JP2006233312A - Method for controlling blowing in converter - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method for controlling blowing in a converter with which the prediction of flux quantity for matching components in molten steel to the target values, can accurately be performed. <P>SOLUTION: When the flux quantity in the control of blowing in the converter is decided, a blowing conditional vector composed of physical quantities etc., indicating the peculiarity of blowing in the converter, is decided and the past blowing actual result vector resembled to the planned blowing conditional vector, is selected from the past blowing actual result data-base. A flux quantity calculating model for assuming the flux quantity is made based on the selected blowing actual result vector, and the flux quantity predicting value for the planned blowing condition is calculated from the flux quantity calculating model. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a converter blowing control method in which the molten steel temperature at the end of the converter blowing process and each component are matched with the end point target values.

  Conventionally, the calculation of the amount of solvent (lime content) in the blowing of converters is necessary for the refining reaction from the static material balance calculation in order to match the molten steel components such as C, P, Si, and Mn to the target values. I was asking for the amount. However, in the converter blowing control based on the amount of solvent obtained in the above calculation, the operational fluctuations and changes over time cannot be expressed sufficiently in the model, so that sufficient control performance cannot be obtained, that is, the molten steel component is set to the target value. It was difficult to match.

On the other hand, for example, as disclosed in Patent Document 1, a method of correcting a theoretical formula using a learning term or the like is disclosed. The calculation by the static model, for example x _1, x _2, ..., when the x _n and blowing conditions, medium solvent content y is a function of these blowing conditions, i.e. the function f (x _1, x _2, ..., x _n ) Can be described by the following expression (1).

y = f (x ₁ , x ₂ , ..., x _n ) (1)
In this method, a model correction term Δa based on results is added to the equation (1) and rewritten as the following equation (2).

y = f (x ₁ , x ₂ , ..., x _n , Δa) (2)
The learning term Δa is updated every time blowing is completed.
JP 2000-309817 A

    However, in the technique described in Patent Document 1, the correction based on the learning term Δa is centered on the correction based on the previous blowing performance, and therefore, when the blowing conditions are greatly different, the error of the calculated amount of solvent is large. There is a problem that sufficient converter blowing control performance cannot be exhibited.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a converter blowing control method capable of accurately predicting the amount of solvent for making molten steel components coincide with target values. And

  In the invention according to claim 1 of the present invention, when determining the amount of solvent in the converter blowing control, a blowing condition vector consisting of physical quantities representing the characteristics of converter blowing is determined, and from the past blowing record database, Select a past blowing performance vector similar to the planned blowing condition vector, and create a solvent amount calculation model for estimating the amount of solvent based on the selected blowing performance vector. From this, it is a converter blowing control method characterized in that a predicted amount of solvent for the blowing conditions to be implemented is calculated.

  Further, in the invention according to claim 2 of the present invention, when determining the amount of the solvent in the converter blowing control, a blowing condition vector consisting of a physical quantity representing the characteristics of the converter blowing is determined, and a past blowing record database is used. Then, a past blowing performance vector similar to the scheduled blowing condition vector is selected, and a weighting coefficient corresponding to the distance between the selected blowing performance vector and the scheduled blowing condition vector is obtained. It is a converter blowing control method characterized in that a predicted solvent amount for a blowing condition to be implemented is calculated from a load sum of actual solvent amount using a coefficient.

  Further, in the invention according to claim 3 of the present invention, in the converter blowing control method according to claim 1 or 2, selection of a past blowing performance vector similar to a scheduled blowing condition vector is A converter blowing control method characterized in that a norm of a difference between a scheduled blowing condition vector and a past blowing performance vector is calculated, and the calculated norm is selected in ascending order.

  The invention according to claim 4 of the present invention is the converter blowing control method according to claim 1 or claim 3, wherein the solvent amount calculation model is a regression equation based on the selected blowing performance vector. It is a converter blowing control method characterized by being a model.

  According to the present invention, not only the effect of the previous blowing but also the prediction from past blowing groups with similar blowing conditions, the solvent calculation is accurately performed and the molten steel component accuracy is improved. Can improve product quality. Furthermore, this also provides the effect of extending the life of the refractory and reducing the load of secondary refining.

As described above, the determination of the amount of the solvent to be carried out before the converter blowing is performed mainly on the calculation of the material balance in the molten steel and slag. The solvent amount calculation is calculated by an equilibrium equation of the P ₂ O ₅ concentration present in the slag and the P concentration present in the molten steel, a conservation law of the P amount charged into the converter, and a slag amount estimation equation.

  The present invention is a method for creating an approximate model for directly calculating the amount of solvent, and in this case, past blowing similar to the calculation target blowing is collected, and an appropriate model is constructed from the collected results. In the present invention, this model construction and one of its learning methods are given. The technique is a technique in which the neighborhood of the blowing condition is determined and a model established in the neighborhood is constructed each time.

The best mode for carrying out the present invention will be described below with reference to the drawings and mathematical expressions. FIG. 1 is a flowchart showing an example of a processing procedure according to the present invention.
First, an n-dimensional blowing condition vector X (element x _i ) consisting of physical quantities (feet temperature, hot metal amount, hot metal component, target component, etc.) representing the characteristics of blowing is determined (step S01).

