Steel mechanical property prediction method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a steel mechanical property prediction method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining molten steel components and rolling process parameters of to-be-predicted steel in a production process as to-be-predicted feature data; inputting the to-be-predicted feature data into two pre-trained prediction models with different model types to obtain two prediction results, and comprehensively determining a predicted mechanical property result of the to-be-predicted steel based on the two prediction results. Through the mode, the mechanical property of the produced steel can be predicted and evaluated based on related data in the production process, a preliminary result is obtained, the cost is relatively low, the speed is higher, and once the steel with the unqualified mechanical property is found, the steel can be fed back to a production line in time so as to be adjusted in time, and the enterprise loss is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of prediction of mechanical properties of steel materials, and in particular to a method, device, electronic equipment and storage medium for predicting mechanical properties of steel materials. Background Art

[0002] After the steel is produced, the mechanical properties of the steel are often tested by random sampling. Since the testing process is cumbersome and costly, it is often necessary to increase the density of random sampling to further understand the mechanical properties of the steel while ensuring the quality of the products, which greatly increases the cost. On the other hand, since testing also takes a certain amount of time, once the mechanical properties of the steel do not meet the standards, the feedback to the production line will be delayed, resulting in greater losses for the company. Summary of the invention

[0003] The present invention provides a method, device, electronic device and storage medium for predicting the mechanical properties of steel to solve the technical problems in related technologies that the produced steel needs to be actually tested to obtain the mechanical properties, which is time-consuming, costly, and has a delayed result, which may cause great losses to the enterprise.

[0004] The present invention provides a method for predicting the mechanical properties of steel, which comprises: obtaining feature data to be predicted of the steel to be predicted, wherein the feature data to be predicted includes the composition of molten steel and rolling process parameters of the steel to be predicted during the production process; inputting the feature data to be predicted into a pre-trained first prediction model to obtain a first prediction result, and inputting the feature data to be predicted into a pre-trained second prediction model to obtain a second prediction result, wherein the first prediction model and the second prediction model are of different model types, the first prediction model and the second prediction model are obtained by training a sample training data set, the sample training data in the sample training data set includes training sample feature data and training sample mechanical property labels, and the training sample feature data includes the composition of molten steel and rolling process parameters of the sample steel during the production process; and determining the predicted mechanical property result of the steel to be predicted based on the first prediction result and the second prediction result.

[0005] In one embodiment of the present invention, before the feature data to be predicted is input into a pre-trained first prediction model to obtain a first prediction result, and before the feature data to be predicted is input into a pre-trained second prediction model to obtain a second prediction result, the method includes: obtaining a first sample data subset, wherein the first sample data in the first sample data subset includes first sample feature data of a first sample steel and a real mechanical property label of the first sample steel, and the first sample feature data includes molten steel composition and rolling process parameters of the first sample steel during the production process; if the number of first sample data in the first sample data subset is less than a preset real sample number threshold, obtaining a second sample data subset, wherein the second sample data in the second sample data subset includes second sample feature data, and the second sample feature data includes molten steel composition and rolling process parameters of the second sample steel during the production process; training the first initial model and the second initial model for the first time respectively through the first sample data subset to obtain a first initial model after the first training and a second initial model after the first training; inputting at least part of the second sample feature data in the second sample data subset into the first initial model after the first training and the second initial model after the first training respectively to obtain the The first predicted mechanical property label output by the first initial model after the first training, and the second predicted mechanical property label output by the second initial model after the first training; determine the first predicted utility value of each first predicted mechanical property label, and determine the second predicted utility value of each second predicted mechanical property label; determine the first predicted mechanical property label whose first predicted utility value is greater than a preset utility threshold as a first sample mechanical property pseudo label, generate a first pseudo sample training sub-dataset based on the first sample mechanical property pseudo-label and the second sample feature data corresponding to the first sample mechanical property pseudo-label, and generate a first sample training data subset for training the first prediction model based on the first pseudo sample training sub-dataset and the first sample data subset; determine the second predicted mechanical property label whose second predicted utility value is greater than the preset utility threshold as a second sample mechanical property pseudo-label, generate a second pseudo sample training sub-dataset based on the second sample mechanical property pseudo-label and the second sample feature data corresponding to the second sample mechanical property pseudo-label, and generate a second sample training data subset for training the second prediction model based on the second pseudo sample training sub-dataset and the first sample data subset; generate the sample training data set based on the first sample training data subset and the second sample training data subset.

[0006] In one embodiment of the present invention, at least part of the second sample feature data in the second sample data subset is inputted into the trained first initial model and the trained second initial model, respectively, including: selecting a preset number of second sample feature data from the second sample data subset, and inputting the selected second sample feature data into the trained first initial model and the trained second initial model, respectively; if the number of the first sample mechanical property pseudo-labels is less than the preset pseudo-sample number threshold, and the number of the second sample mechanical property pseudo-labels is less than the preset pseudo-sample number threshold, reselecting a preset number of second sample feature data from the second sample data subset, and inputting the selected second sample feature data into the trained first initial model and the trained second initial model, respectively, until the number of reselected second sample feature data is greater than the preset number threshold, or the number of the first sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold, or the number of the second sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold; if If the number of pseudo-labels of mechanical properties of the first samples is less than the preset pseudo-sample number threshold, and the number of pseudo-labels of mechanical properties of the second samples is greater than or equal to the preset pseudo-sample number threshold, a preset number of second sample feature data is reselected from the second sample data subset, and the selected second sample feature data is input into the trained first initial model until the number of reselected second sample feature data is greater than the preset number threshold, or the number of pseudo-labels of mechanical properties of the first samples is greater than or equal to the preset pseudo-sample number threshold; if the number of pseudo-labels of mechanical properties of the second samples is less than the preset pseudo-sample number threshold, and the number of pseudo-labels of mechanical properties of the first samples is greater than or equal to the preset pseudo-sample number threshold, a preset number of second sample feature data is reselected from the second sample data subset, and the selected second sample feature data is input into the trained second initial model until the number of reselected second sample feature data is greater than the preset number threshold, or the number of pseudo-labels of mechanical properties of the second samples is greater than or equal to the preset pseudo-sample number threshold.

