JP7147326B2 - Component concentration calculation device and component concentration calculation method - Google Patents

Description

TECHNICAL FIELD The present invention relates to a component concentration calculation device and a component concentration calculation method for predicting component concentrations in molten iron refining processing.

In the molten iron refining process, molten pig iron is desulfurized. Desulfurization in hot metal pretreatment is carried out using, for example, a mechanical stirring desulfurization apparatus (hereinafter also referred to as "KR", 1 as shown in FIG. 10). A desulfurization agent is added to the hot metal 5 in the ladle 10, and the impeller 21 is rotated by the driving device 25 to accelerate the desulfurization reaction. Therefore, it is important to reduce the S concentration in the hot metal 5 after desulfurization treatment by KR1 (hereinafter also referred to as “post-treatment S concentration”) to the target S concentration when delivered to the converter. Therefore, if the desulfurization agent is excessively added in order to ensure that the post-treatment S concentration is below the target S concentration in the KR treatment, the production cost will increase. In setting the conditions, it is desirable to be able to accurately estimate the post-treatment S concentration.

For example, Patent Literature 1 describes the charging conditions of the treating agent in the hot metal pretreatment operation, which are capable of accurately guiding the charging condition of the treatment agent in the hot metal pretreatment operation, and taking into account practical variations. A guidance method is disclosed. In Japanese Patent Laid-Open No. 2004-100000, the operating conditions including the processing agent input conditions are changed and repeatedly input to a neural network to determine the processing agent input conditions, and by adding practical variations to the evaluation, an appropriate processing agent is determined. Find the input conditions. As a result, the estimation accuracy is enhanced regardless of whether the correspondence characteristic between the processing agent input condition and the component concentration after processing exhibits linearity or nonlinearity.

Further, Patent Document 2 discloses a method of appropriately guiding the oxygen supply amount in the final stage of blowing by constructing a model for predicting the molten iron carbon concentration in the converter process. In Patent Document 2, a highly readable prediction model that conforms to physical phenomena is constructed by constructing a model for estimating component concentrations and reaction rate coefficients in an equilibrium state in a first-order reaction model by multiple regression analysis.

JP-A-8-269518 Japanese Patent Application Laid-Open No. 2001-11520

Kengo Kato, and two others, "Theory of Lasso quantile regression and its application to non-life insurance," Journal of the Japan Statistical Society, March 2009, Vol. 38 No. 2, p121-149

However, since the prediction model of Patent Document 1 is constructed by black-box modeling that does not involve physical laws in the relationship between input and output, there is a high risk of making predictions that do not conform to physical phenomena, and it is not suitable for control. do not have. In addition, since it is black box modeling, there is a problem that the readability of the model is low and it is difficult to get a sense of satisfaction from the prediction results.

Furthermore, in the technique of Patent Document 2, in order to estimate the component concentration in the equilibrium state, a prediction model is constructed using data assuming that the final stage of the treatment is in the equilibrium state. Therefore, if the assumption is wrong, or if sufficient data that can be assumed to be in equilibrium cannot be collected, there is a risk that the accuracy of the prediction model will be significantly degraded.

Therefore, the present invention has been made in view of the above problems, and the object of the present invention is to convert the concentration of components in the equilibrium state and the reaction rate coefficient in the primary reaction model into operation data in the refining process of molten iron. It is an object of the present invention to provide a new and improved component concentration calculation device and component concentration calculation method capable of accurately predicting post-processing component concentrations by estimating accordingly.

In order to solve the above problems, according to one aspect of the present invention, there is provided a component concentration calculation device for predicting the component concentration in the molten iron refining process, which is represented by the following formula (A) with actual data in past operations. A dynamic parameter estimation model for estimating the equilibrium component concentration [A] _e and the reaction rate coefficient k based on the first-order reaction equation and the preconditions before treatment and the operating conditions related to treatment as inputs, and , a model building unit that builds a component concentration estimation model based on the component concentration [A] _e in equilibrium state estimated using the dynamic parameter estimation model and the reaction rate coefficient k, and using the component concentration estimation model, the reaction and a component concentration estimator for estimating the component concentration [A] after processing.

