted to small samples is adopted. By conducting hyperparameter search and optimization, and combining the target hyperparameter combination to fully retrain the regression chain model to obtain a joint prediction model and output the steel performance indicators to be predicted, this approach can effectively solve the pain points of large evaluation variance under small samples, easy occurrence of accidental optimality in hyperparameter search, and lack of engineering specifications for chain order configuration when the regression chain is directly applied to steel performance prediction. This approach can significantly improve the robustness and prediction accuracy of the steel performance prediction model. Furthermore, through an engineering-oriented chain order configuration, hyperparameter optimization, and model retraining process, the model can meet the core requirements of reproducibility and practicality in industrial scenarios, achieving accurate joint prediction of multi-dimensional mechanical performance indicators of steel.

[0049] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, and to fully illustrate the implementation of this embodiment, this embodiment also provides another method for joint prediction of multiple indicators of steel quality under small sample conditions, such as... Figure 2 As shown, the method includes:

[0050] Step 210: Determine the standardized dataset for steel materials. The standardized dataset includes steel chemical composition, process parameter characteristics, and multi-dimensional mechanical property index data. The multi-dimensional mechanical property index includes at least two of the following: yield strength, tensile strength, elongation, reduction of area, and impact energy.

[0051] In this embodiment of the disclosure, raw data containing steel chemical composition, process parameters and corresponding multi-dimensional mechanical performance indicators can be obtained first to form a raw dataset of steel materials. Then, a standardization preprocessing operation is performed on the raw dataset. First, outliers in the data are detected and processed, and missing parts in the data are filled. Then, the differences caused by different dimensions and numerical scales of different features and indicators are eliminated through standardization processing. Finally, a standardized dataset containing steel chemical composition, process parameter features and at least two multi-dimensional mechanical performance indicators is obtained.

[0052] Accordingly, the implementation steps may include: obtaining the original dataset of steel materials, which contains the original data of the chemical composition, process parameters and corresponding multi-dimensional mechanical performance indicators of steel materials; performing standardization preprocessing on the original dataset to obtain a standardized dataset, wherein the standardization preprocessing includes at least missing value imputation, outlier detection and standardization processing.

[0053] Step 220: Configure the target order of multi-dimensional mechanical performance indicators based on preset rules or traversal methods, and construct a regression chain model based on the target order. In the regression chain model, the predicted value of the preceding mechanical performance indicator is used as the additional input feature of the regressor corresponding to the subsequent mechanical performance indicator.

[0054] In the embodiments of this disclosure, when configuring the target order of multi-dimensional mechanical performance indicators based on preset rules or traversal methods, one possible implementation is to analyze and determine the causal dependencies between various mechanical performance indicators based on the industrial mechanism of steel performance formation, and then directly configure a reasonable target order based on this relationship. Another possible implementation is to comprehensively traverse all permutations and combinations of multi-dimensional mechanical performance indicators to generate all feasible target orders that conform to the formation law of steel performance, thereby providing a scientific and suitable indicator ranking basis for the subsequent construction of the regression chain model. Yet another possible implementation is to combine the order determined by the above two methods to comprehensively determine the final target order. Accordingly, the implementation steps may include: determining the causal dependencies of multi-dimensional mechanical performance indicators based on the industrial mechanism of steel performance formation, configuring the target order based on the causal dependencies; and / or, traversing all permutations and combinations of multi-dimensional mechanical performance indicators to generate all feasible target orders.

[0055] By analyzing industrial mechanisms to determine the causal dependencies between indicators and configuring the target order, the indicator order aligns with the objective laws governing the formation of steel properties, ensuring the scientific validity and rationality of the target order. Furthermore, by traversing all permutations and combinations of indicators to generate all feasible target orders, the optimal indicator order suitable for subsequent modeling can be fully explored. Combining these two methods ensures that the configuration of the target order has both a theoretical basis for engineering practice and can optimize the order through comprehensive exploration. This lays the foundation for the subsequent explicit modeling of the conditional dependencies between indicators in the regression chain model, effectively avoiding the problem of model prediction logic deviation caused by unreasonable indicator order, and improving the scientific validity and adaptability of subsequent model construction.

[0056] In the embodiments of this disclosure, when constructing a regression chain model based on the target sequence, the target sequence of the configured multi-dimensional mechanical performance indicators can be mapped to the link index of the regression chain first. The specific mechanical performance indicators corresponding to each node in the regression chain are clarified through the link index. Then, an adapted single-target regressor is configured for each node of the regression chain. At the same time, the information transmission rules between each node of the regression chain are defined. Finally, according to the information transmission rules, the predicted value of the mechanical performance indicator corresponding to the preceding node is used as the additional input feature of the single-target regressor of the following node. This feature, together with the steel chemical composition and process parameter features, participates in the index prediction of the following node, thereby completing the overall construction of the regression chain model.