[0007] In one embodiment of the present invention, determining the first predictive utility value of each first predictive mechanical property label and determining the second predictive utility value of each second predictive mechanical property label include: adding all first predictive mechanical property labels and all second sample feature data corresponding to the first predictive mechanical property labels to the first sample data subset to obtain a first sample expansion data set, performing a second training on the second initial model after the first training using the first sample expansion data set to obtain a second initial model after the second training; determining the first k-nearest neighbor sample feature data of the second sample feature data corresponding to the first predictive mechanical property label from all the first sample feature data in the first sample data subset, inputting the first k-nearest neighbor sample feature data into the second initial model after the first training and the second initial model after the second training, respectively, to obtain a first sample predictive label and a second sample predictive label; determining the first sample predictive label and the second sample predictive label based on the first sample predictive label and the second sample predictive label respectively compared with the first k-nearest neighbor sample feature data in the first sample data subset ... sample predictive label and the second sample predictive label respectively compared with the first k-nearest neighbor sample feature data in the first sample data subset; determining the first sample predictive label and the second sample predictive label respectively compared with the first k-nearest neighbor sample feature data in the first sample data subset; determining the first sample predictive label and the second sample predictive label respectively compared with the first k-nearest neighbor sample feature data in the first sample data subset; determining the first sample predictive label and the second sample predictive label The first predictive utility value is determined by the degree of label difference between the real mechanical property labels corresponding to the second predicted mechanical property labels; all the second predicted mechanical property labels and all the second sample feature data corresponding to the second predicted mechanical property labels are added to the first sample data subset to obtain a second sample expansion data set, and the first initial model after the first training is trained for a second time through the second sample expansion data set to obtain the first initial model after the second training; the second k-nearest neighbor sample feature data of the second sample feature data corresponding to the second predicted mechanical property label is determined from all the first sample feature data in the first sample data subset, and the second k-nearest neighbor sample feature data is input into the first initial model after the first training and the first initial model after the second training respectively to obtain a third sample predictive label and a fourth sample predictive label; the second predictive utility value is determined based on the degree of label difference between the third sample predictive label and the fourth sample predictive label and the real mechanical property labels corresponding to the second k-nearest neighbor sample feature data in the first sample data subset.

[0008] In one embodiment of the present invention, determining a first prediction utility value of each first prediction mechanical property label and determining a second prediction utility value of each second prediction mechanical property label include:

[0009]

[0010] Among them, utility 1_i is the first predicted utility value of the first predicted mechanical property label i, x 1kis the first k-nearest neighbor sample feature data of the second sample feature data corresponding to the first predicted mechanical property label i in all the first sample feature data of the first sample data subset, N1 is the set of the first k-nearest neighbor sample feature data of the second sample feature data corresponding to all the first predicted mechanical property labels in all the first sample feature data of the first sample data subset, y 1k is the first k-nearest neighbor sample feature data x 1k The corresponding true mechanical property label in the first sample data subset, h1 is the second initial model after the first training, h1(x 1k ) is the predictive label of the first sample, h'1 is the second initial model after the second training, h'1(x 1k ) is the predictive label of the second sample; utility 2_i is the second predicted utility value of the second predicted mechanical property label i, x 2k is the second k-nearest neighbor sample feature data of the second sample feature data corresponding to the second predicted mechanical property label i among all the first sample feature data of the first sample data subset, N2 is the set of the second k-nearest neighbor sample feature data of the second sample feature data corresponding to all the second predicted mechanical property labels among all the first sample feature data of the first sample data subset, y 2k is the second k-nearest neighbor sample feature data x 2k The corresponding true mechanical property label in the first sample data subset, h2 is the first initial model after the first training, h2(x 2k ) is the predictive label of the third sample, h'2 is the first initial model after the second training, h'2(x 2k ) is the predictive label of the fourth sample.

[0011] In one embodiment of the present invention, the predicted mechanical property result of the steel to be predicted is determined based on the first prediction result and the second prediction result, including: obtaining the first preset model coefficient of the pre-trained first prediction model and the second preset model coefficient of the pre-trained second prediction model; performing weighted averaging processing on the first prediction result and the second prediction result according to the first preset model coefficient and the second preset model coefficient, respectively, to obtain the predicted mechanical property result of the steel to be predicted.

[0012] In one embodiment of the present invention, before obtaining the first preset model coefficient of the pre-trained first prediction model and the second preset model coefficient of the pre-trained second prediction model, the method further includes: obtaining a third sample data subset, the third sample data in the third sample data subset including third sample feature data of a third sample steel and a real mechanical property label of the third sample steel, the third sample feature data including molten steel composition and rolling process parameters of the third sample steel during production; inputting the third sample feature data in the third sample data subset into the pre-trained first prediction model and the pre-trained second prediction model respectively to obtain a third prediction result output by the pre-trained first prediction model and a fourth prediction result output by the pre-trained second prediction model; determining a first accuracy rate of the pre-trained first prediction model and a second accuracy rate of the pre-trained second prediction model through matching results of the third prediction result and the fourth prediction result with the real mechanical property label of the corresponding third sample steel; determining the first accuracy rate as the first preset model coefficient, and determining the second accuracy rate as the second preset model coefficient.

[0013] The present invention also provides a device for predicting the mechanical properties of steel, the device for predicting the mechanical properties of steel comprising: an acquisition module for acquiring characteristic data to be predicted of the steel to be predicted, the characteristic data to be predicted including the molten steel composition and rolling process parameters of the steel to be predicted during the production process; a pre-trained first prediction model for inputting the characteristic data to be predicted into the pre-trained first prediction model to obtain a first prediction result; a pre-trained second prediction model for inputting the characteristic data to be predicted into the pre-trained second prediction model to obtain a second prediction result; a result determination module for determining the predicted mechanical property result of the steel to be predicted based on the first prediction result and the second prediction result; wherein the first prediction model and the second prediction model have different model types, the first prediction model and the second prediction model are obtained by training a sample training data set, the sample training data in the sample training data set includes training sample characteristic data and training sample mechanical property labels, the training sample characteristic data includes the molten steel composition and rolling process parameters of the sample steel during the production process.

[0014] The present invention also provides an electronic device, comprising a processor, a memory and a communication bus; the communication bus is used to connect the processor and the memory; the processor is used to execute a computer program stored in the memory to implement a method provided in any one of the above embodiments.

[0015] The present invention also provides a computer-readable storage medium on which a computer program is stored, wherein the computer program is used to enable a computer to execute the method provided in any one of the above embodiments.