Here, [A] ₀ is the component concentration before the reaction treatment.

The component concentration estimation model outputs the component concentration [A] after the reaction treatment, with the input of the component concentration [A] _e in the equilibrium state calculated by the dynamic parameter estimation model, the reaction rate coefficient k, and the processing time t. You may

A dynamic parameter estimation model may be constructed by a neural network.

The component concentration estimator may output the quantile of the post-processing component concentration.

The component concentration calculation device may further include an optimum solution search unit that searches for operating conditions under which the predicted component concentration is equal to or less than a preset component concentration based on the component concentration estimation model.

Further, in order to solve the above problems, according to another aspect of the present invention, there is provided a component concentration calculation method for predicting component concentrations in a molten iron refining process, wherein actual data in past operations and the following formula (A) Dynamic parameter estimation for estimating the equilibrium component concentration [A] _e and the reaction rate coefficient k based on the first-order reaction equation represented by A model building step of building a component concentration estimation model based on the component concentration [A] _e in equilibrium state estimated using the model and the dynamic parameter estimation model and the reaction rate coefficient k, and using the component concentration estimation model and a component concentration estimation step of estimating the component concentration [A] after reaction processing.

Here, [A] ₀ is the component concentration before the reaction treatment.

As described above, according to the present invention, in the refining process of molten iron, by estimating the component concentration and reaction rate coefficient in the equilibrium state in the primary reaction model according to the operation data, the post-treatment component concentration can be accurately predicted. can do.

1 is a block diagram showing the functional configuration of an operating condition calculation device according to one embodiment of the present invention; FIG. It is explanatory drawing which shows an example of the prediction model which concerns on the same embodiment. It is a flow chart which shows operating condition calculation processing concerning the embodiment. FIG. 3 is a graph showing evaluation results of prediction accuracy of post-prediction S concentration by the prediction model shown in FIG. 2. FIG. 4 is a graph showing a predicted processed S concentration obtained by a prediction model when only the treatment time is changed with respect to the performance of one charge. 4 is a graph showing the relationship between R-dip and predicted post-treatment S concentration when only the treatment time is changed with respect to the performance of one charge. FIG. 10 is a graph showing the evaluation result of prediction accuracy of post-prediction S concentration when a 99% quantile regression model is applied to the prediction model. FIG. 7 is a graph showing an example of a search range for a solution set based on a probability density function estimated from past performance data by a mixed Gaussian distribution with respect to an input amount of secondary material; It is a block diagram for demonstrating the hardware constitutions of the operating-conditions calculating apparatus which concerns on the same embodiment. It is an explanatory view showing a schematic configuration of a mechanical stirring desulfurization apparatus (KR).

A preferred embodiment of the present invention will be described in detail below with reference to the accompanying drawings, taking as an example post-treatment S concentration prediction in a mechanical stirring desulfurization apparatus (KR). In the present specification and drawings, constituent elements having substantially the same functional configuration are denoted by the same reference numerals, thereby omitting redundant description.

<1. Functional Configuration of Operating Condition Calculator>
First, based on FIG.1 and FIG.2, the operating condition calculating apparatus which concerns on one Embodiment of this invention is demonstrated. FIG. 1 is a block diagram showing the functional configuration of an operating condition calculation device 100 according to this embodiment. FIG. 2 is an explanatory diagram showing an example of a prediction model according to this embodiment.

The operating condition calculation device 100 according to the present embodiment is an information processing device for performing optimization processing of operating conditions in the desulfurization treatment of hot metal. The operating condition calculation device 100 includes, as shown in FIG.

The performance data acquisition unit 110 acquires from the performance data storage unit 200 performance data in past operations used for building a model for estimating the post-refining component concentrations. The performance data includes measured values measured in the KR process or values calculated based on the measured values in constructing an estimation model for the S concentration after the KR process as the component concentration after the refining process. For example, hot metal temperature, hot metal weight, pre-treatment C concentration, pre-treatment Si concentration, pre-treatment Mn concentration, pre-treatment P concentration, pre-treatment S concentration, surface height, function of impeller height and stirring speed (R-dip ), the number of times the impeller is used, the powder CaO charging amount, the secondary slag charging amount, the Al concentration, etc. are obtained as performance data. The performance data acquisition unit 110 acquires performance data of a predetermined number of past operations, and outputs the data to the model construction unit 120 and the constraint condition setting unit 130 .