[0057] Accordingly, the implementation steps may include: mapping the target sequence to the link index of the regression chain, and determining the mechanical performance index corresponding to each node in the regression chain; configuring an independent single-objective regressor for each node in the regression chain, defining the information transmission rules of the regression chain, wherein the single-objective regressor is one or more of gradient boosting regressor, extreme gradient boosting tree, random forest, and support vector regression; using the predicted value of the mechanical performance index corresponding to the preceding node as the additional input feature of the single-objective regressor configured for the subsequent node, and combining the characteristics of steel chemical composition and process parameters to construct a regression chain model.

[0058] By mapping the target sequence to a link index, the correspondence between nodes and indicators is clearly defined, ensuring a high degree of fit between the model structure and the indicator order. This provides a reasonable model framework for multi-indicator correlation prediction. Configuring independent single-objective regressors for each node and flexibly selecting different types of regressors allows each indicator's prediction to match the optimal learner, improving the accuracy of single-indicator prediction. Furthermore, using the predicted values ​​of preceding nodes as additional input features for subsequent nodes, combined with core steel features, enables explicit modeling of the conditional dependencies between multi-dimensional mechanical performance indicators. This fully utilizes the correlation information between indicators, making the model's prediction results more consistent with the objective logic of steel performance formation. This effectively solves the problem of neglecting indicator correlation in single-objective modeling, significantly improving the overall accuracy and logical consistency of joint prediction of multi-dimensional mechanical performance indicators in steel.

[0059] Step 230: Divide the standardized dataset into multiple sets of outer training fold datasets and corresponding multiple sets of outer test fold datasets according to the repeated K-fold rule, and build a nested cross-validation framework of outer evaluation and inner hyperparameter search.

[0060] The repeated K-fold rule is a dataset partitioning rule that divides the dataset into K mutually exclusive subsets and repeatedly executes the partitioning and validation process. It is a dataset partitioning method adapted to small sample scenarios. The outer training fold dataset is a subset of the dataset used for model training and hyperparameter search after partitioning according to the repeated K-fold rule. It is the training data foundation for the outer stage of nested cross-validation. The outer test fold dataset is a subset of the dataset used to evaluate the model's generalization performance, corresponding to the outer training fold dataset after partitioning according to the repeated K-fold rule. The nested cross-validation framework is a two-layer cross-validation system that includes outer evaluation and inner hyperparameter search. The outer layer evaluates the model's generalization performance, while the inner layer searches for and selects model hyperparameters. The outer evaluation is the outer stage of the nested cross-validation framework. It verifies the model's performance using the outer test fold dataset, achieving an unbiased evaluation of the hyperparameter configuration effect. The inner hyperparameter search is the inner stage of the nested cross-validation framework. It searches, selects, and optimizes model hyperparameters based on the outer training fold dataset.

[0061] In this embodiment of the disclosure, the standardized dataset of steel materials can be divided multiple times according to the repeated K-fold rule. Each division generates multiple sets of corresponding outer training fold datasets and outer test fold datasets. Based on the multiple sets of outer training fold datasets and outer test fold datasets obtained from the multiple divisions, a two-layer nested cross-validation framework including outer evaluation and inner hyperparameter search is built. It is clear that the outer layer of the framework uses multiple sets of outer test fold datasets to evaluate the generalization performance of the model, and the inner layer uses multiple sets of outer training fold datasets to search for and optimize the model hyperparameters, forming a clear hierarchical validation system in which training and evaluation complement each other.

[0062] The standardized dataset is partitioned multiple times according to the repeated K-fold rule, and a nested cross-validation framework of outer evaluation and inner hyperparameter search is built. The repeated K-fold data partitioning makes full use of the dataset, effectively adapting to the problem of limited data volume in small sample scenarios. The two-layer architecture of outer evaluation and inner hyperparameter search can separate training and evaluation, avoiding evaluation bias caused by a single data partition, making the hyperparameter search process more scientific. At the same time, the outer layer completes performance evaluation with an independent test fold dataset, which can achieve unbiased verification of the effect of hyperparameter configuration, significantly reducing the variance of model evaluation in small sample scenarios, avoiding the problem of accidental optimality in hyperparameter search, and laying a solid validation foundation for the accurate selection and optimization of subsequent hyperparameters, thus improving the scientificity and reliability of the overall modeling process.

[0063] Step 240: Input the outer training dataset into the regression chain model, and define the hyperparameter search space that matches the steel features for the single-objective regressor of each node in the regression chain model based on the characteristics of steel chemical composition and process parameters.

[0064] Among them, the hyperparameter search space is the set of reasonable value ranges and combinations of values ​​set for the hyperparameters of the regressor. It is the basic boundary for carrying out hyperparameter search, and its delineation needs to match the data characteristics and model training requirements. Steel characteristics are the core characteristics of steel materials, specifically including the chemical composition and process parameter characteristics of steel. They are the key attributes that determine the mechanical properties of steel and are also the core input basis for modeling.