[0016] The beneficial effects of the present invention are as follows: a method, device, electronic device and storage medium for predicting the mechanical properties of steel proposed in the present invention. The method obtains the molten steel composition and rolling process parameters of the steel to be predicted in the production process as the characteristic data to be predicted, and then inputs the characteristic data to be predicted into two pre-trained prediction models of different model types to obtain two prediction results, and then comprehensively determines the predicted mechanical property results of the steel to be predicted based on the two prediction results. In this way, the mechanical properties of the produced steel can be predicted and evaluated based on the relevant data in the production process to obtain preliminary results. The cost is relatively low and the speed is faster. Once it is found that the mechanical properties of steel do not meet the standards, the production line can be fed back in time for timely adjustment to reduce enterprise losses. Accurate prediction of the mechanical properties of steel can help steel companies reduce inspection and testing costs and predict batch performance quality problems in advance, which plays an important role in reducing costs and quality objection rates for steel companies. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a method for predicting the mechanical properties of steel provided in one embodiment of the present invention;

[0018] Figure 2 A schematic diagram of prediction results of a prediction model trained by a conventional machine learning method provided in one embodiment of the present invention;

[0019] Figure 3 A schematic diagram of prediction results of a prediction model trained by the method provided in an embodiment of the present invention is provided in an embodiment of the present invention;

[0020] Figure 4 It is a structural schematic diagram of a steel mechanical property prediction device provided by an embodiment of the present invention;

[0021] Figure 5 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0023] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0024] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0025] See also Figure 1 , Figure 1 A schematic diagram of a flow chart of a method for predicting the mechanical properties of steel provided in one embodiment of the present invention, referring to Figure 1 , the steel mechanical properties prediction method includes the following steps:

[0026] Step S101: Acquire the characteristic data of the steel material to be predicted.

[0027] The characteristic data to be predicted include the molten steel composition and rolling process parameters of the steel to be predicted during the production process.

[0028] Testing the mechanical properties of steel is an important means of evaluating its behavior under stress conditions, including the determination of hardness, tensile strength, yield strength, elongation, impact toughness, etc. Steel is produced based on molten steel. The mechanical properties of steel produced from molten steel are related to the composition of the molten steel and the rolling process parameters during the production process. Therefore, the mechanical properties of the produced steel can be predicted by obtaining these data.

[0029] Step S102: inputting the feature data to be predicted into a pre-trained first prediction model to obtain a first prediction result, and inputting the feature data to be predicted into a pre-trained second prediction model to obtain a second prediction result.

[0030] The first prediction model and the second prediction model have different model types, for example, a generalized additive model and a GBDT (Gradient Boosting Decision Tree) model can be used respectively. The specific model type can be set and selected by those skilled in the art as needed. By selecting models of different model types, the defects caused by a single type of model can be further avoided and the reliability of the results can be improved. Accordingly, the model parameters of the first prediction model and the second prediction model are also different.

[0031] Among them, the first prediction model and the second prediction model are obtained by training a sample training data set. The sample training data in the sample training data set includes training sample feature data and training sample mechanical property labels. The training sample feature data includes the molten steel composition and rolling process parameters of the sample steel during the production process.

[0032] The sample training data set can be a sample set with real test results. At this time, the training sample feature data and the training sample mechanical property labels are all real sample data. For mass-produced steel, such sample data is relatively easy to obtain, because a large amount of test data accumulated in the previous production process can be used for model training. However, for steel types with small production volumes, there is a lack of sample data with real mechanical property test results. If the number of sample training data is too small, the prediction accuracy of the trained first prediction model and the second prediction model will not be high. To solve the above problems, this embodiment provides a pseudo-sample expansion method. Through the scheme provided by the following embodiment, the number of sample training data can be effectively expanded, thereby improving the prediction accuracy of the trained first prediction model and the second prediction model. A pseudo-label with high credibility is constructed by collaborative regression, which can effectively solve the problem of mechanical property prediction of steel types with too few test results, and improve the mechanical property prediction accuracy of small sample steel types, so that the scheme can be applied to the mechanical property prediction of new steels and the mechanical property prediction of small sample steels.

[0033] In one embodiment, the method includes: obtaining a first sample data subset, wherein the first sample data in the first sample data subset includes first sample feature data of a first sample steel and a real mechanical property label of the first sample steel, and the first sample feature data includes molten steel composition and rolling process parameters of the first sample steel during production; if the number of first sample data in the first sample data subset is less than a preset real sample number threshold, obtaining a second sample data subset, wherein the second sample data in the second sample data subset includes second sample feature data, and the second sample feature data includes molten steel composition and rolling process parameters of the second sample steel during production; training the first initial model and the second initial model for the first time respectively through the first sample data subset to obtain a first initial model after the first training and a second initial model after the first training; inputting at least part of the second sample feature data in the second sample data subset into the first initial model after the first training and the second initial model after the first training respectively to obtain a first The first predicted mechanical property label output by the first initial model after the first training, and the second predicted mechanical property label output by the second initial model after the first training; determine the first predicted utility value of each first predicted mechanical property label, and determine the second predicted utility value of each second predicted mechanical property label; determine the first predicted mechanical property label whose first predicted utility value is greater than the preset utility threshold as the first sample mechanical property pseudo-label, generate a first pseudo-sample training sub-dataset based on the first sample mechanical property pseudo-label and the second sample feature data corresponding to the first sample mechanical property pseudo-label, and generate a first sample training data subset for training the first prediction model based on the first pseudo-sample training sub-dataset and the first sample data subset; determine the second predicted mechanical property label whose second predicted utility value is greater than the preset utility threshold as the second sample mechanical property pseudo-label, generate a second pseudo-sample training sub-dataset based on the second sample mechanical property pseudo-label and the second sample feature data corresponding to the second sample mechanical property pseudo-label, and generate a second sample training data subset for training the second prediction model based on the second pseudo-sample training sub-dataset and the first sample data subset; generate a sample training data set based on the first sample training data subset and the second sample training data subset.

[0034] It can be understood that the first sample data in the first sample data subset is sample data with real test results, and the second sample data in the second sample data subset is sample data without real test results. Then, the first initial model and the second initial model constructed are trained for the first time through the first sample data subset, and then the collaborative regression semi-supervised learning method can be used to make the first initial model and the second initial model select appropriate samples from the first sample data subset to add pseudo labels to each other (continuously increase the number of data in the first pseudo sample training sub-dataset and the second pseudo sample training sub-dataset), and repeat the pseudo label adding step until the number of repetitions reaches the preset number threshold or the number of data in the first pseudo sample training sub-dataset and the second pseudo sample training sub-dataset reaches the preset pseudo sample number threshold.

[0035] The first initial model and the second initial model can be KNN models. The two models need to use different distance measurement methods, such as Minkowski distances of different orders, so as to improve the reliability of subsequent output data. It should be noted that the first initial model, the second initial model and the subsequent first prediction model and the second prediction model are not the same model, that is, two sets of models are used in the sample expansion stage and the subsequent prediction stage.

