CN-122021288-A - Mechanism and model compensation-based converter carbon content real-time prediction method

CN122021288ACN 122021288 ACN122021288 ACN 122021288ACN-122021288-A

Abstract

The invention discloses a method for predicting the carbon content of a converter in real time based on mechanism and model compensation, which belongs to the technical field of monitoring of converter steelmaking process, and comprises the steps of collecting process data influencing the carbon content of the converter; the method comprises the steps of constructing a mechanism model, calculating a carbon content mechanism simulation value, constructing and training a compensation model based on a preset machine learning algorithm, predicting and obtaining a carbon content compensation value of a corresponding heat by using the compensation model, and adding the carbon content mechanism simulation value and the corresponding carbon content compensation value to obtain the final carbon content of the converter. The converter carbon content real-time prediction method effectively combines the clear physical meaning of the mechanism model and the environmental adaptability of the data driving method, and improves the description capability of the model to real working conditions, thereby improving the accuracy of the model to terminal control.

Inventors

FU MEIXIA
JI JINGFENG
LI WEI
WANG JIANQUAN
LI SHENG
HE JIANJUN
WANG HONGBING
WANG QU

Assignees

北京科技大学

Dates

Publication Date: 20260512
Application Date: 20260123

Claims (8)

1. A converter carbon content real-time prediction method based on mechanism and model compensation is characterized by comprising the following steps: Collecting process data influencing the carbon content of a converter in the steelmaking decarburization process of the converter; constructing a mechanism model for describing decarburization rate in the converter steelmaking decarburization process, and calculating the carbon content of the converter in real time based on the mechanism model and combining the process data to obtain a carbon content mechanism simulation value; Constructing and training a compensation model based on a preset machine learning algorithm, wherein training data of the compensation model takes process data and a carbon content mechanism simulation value as samples and takes a carbon content compensation value as a label, wherein the carbon content compensation value refers to a difference value between a carbon content true value and the carbon content mechanism simulation value; based on the compensation model, combining the process data and the carbon content mechanism simulation value, and predicting to obtain a carbon content compensation value of the corresponding heat; and adding the carbon content mechanism simulation value and the corresponding carbon content compensation value to obtain the final carbon content of the converter.
2. The method for predicting carbon content of a converter in real time based on mechanism and model compensation according to claim 1, wherein the process data comprises static data and dynamic data, wherein, The static data comprise furnace charging temperature, scrap weight, semisteel weight, furnace age, slag quantity, total converting time, slag melting dosage and initial carbon content; the dynamic data includes lance height at each time and oxygen blow data accumulated over time.
3. The method for predicting the carbon content of a converter in real time based on mechanism and model compensation according to claim 1, wherein the constructing a mechanism model for describing the decarburization rate in the decarburization process of steelmaking of the converter comprises: Dividing the converter steelmaking decarburization process into three different stages in sequence, and respectively constructing a mechanism model corresponding to each stage, wherein the first stage is a stage in which the decarburization rate gradually increases, the second stage is a stage in which the decarburization rate reaches and maintains the maximum value, and the third stage is a stage in which the decarburization rate gradually decreases; Correspondingly, when the converter carbon content is calculated in real time based on the mechanism model and in combination with the process data, the mechanism model of the corresponding stage is selected for calculation according to the stage corresponding to the process data.
4. The method for predicting the carbon content of the converter in real time based on mechanism and model compensation according to claim 3, wherein the inflection point between the first stage and the second stage is determined according to the total oxygen blowing time and the charging temperature, and the ending time of the first stage is between 28% and 45% of the total oxygen blowing time; The inflection point between the second stage and the third stage is determined according to the furnace charging temperature and the furnace age.
5. The method for predicting carbon content of converter in real time based on mechanism and model compensation as claimed in claim 4, wherein the end time of the first stage The calculation formula of (2) is as follows: ; Wherein, the Is the total oxygen blowing time; , in order to achieve the temperature of the furnace entering, Is the reference temperature.
6. The method for predicting the carbon content of a converter in real time based on mechanism and model compensation as claimed in claim 3, wherein the mechanism model in the first stage Expressed as: ; Mechanism model of the second stage Expressed as: ; Mechanism model of third stage Expressed as: ; Wherein, the In order to achieve the decarburization factor, , In order to achieve the utilization rate of oxygen, Is made of the weight of the semi-steel, As the height of the reference oxygen lance, The height of the oxygen lance at the time t, The oxygen blowing rate at the time t is, As a reference to the oxygen blowing rate, Is used as a scrap steel factor, and the steel is processed into a steel product, , In order to obtain the weight of the scrap steel, In order to obtain the temperature factor of the furnace, , In order to achieve the temperature of the furnace entering, As a result of the reference temperature, In order to provide the additive factor(s), , B is a stage adjustment factor; Is the decarburization coefficient of the first stage; Is the initial carbon content; Is the carbon content of the last moment; is the decarburization coefficient of the second stage; carbon content corresponding to inflection points between the second stage and the second stage; representing a parameter related to the ideal phase time, , For the end time of the first phase, Is the end time of the second stage; Is the decarburization coefficient of the third stage; Is the target carbon content; carbon content corresponding to the inflection point between the second stage and the third stage; for the correction parameters relating to the total converting time, for correcting the decarburization tendency of the third stage, , Is the total oxygen blowing time.
7. The method for predicting the carbon content of the converter in real time based on mechanism and model compensation according to claim 2, wherein the process of the compensation model on input data comprises the following steps: the dynamic data and the carbon content mechanism simulation value are spliced and then input into a two-way long-short-term memory network, and the dynamic data and the carbon content mechanism simulation value are processed by adopting the two-way long-short-term memory network to obtain the dynamic characteristic data; and after the static characteristic data and the dynamic characteristic data are fused in the time dimension, inputting the static characteristic data and the dynamic characteristic data into a multi-layer perceptron model, and outputting carbon content compensation values of corresponding furnace times by using the multi-layer perceptron model.
8. The method for predicting carbon content of a converter in real time based on mechanism and model compensation as set forth in claim 7, wherein said compensation model has a loss function Expressed as: ; Wherein, the Representing a weight factor for making the prediction error weight in the later stage of converting larger; representing the predicted carbon content of the ith sample at time t, And (3) representing the real carbon content of the ith sample at the time t, wherein p is a constant, if a missing value occurs at a certain time, p is taken as 0, otherwise p is taken as 1;N, and the total sample amount is taken.