[0065] In this embodiment of the disclosure, the outer training dataset obtained by partitioning can be input into the constructed regression chain model. Taking the core input basis of model training, namely the characteristics of steel chemical composition and process parameters, as the core reference, and combined with the structural characteristics of the regression chain model, a hyperparameter search space that fits the steel material properties, adapts to the data characteristics and the prediction requirements of the regressor is defined for each independently configured single-objective regressor in the model, so that the hyperparameter search of each single-objective regressor has a clear and matching value boundary.

[0066] The outer training dataset is input into the regression chain model, and a matching hyperparameter search space is defined for each node's single-objective regressor based on the characteristics of steel chemical composition and process parameters. This ensures that the definition of the hyperparameter search space is no longer an unfounded generalization setting, but closely matches the core features of steel and the actual modeling needs of each node's regressor. This effectively reduces the invalid search range of hyperparameters, improves the efficiency and accuracy of subsequent hyperparameter searches, and ensures that the range of hyperparameter values ​​is highly adapted to the features of steel. This avoids the problem of poor hyperparameter adaptability caused by unreasonable search space settings, lays a reasonable boundary foundation for efficient hyperparameter search and optimization, and ensures that the hyperparameter selection results can better adapt to the modeling needs of steel performance prediction, thereby improving the prediction performance of each node in the regression chain model.

[0067] Step 250: In the inner cross-validation of the nested cross-validation framework, the hyperparameters are sampled and searched based on the hyperparameter search space, and multiple sets of hyperparameter configurations suitable for small samples are obtained by using dimensionless evaluation indicators as the screening criteria.

[0068] Among them, inner cross-validation is the internal validation system built on the outer training dataset in the nested cross-validation framework, and it is the core link in carrying out hyperparameter sampling search and selection; sampling search is the search process of selecting hyperparameter combinations and verifying their model fit within the defined hyperparameter search space through scientific sampling methods; dimensionless evaluation index is a model performance evaluation index that eliminates the influence of dimensions and numerical scales, and can realize a unified evaluation of the prediction effect of different types of mechanical performance indexes; hyperparameter configuration is the combination of hyperparameter values ​​selected for the single-objective regressors of each node of the regression chain model, and it is the key parameter setting that determines the model training effect and prediction performance.

[0069] In the embodiments of this disclosure, in the inner cross-validation stage of the nested cross-validation framework, the hyperparameters are systematically sampled and searched using the hyperparameter search space predefined for each node's single-objective regressor as the boundary. This generates multiple sets of different hyperparameter combinations, which are then applied to the regression chain model for training and validation. Simultaneously, a dimensionless evaluation index is used as a unified screening criterion for model performance to objectively evaluate and screen the model prediction effects under different hyperparameter configurations. Ultimately, multiple sets of hyperparameter configurations with good adaptability in small sample data scenarios are obtained.

[0070] In the inner cross-validation of the nested cross-validation framework, multiple sets of hyperparameter configurations suitable for small samples are obtained by sampling the hyperparameter search space and screening them with dimensionless evaluation indicators. This ensures that the hyperparameter search is always carried out within a reasonable boundary that fits the characteristics of steel, avoiding efficiency losses caused by invalid searches. At the same time, the dimensionless evaluation indicators can achieve a unified and fair evaluation of the prediction effect of multi-dimensional mechanical performance indicators, avoiding screening bias caused by differences in dimensions, and ensuring the objectivity and scientific nature of hyperparameter configuration screening. The multiple sets of hyperparameter configurations suitable for small samples obtained by screening enable the regression chain model to effectively avoid overfitting problems in small sample data scenarios, maintain good model performance, and provide a high-quality parameter foundation for further optimization of hyperparameter configurations and full model retraining, thereby improving the pertinence and effectiveness of model hyperparameter optimization in small sample scenarios.

[0071] Step 260: Input the outer test fold datasets of each group into the regression chain model trained with the corresponding hyperparameter configurations to complete the generalization performance evaluation of the hyperparameter configurations. Based on the evaluation results, sort the multiple hyperparameter configurations according to the model evaluation effect and determine the multiple target hyperparameter combinations.

[0072] Among them, generalization performance evaluation is the process of verifying the model's predictive ability on unseen data through independent test datasets and evaluating the model's performance stability and adaptability; model evaluation effect is the result of quantitatively judging the model's prediction accuracy, stability and other performance based on preset evaluation criteria, which is the core basis for selecting hyperparameter configurations; target hyperparameter combination is the hyperparameter configuration with better model evaluation effect and adaptability to the needs of steel performance prediction after generalization performance evaluation and ranking, which is the core parameter basis for subsequent model retraining.

