JP2014201770A - Estimation apparatus and estimation method of prediction model of converter - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable a prediction model to be estimated, the prediction model deciding a charge amount of auxiliary feedstock in a converter and enabling a further optimal charge amount or the like of the auxiliary feedstock to be obtained.SOLUTION: An estimation apparatus 1 of a prediction model of a converter is used to calculate a charge amount of an auxiliary feedstock to be charged for component adjustment in a converter 2, and is the apparatus that estimates the prediction model of the converter 2 obtained based on an operation result value of the converter 2, and includes: an input part 6 that inputs a generation condition of the prediction model; a model generation part 7 which extracts data so as to coincide with the condition input in the input part 6 and generates two or more prediction models; a model verification part 8 which performs simulation using two or more prediction models obtained in the model generation part 7 and verifies the result of the simulation, thereby estimates the surest prediction model in two or more prediction models and selects the surest prediction model; and an output part 9 which outputs the prediction model selected in the model verification part 8.

Description

  The present invention relates to an apparatus and a method for estimating a prediction model used for calculating an input amount of an auxiliary material input for component adjustment in a converter.

  Conventionally, in hot metal refining performed in a converter, target values are set for each concentration of carbon (C), phosphorus (P), manganese (Mn), etc. in molten steel, and oxygen (gasic acid, solid By adjusting the supply amount of the acid) and the input amount of the auxiliary material, each component is adjusted to satisfy this target value. The input of the auxiliary raw material is performed by considering the cost of the auxiliary raw material in consideration of the quality of the molten steel, but the method of obtaining the input amount is obtained from a long experience of operation.

On the other hand, techniques for determining the input amount of secondary raw materials by applying various mathematical methods and prediction models have been developed (Patent Documents 1 and 2).
For example, the converter blowing control method and the converter blowing control system disclosed in Patent Document 1 include a material balance equation, a heat balance equation, and a distribution ratio estimation that regulates the ratio of the molten steel distributed to the molten steel and slag. Discloses a method and device that optimizes the amount of oxygen blown in and the amount of auxiliary material input, using the formulas, formulas related to operating conditions, etc. as constraints, and distribution that has been insufficient for multiple regression based on conventional physical models Highly accurate estimation is realized by constructing the ratio estimation formula using a regression tree model.

  Further, Patent Document 2 discloses a model parameter determination method, and even when there is little past performance data of an object belonging to the same group as the model object, it is possible to add a model with high accuracy by supplementing other appropriate data. The parameter is obtained.

JP 2009-52109 A JP 2005-36289 A

As described above, in the hot metal refining performed in the converter, various auxiliary materials are input according to the blowing conditions, but the determination of the input amount is very complicated and difficult.
For this reason, it is conceivable to employ the technique of Patent Document 1. However, Patent Document 1 is a technique that can construct a highly accurate distribution ratio estimation formula for data used for estimation by using a regression tree model. However, it is a technology that has the disadvantage that it is difficult to reflect physical and metallurgical laws based on previous experience in the estimation formula. Therefore, in order to achieve higher accuracy, a large amount of data is required for a wide range of operating conditions, but there are few data that are close to operating conditions in high-mix low-volume production, and data with large measurement errors is included. In the range, the model accuracy deteriorates.

  Further, in the model parameter determination method of Patent Document 2, an attempt is made to compensate for the small amount of data by supplementing similar data. In particular, a model in the converter operation (for example, a P distribution estimation model formula or the like) is used. The slag component necessary for obtaining is not always measured every time. Therefore, there is no confirmation that sufficient data can be obtained, data that does not have an observed value corresponding to the model predicted value cannot be used, and the model accuracy under the operating conditions is not guaranteed.

  In other words, due to the characteristics of the converter process, the data for estimating the prediction model may contain a large amount of error, and it is difficult to collect a large amount of data. Predictive models with high accuracy are indispensable for the operation of converters, but the purpose is not to obtain the predicted values of each model with high accuracy, but to optimize the amount of auxiliary material input, which is the final calculation result, etc. It is to obtain a value.

  In view of the above problems, the present invention provides a model estimation technique for constructing a prediction model for determining the input amount of secondary raw material in a converter, and for obtaining a more optimal input amount of secondary raw material. For the purpose.

In order to achieve the above-described object, the present invention takes the following technical means.
The converter predictive model estimation apparatus according to the present invention is used to calculate the input amount of the auxiliary raw material that is input for component adjustment in the converter, and is obtained based on the actual operation value of the converter. An apparatus for estimating a predicted model of a converter, an input unit for inputting a generation condition of the prediction model, and a model generation unit for generating a plurality of prediction models that match the conditions input by the input unit; A model verification unit that estimates and selects the most probable prediction model from the plurality of prediction models by performing simulation using the plurality of prediction models obtained by the model generation unit and verifying the results; And an output unit that outputs a prediction model selected by the model verification unit.

