EP1593090B1 - Multivariate, predictive regulation of a direct reduction process - Google Patents

Abstract

A property is measured for each of several products to be made at different points in time and from these measurements, the value of the product property is postulated. This is taken into consideration by determining a value for an input neurone of a neural network. Using the neural network, the property of the product to be manufactured is predicted. Independent claims are included for the following ; (1) plant; and (2) computer program product.

Description

The direct reduction of iron ore to the product sponge iron in the form of DRI (Direct Reduced Iron) or HBI (Hot Briquetted Iron) takes place using heated process gases, preferably in a shaft furnace. In simple terms, chemical reactions that convert iron ore (Fe ₂ O ₃ ) and natural gas (CH ₄ ) into iron Fe, carbon dioxide CO _2, and water H ₂ O occur.

The shaft furnace is operated continuously by continuously adding raw material in the form of iron ore pellets from above and continuously removing the sponge iron from below.

For example, processes for making sponge iron are known U.S. 4,093,455 . US 4,178,151 . US 4,234,169 and WO 02/097138 A1 known. Other publications are Thompson, M .: "Control Innovations in MIDREX Plants: An Introduction" in "Direct from MIDREX", 1st Quarter 2001, pp. 3-4, 2001 and Görner, F., Bacon, F .: "Development of Process Automation for the MIDREX Process" in "Direct from MIDREX", 1st Quarter 2002, p. 3-5, 2002 ,

In the production of sponge iron, it is desirable to produce a product having as constant as possible, precisely specified properties. For this purpose, it is known to keep all influencing factors for the product to be produced sponge iron as constant as possible and thus to operate the process at a known operating point. However, for example, the assumption of a completely homogeneous iron ore as a raw material in practice is often not fulfilled.

Based on this, the object of the invention is to predict a property of a product to be produced, in particular of sponge iron produced by direct reduction.

This object is achieved by the inventions specified in the independent claims. Advantageous embodiments emerge from the subclaims.

The inventions are based on the general idea, when predicting a value of a property to be predicted with the aid of a neural network, to take into account the product history by incorporating values into the prognosis which have the property to be predicted for already manufactured products. Such values can be taken into account when determining an input quantity of the neural network. Alternatively or additionally, however, it is also possible, when determining the same input variable or another input variable of the neural network, to take into account the difference between a value of the property measured at a specific time and a value of the property predicted for this time.

Accordingly, in a method, in particular a production method, a prognosis of a value of a property of a product to be produced in the present and / or future with the help of a neural network, especially in the future, can be predicted. In this case, a value of the property, either completely or with the aid of random samples, is measured in each case for several products produced at different times in the past. Thereafter, the value of the property of the product to be produced in the present and / or future is predicted in a prediction from the values of the property measured for the times in the past. It is, so to speak a forecast in the impure, in which essentially only the values of the property measured for the times in the past are included.

This value predicted in the prediction is taken into account when determining an input variable for an input neuron of the neural network. Finally, with the help of the neural network, the value of the property of the product to be produced in the present and / or future is predicted in the actual prognosis.

The prediction can preferably be carried out with the aid of a recursive filter. The recursive filter can be realized relatively simply by performing a linear extrapolation. More accurate predictions are possible by using a second neural network, in particular a recurrent neural network, as the recursive filter. For this, an additional mathematical description of the cause-and-effect relationship between process parameters and the property is expedient.

In order to determine an input quantity of the neural network, however, a comparison between a value of the property measured for a time in the past and a value of the property predicted for this time can also be taken into account when determining an input quantity for the neural network.

Accordingly, a prognosis is made of a value of a property of a product to be produced in the present and / or future with the aid of a neural network, especially in the future. For this purpose, a value of the property is measured at a product manufactured at a certain time. Furthermore, a value of the property in the product produced at that time is predicted with the prognosis. This requires the forecast for the past with random or meaningful values be initialized from an even later past. Then, the difference between the measured and the predicted value of the property of the product at that time is formed. This difference then goes into the determination of an input quantity for an input neuron of the neural network. Finally, with the help of the neural network, the value of the property of the product to be produced in the present and / or future is predicted in the prognosis.

The goal of a good prognosis is to intelligently control the manufacturing process of the product. In the method or by the method, therefore, parameters in the production process of the product to be produced are preferably changed until the value predicted in the prognosis of the property of the product to be produced at least approximately corresponds to a nominal value of the property of the product to be produced.

Although the described method is generally suitable for all possible products, it is particularly relevant for those products whose properties can not be measured during their production or shortly thereafter, but only with a time delay of possibly several hours. The process can be used particularly advantageously in continuous production processes.

