JPH0587462A - Burning controller of sintered ore - Google Patents

Abstract

PURPOSE:To reduce a variatin in the quality of sinterecl ores by leaning actual results of operations in the past by means of a neural network and, at the same time, estimating the progress in the quality of sintered ores by using the results of the leaning in order to carry out control of the calcination that is most suitable. CONSTITUTION:In the data of sintering operation that is based on measurement data by various sensors, data 1 before a change in operation such as the quantity of charged raw materials, moisture, temperature, pressure, etc., at various sections of a sintering machine, and data 2 after the change in operation such as target values in the sintering operation data after change on grain size, strength, FeO R1 and RDI of sintered ores are respectively inputted to a data working means 4 through a data input means 3, and here the data are worked into information that contains temporal elements such as the value of change in a certain time from the past to the present. Next each data is handed over to a neural network 5 with reference input data 8 and instructor's input data 9, and the control quantity 6 that comprises the ratio of powdered coke composition, pallet speed is sought by a forward direction device learned in the past to control a controller 7.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention uses a neural network to learn the operating results and predict the action means when the operation is changed to calcine the sintered ore for the purpose of stabilizing the operation. The present invention relates to a control device.

[0002]

2. Description of the Related Art Strength and grain size of sinter and FeO, RI, R
The sinter ore grade represented by DI qualitatively evaluates various sensor information installed in the sintering and firing process by an operator,
It has been managed and controlled by estimating and evaluating the firing situation and its transition, and by optimally adjusting the operating factors such as the mixing ratio of powder coke. However, there are individual differences in firing control based on human judgment, it was difficult to standardize control methods and quantify evaluations, and it was difficult to always continue appropriate actions.

Some control devices applying knowledge engineering are also reported in "Materials and Processes" issued by the Iron and Steel Institute.
For example, Japanese Unexamined Patent Publication No. 63-100138 and "Materials and Processes: 1986-S799" disclose a report entitled "Application of fuzzy control to sintering operation".
This report aims to stabilize the quality of the sintered ore by applying fuzzy control to the level control of the return tank, but it focuses only on the return tank level.
It is not a comprehensive judgment of firing control. With this control described above, the quality of the sintered ore was slightly stable, but it was far from ideal. There is also a problem that it is difficult to learn the membership function and the maintenance load is large.

In addition, "Materials and Processes: Vol. 2 (19
89) -974, Vol. 4 (1991) -123 ", a report entitled" Tobata No. 3 Sintering Machine Expert System Online Application Method "is disclosed. This report is based on the rule-based knowledge of the sintering operation. It is intended to standardize the operation by systematizing
Rules have no learning function. Therefore, when operating conditions change, it is always necessary to manually modify the rules. In addition, when systematizing the operation technology of skilled operators based only on the rule base, there is a problem that it is not possible to flexibly deal with new phenomena like human beings.

Furthermore, "Materials and Process Vol. 4 (19
91) -331 ”, a report entitled" Sintered Heat Pattern Recognition System Using Neural Network "is disclosed. This report is an example of application to a heat pattern control of a neural network. Is disclosed. This example is effective in optimizing the temperature distribution in the width and machine length directions of the sintering machine, but the heat pattern is not simply determined by the windbox temperature or the temperature below the grate. Further, the quality of the sintered ore is not stable if the heat pattern is simply optimized. For this reason, appropriate operational actions are not always taken.

[0006]

The above-described conventional sintering control apparatus for sinter using the AI technology (fuzzy control, expert system, neural network) has the following problems. (1) Fuzzy control and neural network are applied to one element of firing control, and cannot control and evaluate the firing situation comprehensively. (2) The fuzzy control and expert system cannot be used for new equipment (equipment addition, remodeling) and accompanying operation changes because it is made based on human experience. Also, maintenance of membership functions and rules is difficult.

The present invention has been made in order to solve the above-mentioned problems, and learns the past operation results by using a neural network and uses the learning results to determine the transition of the sinter quality. The purpose is to develop a device that predicts, controls firing optimally, and reduces fluctuations in sinter quality.