The model construction unit 120 constructs a prediction model for outputting the post-processing S concentration based on the performance data acquired by the performance data acquisition unit 110 . The prediction model constructed by the model construction unit 120 according to this embodiment is a gray box model based on a desulfurization reaction model that is easy to physically interpret. Specifically, for example, a prediction model as shown in FIG. 2 is constructed.

A desulfurization rate formula is shown in the following formula (1). Note that K is a reaction rate coefficient and is represented by the following formula (1-1). A is the reaction interfacial area, _km is the mass transfer coefficient, V is the molten iron weight, [%S] is the molten iron S concentration, and [%S] _e is the molten iron S concentration at the equilibrium point.

From the above formula (1), the molten iron S concentration at the desulfurization time t can be expressed by the following formula (2). [%S] ₀ is the concentration of molten iron S at the start of KR treatment, and t ₀ is the completion time of entrainment.

In the above formula (2), the molten iron S concentration [% S] _e at the equilibrium point and the coefficient A· _km are unknown values that change depending on the operating conditions. Here, assuming that the influence of the entrainment completion time t ₀ is included in the coefficient K and t ₀ = 0, the dynamic parameters to be estimated are the molten iron S concentration [% S] _e at the equilibrium point and the coefficient A・It becomes two of _km . In general, the value of the molten iron S concentration [%S] _e at the equilibrium point is calculated, for example, [%S] _e = 4 [e ⁻³ % ] or fixed at [%S] _e =0 on the assumption that the KR process does not reach an equilibrium state, the coefficient A· _km was estimated. On the other hand, the inventors of the present application focused on the fact that the molten iron S concentration [% S] in the above formula (2) can be expressed by a combination of two functions F(x ₁ ) and G(x ₂ ), and the following formula As shown in (3), we came up with the idea of estimating functions simultaneously.

The function F(x ₁ ) is a function related to the equilibrium theory contribution and simulates the molten iron S concentration at the equilibrium point. The molten iron S concentration at the equilibrium point depends on the molten iron temperature, slag composition, desulfurization material, composition and metal weight. Furthermore, it is said that the molten iron S concentration at the equilibrium point also depends on the stirring method because the equilibrium point changes as the grains become finer.

Function G(x ₂ ) is a function corresponding to the kinetic contribution. From the viewpoint of kinetics, the molten iron S concentration is considered to depend not only on the stirring method, but also on the molten iron temperature and composition from the viewpoint of entrainment completion time.

In this way, multiple functions (functions F(x ₁ ) and G(x ₂ ) in this embodiment) must be estimated, but the true values of F(x ₁ ) and G(x ₂ ) are unknown. Therefore, it is not possible to estimate functions individually using a machine learning method or the like. Therefore, in the present embodiment, as shown in FIG. 2, first, the function F(x ₁ ; w ₁ ) and the function G(x ₂ ; w ₂ ) are combined into a combined model (component concentration estimation parameters w ₁ , w that minimize the loss function J(y, y ) expressed by the predicted value y^ of the post-treatment S concentration obtained by synthesis using the model) and the actual post-treatment S concentration y ₂ , F(x ₁ ) and G(x ₂ ) are estimated at the same time. This makes it possible to simultaneously estimate two dynamic parameters, functions F(x ₁ ) and G(x ₂ ), by preparing only input and output training data.

Functions representing functions F (x ₁ ; w ₁ ) and G (x ₂ ; w ₂ ), which are dynamic parameter estimation models for estimating dynamic parameters, include, for example, neural networks, multiple regression models, and polynomial regression models. You can use it. Parameters to be estimated by the machine learning model _are parameters W ₌ [ _w1 , w2] of functions F(x1; _w1 ) and G ₍ x2; _w2 ). Here, the loss function is defined as the sum of squares of errors as in the following equation (4). Note that y _i represents the post-actual S concentration, and y _i ̂ represents the post-prediction S concentration, which is the output of the prediction model. In this specification, the post-prediction S concentration is expressed as "y^", which is the same as the output of the prediction model shown in FIG. 2 with "^" above "y". be.