• Der Metallisierungsgrad, das heißt das Verhältnis zwischen dem absoluten Eisengehalt im Eisenerz und dem freigesetzten Eisen (Fe),
• der Gewichtsanteil des Eisenschwamms, der als metallisches Eisen (Fe) vorliegt,
• der Kohlenstoffgehalt im Eisenschwamm.

The forecasting method described is particularly suitable for the prognosis of properties of sponge iron produced in the direct reduction process. Accordingly, the property may consist of one or more of the following quantities:

• The degree of metallization, ie the ratio between the absolute iron content in iron ore and the released iron (Fe),
• the proportion by weight of the sponge iron, which is present as metallic iron (Fe),
• the carbon content in the sponge iron.

Of course, within the scope of the invention, with the aid of the prognosis method, not only one property but several properties of the product to be produced are to be predicted.

In the prognosis with the aid of the neural network, in addition to the history of the products already produced, further parameters in the production process of the product to be produced should also be taken into account by influencing or representing input variables of neuronal network input neurons. Such parameters are in particular process temperatures, gas compositions of used process gases and / or properties of raw materials.

The invention further relates to an arrangement adapted to carry out one of the aforementioned methods. Such an arrangement can be realized, for example, by appropriately programming and setting up a computer or a computer system. The arrangement may also include a direct reduction plant, in particular with a shaft furnace.

A program product for a data processing system, which contains code sections with which one of the described methods can be executed on the data processing system, can be implemented by suitable implementation of the method in a programming language and translation in code executable by the data processing system. The code sections are stored for this purpose. In this case, the program is understood as a tradable product under a program product. It can be in any form, such as on paper, a computer-readable medium or distributed over a network.

Figur 1

einen Schachtofen zur Herstellung von Eisenschwamm;

Figur 2

eine Vorprognose eines Wertes einer Eigenschaft eines herzustellenden Produktes aufgrund von Werten der Eigenschaften bei bereits hergestellten Produkten.

Figur 3

eine Vörprognose eines Wertes einer Eigenschaft eines herzustellenden Produktes aufgrund eines gemessenen Wertes der Eigenschaft bei einem bereits hergestellten Produkt und eines prognostizierten Wertes der Eigenschaft bei diesem bereits hergestellten Produkt.

Further essential advantages and features of the invention will become apparent from the description of an embodiment with reference to the drawing. Showing:

FIG. 1

a shaft furnace for producing sponge iron;

FIG. 2

a prediction of a value of a property of a product to be manufactured based on values of the properties of already manufactured products.

FIG. 3

a prediction of a value of a property of a product to be produced based on a measured value of the property of an already manufactured product and a predicted value of the property of that already manufactured product.

In FIG. 1 A shaft furnace 1 with a feed 2 for supplying iron ore pellets, which accumulate in the interior of the shaft furnace 1 on one or more piles 3, can be seen. The shaft furnace 1 further has means 4 for supplying reaction gases, in particular hydrogen and carbon monoxide, and means 5 for discharging reaction gases, in particular carbon dioxide and water.

In the interior of the shaft furnace 1, the iron ore is reduced under high temperatures and under the action of the process gases in a direct reduction to sponge iron, which is removed at the lower end of the furnace via a removal device 6 and transported by a conveyor belt 7.

At regular intervals, for example every two to four hours, samples of cooled products are taken and examined in the laboratory for the properties metallization or carbon content. The time difference from completion of the development process to the evaluated sample is about 5 to 9 hours in the metallization and about 3.5 to 7.5 hours in the enrichment with carbon. This is because the metallization is completed as a sub-process of the manufacturing process, such as in the middle of the shaft furnace 1. The product still remains for about 3 hours for cooling in the shaft furnace 1 until it leaves it at its removal device 6. The sub-process enrichment with carbon is completed after about three quarters of the time, ie after about 4.5 hours. In addition, it takes about 2 hours in the laboratory to measure the value of the metallization and to measure the value of carbon enrichment. Assuming that the measurement is made approximately every 4 hours, so there is a dead time of 5 to 9 or 3.5 to 7.5 hours.

With this, assuming that all influencing factors have remained constant, it was possible to carry out control interventions in the current process. For the preceding 3.5 to 9 hours of the process, however, there is no longer an opportunity to intervene. In addition, for various reasons, it is not possible in practice to keep all influencing factors constant.

In FIG. 2 it is seen that at time t _-3 present value W _-3 a property, that is, for example, the metallization or the carbon content, of the product until the time t _-3 + 5 can be measured. The same applies mutatis mutandis to the value W _{-2 of} the property of the product at time t _-2 and the value W _{-1 of} the product at time t _-1 , which can also be measured only 5 hours after manufacture.