[0008]

[Means for Solving the Problems] The control device of the present invention for predicting the transition of the sinter ore grade and performing the optimum operation action includes (1) data before operation change and data after operation change. It has an input layer that can be input to a new unit, and also has an output layer that outputs a controlled variable. A multilayer neural network of two or more layers learned in advance based on operation and sensor data is used for firing in the sintering process. Means for preparing in the forward direction of the process (hereinafter referred to as forward device), (2) data before the operation change (current value) and data after operation change (target value) in the input layer of the "forward device" Qualitatively by using the means for inputting
The above three means are used for quantitatively controlling the firing condition. Furthermore, when the actual operation data obtained as a result of controlling the operation amount output from the output layer of the forward direction device and the formation situation are different from those before the operation change input to the input layer and when the operation is changed. The sinter ore sintering control device is characterized in that the control accuracy of the device is improved by learning the self-organization of the neural network.

[0009]

In the present invention, the variation of the quality of the sinter is reduced by controlling the firing of the sinter by the neural network constructed in multiple layers based on the data and the operation amount before and after various operation changes. .. Further, by learning the neural network, the accuracy of the apparatus can be maintained by automatically generating or modifying the mutual coefficient of the neural network.

[0010]

EXAMPLES Next, examples will be described. FIG. 1 is a block diagram illustrating input / output and processing procedures of a firing control device for carrying out the present invention. In FIG. 1, a circular frame indicates an input and an output of the apparatus, a short block indicates a process inside the apparatus, 1 is a data value before the operation is changed, 2 is a data value after the operation is changed, and 3 is a data. Data input means, 4 data processing means, 5 a neural network (forward device), 6 a control amount, (a) a powder coke mixing ratio as a control amount,
(B) shows the pallet speed as a controlled variable, (c) shows the layer thickness as a controlled variable, 7 is a control controller for the controlled variable, that is, CFW control device, pallet control device, layer thickness control device in broken lines respectively. , 8 is reference input data, and 9 is teacher input data.
Reference numerals 10 and 10 represent learning pattern creation processing, respectively.

In FIG. 1, data 1 before operation change is sintering operation data based on the data measured by various sensors, and is, for example, the blending amount of raw materials, water, temperature and pressure of each part of the sintering machine. The data 2 after the operation change is the target value of the sintering operation data after the operation, and is the particle size, strength, FeO, RI, R of the sintered ore.
It is DI. These are taken into the apparatus through the data inputting means 3, and the data processing means 4 removes noise and the like by smoothing processing and linear regression processing, and then the data necessary for judging the control amount of the firing control. For example, powder
It is processed into information that includes time factors such as the mixing ratio, sintering machine length, width, temperature and pressure distribution in the height direction, and the amount of change over a certain period from the past to the present.

These data are passed to the neural network 5, and the neural network 5 outputs the controlled variable 6 by the forward device learned in the past. In the embodiment of the present invention, the control amount 6 is (a) powder coke blending ratio,
(B) Pallet speed, (c) Three types of information, layer thickness,
The operating policy at that time was given in advance, and any one of them was selectively controlled. Each of these items is calculated as a value normalized to -1 to -1 so that it can be handled quantitatively, and then converted into a specific control amount 6 to instruct the operator or to transmit it to the control device 7. Connect to automatic control.

Separately from this, reference input data 8 and teacher input data 9 for learning the network are given to the apparatus. The reference input data 8 in this case is formally the same as the sintering operation data 1, and the teacher input data 9 is formally the same as the control amount 6, but the teacher input data 9 is the reference input data 8 Exact information is required as output to. Therefore, the operation results on the day when the operation situation is stable become a teacher,
Teacher input data 9 will be given.

In the learning pattern forming process 10, the reference input data 8 and the teacher input data 9 are input to the neural network 5.
It is simply converted into a pattern that is easy to learn in.