Preferably, the input unit inputs a function used for a prediction model and a parameter range in the prediction model, and the model generation unit determines a plurality of parameter sets used for the prediction model using a PLS method. It is good to be configured.
Preferably, the model verification unit uses the evaluation function that uses a deviation between the input amount obtained by simulation of the prediction model and the actual input amount in the operation to predict the most probable prediction among the plurality of prediction models. It may be configured to estimate and select a model.

  The estimation method of the prediction model of the converter according to the present invention is used to calculate the input amount of the auxiliary material input for component adjustment in the converter, and is obtained based on the operation result value of the converter. A method for estimating a predicted model of a converter, wherein an input step of inputting a generation condition of the prediction model, and a model generation step of generating a plurality of prediction models that match the condition input in the input unit, A model verification step of estimating and selecting a most probable prediction model from the plurality of prediction models by performing simulation using the plurality of prediction models obtained in the model generation step and verifying the result; And an output step of outputting the prediction model selected by the model verification unit.

Preferably, in the input step, a function used in the prediction model and a parameter range in the prediction model are input, and in the model generation step, a plurality of parameter sets used in the prediction model may be determined using a PLS method. .
Preferably, in the model verification step, the most probable prediction among the plurality of prediction models is performed using an evaluation function using a deviation between the input amount obtained by simulation of the prediction model and the actual input amount in operation. A model should be estimated and selected.

  According to the estimation apparatus and the estimation method for estimating the prediction model of the converter according to the present invention, the prediction model is for determining the input amount of the auxiliary raw material in the converter, and the prediction for obtaining the optimal input amount of the auxiliary raw material and the like. A model can be obtained.

It is the figure which showed the system configuration | structure of the estimation apparatus. It is a flowchart which shows the process in an input step. It is a flowchart which shows the process in a model production | generation step. It is a flowchart which shows the process in a model verification step. It is a flowchart which shows the process in an output step. It is a figure which shows the outline of a converter.

Hereinafter, embodiments of an estimation apparatus 1 and an estimation method for estimating a prediction model of a converter 2 according to the present invention will be described with reference to the drawings.
Before describing the details of the estimation apparatus 1 and the estimation method of the present invention, an outline of the converter 2 to which the technology of the present invention is applied will be described.
As shown in FIG. 6, the converter 2 for performing the dephosphorization process is an upper bottom blowing type provided with an upper blowing lance 3 for blowing gaseous oxygen into the molten iron and a tuyere 4 for blowing oxygen or an inert gas from the furnace bottom into the molten iron. Then, oxygen is supplied by gaseous oxygen from the top blowing lance 3 and the molten metal is stirred by oxygen from the tuyere 4 or by an inert gas. Further, the converter 2 includes a supply device 5. The supply device 5 supplies auxiliary materials (quick lime, solid oxygen source, etc.) and is, for example, a hopper or a chute.

FIG. 1 shows a system configuration of an apparatus 1 that estimates a prediction model for accurately predicting the amount of auxiliary raw material to be charged in the converter 2 described above.
The prediction model estimation apparatus 1 includes an input unit 6 that inputs a generation condition of a prediction model, and a converter model that extracts data so as to match the conditions input by the input unit 6 and generates a plurality of prediction models. A simulation is performed using a plurality of prediction models obtained by the generation unit 7 and the converter model generation unit 7, and the result is verified to estimate the most probable prediction model from among the plurality of prediction models. A converter model verification unit 8 to be selected and an output unit 9 to output a prediction model selected by the converter model verification unit 8 are provided.

  In the input unit 6 described above, a function used for the prediction model and a parameter range in the prediction model are input. In the converter model generation unit 7, a PLS method (Partial Least Squares Regression, partial least square method) is used. And a plurality of parameter sets used in the prediction model are determined. Note that the PLS method is an example of quantitative chemistry as disclosed in documents such as “Model selection in PLS regression, Hashimoto, Tanaka, Academia, Information Science and Engineering 10, 39-49, 2010-03, Nanzan University”. It is a regression analysis technique that is often used in the field of In quantitative chemistry, it is useful when the number of wavelengths (variables) is much larger than the sample size, such as spectral calibration, or when the collinearity between variables is high. In recent years, it has attracted attention not only for the purpose of improving the accuracy of regression analysis but also for usage such as dimension reduction or extraction of related factors. PLS regression has a feature of converting high-dimensional data into low-dimensional data strongly related to the dependent variable. When this PLS method is used, a plurality of sets of regression result data can be obtained instead of one.