From the measured values W _-3 , W _-2 and W _-1 , which indicate, for example, a metallization of 94.1%, 94.2% and 94.3%, the value W ^V _{P, 0} is now in a prediction V the property of the product to be produced at the current time t ₀ predicts. Im in FIG. 2 shown Embodiment, this is done simply by linear extrapolation, which is indicated by the dashed line. However, even more complicated algorithms can be used here.

The predicted value W ^v ₀ in the prediction is now taken into account when determining an input variable for an input neuron of a neural network.

In addition, as in FIG. 3 also takes into account the difference between a value W _-1 measured in the past for a time t _-1 and a value W _{P, -1} predicted for this time with the actual prognosis P in a second prediction V '.

Mit α = {0; 1}, beispielsweise α = 0,25.This is done with the help of the formula: $${{\mathrm{<figure-callout\; id="0"\; label="W\; V"\; filenames="imgf0002.png"\; state="\{\{state\}\}">W\; </figure-callout>}}^{\mathrm{V}}}_{\mathrm{0}}\mathrm{=}{\mathrm{W}}_{\mathrm{-}\mathrm{1}}\mathrm{+}\mathrm{\alpha }\mathrm{\cdot }\left({{\mathrm{W}}^{\mathrm{P}}}_{\mathrm{-}\mathrm{1}}\mathrm{-}{\mathrm{W}}_{\mathrm{-}\mathrm{1}}\right)\mathrm{;}$$

With α = {0; 1}, for example α = 0.25.

Taking into account the difference between the measured value W _-1 and the predicted value W ^P _{-1 of} the property of the product at time t _-1 , the value W ^{V '} _{0 of} a second prediction now becomes an input value for a second input neuron neural network.

• Die Gaszusammensetzung aller Prozessgase (trocken und nass), die in der Direktreduktionsanlage zum Einsatz kommen,
• quantitative und zeitliche Aussagen bei Gasen zu Durchflüssen (beispielsweise Tonnen pro Stunde) und Temperaturen,
• alle Temperaturmessungen im und am Schachtofen,
• Eigenschaften, die das Rohmaterial beschreiben (Porosität, chemische Zusammensetzung, Größe und Form der.Pellets, Dichte, Temperatur),
• Eigenschaften, die das hergestellte oder herzustellende Produkt beschreiben (Dichte, chemische Zusammensetzung, Kohlenstoffgehalt, Eisengehalt, Gehalt des metallischen Eisens, Metallisierungsgrad),
• Massenflüsse des Rohmaterials und des Endproduktes sowie
• Luft (Temperatur und Feuchtigkeit über der Zeit).

The input variables of further input neurons of the neural network are formed by further process parameters or calculated taking into account further process parameters. Such process parameters are:

• The gas composition of all process gases (dry and wet) used in the direct reduction plant
• quantitative and temporal statements for gases at flow rates (eg tonnes per hour) and temperatures,
• all temperature measurements in and at the shaft furnace,
• Properties describing the raw material (porosity, chemical composition, size and shape of pellets, density, temperature),
• Characteristics describing the product produced or to be produced (density, chemical composition, carbon content, iron content, content of metallic iron, degree of metallization),
• Mass flows of the raw material and the final product as well
• Air (temperature and humidity over time).

The modeling advantageously involves about 100 measured and calculated input variables.

In the neural network, it has proven to be particularly advantageous to use an ensemble of neural networks and to evaluate its median. A combination of feed-forward networks with recursive filters is also advantageous.

Preferred training methods for the neural network include the use of digital filters for the training data, automatic outlier removal, bagging, compliance with monotonic constraints on relevant control variables, and target sizes.

The input variables of the neural network are preferably modeled analytically from the named process parameters. This is done, for example, by the integration of quantities, the quantitative description of chemical conversion including the reaction kinetics with the help of differential equations and a calculation of which piece of material where in the shaft furnace, with the aid of the product output per hour.

For execution, it is advantageous to use various interlinked software programs in Fortran, C, C ++ and MATLAB. A user interface can be programmed in Visual Basic.