[0036] Continuing from the above-mentioned embodiment, determining the first predictive utility value of each first predictive mechanical property label and determining the second predictive utility value of each second predictive mechanical property label include: adding all first predictive mechanical property labels and all second sample feature data corresponding to the first predictive mechanical property labels to the first sample data subset to obtain a first sample expansion data set, performing a second training on the second initial model after the first training through the first sample expansion data set to obtain a second initial model after the second training; determining the first k-nearest neighbor sample feature data of the second sample feature data corresponding to the first predictive mechanical property label from all the first sample feature data in the first sample data subset, inputting the first k-nearest neighbor sample feature data into the second initial model after the first training and the second initial model after the second training, respectively, to obtain a first sample predictive label and a second sample predictive label; determining the first k-nearest neighbor sample feature data of the second sample feature data corresponding to the first predictive mechanical property label in the first sample data subset based on the first sample predictive label and the second sample predictive label, respectively, and the first k-nearest neighbor sample feature data in the first sample data subset. The first prediction utility value is determined by the degree of label difference between the corresponding true mechanical property labels; all the second predicted mechanical property labels and all the second sample feature data corresponding to the second predicted mechanical property labels are added to the first sample data subset to obtain a second sample expansion data set, and the first initial model after the first training is trained for a second time through the second sample expansion data set to obtain the first initial model after the second training; the second k-nearest neighbor sample feature data of the second sample feature data corresponding to the second predicted mechanical property label is determined from all the first sample feature data in the first sample data subset, and the second k-nearest neighbor sample feature data is input into the first initial model after the first training and the first initial model after the second training, respectively, to obtain a third sample predictive label and a fourth sample predictive label; the second prediction utility value is determined based on the degree of label difference between the third sample predictive label and the fourth sample predictive label and the true mechanical property labels corresponding to the second k-nearest neighbor sample feature data in the first sample data subset.

[0037] Following the above embodiment, at least part of the second sample feature data in the second sample data subset is input into the trained first initial model and the trained second initial model respectively, including: selecting a preset number (wherein the preset number can be set by technicians in this field as needed, and the preset number of each batch can be the same or different) of second sample feature data from the second sample data subset, and inputting the selected second sample feature data into the trained first initial model and the trained second initial model respectively; if the number of pseudo-labels of mechanical properties of the first sample is less than the preset pseudo-sample number threshold, and the number of pseudo-labels of mechanical properties of the second sample is less than the preset pseudo-sample number threshold, re-selecting a preset number of second sample feature data from the second sample data subset, and inputting the selected second sample feature data into the trained first initial model and the trained second initial model respectively, until the number of re-selected second sample feature data is greater than the preset number threshold, or the number of pseudo-labels of mechanical properties of the first sample is greater than or equal to the preset pseudo-sample number threshold, or the second sample mechanical property is greater than or equal to the preset pseudo-sample number threshold, The number of performance pseudo-labels is greater than or equal to the preset pseudo-sample number threshold; if the number of first sample mechanical property pseudo-labels is less than the preset pseudo-sample number threshold, and the number of second sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold, reselect a preset number of second sample feature data from the second sample data subset, and input the selected second sample feature data into the trained first initial model until the number of reselected second sample feature data is greater than the preset number threshold, or the number of first sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold; if the number of second sample mechanical property pseudo-labels is less than the preset pseudo-sample number threshold, and the number of first sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold, reselect a preset number of second sample feature data from the second sample data subset, and input the selected second sample feature data into the trained second initial model until the number of reselected second sample feature data is greater than the preset number threshold, or the number of second sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold.

[0038] It can be understood that the generation process of the first sample mechanical property pseudo label and the second sample mechanical property pseudo label is controlled from two dimensions, one is the number of cycles, and the other is the quantity. The process ends when the condition is met first. The generation of the first sample mechanical property pseudo label and the second sample mechanical property pseudo label may not be simultaneous, so whichever one has not been completed can be carried out.

[0039] In order to improve the richness of the data, it can be limited that each time a preset number of second sample feature data is selected from the second sample data subset, the second sample feature data are all different. In actual scenarios, there are enough second sample feature data, so enough different second sample feature data can be selected to generate pseudo labels.

[0040] As an example, the method of expanding sample training data is as follows: first construct the first initial model and the second initial model, and then initialize the training sets L1 and L2 of the two models as sample sets with test results, that is, L1=L, L2=L, where L is also the first sample data subset. Then perform the first model training, perform the first training on the first initial model through L1, and perform the first training on the second initial model through L2, and obtain the first initial model h1 after the first training and the second initial model h2 after the first training. Then use the collaborative regression semi-supervised learning method to make h1 and h2 select appropriate samples from the second sample data subset U to add pseudo-labels to each other, which specifically includes the following steps:

[0041] 1) Select a random subset U' (a set formed by selecting a preset number of second sample feature data from the second sample data subset) from the sample set U (the second sample data subset) without test results, and for each sample u of U' i (the second sample feature data currently selected i), respectively use the first initial model h1 after the first training and the second initial model h2 after the first training to predict the predicted value h1 (u i )(first predicted mechanical properties label) and h2(u i )(Second predicted mechanical properties label), (u i ,h1(u i )) and (u i ,h2(u i )) add the training sets L2 and L1 respectively, and train the first initial model h1 after the first training and the second initial model h2 after the first training respectively through the added training sets L2 and L1, so as to obtain the first initial model h'1 after the second training and the second initial model h'2 after the second training;

[0042] 2) Select samples u that do not have test results in L1 and L2 iThe k-nearest neighbor sample sets N1 and N2 are respectively predicted by h1 and h'1 for the label value of N1, and h2 and h'2 for the label value of N2. In this process, since L1 and L2 include feature data with true mechanical property labels and feature data without true mechanical property labels, the feature data without true mechanical property labels (that is, the selected second sample feature data) is used to select a first sample feature data with the highest similarity from the feature data with true mechanical property labels (the first k-nearest neighbor sample feature data, the second k-nearest neighbor sample feature data), and then the corresponding first k-nearest neighbor sample set N1 and the second k-nearest neighbor sample set N2 are obtained;

[0043] 3) Calculate each u in U' separately i The utility value of using models h1 and h2 in prediction 1_i and utility 2_i , if there is a sample with a utility value greater than 0 (taking the preset utility threshold as 0 as an example), then select the unlabeled sample u with the largest utility maxj , add the sample and its prediction results to the other party's training set, that is:

[0044] L1=L1∪{(u max2 ,h2(u max2 ))}, L2=L2∪{(u max1 ,h1(u max1 ))}Formula (1);

[0045] If there is no sample with a utility value greater than 0, enter the next cycle and repeat steps 1)-3).