Further, the converter model verification unit 8 uses the evaluation function using the input amount obtained by the simulation of the prediction model and the actual input amount in the operation to determine the most probable prediction model from the plurality of prediction models. Estimated and elected.
Hereinafter, the estimation apparatus 1 and the estimation method of the present invention will be described in detail.
As shown in FIG. 1, the estimation apparatus 1 includes four parts: an input unit 6 (data input unit), a converter model generation unit 7, a converter model verification unit 8, and an output unit 9 (converter model output unit). Estimate the optimal prediction model.

The prediction model generated by the converter model generation unit 7 is assumed to be represented by the following formula (1).
y = a ₁ F ₁ (x ₁ ) + a ₂ F ₂ (x ₂ ) +... + a _n F _n (x _n ) + b (1)
Where y is the objective variable of the prediction model,
x 1 to _n are explanatory variables of the prediction model,
F 1 to _n are arbitrary functions,
a 1 to _n and b are model parameters.

First, in the data input part 6, the range of arbitrary functions F1- _n and model parameters a1- _n , b used for calculating a prediction model in the converter model production | generation part 7 is input into the input part 6 ( S1-1 in FIG. 1, data input step). In S1-1, the prediction model may be estimated by inputting some values of the model parameters a 1 to _n , b as arbitrary fixed values instead of ranges. For example, if a fixed value is input to a1, the prediction model: y−a ₁ F ₁ (x ₁ ) = a ₂ F ₂ (x ₂ ) +... + A _n F _n (x _n ) + b Data sets a 2 to _n and b are estimated by the following processing.

  Next, the converter model generation unit 7 first selects one of a plurality of methods such as multiple regression analysis and PLS prepared in the prediction model calculation unit 10 (S2-1), and the data extraction unit 11 performs operation results. A part of the teacher / learning data recorded in the storage unit 16 is extracted (S2-2). At this time, the already extracted data recorded in the OK data storage unit 12 or the NG data storage unit 13 is not extracted.

Next, the prediction model calculation unit 10 calculates model parameters of the prediction model using a method selected from the extracted teacher / learning data and data recorded in the OK data storage unit 12 (S2-4). Then, the prediction model determination unit 14 determines whether the calculated model parameter is within the input range (S2-5). If it is within the range, the extracted data is recorded as OK data in the OK data storage unit 12 (S2-6). Then, the calculated model parameters are recorded in the prediction model storage unit 15 (S2-7), and the process returns to the data extraction unit 11.

  If it is not within the range, it is determined whether the data currently extracted is one (S2-8). If there is one, the data is recorded as NG data in the NG data storage unit 13 and the data extraction unit 11 Return to (S2-9). Otherwise, return to the data extraction unit 11 without doing anything. These operations are repeated until data cannot be extracted (S2-3). It should be noted that the extraction may be aborted when the number of extracted data becomes less than a certain number, not until the extracted data disappears in S2-3.

After determining the model parameters for one method, if the prepared method still remains (S2-10), the contents of the OK data storage unit 12 and the NG data storage unit 13 are reset (S-11), Return to method selection. In this way, model parameters are determined for all prepared methods.
According to the process (converter model generation step) by the converter model generation unit 7 described above, a plurality of sets of model parameters can be obtained with respect _to the parameters a 1 to _n and b of the prediction model represented by the equation (1). become.

  Thereafter, the converter model verification unit 8 first selects one prediction model from a plurality of sets of model parameters obtained by the converter model generation unit 7, in other words, one prediction model (S3-1). . Thereafter, using the selected prediction model, a simulation is performed under the past operation conditions recorded in the operation result storage unit 16 to calculate the input amount of the auxiliary material (S3-2). This simulation is performed by a simulation calculation unit 17 having a function equivalent to that of a subsidiary material input amount calculation device or a process computer that calculates the amount of subsidiary material input in actual operation.

Thereafter, the prediction model evaluation unit 18 calculates a deviation between the input amount calculated in the simulation and the actual input amount recorded in the operation result storage unit 16 as an evaluation function, and stores the calculation result in the model evaluation storage unit 20 ( S3-3). In addition, when calculating an evaluation function in S3-3, you may use the input amount corrected to the value considered desirable instead of the actual input amount.
The same operation is repeated for all prediction models obtained by the converter model generation unit 7 (S3-4), and an optimum prediction model determination unit 19 determines a prediction model that minimizes the evaluation function (S3-4). S3-5).

According to the process (converter model verification step) by the converter model verification unit 8 described above, the prediction model that minimizes the evaluation function (model parameters a 1 to _n , b) is determined.
The converter model output unit 9 outputs the optimum prediction model obtained by the converter model verification unit 8 to the output unit 9 (S4-1, output step). Note that without determining the optimal prediction model in S3-5, the simulation and evaluation results of all prediction models may be displayed as a chart or the like in S4-1, and the model may be determined by the operator.