[0073] In this embodiment of the disclosure, the outer test fold datasets obtained by the repeated K-fold rule can be input into the regression chain model trained by the corresponding hyperparameter configuration. The prediction performance of each model can be tested by the outer test fold dataset to complete the generalization performance evaluation of each hyperparameter configuration. Then, based on the results of the generalization performance evaluation, all the multiple hyperparameter configurations adapted to small samples are ranked according to the model evaluation effect. The multiple hyperparameter configurations with better performance are selected from the ranking results and determined as the target hyperparameter combination for subsequent full retraining of the model.

[0074] Evaluation using an independent outer test dataset effectively avoids overfitting of the model on the training data, objectively assesses the actual predictive ability of the model under each hyperparameter configuration, and the ranking and selection based on the evaluation results provides a quantitative scientific basis for determining the target hyperparameter combination. This ensures that the selected hyperparameter combination enables the regression chain model to maintain good generalization performance and prediction accuracy. At the same time, the determination of multiple sets of target hyperparameter combinations provides a high-quality parameter foundation for the subsequent integration and fusion of models, further improving the robustness and accuracy of the joint prediction model of steel performance in small sample scenarios, and effectively solving the problem that hyperparameter search in small sample scenarios is prone to accidental optima.

[0075] Step 270: Based on the target hyperparameter combination, the regression chain model is fully retrained using a standardized dataset to obtain the trained joint prediction model for steel performance.

[0076] In this embodiment of the present disclosure, each set of target hyperparameters obtained through screening can be used to independently initialize the parameters of the regression chain model. Then, the complete standardized dataset is used as a unified training basis, and full training is performed on all regression chain models that have completed independent initialization configuration in sequence to obtain multiple trained regression chain sub-models. Subsequently, a comprehensive performance evaluation is carried out on the overall performance of each regression chain sub-model based on the standardized dataset. According to the evaluation results, each sub-model is assigned a corresponding performance weight. Finally, all regression chain models are integrated and fused according to the performance weight to form a trained joint prediction model for steel performance. The fusion method can be other applicable fusion methods such as arithmetic mean fusion and weighted average fusion.

[0077] Accordingly, the implementation steps may include: independently initializing and configuring the regression chain model using each of the multiple sets of target hyperparameter combinations; using the standardized dataset as the training basis, performing full training on each independently initialized regression chain model to obtain multiple trained regression chain sub-models; performing a comprehensive performance evaluation on each regression chain sub-model based on the standardized dataset to determine the performance weight of each regression chain sub-model; and integrating and fusing the multiple regression chain models based on the performance weights to obtain a joint prediction model for steel performance.

[0078] Among them, full retraining is the complete training process of the model using the complete standardized dataset instead of the partitioned subset, which allows the model to fully learn all the feature associations in the data; the regression chain sub-model is the regression chain model that has been initialized and configured with a single objective hyperparameter combination and has completed full training, and it is the basic model unit that constitutes the final joint prediction model; the performance weight is the weight value assigned according to the comprehensive performance evaluation result of the regression chain sub-model, reflecting the contribution of each sub-model in the integration and fusion; integration and fusion is the modeling process of integrating multiple regression chain sub-models according to their corresponding performance weights, and the integrated model forms a comprehensive prediction model with better performance.

[0079] Based on the target hyperparameter combination, a standardized dataset is used to fully retrain the regression chain model and then integrate it to obtain a joint prediction model for steel performance. Through independent initialization configuration and full training, each regression chain sub-model fully learns the feature correlation rules in the data under the optimal parameter configuration, maximizing the value of the data and adapting to the problem of limited data volume in small sample scenarios. The comprehensive performance evaluation based on the standardized dataset provides an objective and scientific basis for determining the performance weights. The integration of performance weights can combine the prediction advantages of each sub-model, avoid the prediction bias and overfitting risk of a single model, and significantly improve the generalization ability and prediction accuracy of the model. The final joint prediction model for steel performance can achieve accurate and stable joint prediction of multi-dimensional mechanical performance indicators of steel, ensuring the reliability and adaptability of the model in industrial applications.

[0080] Step 280: Input the chemical composition and production process parameters of the steel to be predicted into the joint prediction model of steel performance to obtain the joint prediction results of each steel performance index corresponding to the steel to be predicted.

[0081] In this embodiment of the present disclosure, the relevant data of chemical composition and production process parameters of the steel to be predicted can be obtained first, and standardized preprocessing can be carried out to fill missing values ​​and detect and process outliers in the data. Then, the data scale difference can be eliminated through standardization processing to obtain standardized feature data to be predicted. Subsequently, the feature data to be predicted is input into the trained steel performance joint prediction model. The model completes the prediction calculation of the mechanical performance index corresponding to each node in the order of the internal regression chain nodes and outputs the results. The prediction values ​​of all nodes together constitute the joint prediction result of each steel performance index corresponding to the steel to be predicted.