Parameters w ₁ and w ₂ are updated as shown in Equation (5) below. α is a learning coefficient, and n ₁ and n ₂ are the numbers of parameters of F(x ₁ ; w ₁ ) and G(x ₂ ; w ₂ ).

From the above, the post-prediction processing S concentration y _i ̂, which is the output of the prediction model shown in FIG.

In this way, the model construction unit 120 constructs a prediction model that takes physical phenomena into account based on the desulfurization reaction model as shown in FIG. For example, in Patent Literature 1, the conditions for adding a treatment agent in hot metal pretreatment are estimated by a prediction model, and this prediction model is a so-called black box model constructed by a neural network. On the other hand, the prediction model constructed by the model construction unit 120 according to the present embodiment is a gray box model based on the desulfurization reaction model, which not only improves the prediction accuracy but also makes it a model that is physically easy to interpret. It's becoming

Since a general regression model is a model for predicting expected values, controlling the KR treatment based on predicted values obtained from such a model may lead to insufficient desulfurization. Therefore, in order to prevent insufficient desulfurization in the KR treatment, the prediction model shown in FIG. You may A quantile regression model is a regression model that explains the quantiles of the distribution of the dependent variable (see, for example, Non-Patent Document 1). A quantile regression model uses I am building a linear model with When estimating, the linear model corresponding to the quantile to be estimated is used to estimate the quantile of the distribution of the dependent variable for any explanatory variable. For example, by setting the 99% quantile regression model after the component concentration estimation model shown in FIG. Insufficient desulfurization in treatment can be prevented more reliably.

The constraint setting unit 130 sets constraints for the operating conditions using a probability density function, and sets constraints for the post-treatment component concentrations. The constraint on the operating conditions is also referred to as "likelihood constraint", and the constraint on post-treatment component concentration is also referred to as "post-treatment component concentration constraint". A feasible operating condition can be obtained by searching for an operating condition to be a solution within a reliable range of the post-treatment component concentration estimation model. Therefore, the constraint condition setting unit 130 sets a range for searching for a solution as a constraint condition.

To set the likelihood constraint, specifically, the constraint setting unit 130 first performs simultaneous It is estimated from a probability density function (eg Gaussian mixture distribution). Then, the constraint condition setting unit 130 calculates the logarithmic likelihood of the training data based on the estimated density function, and the logarithm Based on the one-dimensional probability density function of the likelihood, a control limit with which the estimation model can be trusted is obtained. This control limit is set as a likelihood constraint. At least one of the upper limit value and the lower limit value of the post-treatment component concentration is defined in the post-treatment component concentration constraint condition.

Note that the constraint condition setting unit 130 may further set the operating cost to be minimized as an evaluation function. The constraint condition setting unit 130 sets an evaluation function for the operating cost and outputs it to the optimum solution searching unit 140 only when the operating cost is considered when the optimum solution searching unit 140 described later searches for operating conditions.

The optimum solution searching unit 140 searches for operating conditions that satisfy the likelihood constraint and the post-treatment component concentration constraint using the prediction model constructed by the model construction unit 120 . In addition, when searching for operating conditions in consideration of operating costs, the optimum solution search unit 140 uses an estimation model to satisfy the likelihood constraint and the post-processing component concentration constraint, and to minimize the cost Search for the operating conditions that minimize the evaluation function that has been set. The optimum solution searching unit 140 estimates the component concentration after the reaction process using the prediction model constructed by the model construction unit 120 in searching for the operating conditions, and has a function as a component concentration estimating unit. ing. The optimum solution searching unit 140 sets the optimum and executable operating conditions as the optimum operating conditions as a result of the search. The optimum operating conditions may be output via the output unit 150. The output unit 150 is, for example, a display capable of displaying information, a speaker capable of outputting audio, or the like.

<2. Operating condition calculation processing>
Based on FIG. 3, the operating condition calculation processing by the operating condition calculating device 100 according to this embodiment will be described. FIG. 3 is a flowchart showing operating condition calculation processing according to the present embodiment.

