CN-121980262-A - Soft measurement method for ferrous oxide content of sinter and related device

CN121980262ACN 121980262 ACN121980262 ACN 121980262ACN-121980262-A

Abstract

The invention discloses a soft measurement method for ferrous oxide content of a sinter and a related device. The method comprises the steps of obtaining time sequence data and image data of different sinter samples under different sintering stage conditions, constructing a labeled data set and a non-labeled data set, dividing the labeled data set into a training set, a verification set and a test set according to preset proportion, adding the non-labeled data set into the training set as expansion, carrying out iterative training and evaluation on MTFIF-STMM models based on the training set, the verification set and the test set to obtain a final MTFIF-STMM model, inputting sinter multi-source data acquired in real time into the final MTFIF-STMM model for soft measurement, and obtaining a real-time evaluation result of the ferrous oxide content of the sinter. The invention realizes the real-time soft measurement of the content of the ferrous oxide in the sintering process and improves the modeling precision.

Inventors

XIONG HUI
YANG XIBING
YANG CHONG
WEI LI
LIU JINYI
FU LIRONG

Assignees

海南大学

Dates

Publication Date: 20260505
Application Date: 20260112

Claims (9)

1. A soft measurement method for the ferrous oxide content of a sinter is characterized by comprising the following steps: acquiring time sequence data and image data of different sinter samples under different sintering stage conditions, constructing a labeled data set and a non-labeled data set, dividing the labeled data set into a training set, a verification set and a test set according to a preset proportion, and adding the non-labeled data set into the training set as expansion; Inputting a training set into a MTFIF-STMM model for iterative training, optimizing the model by minimizing the similarity loss of two branch time sequence characteristics for unlabeled data, optimizing the model by a mean square error loss and an integral loss function combined by the similarity loss of the two branch time sequence characteristics for labeled data, updating model parameters by adopting a counter propagation mechanism, selecting an optimal MTFIF-STMM model by utilizing a verification set, performing performance evaluation on the optimal MTFIF-STMM model by utilizing the test set until the performance evaluation meets a set threshold value, and obtaining a final MTFIF-STMM model; and inputting the multi-source data of the sinter collected in real time into a final MTFIF-STMM model for soft measurement, and obtaining a real-time evaluation result of the ferrous oxide content of the sinter.
2. The method for soft measurement of ferrous oxide content in sinter as claimed in claim 1, wherein the MTFIF-STMM model comprises a multi-source data feature encoder, a feature interactor, and a decoding and predictor, wherein, In the multi-source data feature encoder, a transducer encoder is utilized to encode time sequence data to obtain features Extracting and encoding the image data by using MobileViT module and Semi-MobileViT module to obtain the features ; In the feature interactors, the features are respectively connected by using a full connection layer And features After downsampling, performing feature interaction by using cosine similarity to obtain features And features The features are And features The full connection layer is respectively utilized to reconstruct and restore the characteristics into the original dimension, and the characteristics are obtained And features The features are And features Respectively and characteristic of And features Residual connection to obtain characteristics And features And combining the features And features Performing feature fusion to obtain multi-mode fusion features; And in the decoding and predicting device, decoding and predicting the multi-mode fusion characteristic to obtain a predicting result.
3. The method for soft measurement of ferrous oxide content in sinter as claimed in claim 2, wherein the image data is extracted and encoded by MobileViT module and Semi-MobileViT module to obtain features Extracting features of the image data through the convolution layer to obtain initial features; Extracting the initial feature by using MobileNetV style of inverted residual block to obtain local feature, expanding the local feature into sequence by MobileViT module, inputting lightweight transform to make global relation modeling, and folding back to space feature to obtain feature , 、 And The height, width and channel number of the feature map input to the Semi-MobileViT module are respectively shown; the Semi-MobileViT module follows the design of MobileViT module from local characterization to global characterization, and features the Conversion to , 、 And Respectively representing the number of pixel points, the number of patches and the number of channels in each patch, adopting a compression-excitation module to carry out the following steps Conversion to 。
4. The method for soft measurement of ferrous oxide content of sinter as set forth in claim 2, wherein the decoding and predictor includes a plurality of decoding layers of the same structure stacked, the decoding layers including two multi-headed attention sublayers and a point-by-point feed-forward network sublayer.
5. The method for soft measurement of ferrous oxide content in sinter as claimed in claim 2, wherein the mean square error loss The definition is: In the formula, For the tag value of sample i, The predicted value of the sample i, and n is the number of samples.
6. The method for soft measurement of ferrous oxide content in sinter as claimed in claim 5, wherein the similarity between the two branched timing characteristics is lost The definition is: In the formula, cosine similarity of two branch time sequence characteristics is expressed as Entropy regularization constraint is expressed as , wherein, Is a regularization coefficient; Representing the probability of a distribution of cosine similarities, Is that Is the first of (2) The elements. The overall loss function is: In the formula, Is that For adjusting weights of (2) And (3) with The ratio between them.
7. A soft measurement system for ferrous oxide content of sinter, comprising: The acquisition module is used for acquiring time sequence data and image data of different sinter samples under different sintering stage conditions, constructing a labeled data set and a non-labeled data set, dividing the labeled data set into a training set, a verification set and a test set according to a preset proportion, and adding the non-labeled data set into the training set as an extension; The training module is used for inputting a training set into the MTFIF-STMM model for iterative training, optimizing the model by minimizing the similarity loss of the two-branch time sequence characteristics for the unlabeled data, optimizing the model by a mean square error loss and an integral loss function combined by the similarity loss of the two-branch time sequence characteristics for the labeled data, updating model parameters by adopting a back propagation mechanism, selecting an optimal MTFIF-STMM model by utilizing a verification set, and performing performance evaluation on the optimal MTFIF-STMM model by utilizing the test set until the performance evaluation meets a set threshold value, thereby obtaining a final MTFIF-STMM model; And the prediction module is used for inputting the multi-source data of the sintering ore acquired in real time into a final MTFIF-STMM model for soft measurement, and obtaining a real-time evaluation result of the ferrous oxide content of the sintering ore.
8. A computer device comprising a memory for storing a computer program, and a processor for implementing the method according to any one of claims 1 to 6 when the computer program is executed.
9. A readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the method according to any of claims 1 to 6.