  By using the estimation device 1 and the estimation method for estimating the prediction model of the converter 2 described above, by inputting the generation conditions of the converter model, the physics and metallurgical laws based on past experience can be obtained. A prediction model can be taken in, and a highly accurate prediction model can be generated even when estimated from a small number of data. Data that does not match the generation conditions can be excluded as data with a large error.

In addition, by comparing the simulation results obtained with the obtained prediction model and the actual operation results, the prediction model that requires the optimum input can be selected even under the operating conditions where the data required for model estimation was not measured. Can do.
Regarding the estimation of the prediction model, there can be a model estimation method in which the simulation result and the operation result coincide with each other. However, since the operation result is not necessarily the optimum operation, this method uses the model for determining the optimum operation condition. Does not lead to estimation. By combining the converter model generation step and the converter model verification step as in the present invention, the estimation of a more optimal converter model close to the current operation → the more optimal calculation of the auxiliary material input amount → the input of the auxiliary material through the operation improvement It is expected that the improvement loop of optimization of quantity → estimation of more optimal converter model will be born. In the present embodiment, by using the distance between the input amount obtained in the simulation and the actual input amount as an evaluation function, the auxiliary material input amount close to the current operation is calculated using the obtained converter model. Therefore, it can be expected to be easily accepted by operators when changing models.

  By the way, this invention is not limited to said each embodiment, The shape, structure, material, combination, etc. of each member can be suitably changed in the range which does not change the essence of invention. Further, in the embodiment disclosed this time, matters that are not explicitly disclosed, for example, operating conditions and operating conditions, various parameters, dimensions, weights, volumes, and the like of a component deviate from a range that a person skilled in the art normally performs. However, matters that can be easily assumed by those skilled in the art are employed.

For example, as a prediction model, y = a ₁ F ₁ (x ₁ ) + a ₂ F ₂ (x ₂ ) +... + A _n F _n (x _n ) + b, and a part of input F _{1 to} F _n A prediction model using only the function may be constructed. Moreover, arbitrary functions F 1 to _n used for calculating the prediction model may include a plurality of explanatory variables, for example, F ₁ (x ₁ , x ₂ ) = x ₁ · x _2. .

DESCRIPTION OF SYMBOLS 1 Estimation apparatus 2 Converter 3 Top blowing lance 4 Tuyere 5 Supply apparatus 6 Input part 7 Model generation part 8 Model verification part 9 Output part 10 Prediction model calculation part 11 Data extraction part 12 OK data storage part 13 NG data storage part 14 Prediction model determination unit 15 Prediction model storage unit 16 Operation result storage unit 17 Simulation calculation unit 18 Prediction model evaluation unit 19 Optimal prediction model determination unit 20 Model evaluation storage unit

Claims (6)

An apparatus for estimating a prediction model of a converter that is used to calculate an input amount of a secondary material that is input for component adjustment in the converter, and that is obtained based on an actual operation value of the converter. ,
An input unit for inputting generation conditions of the prediction model;
A model generation unit that generates a plurality of prediction models that match the conditions input by the input unit;
A model verification unit that estimates and selects the most probable prediction model from the plurality of prediction models by performing simulation using a plurality of prediction models obtained by the model generation unit and verifying the result;
An output unit for outputting a prediction model selected by the model verification unit;
An apparatus for estimating a predictive model of a converter. In the input unit, a function used for the prediction model and a parameter range in the prediction model are input,
The prediction model estimation device for a converter according to claim 1, wherein the model generation unit is configured to determine a plurality of parameter sets used for the prediction model using a PLS method.   The model verification unit estimates the most probable prediction model from the plurality of prediction models by using an evaluation function that uses a deviation between the input amount obtained by simulation of the prediction model and the actual input amount in operation. 3. The prediction model estimation apparatus for a converter according to claim 1, wherein the prediction model estimation apparatus is configured to be selected. A method of estimating a prediction model of a converter that is used to calculate an input amount of auxiliary raw material that is input for component adjustment in a converter and that is obtained based on an actual operation value of the converter. ,
An input step of inputting a generation condition of the prediction model;
A model generation step of generating a plurality of prediction models that match the conditions input by the input unit;
A model verification step of estimating and selecting a most probable prediction model from the plurality of prediction models by performing simulation using a plurality of prediction models obtained in the model generation step and verifying the result;
An output step of outputting the prediction model selected by the model verification unit;
A method for estimating a predictive model of a converter. In the input step, a function used for the prediction model and a parameter range in the prediction model are input,
5. The converter predictive model estimation method according to claim 4, wherein in the model generation step, a plurality of parameter sets used for the prediction model is determined using a PLS method. In the model verification step, the most probable prediction model is estimated from the plurality of prediction models by using an evaluation function using a deviation between the input amount obtained by simulation of the prediction model and the actual input amount in operation. The predictive model estimation method for a converter according to claim 4 or 5, wherein the prediction model is selected.

Images