(S100: Prediction model construction processing)
First, when the performance data acquisition unit 110 of the operating condition calculation device 100 acquires the performance data of the past operation from the performance data storage unit 200, the model construction unit 120 constructs a prediction model based on the desulfurization reaction model. Specifically, for example, a prediction model as shown in FIG. 2 is constructed.

The prediction model shown in FIG. 2 outputs the S concentration after prediction processing by the component concentration estimation model represented by the above formula (6) according to the physical formula, but as described above, the molten iron S concentration at the desulfurization time t is For the output, two dynamic parameters must be set: the molten iron S concentration [%S] _e at the equilibrium point and the coefficient A· _km . Conventionally, these dynamic parameters were assumed to be certain values in the _{desulfurization} reaction model, and the post-treatment S concentration was calculated. The two dynamic parameters of the coefficient A· _km are also configured to be jointly estimated by the machine learning model.

That is, as shown in FIG. 2, the output value of the function F (x ₁ ) related to the equilibrium contribution, the output value of the function G (x ₂ ) related to the kinetic contribution, the processing time t, the molten iron weight V, and KR The S concentration [%S] ₀ before processing is input to the component concentration estimation model, and the S concentration after prediction processing is output. At this time, the output value of the function F (x ₁ ; w ₁ ) regarding the equilibrium contribution and the output value of the function G (x ₂ ; w ₂ ) regarding the kinetic contribution are calculated by the dynamic parameter estimation model. be. The model construction unit 120 updates these machine learning models based on the above equation (5), and determines the functions F(x ₁ ; w ₁ ) and G(x ₂ ; w ₂ ) when the loss function is minimized. Then, a prediction model to be used in the optimum solution searching unit 140 is constructed.

Here, FIG. 4 shows the evaluation result of the prediction accuracy of the post-prediction S concentration by the prediction model shown in FIG. In constructing the prediction model, of the performance data of all 5,780 charges, the performance data of the first 4,000 charges was used as training data for constructing the prediction model, and the performance data of the remaining 1,780 charges was used as evaluation data for model evaluation. The dynamic parameter estimation models representing the functions F(x ₁ ; w ₁ ) and G(x ₂ ; w ₂ ) are neural networks with one intermediate layer. As shown in the prediction results for the evaluation data in FIG. 4, the predicted post-treatment S concentration and the actual post-treatment S concentration roughly match, and the RMSE (Root Mean Squared Error) also shows a low value of 1.644. rice field.

Further, when examining the correlation between the value G(x ₂ )/V calculated based on the function G(x ₂ ) related to the kinetic contribution estimated by the neural network and the expected hot metal temperature, a strong positive correlation was found. One thing has been confirmed. On the other hand, in experiments, it was confirmed that there is a positive correlation between the reaction rate constant K and the hot metal temperature, and it was also confirmed that the order of the reaction rate constant K is close to the value G(x ₂ )/V. . From this, it can be considered that the value G(x ₂ )/V simulates the reaction rate constant K.

FIG. 5 shows the predicted processed S concentration obtained by the prediction model when only the treatment time is changed with respect to the actual result of one charge. As shown in FIG. 5, it can be confirmed that the predicted post-treatment S concentration decreases as the treatment time elapses and gradually reaches an equilibrium point. Similarly, FIG. 6 shows the relationship between R-dip and the predicted post-treatment S concentration when only the treatment time is changed with respect to the performance of one charge. As shown in FIG. 6, the relationship between R-dip and the predicted post-treatment S concentration is represented by a curve like a quadratic function, which agrees with known physical knowledge. From this, it can be seen that by using a gray box model based on a neural network as a prediction model, it is possible to construct a model that is easy to interpret physically while improving the prediction accuracy.

Furthermore, in order to make the prediction model shown in FIG. FIG. 7 shows the prediction accuracy evaluation results of . FIG. 7 uses the same historical data as in FIG. 4 to build a predictive model, including a quantile regression model. As shown in the prediction results for the evaluation data in FIG. 7, it can be seen that the actual post-treatment S concentration can be made larger than the predicted post-treatment S concentration with a probability of 99%.