[0046] Following the above embodiment, the first predicted utility value of each first predicted mechanical property label is determined 1_i , and determining a second predicted utility value utility for each second predicted mechanical property label 2_i ,include:

[0047]

[0048] Among them, utility 1_i is the first predicted utility value of the first predicted mechanical property label i, x 1k is the first k-nearest neighbor sample feature data of the second sample feature data corresponding to the first predicted mechanical property label i in all the first sample feature data of the first sample data subset, N1 is the set of the first k-nearest neighbor sample feature data of the second sample feature data corresponding to all the first predicted mechanical property labels in all the first sample feature data of the first sample data subset, y 1k is the first k-nearest neighbor sample feature data x 1kThe corresponding true mechanical property label in the first sample data subset, h1 is the second initial model after the first training, h1(x 1k ) is the predictive label of the first sample, h'1 is the second initial model after the second training, h'1(x 1k ) is the predictive label of the second sample; utility 2_i is the second predicted utility value of the second predicted mechanical property label i, x 2k is the second k-nearest neighbor sample feature data of the second sample feature data corresponding to the second predicted mechanical property label i among all the first sample feature data of the first sample data subset, N2 is the set of the second k-nearest neighbor sample feature data of the second sample feature data corresponding to all the second predicted mechanical property labels among all the first sample feature data of the first sample data subset, y 2k is the second k-nearest neighbor sample feature data x 2k The corresponding true mechanical property label in the first sample data subset, h2 is the first initial model after the first training, h2(x 2k ) is the predictive label of the third sample, h'2 is the first initial model after the second training, h'2(x 2k ) is the predictive label of the fourth sample.

[0049] It can be understood that the first predicted mechanical property label output by the first initial model and the second sample feature data corresponding to the first predicted mechanical property label and input into the first initial model are added as a group of expanded sample data to be selected into the first sample data subset to obtain the first sample expanded data set, and then the second initial model after the first training is trained for the second time based on the first sample expanded data set to obtain the second initial model after the second training. Correspondingly, the second predicted mechanical property label output by the second initial model and the second sample feature data corresponding to the second predicted mechanical property label and input into the second initial model are added as a group of expanded sample data to be selected into the first sample data subset to obtain the second sample expanded data set, and then the first initial model after the first training is trained for the second time based on the second sample expanded data set to obtain the first initial model after the second training. Then, the first sample feature with the highest similarity to the feature corresponding to the predicted mechanical performance label is found from the features with the real mechanical performance label as the k-nearest neighbor sample feature data, and then the k-nearest neighbor sample feature data is predicted based on the initial models of the first training and the second training respectively. If the difference between the two predicted performance labels and the real mechanical performance labels corresponding to the k-nearest neighbor sample feature data is not large, and the difference between the result obtained by the second training model and the real result is smaller than the difference between the result obtained by the first training model and the real result, then the predicted utility value will be greater than 0. A utility value greater than 0 indicates that the unlabeled sample u i and its prediction results, so that u i The prediction error of the neighboring sample set is reduced, so it can be considered that the unlabeled sample u i The prediction results are pseudo labels with higher credibility, and adding them to the training set helps improve the accuracy. In this way, these first sample mechanical property pseudo labels and second sample mechanical property pseudo labels can be regarded as real detection labels to expand the samples with fewer real detection values ​​for subsequent prediction model training.

[0050] Step S103: Determine the predicted mechanical property results of the steel to be predicted based on the first prediction result and the second prediction result.

[0051] In one embodiment, the predicted mechanical property result of the steel to be predicted is determined based on the first prediction result and the second prediction result, including: obtaining a first preset model coefficient of a pre-trained first prediction model and a second preset model coefficient of a pre-trained second prediction model; performing weighted average processing on the first prediction result and the second prediction result according to the first preset model coefficient and the second preset model coefficient, respectively, to obtain the predicted mechanical property result of the steel to be predicted. Since the credibility of different types of models may vary, and the accuracy of the models obtained after training may also vary, the model weight (preset model coefficient) can be set with reference to the above factors, and then the output results of the prediction model are weighted averaged based on the model weight to obtain the final result.

[0052] Continuing from the above embodiment, before obtaining the first preset model coefficient of the pre-trained first prediction model and the second preset model coefficient of the pre-trained second prediction model, the method also includes: obtaining a third sample data subset, the third sample data in the third sample data subset includes third sample feature data of the third sample steel and the real mechanical property label of the third sample steel, and the third sample feature data includes the molten steel composition and rolling process parameters of the third sample steel during the production process; inputting the third sample feature data in the third sample data subset into the pre-trained first prediction model and the pre-trained second prediction model respectively, to obtain the third prediction result output by the pre-trained first prediction model and the fourth prediction result output by the pre-trained second prediction model; determining the first accuracy of the pre-trained first prediction model and the second accuracy of the pre-trained second prediction model through the matching results of the third prediction result and the fourth prediction result with the corresponding real mechanical property label of the third sample steel; determining the first accuracy as the first preset model coefficient, and determining the second accuracy as the second preset model coefficient.

[0053] The above embodiment verifies the trained prediction model through sample data with real mechanical property labels, and uses the accuracy obtained in the verification process as the preset model coefficient of the corresponding model. The prediction results are predicted through multiple models, and the final prediction results are obtained by combining the output results of multiple models to further improve the accuracy of the prediction.

[0054] The following is an exemplary description of the above-mentioned steel mechanical property prediction method with a specific embodiment. In the specific implementation, this embodiment selects the 610L steel grade of automobile beam steel of a domestic hot-rolled plate and strip production line for modeling and analysis. This steel grade has a total of 1022 production data with inspection and test results. In order to facilitate the evaluation of the algorithm effect, 10% is taken as test data, a total of 102 data, 20% is taken as a sample set L with inspection and test results, a total of 205 data, and the remaining 70% is taken without the inspection and test results as an unlabeled sample set U. Models h1 and h2 use the Minkowski distance with p values ​​of 1 and 2 as the measure of the KNN model. The algorithm is executed 200 times, and 239 pseudo labels are added to the sample set L1 corresponding to the model h1, and 239 pseudo labels are added to the sample set L2 corresponding to the model h2, so that L1 and L2 finally contain 444 training data. In the final prediction, both data sets use linear regression models for training and prediction, and the comparison algorithm directly uses the linear regression model for training and prediction on L.