X = (x ₁ , x ₂ , x ₃ , ..., x _n ) (3)
Next, a process of extracting similar data from a blowing performance database in which data for a predetermined period is stored is performed. Therefore, first, the average μ _i of each item x _i of the blowing condition vector X and the standard deviation σ _i (i = 1 to n) are calculated from the blowing performance database (step S02). Next, a normalized vector X ^j ′ is created by the conversion shown in the following equation (4) (step S03).

x _i ^j '= (x _i −μ _i ) / σ _i (4)
Then, a blowing condition vector X ⁰ to be executed from now is created, and a normalized blowing condition vector X ⁰ ′ is created using (μ _i , σ _i ) in the same manner as the transformation shown in the equation (4). At this time, the physical values such as the molten steel temperature and components at the end of the treatment use the target values, and the auxiliary raw materials (lime, dolomite, coke, iron ore, Mn ore, etc.) use the scheduled input amounts.

After the above preparation, a normalized vector X ^j ′ similar to the normalized blowing condition vector X ⁰ ′ to be implemented is extracted from the past performance database (step S04). Various methods can be considered for determining the degree of similarity, such as the size of each item x _i ^j ′ of X ^j ′ falling within a predetermined range. A method of defining the similarity by the norm of the vector difference will be described below.

First, a deviation ΔX ′ from the normalized vector X ^j ′ of the database is calculated by the following equation (5) with reference to the normalized vector X ⁰ ′ of blowing that will be performed.

ΔX ′ = X ^j ′ −X ⁰ ′ (5)
Then, the norm of the deviation ΔX ′ is calculated by the following equation (6).

Further, in order to give weight to each item, the norm shown in the following equation (7) in which weight coefficients w _i (i = 1,..., N) are individually introduced may be used.

  Various norms and distance functions (an example will be described later) other than the norm exemplified here may be considered, but they may be applied.

Then, by determining the number of neighbors k of the data, similar data is extracted by collecting k past performance data from the smallest | ΔX ^j ′ |. The number of neighbors k may be determined by Akaike's information criterion, prediction error, cross-validation method, or the like.

Next, a model for calculating the amount of solvent is created using the collected similar blowing data (step S05). The model for determining the amount of solvent is based on the material balance calculation, and various forms are possible. As an example, we introduce a regression model for each item.
Assuming that the amount of solvent is Hd, it is given as a primary expression for each item as shown in the following expression (8).

Hd = f (x ₁ , x ₂ , ..., x _n )
≒ a ₁ * x ₁ + a ₂ * x ₂ + ... + a _n * x _n + b ₀ (8)
Here, the coefficients a ₁ , a ₂ ,..., A _n , b ₀ in the above approximate expression are ^obtained by the least square method or the like using the data X ^{1 to} X ^k collected as the neighborhood data of X ^0. To. Since the coefficient obtained in this manner uses data similar to the blowing data to be performed from now on as the original data, the estimation accuracy of the solvent amount Hd can be increased.

Estimate of medium solvent amount Hd ⁰ for the vector X ⁰ of blowing implementing now can be obtained by substituting each item x ⁰ of X ⁰ to regression equation (8).

The following method does not use the above-described solvent amount calculation model but uses | ΔX ^j '| which is the distance between the blowing vector X ^{0 to be performed} and each neighboring data.

Assume that X ¹ , X ² ,..., X ^k are obtained as neighborhood data vectors. At this time, distances d ¹ , d ² ,..., D ^k between X ⁰ and each vector are obtained as shown in equation (9).

d ^j = | ΔX ^j '| (9)
Here, d _max is the maximum of d ^j (Equation (10)).

d _max = max (d ¹ , d ² ,..., d ^k ) (10)
Further, the weighting coefficient k _j using each d ^j is defined by the following equation (11).

The actual solvent amount corresponding to each X ^j is Hd ^j . The medium solvent amount prediction value Hd ⁰ for the time X ^0, given by the following equation (12).

The above processing / calculation procedure makes it possible to accurately estimate the amount of solvent during blowing based on past similar results.

It is the flowchart which showed the example of the process sequence which concerns on this invention.

Explanation of symbols

S01 Definition of blowing condition vector X S02 Calculation of mean μ _i and standard deviation σ _i S03 Conversion to normalized vector X ′ Extract data X ^{j ′} similar to S04 X ^{0 ′} from past performance database S05 Similar data Using the S06 model and X ⁰ to calculate the solvent / solvent amount prediction value Hd ⁰

Claims (4)

When deciding the amount of solvent in converter blowing control, a blowing condition vector consisting of physical quantities representing the characteristics of converter blowing is determined, and past similar to the scheduled blowing condition vector is determined from the past blowing performance database. A medium solvent amount calculation model is established for estimating the amount of solvent based on the selected blowing performance vector, and the medium for the blowing condition to be implemented is determined from the medium amount calculation model. A converter blowing control method, wherein a predicted solvent amount is calculated. When deciding the amount of solvent in converter blowing control, a blowing condition vector consisting of physical quantities representing the characteristics of converter blowing is determined, and past similar to the scheduled blowing condition vector is determined from the past blowing performance database. A weighting coefficient corresponding to the distance between the selected blowing performance vector and the planned blowing condition vector, and from the load sum of the solvent amount performance using this weighting coefficient. A converter blowing control method, characterized by calculating a predicted solvent amount for a blowing condition to be implemented. In the converter blowing control method according to claim 1 or 2,
The selection of the past blowing performance vector similar to the scheduled blowing condition vector is performed by calculating the norm of the difference between the scheduled blowing condition vector and the past blowing performance vector. A converter blowing control method characterized by selecting in ascending order. In the converter blowing control method of Claim 1 or Claim 3,
The converter blowing method according to claim 1, wherein the solvent amount calculation model is a regression model based on the selected blowing performance vector.

Images