[0082] Accordingly, the implementation steps may include: performing standardized preprocessing on the chemical composition and production process parameters of the steel to be predicted to obtain the feature data to be predicted. The standardized preprocessing includes at least missing value filling, outlier detection and standardization; inputting the feature data to be predicted into the joint prediction model of steel performance, and sequentially outputting the predicted values ​​of the mechanical performance indicators of each node as the joint prediction results of the steel performance indicators corresponding to the steel to be predicted.

[0083] Among them, the steel to be predicted refers to steel materials whose mechanical performance indicators have not yet been tested and whose comprehensive mechanical properties need to be predicted by the model, and is the target object of performance prediction; the production process parameters are the various process control parameters actually used in the production and processing of the steel to be predicted, and are the core input features for predicting its mechanical properties; the joint prediction result is the set of multi-dimensional mechanical performance indicator prediction values ​​output by the steel performance joint prediction model for the steel to be predicted, which can fully reflect the comprehensive mechanical properties of the steel material; the feature data to be predicted is the standardized data obtained after the chemical composition and production process parameters of the steel to be predicted have been standardized and preprocessed, and is the effective feature data input to the joint prediction model; the node mechanical performance indicators are the single mechanical performance indicators corresponding to each regression chain node in the steel performance joint prediction model, and the model will complete the prediction of each indicator in the order of the nodes.

[0084] The chemical composition and production process parameters of the steel to be predicted are standardized and preprocessed before being input into the joint prediction model for steel performance to obtain joint prediction results. Standardized preprocessing ensures that the feature data to be predicted and the model training data maintain a consistent scale and standard, which can guarantee the effectiveness of the input features of the model and avoid prediction bias caused by inconsistent data formats and scales. The model outputs the predicted values ​​of each mechanical performance index in the order of nodes, which can give full play to the advantages of the regression chain model in explicitly modeling the conditional dependencies between indicators. This makes the joint prediction results more consistent with the objective logic of steel performance formation, realizing rapid and accurate joint prediction of multi-dimensional mechanical performance indicators of steel. Comprehensive steel performance data can be obtained without extensive experimental testing, which can significantly reduce the time and cost of steel performance testing. At the same time, the joint prediction results can fully reflect the mechanical properties of the steel to be predicted, providing an efficient and reliable reference for steel production process optimization, quality control and product performance judgment, meeting the actual needs of industrial scenarios for rapid prediction of steel performance.

[0085] In specific application scenarios, as an optional approach, the implementation steps may also include model diagnosis and traceability steps: based on the actual true values ​​of multi-dimensional mechanical performance indicators and the predicted values ​​of the model output, calculate the target-specific evaluation indicators and the overall evaluation indicators reflecting the overall performance of the model for each indicator, and then generate a true value-predicted value comparison chart according to the correspondence between the true values ​​and the predicted values. At the same time, mine and present the degree of influence of each input feature on the prediction results, generate a feature importance chart, and completely save the key information in the model construction and training process, including the optimal target order, target hyperparameter combination, random number seed, and nested cross-validation parameters. A comprehensive diagnosis of model performance is completed through quantitative indicators and visualization charts, and the audit and reproduction of the entire model construction and training process is realized based on the saved key information.

[0086] By calculating target-specific and overall evaluation indicators, the predictive performance of the model can be accurately quantified from both local and global dimensions. The comparison chart of actual and predicted values ​​and the feature importance chart visualize the model performance and the influence of features, facilitating the rapid identification of model prediction biases and problems, and improving the targeting and efficiency of model optimization. The complete preservation of key information such as the optimal target order and target hyperparameter combination not only enables auditability of the entire model construction and training process, making the model development process verifiable, but also ensures the reproducibility of the model by relying on information such as fixed random number seeds. This solves the problems of information loss and untraceability in the model development process, making the construction and application of the steel performance joint prediction model more standardized and scientific. At the same time, it provides a complete reference for the subsequent optimization, iteration and cross-scenario migration of the model, which can significantly improve the engineering application value of the model in industrial scenarios.

[0087] In summary, the technical solution in this application, by determining a standardized dataset containing steel chemical composition, process parameter characteristics, and multiple mechanical property indicators, and configuring the target order of indicators according to preset rules or traversal methods, and constructing a regression chain model with the predicted values ​​of preceding indicators as additional inputs to the subsequent regressor, can specifically address the problems of insufficient multi-indicator joint modeling capability of physical / empirical models, inconsistent prediction logic caused by single-objective data-driven models ignoring conditional dependencies between indicators, and the tendency of multi-output regression models to implicitly mine indicator correlations and suffer from training imbalance under small sample conditions. Simultaneously, a nested cross-validation strategy adapted to small samples is employed for hyperparameter search. The optimization method involves fully retraining the regression chain model using a combination of target hyperparameters to obtain a joint prediction model and output the predicted steel performance indicators. This effectively addresses the pain points of directly applying regression chains to steel performance prediction, such as large evaluation variance under small samples, the tendency for hyperparameter search to result in accidental optima, and the lack of engineering specifications for chain order configuration. It can significantly improve the robustness and prediction accuracy of the steel performance prediction model. Furthermore, through an engineering-oriented chain order configuration, hyperparameter optimization, and model retraining process, the model can meet the core requirements of reproducibility and practicality in industrial scenarios, achieving accurate joint prediction of multi-dimensional mechanical performance indicators of steel.