(S110: Constraint condition setting process)
Next, the constraint condition setting unit 130 sets a probability density function that defines the distribution area of the performance data as a constraint condition (S110). In order to search for a solution within the reliable range of the predictive model, the present embodiment implements reliability determination by a mixed Gaussian distribution that can estimate complex joint probability density functions.

Specifically, first, a joint probability density function (assuming Gaussian mixture distribution) of all or part of the explanatory variables is estimated from the training data used to construct the expected value prediction model. Then, the constraint condition setting unit 130 calculates the logarithmic likelihood of the training data based on the estimated density function, and the logarithm Based on the probability density function of the likelihood, the control limits that the prediction model can rely on are obtained. Constraint condition setting section 130 sets this control limit as a constraint condition and outputs it to optimum solution search section 140 .

FIG. 8 shows an example of a solution search range (that is, constraint conditions) set based on a probability density function estimated from past performance data by a mixed Gaussian distribution with respect to the secondary material input amount. In FIG. 8, distribution of training data for distribution 1: powder CaO charging amount - secondary slag charging amount, distribution 2: powder CaO charging amount - treatment time, distribution 3: powder CaO charging amount - B-Al and the solution search range set by the estimated probability density function. As shown in FIG. 8, the solution search range set by the probability density function matches the training data distribution. Constraint condition setting section 130 generates a constraint condition for excluding performance data outside the solution search range set by the probability density function, and outputs the constraint condition to optimum solution search section 140 . Constraint condition setting section 130 also sets at least one of the upper limit value and the lower limit value of the post-treatment component concentration as a post-treatment concentration constraint condition, and outputs it to optimum solution search section 140 . When operating costs are considered when searching for operating conditions, the operating costs to be minimized are set by the constraint setting unit 130 as an evaluation function together with the above constraints, and output to the optimum solution search unit 140. Just do it.

(S120: optimum solution search process)
After that, the optimal solution search unit 140 searches for the optimal operating conditions based on the prediction model constructed by the constraint setting unit 130 and the constraint conditions set by the constraint setting unit 130 . That is, the optimal solution searching unit 140 solves the optimization problem represented by the following equations (7) to (10) to find the optimal solution. In Equation (7) below, function f(u, v) is a function representing a quantile regression model, and function g(a) is a function representing a constraint. Note that the function f(u, v) representing the quantile regression model and the probability density function p(u) are non-linear functions and cannot be solved by mathematical methods. Therefore, in this embodiment, a differential evolution method, which is one of evolutionary algorithms, is used as an optimization method.

The above variables are as follows.

The decision variable u is a target variable to be obtained as the optimum operating conditions, for example, impeller initial height, treatment time, stirring speed, powdered CaO charging amount, secondary slag charging amount, and B-Al. Weights are respectively set for each decision variable u in terms of the cost of KR processing. Also, in order to search for a solution within the area where the training data is distributed, the logarithmic degree function (that is, the probability density function) p(u) is less than or equal to the control limit α, which is the logarithmic likelihood threshold, As the likelihood constraint, the constraint expression of Equation (8) is set. Furthermore, the constraint expression (9) is set as the post-treatment component concentration constraint condition so that the post-treatment component concentration is equal to or less than the target upper limit S concentration. Furthermore, the constraint expression of Equation (10) is set as a likelihood constraint so that the value of each decision variable falls within the specified range of upper and lower limits [u _i ^lower , u _i ^upper ]. Specifically, for example, values as shown in Table 1 below are set.

The evaluation function J represented by the formula (1) expresses the secondary material cost as the operating cost in the KR treatment. The optimum solution search unit 140 solves the optimization problem represented by the above equations (7) and (8) to (10), and finds the optimum operating conditions for improving the component predictive value of the post-treatment S concentration and achieving cost reduction. Ask for

<3. Hardware Configuration>
Next, the hardware configuration of the operating condition calculation device 100 according to the embodiment of the present invention will be described in detail with reference to FIG. 9 . FIG. 9 is a block diagram for explaining the hardware configuration of the operating condition calculation device 100 according to this embodiment.

