JP7328142B2 - Plant control device and its control method, rolling mill control device and its control method, and program - Google Patents

Description

The present invention relates to a plant control device and its control method, a rolling mill control device and its control method, and a program in real-time feedback control performed using artificial intelligence technology such as a neural network.

2. Description of the Related Art Conventionally, in various plants, plant control based on various control theories has been carried out in order to obtain desired control results through the control.

As an example of a plant, for example, in the control of a rolling mill, fuzzy control and neuro-fuzzy control have been applied as control theories for shape control to control the corrugated state of a strip as an example of control. Fuzzy control is applied to shape control using coolant, and neuro-fuzzy control is applied to shape control of Sendzimir rolling mill. Of these, the shape control to which neuro-fuzzy control is applied is, as shown in Patent Document 1, the difference between the actual shape pattern detected by the shape detector and the target shape pattern, and the similarity with a preset reference shape pattern. A ratio is obtained, and from the similarity ratio, a control output amount for an operation terminal is obtained according to a control rule expressed by a control operation terminal operation amount for a preset reference shape pattern. In the following, as a conventional technique, shape control of a Sendzimir rolling mill using neuro-fuzzy control is used.

FIG. 5 shows shape control of the Sendzimir rolling mill described in FIG. Neuro-fuzzy control is used in the shape control of the Sendzimir rolling mill. In this example, the pattern recognition mechanism 51 performs pattern recognition of the shape from the actual shape detected by the shape detector 52, and calculates which of the preset reference shape patterns the actual shape is closest to. The control calculation mechanism 53 performs control using a control rule composed of control operation terminal manipulated variables for preset shape patterns as shown in FIG. More specifically, in the pattern recognition mechanism 51, the difference (.DELTA..epsilon.) between the actual shape detected by the shape detector 52 and the target shape (.epsilon.ref) is calculated as one of 1 to 8 shape patterns (.epsilon.). Which one is the closest is calculated, and the control calculation mechanism 53 selects and executes one of the control methods 1 to 8.

However, in the method of Patent Document 1, in order to verify the control rule, there are cases where the operator performs manual operation during rolling to verify the control rule, etc., but there are cases where unexpected shape change is shown. In other words, the control rule determined as described above may not conform to reality. This is due to insufficient investigation of mechanical properties and changes in rolling mill operating conditions and machine conditions. Many conditions are difficult. Therefore, once a control rule is set, it is often left as it is unless there is a problem.

If the control rule does not conform to reality due to changes in operating conditions, etc., it becomes difficult to achieve a certain level of control accuracy because the control rule is fixed. In addition, once the shape control is activated, the operator does not perform manual operations (there is a disturbance to the control), so it is difficult for the operator to manually intervene to find new control rules. Furthermore, it is difficult to set the control rule according to the material when rolling a material of a new standard.

As described above, the conventional shape control has a problem that it is difficult to modify the control rule because the control rule is set in advance.

In order to solve this problem, as shown in Patent Document 2, by randomly changing the control rule while performing shape control and learning the rule that improves the shape,
1) Discover new control rules while performing shape control during rolling.
2) New control rules cannot be predicted in advance, and there are cases where completely unpredictable control rules are optimal. go.
We have achieved that.

Patent No. 2804161 Patent No. 4003733

In the prior art described above, a representative shape is set in advance as a reference shape pattern, and control is performed based on a control rule that indicates the relationship between the reference waveform pattern and the manipulated variable of the control operation terminal. The learning of the control rule also relates to the manipulated variable of the control operation terminal with respect to the reference waveform pattern, and the predetermined typical reference shape pattern is used as it is.

However, for the same shape pattern, the manipulated variable of the control terminal is diverse, and if the control method is learned based on the evaluation criterion that the shape deviation is reduced by the manipulation of the manipulated terminal, the manipulated variable of the control terminal is learned. Therefore, the control output becomes the average value, and the control effect may become small. Although it is possible to select and learn only the operation method with a large control effect, in that case, the control output with a small control effect is not learned, which may also reduce the control effect.

Further, in the above-described prior art, the manipulated variable of the control terminal for the reference waveform pattern is determined, but the mechanical life of the control terminal when the control operation is performed according to the determined manipulated variable of the control terminal is taken into consideration. do not have.

From the above, in the present invention, a plant control device and its control method that can be expected to have effects such as an increase in the mechanical life of a rolling mill to be controlled, an improvement in control accuracy by avoiding ineffective control, etc. An object of the present invention is to provide a rolling mill control device, its control method, and a program.