Description

Soft measurement method for ferrous oxide content of sinter and related device Technical Field The invention relates to the technical field of soft measurement, in particular to a soft measurement method and a related device for ferrous oxide content of a sinter. Background The sintering process is used as an iron ore pretreatment unit in the blast furnace ironmaking process, provides main raw materials for blast furnaces, is a key link of the steel production, and has direct influence on the reduction efficiency, fuel consumption and molten iron components of the iron ore. The low-quality sinter causes the consequences of increased return rate, reduced yield, increased energy consumption, increased equipment loss and the like. Therefore, the quality index of the sinter is measured in real time and effectively estimated, which is beneficial to the implementation of parameter optimization and control decision in the sintering process, so that the method has important significance and value in stabilizing the operation of the blast furnace, improving the production efficiency, reducing the consumption of raw materials and energy sources and improving the quality and economic benefit of the product. However, various important quality indexes of the sinter, including chemical components, physical properties, metallurgical properties and the like, have diversity and complexity, the quality detection of the traditional sinter is obtained by manually performing a series of physicochemical experiments, the speed is low, the hysteresis is strong, the requirement of process strategy adjustment on timeliness cannot be met, and the cost is high. In recent years, data-driven soft measurement models have been widely used in complex problems. The related scholars obtain satisfactory effects in different industrial soft measurement fields through a mechanism mathematical model, machine learning and deep learning method. However, high performance deep learning models rely heavily on the quality, richness, distribution consistency and relevance of training and test data. To ensure reliability of the quality evaluation of the sintered ore, it is necessary to consider multiple measurement data, even multiple heterogeneous data, at the same time, which may cause poor performance or even failure of the model when the relationship between the multiple data and the heterogeneous data cannot be processed. In addition, the manufacturing cost of the agglomerate quality index data label is high, the period is long, and a large amount of multi-source label-free data information cannot be fully utilized. Therefore, how to realize the effective interactive fusion among various data and multi-source heterogeneous data and make full use of the information contained in the rich unlabeled data is important to fine modeling and improving the soft measurement performance of the model. Disclosure of Invention In order to solve the technical problems, the invention provides a soft measurement method and a related device for ferrous oxide content of a sintering ore, which can fuse and consider complementary information and multidimensional information of multi-source heterogeneous data in a sintering process, fully excavate the relationship between different dimensional data and quality indexes of the sintering ore on the basis of multi-mode fusion, and expand and learn unlabeled time sequence feature similarity. In order to achieve the above purpose, the technical scheme of the invention is as follows: A soft measurement method for ferrous oxide content of sinter comprises the following steps: acquiring time sequence data and image data of different sinter samples under different sintering stage conditions, constructing a labeled data set and a non-labeled data set, dividing the labeled data set into a training set, a verification set and a test set according to a preset proportion, and adding the non-labeled data set into the training set as expansion; Inputting a training set into a MTFIF-STMM model for iterative training, optimizing the model by minimizing the similarity loss of two branch time sequence characteristics for unlabeled data, optimizing the model by a mean square error loss and an integral loss function combined by the similarity loss of the two branch time sequence characteristics for labeled data, updating model parameters by adopting a counter propagation mechanism, selecting an optimal MTFIF-STMM model by utilizing a verification set, performing performance evaluation on the optimal MTFIF-STMM model by utilizing the test set until the performance evaluation meets a set threshold value, and obtaining a final MTFIF-STMM model; and inputting the multi-source data of the sinter collected in real time into a final MTFIF-STMM model for soft measurement, and obtaining a real-time evaluation result of the ferrous oxide content of the sinter. Preferably, the MTFIF-STMM model includes a multi-source data feature encoder, a feature inte