[0088] Furthermore, as Figure 1 and Figure 2 The specific implementation of the method shown in this embodiment provides a device for joint prediction of multiple indicators of steel quality under small sample conditions, such as... Figure 3 As shown, the device includes: a determination module 31, a construction module 32, a verification module 33, a training module 34, and a prediction module 35.

[0089] The determination module 31 can be used to determine the standardized dataset of steel materials. The standardized dataset includes steel chemical composition, process parameter characteristics and multi-dimensional mechanical property index data. The multi-dimensional mechanical property index includes at least two of yield strength, tensile strength, elongation, reduction of area and impact energy.

[0090] The construction module 32 can be used to configure the target order of multi-dimensional mechanical performance indicators based on preset rules or traversal methods, and to build a regression chain model based on the target order. In the regression chain model, the predicted value of the preceding mechanical performance indicator is used as an additional input feature of the regressor corresponding to the subsequent mechanical performance indicator.

[0091] The validation module 33 can be used to perform hyperparameter search and optimization on the regression chain model using a nested cross-validation strategy adapted to small samples, and to determine the target hyperparameter combination.

[0092] Training module 34 can be used to fully retrain the regression chain model based on the target hyperparameter combination and using a standardized dataset to obtain a trained joint prediction model for steel performance.

[0093] The prediction module 35 can be used to input the chemical composition and production process parameters of the steel to be predicted into the joint prediction model of steel performance to obtain the joint prediction results of each steel performance index corresponding to the steel to be predicted.

[0094] In some embodiments of this application, the determining module 31 can be used to obtain the original dataset of steel materials. The original dataset contains the original data of the chemical composition, process parameters and corresponding multi-dimensional mechanical performance indicators of the steel materials. The original dataset is subjected to standardization preprocessing to obtain a standardized dataset. The standardization preprocessing includes at least missing value imputation, outlier detection and standardization processing.

[0095] In some embodiments of this application, when configuring the target order of multi-dimensional mechanical performance indicators based on preset rules or traversal methods, the construction module 32 can be specifically used to determine the causal dependency of multi-dimensional mechanical performance indicators according to the industrial mechanism of steel performance formation, configure the target order based on the causal dependency; and / or, traverse all permutations and combinations of multi-dimensional mechanical performance indicators to generate all feasible target orders.

[0096] In some embodiments of this application, when constructing a regression chain model based on the target sequence, the construction module 32 can be specifically used to map the target sequence to the link index of the regression chain, determine the mechanical performance index corresponding to each node in the regression chain, configure an independent single-objective regressor for each node in the regression chain, define the information transmission rules of the regression chain, and the single-objective regressor can be one or more of gradient boosting regressor, extreme gradient boosting tree, random forest, and support vector regression; use the predicted value of the mechanical performance index corresponding to the preceding node as the additional input feature of the single-objective regressor configured for the subsequent node, and combine the steel chemical composition and process parameter features to construct the regression chain model.

[0097] In some embodiments of this application, the verification module 33 can be specifically used to divide the standardized dataset into multiple sets of outer training fold datasets and corresponding multiple sets of outer test fold datasets according to the repeated K-fold rule, and build a nested cross-validation framework of outer evaluation and inner hyperparameter search; input the outer training fold dataset into the regression chain model, and define the hyperparameter search space matching the steel features for the single-objective regressor of each node in the regression chain model based on the characteristics of steel chemical composition and process parameters; in the inner cross-validation of the nested cross-validation framework, sample and search for hyperparameters based on the hyperparameter search space, and obtain multiple sets of hyperparameter configurations suitable for small samples by using dimensionless evaluation indicators as the screening criteria; input each set of outer test fold datasets into the regression chain model trained by the corresponding hyperparameter configuration, complete the generalization performance evaluation of the hyperparameter configuration, and sort the multiple sets of hyperparameter configurations according to the model evaluation effect based on the evaluation results to determine multiple sets of target hyperparameter combinations.

[0098] In some embodiments of this application, the training module 34 can be specifically used to independently initialize and configure the regression chain model using each of the multiple sets of target hyperparameter combinations; using the standardized dataset as the training basis, perform full training on each independently initialized regression chain model to obtain multiple trained regression chain sub-models; perform a comprehensive performance evaluation on each regression chain sub-model based on the standardized dataset to determine the performance weight of each regression chain model; and integrate and fuse the multiple regression chain models based on the performance weights to obtain a joint prediction model for steel performance.