The operating condition calculation device 100 mainly includes a CPU 901 , a ROM 903 and a RAM 905 . The operating condition calculation device 100 further includes a bus 907 , an input device 909 , an output device 911 , a storage device 913 , a drive 915 , a connection port 917 and a communication device 919 .

The CPU 901 functions as an arithmetic processing device and a control device, and controls all or part of the operation in the operating condition arithmetic device 100 according to various programs recorded in the ROM 903, the RAM 905, the storage device 913, or the removable recording medium 921. . A ROM 903 stores programs and calculation parameters used by the CPU 901 . The RAM 905 temporarily stores programs used by the CPU 901, parameters that change as appropriate during execution of the programs, and the like. These are interconnected by a bus 907 comprising an internal bus such as a CPU bus.

The bus 907 is connected to an external bus such as a PCI (Peripheral Component Interconnect/Interface) bus via a bridge.

The input device 909 is operation means operated by a user, such as a mouse, keyboard, touch panel, button, switch, and lever. Further, the input device 909 may be, for example, remote control means (so-called remote control) using infrared rays or other radio waves, or an external connection device 923 such as a PDA corresponding to the operation of the operating condition calculation device 100. There may be. Further, the input device 909 is composed of, for example, an input control circuit that generates an input signal based on information input by the user using the operation means described above and outputs the signal to the CPU 901 . A user of the operating condition calculation device 100 can input various data to the operating condition calculation device 100 and instruct processing operations by operating the input device 909 .

The output device 911 is configured by a device capable of visually or audibly notifying the user of the acquired information. Such devices include display devices such as CRT display devices, liquid crystal display devices, plasma display devices, EL display devices and lamps, audio output devices such as speakers and headphones, printer devices, mobile phones, and facsimiles. The output device 911 outputs, for example, results obtained by various processes performed by the operating condition calculation device 100 . Specifically, the display device displays the results obtained by various processes performed by the operating condition calculation device 100 as text or images. On the other hand, the audio output device converts an audio signal including reproduced audio data and acoustic data into an analog signal and outputs the analog signal.

The storage device 913 is a data storage device configured as an example of a storage unit of the operating condition calculation device 100 . The storage device 913 is configured by, for example, a magnetic storage device such as a HDD (Hard Disk Drive), a semiconductor storage device, an optical storage device, or a magneto-optical storage device. The storage device 913 stores programs executed by the CPU 901, various data, and various data obtained from the outside.

The drive 915 is a reader/writer for recording media, and is built in or externally attached to the operating condition calculation device 100 . The drive 915 reads information recorded on a removable recording medium 921 such as a mounted magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, and outputs the information to the RAM 905 . The drive 915 can also write records to a removable recording medium 921 such as an attached magnetic disk, optical disk, magneto-optical disk, or semiconductor memory. The removable recording medium 921 is, for example, a CD medium, a DVD medium, a Blu-ray (registered trademark) medium, or the like. Also, the removable recording medium 921 may be a CompactFlash (registered trademark) (CompactFlash: CF), a flash memory, an SD memory card (Secure Digital memory card), or the like. Also, the removable recording medium 921 may be, for example, an IC card (Integrated Circuit card) equipped with a contactless IC chip, an electronic device, or the like.

The connection port 917 is a port for directly connecting equipment to the operating condition calculation device 100 . Examples of the connection port 917 include a USB (Universal Serial Bus) port, IEEE1394 port, SCSI (Small Computer System Interface) port, RS-232C port, and the like. By connecting the external connection device 923 to the connection port 917 , the operating condition calculation device 100 directly acquires various data from the external connection device 923 and provides various data to the external connection device 923 .

The communication device 919 is, for example, a communication interface configured with a communication device or the like for connecting to the communication network 925 . The communication device 919 is, for example, a communication card for wired or wireless LAN (Local Area Network), Bluetooth (registered trademark), or WUSB (Wireless USB). Further, the communication device 919 may be a router for optical communication, a router for ADSL (Asymmetric Digital Subscriber Line), a modem for various types of communication, or the like. This communication device 919 can, for example, transmit and receive signals to and from the Internet or other communication devices according to a predetermined protocol such as TCP/IP. A communication network 925 connected to the communication device 919 is configured by a wired or wireless network, and may be, for example, the Internet, home LAN, infrared communication, radio wave communication, satellite communication, or the like. .