Description

Mechanism and model compensation-based converter carbon content real-time prediction method Technical Field The invention relates to the technical field of monitoring of converter steelmaking processes, in particular to a method for predicting the carbon content of a converter in real time based on mechanism and model compensation. Background Since the industrial revolution, the iron and steel industry has been a basic stone of national economy, and has been developing across from rail laying steel to high-end equipment manufacturing steel. With the development of technology, the requirements for steel in the advanced fields of automobiles, aerospace, bridges and the like are continuously increased, the quality requirements for steel are also continuously improved, and the steel making technology needs to be improved. Oxygen top-blown converter steelmaking is currently becoming the mainstream steelmaking mode of global iron and steel enterprises because of the advantages of high decarburization efficiency, short production cycle, lower cost and the like. Converter steelmaking is a complex process carried out under high temperature and high pressure conditions, involving multiple influencing factors and intense physicochemical reactions. The modern converter steelmaking is to prepare molten steel with the temperature and chemical components meeting the requirements of various steel types by mixing and melting the blast furnace molten iron, scrap steel, iron slag and other raw materials according to a certain proportion. The carbon content and the temperature of the steelmaking end point are key indexes for evaluating the quality of the molten steel at the end point. Currently, the mainstream converter steelmaking process is an oxygen top-blown process, in which an oxygen lance is installed at the top of a converter, and oxygen is continuously fed into the converter. The injected oxygen reacts with elements such as silicon, manganese, phosphorus, sulfur, carbon and the like in the molten steel to generate slag or gas byproducts, thereby obtaining the molten steel with the chemical components meeting the requirements. Along with the continuous improvement of the control precision requirement of the converter steelmaking end point, various end point judging and controlling methods are generated. Early days, the prior manual operation mode is mainly relied on, for example, a steel-making craftsmen with abundant experience judges whether the steelmaking end point is reached or not by visually observing the color, the shape and the change trend of the flame at the furnace mouth. The method has certain flexibility, almost no equipment investment is needed, the operation is simple and convenient, however, the accuracy is greatly influenced by the experience level of operators and the fluctuation of the working conditions on site, the subjectivity is strong, and the stability is poor. Then, in order to improve the objectivity and accuracy of judgment, a method based on the analysis of the components of the furnace gas is developed gradually. The method is used for monitoring the change of the concentration of gases such as CO, CO 2 and the like in the flue gas in real time in the steelmaking process, so that the change trend of key variables such as carbon content and the like is indirectly reflected. Although the method has certain real-time performance and precision, the required analysis equipment is expensive, and the monitoring data has certain hysteresis, so that the control requirement of quick response is difficult to meet. In order to make up for the shortages of the response speed of the method, the industry introduces a sublance detection technology. In the key stage of blowing process, especially near the end point or when the decarburization rate is obviously changed, the auxiliary gun is inserted into the furnace to measure the key parameters of molten steel temperature, carbon content and the like in real time. The method greatly improves the instantaneity and the accuracy of the data, but as the sublance can only acquire the point location information of the local area, the overall state of the whole reaction in the furnace is difficult to comprehensively reflect, and the equipment cost and the maintenance cost are higher, so that the wide application of the method is limited. In recent years, with the rapid development of artificial intelligence and data driving technology, an intelligent final judgment method based on mechanism and data fusion gradually becomes a research hot spot. The method comprises the steps of combining material balance and a thermodynamic model to establish a process mechanism judging model, utilizing a fire hole flame image to perform visual identification analysis, and training a machine learning and deep learning model based on historical fire condition data. The method has strong interpretability and automation level, has small dependence on hardware and has good populariz