[0055] Test on the test set, the prediction results are as follows Figure 2 and Figure 3 As shown, Figure 2 Schematic diagram of prediction results of a prediction model trained by a conventional machine learning method provided in one embodiment of the present invention, Figure 3 A schematic diagram of prediction results of a prediction model trained by the method provided in an embodiment of the present invention is provided in an embodiment of the present invention. Figure 2 and Figure 3 The horizontal axis is the sample steel identification, and the vertical axis is the tensile strength. Figure 2 and Figure 3 It can be seen that the fit between the true value and the predicted value of the algorithm obtained by the method provided by the above embodiment is significantly better than that of the conventional machine learning algorithm. The mean absolute percentage error MAPE (Mean Absolute Percentage Error) of the algorithm is 2.99%, while the MAPE of the conventional method is 4.19%. The error of the method provided by the embodiment of the present invention is significantly lower than that of the conventional method.

[0056] The method for predicting the mechanical properties of steel proposed in this embodiment obtains the molten steel composition and rolling process parameters of the steel to be predicted during the production process as the characteristic data to be predicted, and then inputs the characteristic data to be predicted into two pre-trained prediction models of different model types to obtain two prediction results, and then comprehensively determines the predicted mechanical property results of the steel to be predicted based on the two prediction results. Through the above method, the mechanical properties of the produced steel can be predicted and evaluated based on the relevant data in the production process to obtain preliminary results. The cost is relatively low and the speed is faster. Once it is found that the mechanical properties of steel do not meet the standards, the production line can be fed back in time for timely adjustment to reduce corporate losses. Accurate prediction of the mechanical properties of steel can help steel companies reduce inspection and testing costs and predict batch performance quality problems in advance, which plays an important role in reducing costs and quality objection rates for steel companies.

[0057] See also Figure 4 , Figure 4 is a structural schematic diagram of a steel mechanical property prediction device provided by an embodiment of the present invention, such as Figure 4 As shown, a steel mechanical property prediction device 400 includes: an acquisition module 401, used to acquire the to-be-predicted characteristic data of the steel to be predicted, the to-be-predicted characteristic data including the molten steel composition and rolling process parameters of the steel to be predicted during the production process; a pre-trained first prediction model 402, used to input the to-be-predicted characteristic data into the pre-trained first prediction model to obtain a first prediction result; a pre-trained second prediction model 403, used to input the to-be-predicted characteristic data into the pre-trained second prediction model to obtain a second prediction result; a result determination module 404, used to determine the predicted mechanical property result of the steel to be predicted based on the first prediction result and the second prediction result; wherein the first prediction model and the second prediction model have different model types, the first prediction model and the second prediction model are obtained by training a sample training data set, the sample training data in the sample training data set includes training sample characteristic data and training sample mechanical property labels, and the training sample characteristic data includes the molten steel composition and rolling process parameters of the sample steel during the production process.

[0058] As an example, the acquisition module can be composed of a large number of IoT sensors to collect various process parameters and performance test results in the steel production process.

[0059] In one embodiment, the device also includes a data processing and model building module, which is deployed on the server and is used for preprocessing the collected data, selecting features, and building a performance prediction model.

[0060] In one embodiment, the device further includes a performance prediction module, which is deployed on a server and is used to generate pseudo labels and predict the mechanical properties of the produced steel products.

[0061] In one embodiment, the steel mechanical property prediction device further includes a sample expansion module, which is used to obtain a first sample data subset before inputting the feature data to be predicted into a pre-trained first prediction model to obtain a first prediction result, and inputting the feature data to be predicted into a pre-trained second prediction model to obtain a second prediction result, wherein the first sample data in the first sample data subset includes the first sample feature data of the first sample steel and the real mechanical property label of the first sample steel, and the first sample feature data includes the molten steel composition and rolling process parameters of the first sample steel in the production process; if the number of the first sample data in the first sample data subset is less than the preset real sample number threshold, obtain a second sample data subset, wherein the second sample data in the second sample data subset includes the second sample feature data, and the second sample feature data includes the molten steel composition and rolling process parameters of the second sample steel in the production process; perform a first training on the first initial model and the second initial model respectively through the first sample data subset to obtain a first initial model after the first training and a second initial model after the first training; input at least part of the second sample feature data in the second sample data subset into the first initial model after the first training and the second initial model after the first training respectively. Second initial model, obtain the first predicted mechanical property label output by the first initial model after the first training, and the second predicted mechanical property label output by the second initial model after the first training; determine the first predicted utility value of each first predicted mechanical property label, and determine the second predicted utility value of each second predicted mechanical property label; determine the first predicted mechanical property label whose first predicted utility value is greater than a preset utility threshold as a first sample mechanical property pseudo label, generate a first pseudo sample training sub-data set based on the first sample mechanical property pseudo-label and the second sample feature data corresponding to the first sample mechanical property pseudo-label, and generate a first sample training data subset for training the first prediction model based on the first pseudo sample training sub-data set and the first sample data subset; determine the second predicted mechanical property label whose second predicted utility value is greater than the preset utility threshold as a second sample mechanical property pseudo-label, generate a second pseudo sample training sub-data set based on the second sample mechanical property pseudo-label and the second sample feature data corresponding to the second sample mechanical property pseudo-label, and generate a second sample training data subset for training the second prediction model based on the second pseudo sample training sub-data set and the first sample data subset; generate the sample training data set based on the first sample training data subset and the second sample training data subset.

[0062] In one embodiment, the sample expansion module is also used to select a preset number of second sample feature data from the second sample data subset, and input the selected second sample feature data into the trained first initial model and the trained second initial model respectively; if the number of the first sample mechanical property pseudo-labels is less than the preset pseudo-sample number threshold, and the number of the second sample mechanical property pseudo-labels is less than the preset pseudo-sample number threshold, re-select a preset number of second sample feature data from the second sample data subset, and input the selected second sample feature data into the trained first initial model and the trained second initial model respectively, until the number of re-selected second sample feature data is greater than the preset number threshold, or the number of the first sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold, or the number of the second sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold; if the number of the first sample mechanical property pseudo-labels is less than the preset pseudo-sample number If the number of the second sample mechanical property pseudo-labels is less than the preset pseudo-sample number threshold, and the number of the first sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold, a preset number of second sample feature data is reselected from the second sample data subset, and the selected second sample feature data is input into the trained first initial model until the number of reselected second sample feature data is greater than the preset number threshold, or the number of the first sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold; if the number of the second sample mechanical property pseudo-labels is less than the preset pseudo-sample number threshold, and the number of the first sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold, a preset number of second sample feature data is reselected from the second sample data subset, and the selected second sample feature data is input into the trained second initial model until the number of reselected second sample feature data is greater than the preset number threshold, or the number of the second sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold.