An example of the hardware configuration capable of realizing the functions of the operating condition calculation device 100 according to the embodiment of the present invention has been described above. Each component described above may be configured using general-purpose members, or may be configured by hardware specialized for the function of each component. Therefore, it is possible to appropriately change the hardware configuration to be used according to the technical level at which the present embodiment is implemented.

Although the preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, the present invention is not limited to such examples. It is obvious that a person having ordinary knowledge in the technical field to which the present invention belongs can conceive of various modifications or modifications within the scope of the technical idea described in the claims. It is understood that these also naturally belong to the technical scope of the present invention.

REFERENCE SIGNS LIST 100 operating condition calculation unit 110 performance data acquisition unit 120 model construction unit 130 constraint condition setting unit 140 optimum solution search unit 150 output unit 200 performance data storage unit

Claims (6)

A component concentration calculation device for predicting component concentrations in molten iron refining,
a model building unit that builds a component concentration estimation model for estimating the component concentration [A] after reaction processing ;
a component concentration estimating unit that estimates the component concentration [A] after reaction processing using the component concentration estimation model;
with
The model construction unit
In the first-order reaction formula represented by the following formula (A), the equilibrium component concentration [A] _e and the reaction rate coefficient k, which are dynamic parameters that change depending on the operating conditions, are respectively represented by the operating conditions x ₁ . replaced by a dynamic parameter estimation model of the function F(x ₁ ) and the function G(x ₂ ) represented by the operating condition x ₂ ,
In the following formula (A) replaced with the function F(x ₁ ) and the function G(x ₂ ), the parameter w ₁ of the function F(x ₁ ) and the parameter w ₂ of the function G(x ₂ ) are Simultaneously estimating the equilibrium component concentration [A] _e and the reaction rate coefficient k by setting to minimize the loss function using actual data in past operations ,
A component concentration calculation device for constructing the component concentration estimation model using the estimated component concentration [A] _e in the equilibrium state and the reaction rate coefficient k.

Here, [A] ₀ is the component concentration before the reaction treatment , and t is the treatment time . The component concentration estimation model inputs the component concentration [A] _e in the equilibrium state calculated by the dynamic parameter estimation model, the reaction rate coefficient k, and the processing time t, and the component concentration after reaction processing [ A]. 3. The component concentration calculation device according to claim 1, wherein said dynamic parameter estimation model is constructed by a neural network. The component concentration calculation device according to any one of claims 1 to 3, wherein the component concentration estimator outputs a quantile of the component concentration after processing. The component concentration calculation according to any one of claims 1 to 4, comprising an optimum solution search unit that searches for operating conditions where the predicted component concentration is less than or equal to a preset component concentration based on the component concentration estimation model. Device. A component concentration calculation method for predicting a component concentration in a molten iron refining process,
a model building step of building a component concentration estimation model for estimating the component concentration [A] after the reaction treatment ;
A component concentration estimation step of estimating the component concentration [A] after the reaction treatment using the component concentration estimation model;
including
In the model building step,
In the first-order reaction formula represented by the following formula (A), the equilibrium component concentration [A] _e and the reaction rate coefficient k, which are dynamic parameters that change depending on the operating conditions, are respectively represented by the operating conditions x ₁ . replaced by a dynamic parameter estimation model of the function F(x ₁ ) and the function G(x ₂ ) represented by the operating condition x ₂ ,
In the following formula (A) replaced with the function F(x ₁ ) and the function G(x ₂ ), the parameter w ₁ of the function F(x ₁ ) and the parameter w ₂ of the function G(x ₂ ) are Simultaneously estimating the equilibrium component concentration [A] _e and the reaction rate coefficient k by setting to minimize the loss function using actual data in past operations ,
A component concentration calculation method for constructing the component concentration estimation model using the estimated component concentration [A] _e in the equilibrium state and the reaction rate coefficient k.

Here, [A] ₀ is the component concentration before the reaction treatment , and t is the treatment time .

Images