The firing control device having the above construction will be described in more detail. First, the data 1 of the sintering operation given to the data input means 3 of FIG. (1) Mixing ratio of powder coke (2) Mixing ratio of return ore amount (3) Grain size and water content of blended raw material (4) Layer thickness (5) Pallet speed (6) Grain size and strength of sintered ore (7) Sintered ore FeO, RI, RDI of (8) Bulk density of charge (9) Temperature below grate (10) Ignition furnace ignition strength (11) Exhaust ore afterglow layer thickness (12) Basicity of blended materials

FIG. 2 shows an example of the configuration of the neural network 5 (forward device) in this embodiment.
The configuration of the neural network 5 in this embodiment is
An input layer for taking in various data of the above-mentioned (1) powder coke mixing ratio to (12) basicity of the mixed raw material processed by the data processing means 4 of FIG. 1, and (a) powder coke mixing as the control amount 6. Ratio, (b) pallet speed, (c) layer thickness, etc., and a three-layer structure in which an intermediate layer is placed between output layers. In general, the neural network 5 is linear in the two-layer structure,
As in the present embodiment, three or more layers are non-linear, and the information processing means is remarkably improved. Therefore, 3 with the intermediate layer
In many cases, networks of multiple layers are used. In the present embodiment, the output layer has 3 units, (a) powder coke mixing ratio, (b)
One unit is associated with each of the three controlled variables of pallet speed and (c) layer thickness.

Next, examples of the powder coke control amount, the pallet speed control amount and the layer thickness control amount when the output value of the output layer is displayed as time series data are shown in FIG. Each is normalized to -1 to 1, and the closer to -1 or 1, the larger the operation amount. This value can be converted into an actual control amount to quantitatively control the firing condition. In the sintering process, the control amount is not simultaneously controlled in order to make it easy to understand the relationship between the operation action and the result, and any one is selected. In this example, only powder coke is selected as the actual controlled variable.

In this apparatus, the change of the control amount during the normal operation is 1 because the quality data of the sinter is obtained at a pitch of about 1 hour.
It was set to a fixed period of time. The learning flow of the neural network is considered to be various such as a pile cycle, a monthly cycle, and a daily cycle.
It was a daily cycle.

FIG. 4 shows a state in which the quality is stable as changes in the strength, grain size, RI and RDI of the sinter. Figure 4
As shown in, the fluctuation was reduced by about 15% from the conventional value (1990 average). As a result of the improvement of the quality of the sintered ore and the decrease in the fluctuation of the quality, the amount of powdered coke used was 2 kg /
T / Sr decreased.

[0020]

According to the sinter ore firing control apparatus of the present invention, it is possible to secure a control amount for performing the optimum sintering operation, and to re-learn at any time by using the learning method of the neural network, The accuracy of this device can be maintained even if the equipment or operating conditions change. As a result, the quality of the sintered ore is stabilized and improved, and it is possible to stably manufacture the quality of the sintered ore required by the blast furnace at low cost.

[Brief description of drawings]

FIG. 1 is a block diagram illustrating an input / output and a processing process of a sintering ore firing control system for carrying out the present invention.

FIG. 2 is an explanatory diagram showing a configuration example of a neural network in the present embodiment.

FIG. 3 is an explanatory diagram showing output value examples of an output layer of a neural network as time series data.

FIG. 4 is a transition diagram of measurement data which is a control result in the example.

[Explanation of symbols]

 1 data before operation change, 2 data after operation change, 3 data input means, 4 data processing means, 5 neural network, 6 controlled variable, 7 control device, 8 reference input data, 9 teacher Input data, 10 learning pattern creation processing, a powder coke mixing ratio, b pallet speed, c layer thickness.

Claims (2)

[Claims] 1. An apparatus for controlling the firing condition of sinter, comprising an input layer for inputting data before and after the operation change of sintering, and an output layer for outputting a controlled variable in advance. Means for preparing a multilayer neural network of two or more layers learned based on operation and sensor data in the forward direction of the firing process of the sintering process, and data before operation change and operation in the input layer of the neural network. A calcination control apparatus for sinter, wherein the calcination condition is controlled using a means for inputting the changed data and a control amount obtained from the output layer of the neural network as a result of the input. 2. When the actual operation data obtained as a result of controlling the firing condition with the control amount output from the output layer is different from the data before the operation change and after the operation change input to the input layer, The firing control apparatus for a sintered ore according to claim 1, wherein the control accuracy of the apparatus is improved by learning the self-organization of the neural network.