[0063] In one embodiment, the result determination module is configured to obtain a first preset model coefficient of the pre-trained first prediction model and a second preset model coefficient of the pre-trained second prediction model; and perform weighted averaging processing on the first prediction result and the second prediction result according to the first preset model coefficient and the second preset model coefficient, respectively, to obtain the predicted mechanical properties result of the steel to be predicted.

[0064] In one embodiment, the result determination module is further configured to obtain a third sample data subset, wherein the third sample data in the third sample data subset includes third sample characteristic data of a third sample steel and a real mechanical property label of the third sample steel, wherein the third sample characteristic data includes molten steel composition and rolling process parameters of the third sample steel during production; the third sample characteristic data in the third sample data subset is respectively input into a pre-trained first prediction model and a pre-trained second prediction model to obtain a third prediction result output by the pre-trained first prediction model and a fourth prediction result output by the pre-trained second prediction model; a first accuracy rate of the pre-trained first prediction model is determined through matching results of the third prediction result and the fourth prediction result with the corresponding real mechanical property label of the third sample steel, and a second accuracy rate of the pre-trained second prediction model is determined; the first accuracy rate is determined as the first preset model coefficient, and the second accuracy rate is determined as the second preset model coefficient.

[0065] In this embodiment, the steel mechanical property prediction device is essentially provided with multiple modules for executing the steel mechanical property prediction method in the above embodiment. The specific functions and technical effects can be referred to the above embodiment and will not be described again here.

[0066] See also Figure 5 An embodiment of the present invention further provides an electronic device 1000, including a processor 1001, a memory 1002 and a communication bus 1003; the communication bus 1003 is used to connect the processor 1001 and the memory 1002; the processor 1001 is used to execute a computer program stored in the memory 1002 to implement one or more of the methods described in the above embodiments.

[0067] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. The computer program is used to enable a computer to execute the method described in any one of the above embodiments.

[0068] The embodiment of the present application also provides a non-volatile readable storage medium, which stores one or more modules (programs). When the one or more modules are applied to a device, the device can execute instructions (instructions) of the steps included in the embodiment 1 of the embodiment of the present application.

[0069] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0070] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0071] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0072] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0073] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A method for predicting the mechanical properties of steel, characterized in that: The steel mechanical property prediction method comprises: Acquiring feature data of the steel to be predicted, wherein the feature data includes molten steel composition and rolling process parameters of the steel to be predicted during production; Inputting the feature data to be predicted into a pre-trained first prediction model to obtain a first prediction result, and inputting the feature data to be predicted into a pre-trained second prediction model to obtain a second prediction result, wherein the first prediction model and the second prediction model are of different model types, and the first prediction model and the second prediction model are trained by a sample training data set, and the sample training data in the sample training data set include training sample feature data and training sample mechanical property labels, and the training sample feature data include molten steel composition and rolling process parameters of the sample steel during the production process; The predicted mechanical property result of the steel to be predicted is determined based on the first prediction result and the second prediction result.

2. The method for predicting the mechanical properties of steel according to claim 1, characterized in that: Before inputting the feature data to be predicted into a pre-trained first prediction model to obtain a first prediction result, and inputting the feature data to be predicted into a pre-trained second prediction model to obtain a second prediction result, the method includes: Acquire a first sample data subset, wherein the first sample data in the first sample data subset includes first sample characteristic data of a first sample steel and a real mechanical property label of the first sample steel, wherein the first sample characteristic data includes molten steel composition and rolling process parameters of the first sample steel during production; If the number of first sample data in the first sample data subset is less than a preset real sample number threshold, obtaining a second sample data subset, wherein the second sample data in the second sample data subset includes second sample characteristic data, and the second sample characteristic data includes molten steel composition and rolling process parameters of the second sample steel in the production process; Performing a first training on the first initial model and the second initial model respectively using the first sample data subset to obtain a first initial model after the first training and a second initial model after the first training; Inputting at least part of the second sample feature data in the second sample data subset into the first initial model after the first training and the second initial model after the first training, respectively, to obtain a first predicted mechanical property label output by the first initial model after the first training, and a second predicted mechanical property label output by the second initial model after the first training; determining a first predicted utility value for each first predicted mechanical property signature, and determining a second predicted utility value for each second predicted mechanical property signature; Determine a first predicted mechanical property label whose first predicted utility value is greater than a preset utility threshold as a first sample mechanical property pseudo label, generate a first pseudo sample training sub-data set based on the first sample mechanical property pseudo label and second sample feature data corresponding to the first sample mechanical property pseudo label, and generate a first sample training data subset for training the first prediction model based on the first pseudo sample training sub-data set and the first sample data subset; Determine a second predicted mechanical property label whose second predicted utility value is greater than a preset utility threshold as a second sample mechanical property pseudo label, generate a second pseudo sample training sub-data set based on the second sample mechanical property pseudo label and the second sample feature data corresponding to the second sample mechanical property pseudo label, and generate a second sample training data subset for training the second prediction model based on the second pseudo sample training sub-data set and the first sample data subset; The sample training data set is generated based on the first sample training data subset and the second sample training data subset.

3. The method for predicting the mechanical properties of steel according to claim 2, characterized in that: Inputting at least part of the second sample feature data in the second sample data subset into the trained first initial model and the trained second initial model respectively, comprising: Selecting a preset number of second sample feature data from the second sample data subset, and inputting the selected second sample feature data into the trained first initial model and the trained second initial model respectively; If the number of the first sample mechanical property pseudo labels is less than the preset pseudo sample number threshold, and the number of the second sample mechanical property pseudo labels is less than the preset pseudo sample number threshold, reselect a preset number of second sample feature data from the second sample data subset, and input the selected second sample feature data into the trained first initial model and the trained second initial model respectively, until the number of reselected second sample feature data is greater than the preset number threshold, or the number of the first sample mechanical property pseudo labels is greater than or equal to the preset pseudo sample number threshold, or the number of the second sample mechanical property pseudo labels is greater than or equal to the preset pseudo sample number threshold; If the number of the first sample mechanical property pseudo labels is less than a preset pseudo sample number threshold, and the number of the second sample mechanical property pseudo labels is greater than or equal to a preset pseudo sample number threshold, reselect a preset number of second sample feature data from the second sample data subset, and input the selected second sample feature data into the trained first initial model until the number of reselected second sample feature data is greater than a preset number threshold, or the number of the first sample mechanical property pseudo labels is greater than or equal to a preset pseudo sample number threshold; If the number of the second sample mechanical property pseudo-labels is less than the preset pseudo-sample number threshold, and the number of the first sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold, reselect a preset number of second sample feature data from the second sample data subset, and input the selected second sample feature data into the trained second initial model until the number of reselected second sample feature data is greater than the preset number threshold, or the number of the second sample mechanical property pseudo-labels is greater than or equal to the preset pseudo-sample number threshold.