[0099] In some embodiments of this application, the prediction module 35 can be used to perform standardized preprocessing on the chemical composition and production process parameters of the steel to be predicted, to obtain the feature data to be predicted. The standardized preprocessing includes at least missing value filling, outlier detection and standardization. The feature data to be predicted is input into the joint prediction model of steel performance, and the predicted values ​​of the mechanical performance indicators of each node are output in sequence as the joint prediction results of the steel performance indicators corresponding to the steel to be predicted.

[0100] It should be noted that other corresponding descriptions of the functional units involved in the multi-index joint prediction device for steel quality under small sample conditions provided in this embodiment can be found in [reference needed]. Figure 1 and Figure 2 The corresponding descriptions in [the document] will not be repeated here.

[0101] Based on the above, Figure 1 and Figure 2 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method. Figure 1 and Figure 2 The method for joint prediction of multiple indicators of steel quality under small sample conditions is shown.

[0102] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause an electronic device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0103] Based on the above, Figure 1 and Figure 2 The method shown, and Figure 3 To achieve the above objectives, the present application also provides an electronic device, specifically a personal computer, tablet computer, server, or other network device, as shown in the virtual device embodiment. This device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described objectives. Figure 1 and Figure 2 The method for joint prediction of multiple indicators of steel quality under small sample conditions is shown.

[0104] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0105] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0106] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0107] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.

[0108] This invention addresses several issues by defining a standardized dataset containing steel chemical composition, process parameters, and various mechanical performance indicators. It configures the target order of indicators using preset rules or traversal methods and constructs a regression chain model where the predicted values ​​of preceding indicators are additional inputs to the subsequent regressor. This specifically solves problems such as insufficient multi-indicator joint modeling capabilities of physical / empirical models, inconsistent prediction logic caused by single-objective data-driven models ignoring conditional dependencies between indicators, and the tendency for training imbalances in small-sample training due to implicitly mining indicator correlations in multi-output regression models. Simultaneously, it employs a nested cross-validation strategy adapted to small samples for hyperparameter search optimization. The regression chain model is then fully retrained using the target hyperparameter combination to obtain a joint prediction model and output the steel performance indicators to be predicted. This effectively solves the pain points of large evaluation variance in small-sample conditions, accidental optimality in hyperparameter search, and lack of engineering specifications for chain order configuration when directly applying regression chains to steel performance prediction. It significantly improves the robustness and accuracy of steel performance prediction models. Furthermore, through an engineered chain order configuration, hyperparameter optimization, and model retraining process, the model meets the core requirements of reproducibility and practicality in industrial scenarios, achieving accurate joint prediction of multi-dimensional mechanical performance indicators of steel.

[0109] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0110] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for joint prediction of multiple indicators of steel quality under small sample conditions, characterized in that, include: A standardized dataset for steel materials is determined. The standardized dataset includes steel chemical composition, process parameter characteristics, and multi-dimensional mechanical property index data. The multi-dimensional mechanical property index includes at least two of the following: yield strength, tensile strength, elongation, reduction of area, and impact energy. The target order of the multi-dimensional mechanical performance indicators is configured based on preset rules or traversal methods, and a regression chain model is constructed based on the target order. In the regression chain model, the predicted value of the preceding mechanical performance indicator is used as an additional input feature of the regressor corresponding to the subsequent mechanical performance indicator. A nested cross-validation strategy adapted to small samples is used to search and optimize hyperparameters in the regression chain model to determine target hyperparameter combinations. This includes: dividing the standardized dataset into multiple sets of outer training fold datasets and corresponding sets of outer test fold datasets according to a repeated K-fold rule, and constructing a nested cross-validation framework for outer evaluation and inner hyperparameter search; inputting the outer training fold datasets into the regression chain model, and defining a hyperparameter search space matching the steel chemical composition and process parameter features for the single-objective regressors of each node in the regression chain model; in the inner cross-validation of the nested cross-validation framework, sampling and searching hyperparameters based on the hyperparameter search space, using dimensionless evaluation indicators as the selection criterion to obtain multiple sets of hyperparameter configurations adapted to small samples; inputting each set of outer test fold datasets into the regression chain model trained with the corresponding hyperparameter configuration to complete the generalization performance evaluation of the hyperparameter configurations, and ranking the multiple sets of hyperparameter configurations according to the model evaluation effect based on the evaluation results to determine multiple sets of target hyperparameter combinations; Based on the target hyperparameter combination, the regression chain model is fully retrained using the standardized dataset to obtain a trained joint prediction model for steel performance. The chemical composition and production process parameters of the steel to be predicted are input into the joint prediction model of steel performance to obtain the joint prediction results of each steel performance index corresponding to the steel to be predicted.