• Die Abweichung von vorgegebenen Sollwerten kann reduziert werden. In der Praxis ergibt sich eine Reduzierung der Standardabweichung um ca. 40% bei der Metallisierung und ca. 30% beim Kohlenstoffgehalt.
• Geringere Abweichungen erlauben einen Betrieb des Schachtofens näher am Optimum, was dem Durchsatz und der Qualität zugute kommt.
• Es wird eine Produktionssteigerung von ca. 1% erzielt.
• Durch die konstantere und höhere Materialgüte können die Abnehmer des Eisenschwamms, nämlich Betreiber von Elektrolichtbogenöfen, ihre Öfen optimierter betreiben. Dieser Vorteil wirkt sich noch stärker aus, als die Vorteile beim Betreiben der Direktreduktionsanlage, so dass sich für das hergestellte Produkt entsprechend höhere Preise erzielen lassen.
• Die Berechnung der Materialeigenschaften kann Laborproben ersetzen.
• Die Berechnungen können jederzeit durchgeführt werden könnten. Das heißt, es können zum Beispiel alle 0,1 Sekunden aktuelle Werte berechnet, statt alle zwei bis vier Stunden im Labor analysiert werden.
• Da maßgebliche chemische Reaktionen bereits in der ersten Hälfte bzw. dem ersten Drittel des Prozesses von ca. 6 Stunden abgeschlossen sind, können die Materialeigenschaften bereits ermittelt werden, bevor der Eisenschwamm dem Schachtofen entnommen wird. Damit verkürzen sich die Reaktionszeiten für eine Regelung der Materialeigenschaften von 3,5 bis 9 Stunden auf die Berechnungsdauer für die Modelle, die unter 0,1 Sekunden liegt.

The invention provides the following advantages:

• The deviation from specified target values can be reduced. In practice, the standard deviation is reduced by about 40% for metallization and about 30% for carbon content.
• Lower deviations allow operation of the shaft furnace closer to optimum, which benefits throughput and quality.
• A production increase of approx. 1% is achieved.
• Due to the more constant and higher material quality, the customers of the sponge iron, namely operators of electric arc furnaces, operate their ovens optimized. This advantage has an even greater effect than the advantages of operating the direct reduction plant, so that correspondingly higher prices can be achieved for the manufactured product.
• The calculation of the material properties can replace laboratory samples.
• The calculations could be done anytime. This means that, for example, current values can be calculated every 0.1 seconds, instead of being analyzed every two to four hours in the laboratory.
• Since decisive chemical reactions have already been completed in the first half or the first third of the process of about 6 hours, the material properties can already be determined before the sponge iron is removed from the shaft furnace. This shortens the reaction times for material properties control from 3.5 to 9 hours to the calculation time for the models, which is less than 0.1 seconds.

Claims (12)

1. Computer-aided method, in particular a production method, for predicting, with the aid of a neuronal network, a value of a property of a product to be produced,

   - wherein measurement values of the property of products produced at different times are measured,

   - wherein a preliminary value of the property of the product to be produced is predicted from the measurement values in a preliminary prognosis,

   - wherein the preliminary value is taken into account when determining an input variable for a first input neuron of the neuronal network,

   - wherein a reference value of the property of a produced product is measured,

   - wherein an estimated value of the property of the produced product is predicted by means of the prognosis,

   - wherein the difference is formed between the reference value and the estimated value of the property of the product,

   - wherein a value calculated from this difference is used as the input variable for a second input neuron of the neuronal network,

   - wherein the value of the property of the product to be produced is predicted in the prognosis with the aid of the neuronal network.
2. Method according to claim 1, wherein the preliminary prognosis is undertaken with the aid of a recursive filter.
3. Method according to claim 2, wherein the recursive filter is and/or includes a linear projection.
4. Method according to claim 2, wherein the recursive filter is and/or includes a second neuronal network, in particular a recurrent neuronal network.
5. Method according to one of the preceding claims, wherein parameters in the production process of the product to be produced are changed, until the value of the property of the product to be produced predicted in the prognosis corresponds at least approximately to a setpoint value of the property of the product to be produced.
6. Method according to one of the preceding claims, wherein the product to be produced is produced with the aid of a direct reduction.
7. Method according to one of the preceding claims, wherein the product to be produced is and/or contains iron sponge.
8. Method according to one of the preceding claims, wherein the property is the metallic iron content of the product to be produced.
9. Method according to one of the preceding claims, wherein the property is the carbon content of the product to be produced.
10. Method according to one of the preceding claims, wherein further parameters in the production process of the product to be produced are taken into account during the prognosis with the aid of the neuronal network, in particular process temperatures, gas compositions of process gases used, properties of raw materials and/or parameters determined through analytical computations of chemical conversions and reaction kinetics.
11. Arrangement, which is set up to execute a method according to at least one of the preceding claims.
12. Programme product, which when loaded onto and executed on a data processing system starts up a method according to one of claims 1 to 10 or a device according to claim 11.

Images