4. The method for predicting the mechanical properties of steel according to claim 2, characterized in that: Determining a first predicted utility value for each first predicted mechanical property signature, and determining a second predicted utility value for each second predicted mechanical property signature, comprises: Add all first predicted mechanical property labels and all second sample feature data corresponding to the first predicted mechanical property labels to the first sample data subset to obtain a first sample expansion data set, and perform a second training on the second initial model after the first training using the first sample expansion data set to obtain a second initial model after the second training; determine the first k-nearest neighbor sample feature data of the second sample feature data corresponding to the first predicted mechanical property label from all the first sample feature data in the first sample data subset, and input the first k-nearest neighbor sample feature data into the second initial model after the first training and the second initial model after the second training, respectively, to obtain a first sample predictive label and a second sample predictive label; determine the first prediction utility value based on the degree of label difference between the first sample predictive label and the second sample predictive label and the real mechanical property label corresponding to the first k-nearest neighbor sample feature data in the first sample data subset; All second predicted mechanical property labels and all second sample feature data corresponding to the second predicted mechanical property labels are added to the first sample data subset to obtain a second sample expansion data set, and the first initial model after the first training is trained for a second time using the second sample expansion data set to obtain the first initial model after the second training; the second k-nearest neighbor sample feature data of the second sample feature data corresponding to the second predicted mechanical property label is determined from all the first sample feature data in the first sample data subset, and the second k-nearest neighbor sample feature data is input into the first initial model after the first training and the first initial model after the second training, respectively, to obtain a third sample predictive label and a fourth sample predictive label; the second prediction utility value is determined based on the degree of label difference between the third sample predictive label and the fourth sample predictive label and the real mechanical property label corresponding to the second k-nearest neighbor sample feature data in the first sample data subset.

5. The method for predicting the mechanical properties of steel according to claim 2, characterized in that: Determining a first predicted utility value for each first predicted mechanical property signature, and determining a second predicted utility value for each second predicted mechanical property signature, comprises: Among them, utility 1_i is the first predicted utility value of the first predicted mechanical property label i, x 1k is the first k-nearest neighbor sample feature data of the second sample feature data corresponding to the first predicted mechanical property label i in all the first sample feature data of the first sample data subset, N1 is the set of the first k-nearest neighbor sample feature data of the second sample feature data corresponding to all the first predicted mechanical property labels in all the first sample feature data of the first sample data subset, y 1k is the first k-nearest neighbor sample feature data x 1k The corresponding true mechanical property label in the first sample data subset, h1 is the second initial model after the first training, h1(x 1k ) is the predictive label of the first sample, h'1 is the second initial model after the second training, h'1(x 1k ) is the predictive label of the second sample; utility 2_i is the second predicted utility value of the second predicted mechanical property label i, x 2k is the second k-nearest neighbor sample feature data of the second sample feature data corresponding to the second predicted mechanical property label i among all the first sample feature data of the first sample data subset, N2 is the set of the second k-nearest neighbor sample feature data of the second sample feature data corresponding to all the second predicted mechanical property labels among all the first sample feature data of the first sample data subset, y 2k is the second k-nearest neighbor sample feature data x 2k The corresponding true mechanical property label in the first sample data subset, h2 is the first initial model after the first training, h2(x 2k ) is the predictive label of the third sample, h'2 is the first initial model after the second training, h'2(x 2k ) is the predictive label of the fourth sample.

6. The method for predicting the mechanical properties of steel according to any one of claims 1 to 5, characterized in that: Determining the predicted mechanical property result of the steel to be predicted based on the first prediction result and the second prediction result includes: Acquire a first preset model coefficient of the pre-trained first prediction model and a second preset model coefficient of the pre-trained second prediction model; The first prediction result and the second prediction result are respectively weighted averaged according to the first preset model coefficient and the second preset model coefficient to obtain the predicted mechanical property result of the steel to be predicted.

7. The method for predicting the mechanical properties of steel according to claim 6, characterized in that: Before obtaining the first preset model coefficient of the pre-trained first prediction model and the second preset model coefficient of the pre-trained second prediction model, the method further includes: Acquire a third sample data subset, wherein the third sample data in the third sample data subset includes third sample characteristic data of the third sample steel and a real mechanical property label of the third sample steel, wherein the third sample characteristic data includes molten steel composition and rolling process parameters of the third sample steel during the production process; Inputting the third sample feature data in the third sample data subset into the pre-trained first prediction model and the pre-trained second prediction model respectively, to obtain the third prediction result output by the pre-trained first prediction model and the fourth prediction result output by the pre-trained second prediction model; Determine a first accuracy rate of the pre-trained first prediction model through a matching result between the third prediction result and the fourth prediction result and a corresponding real mechanical property label of the third sample steel, and determine a second accuracy rate of the pre-trained second prediction model; The first accuracy rate is determined as the first preset model coefficient, and the second accuracy rate is determined as the second preset model coefficient.

8. A device for predicting mechanical properties of steel, characterized in that: The steel mechanical properties prediction device: An acquisition module, used for acquiring characteristic data of the steel to be predicted, wherein the characteristic data includes molten steel composition and rolling process parameters of the steel to be predicted during the production process; A pre-trained first prediction model, used to input the feature data to be predicted into the pre-trained first prediction model to obtain a first prediction result; A pre-trained second prediction model, used for inputting the feature data to be predicted into the pre-trained second prediction model to obtain a second prediction result; A result determination module, used for determining the predicted mechanical property result of the steel to be predicted based on the first prediction result and the second prediction result; Among them, the model types of the first prediction model and the second prediction model are different, and the first prediction model and the second prediction model are obtained by training a sample training data set. The sample training data in the sample training data set includes training sample feature data and training sample mechanical property labels. The training sample feature data includes the molten steel composition and rolling process parameters of the sample steel during the production process.

9. An electronic device, characterized in that: including a processor, a memory and a communication bus; The communication bus is used to connect the processor and the memory; The processor is configured to execute the computer program stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is used to make a computer execute the method according to any one of claims 1 to 7.

Images