2. The method according to claim 1, characterized in that, The determination of the standardized dataset for steel materials includes: Obtain the original dataset of steel materials, which includes the original data of the chemical composition, process parameters and corresponding multi-dimensional mechanical performance indicators of steel materials; The original dataset is subjected to standardization preprocessing to obtain a standardized dataset. The standardization preprocessing includes at least missing value imputation, outlier detection, and standardization.

3. The method according to claim 1, characterized in that, The target order for configuring the multi-dimensional mechanical performance indicators based on preset rules or traversal methods includes: The causal dependencies of the multi-dimensional mechanical performance indicators are determined based on the industrial mechanism of steel performance formation, and the target order is configured based on the causal dependencies; and / or, By iterating through all permutations and combinations of the multi-dimensional mechanical performance indicators, all feasible target sequences are generated.

4. The method according to claim 1, characterized in that, The construction of the regression chain model based on the target order includes: The target sequence is mapped to the link index of the regression chain to determine the mechanical performance index corresponding to each node in the regression chain; Configure an independent single-objective regressor for each node in the regression chain, define the information transmission rules of the regression chain, and the single-objective regressor is one or more of gradient boosting regressor, extreme gradient boosting tree, random forest, and support vector regression. The predicted values ​​of the mechanical performance indicators corresponding to the preceding nodes are used as additional input features of the single-objective regressor configured for the subsequent nodes. Combined with the steel chemical composition and the process parameter features, a regression chain model is constructed.

5. The method according to claim 1, characterized in that, The process involves fully retraining the regression chain model based on the target hyperparameter combination and using the standardized dataset to obtain a trained joint prediction model for steel performance, including: Each of the multiple sets of target hyperparameter combinations is used to independently initialize and configure the regression chain model; Using the standardized dataset as the training basis, full training is performed on each independently initialized regression chain model to obtain multiple trained regression chain sub-models. Based on the standardized dataset, a comprehensive performance evaluation is performed on each regression chain sub-model to determine the performance weight of each regression chain model; Based on the performance weights, multiple regression chain sub-models are integrated and fused to obtain a joint prediction model for steel performance.

6. The method according to claim 1, characterized in that, The process of inputting the chemical composition and production process parameters of the steel to be predicted into the joint prediction model for steel performance to obtain the joint prediction results of various steel performance indicators corresponding to the steel to be predicted includes: The chemical composition and production process parameters of the steel to be predicted are standardized and preprocessed to obtain the feature data to be predicted. The standardized preprocessing includes at least missing value filling, outlier detection and standardization. The feature data to be predicted is input into the joint prediction model of steel performance, and the predicted values ​​of mechanical performance indicators of each node are output in sequence as the joint prediction results of each steel performance indicator corresponding to the steel to be predicted.

7. A device for joint prediction of multiple indicators of steel quality under small sample conditions, characterized in that, include: The determination module is used to determine a standardized dataset for steel materials. The standardized dataset includes steel chemical composition, process parameter characteristics, and multi-dimensional mechanical property index data. The multi-dimensional mechanical property index includes at least two of the following: yield strength, tensile strength, elongation, reduction of area, and impact energy. The construction module is used to configure the target order of the multi-dimensional mechanical performance indicators based on preset rules or traversal methods, and to construct a regression chain model based on the target order. In the regression chain model, the predicted value of the preceding mechanical performance indicator is used as an additional input feature of the regressor corresponding to the subsequent mechanical performance indicator. The validation module is used to perform hyperparameter search and optimization on the regression chain model using a nested cross-validation strategy adapted to small samples, and to determine the target hyperparameter combination. This includes: dividing the standardized dataset into multiple sets of outer training fold datasets and corresponding sets of outer test fold datasets according to a repeated K-fold rule, and constructing a nested cross-validation framework of outer evaluation and inner hyperparameter search; inputting the outer training fold datasets into the regression chain model, and defining a hyperparameter search space matching the steel features for the single-objective regressors of each node in the regression chain model based on the steel chemical composition and the process parameter features; in the inner cross-validation of the nested cross-validation framework, sampling and searching for hyperparameters based on the hyperparameter search space, using dimensionless evaluation indicators as the screening criteria to obtain multiple sets of hyperparameter configurations adapted to small samples; inputting each set of outer test fold datasets into the regression chain model trained with the corresponding hyperparameter configuration, completing the generalization performance evaluation of the hyperparameter configurations, and ranking the multiple sets of hyperparameter configurations according to the model evaluation effect based on the evaluation results, to determine multiple sets of target hyperparameter combinations; The training module is used to fully retrain the regression chain model based on the target hyperparameter combination and the standardized dataset to obtain the trained joint prediction model for steel performance. The prediction module is used to input the chemical composition and production process parameters of the steel to be predicted into the joint prediction model of steel performance, and obtain the joint prediction results of each steel performance index corresponding to the steel to be predicted.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

9. An electronic device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.

Images