Based on the above, in the present invention, "a neural network trained by a control method learning device that learns a combination of actual data of a plant to be controlled and a control operation according to the control effect and forms a plurality of neural networks with different control effects" A plant control device comprising a control execution device that provides a control output for controlling the operation end of the plant to be controlled according to the output of the control execution device, wherein the control execution device is a neural network formed by learning when the control effect is high When there is an output and the control effect is high, the control terminal of the controlled plant is controlled according to the output, and only the output of the neural network formed by learning when the control effect is low exists, and the control terminal Control the operating terminal of the controlled plant according to the output of the neural network formed by learning when the control effect is low when there is a margin in the control terminal position, and further learn when the control effect is low. A plant control device characterized by not controlling an operation terminal of a plant to be controlled when there is only an output of a neural network that has been set and there is no margin in the operation terminal position at the operation terminal.

In addition, in the present invention, "a combination of actual data of a plant to be controlled and a control operation is learned according to the control effect, and a plurality of neural networks with different control effects are formed according to the output of the neural network learned by the learning unit. A plant control method comprising a control unit that provides a control output for controlling an operation terminal of a plant to be controlled, wherein the control unit has an output of a neural network formed by learning when the control effect is high, and the control effect is high. If it is high, the control terminal of the plant to be controlled is controlled according to the output, and there is only the output of the neural network formed by learning when the control effect is low, and there is a margin in the control terminal position at the control terminal. The control terminal of the controlled plant is controlled according to the output of the neural network formed by learning when the control effect is low, and only the output of the neural network formed by learning when the control effect is low exists. and a plant control method characterized by not controlling the operating end of the plant to be controlled when there is no margin in the operating end position at the operating end, and a rolling mill control method to which this plant control method is applied.” It is.

Further, in the present invention, "a program for realizing a plant control device for recognizing a combination pattern of performance data of a plant to be controlled and performing control by a computer system, comprising: The system learns combinations of actual data and control operations of the plant to be controlled, and has a control method learning program that forms multiple neural networks with different control effects. The control execution program has the output of a neural network formed by learning when the control effect is high, and if the control effect is high, the plant to be controlled according to the output , and formed by learning when the control effect is low, and when there is only the output of the neural network and there is a margin in the control end position at the control end, the case where the control effect is low is learned and formed. Controls the control terminal of the plant to be controlled according to the output of the neural network that has been processed, and furthermore, there is only the output of the neural network formed by learning the case where the control effect is low, and there is a margin in the control terminal position at the control terminal If not, the operation end of the plant to be controlled is not controlled."

By using the present invention, the mechanical life of the rolling mill by preventing the operating end position from reaching the limit value by automatically changing the control rule in consideration of the operating end position and the control effect during control. Effects such as an increase in control accuracy and an improvement in control accuracy by avoiding ineffective control can be expected.

The figure which shows the outline|summary of the plant control apparatus which concerns on the Example of this invention. FIG. 4 is a diagram showing a specific configuration example of the control rule execution unit 10 according to the embodiment of the present invention; FIG. 4 is a diagram showing a specific configuration example of the control rule learning unit 11 according to the embodiment of the present invention; The figure which shows the neural-net structure in the case of using this invention for the shape control of a Sendzimir rolling mill. The figure which shows the shape control of the Sendzimir rolling mill described in FIG. 1 of patent document 1. FIG. FIG. 1 is a diagram showing control rules in shape control of the Sendzimir rolling mill described in FIG. 1 of Patent Document 1; FIG. 2 is a diagram showing an overview of an input data creation unit 2; FIG. 3 is a diagram showing an outline of a control output calculation section 3; 4 is a diagram showing an outline of a control output determination unit 5; FIG. The figure which shows shape deviation and a control method. The figure which shows the outline|summary of the control quality determination part 6. FIG. FIG. 4 is a diagram showing in order the relationship between the data of each part and the symbols in the control output calculation part 3; FIG. 4 is a diagram showing processing stages and processing contents in a learning data creation unit 7; The figure which shows the data example preserve|saved in learning data database DB2. FIG. 4 is a diagram showing an example of a neural network management table TB; The figure which shows the example of learning data database DB2. FIG. 4 is a diagram showing an example of the content of determination by a control output selection unit 107; The figure which shows the relationship between a shape evaluation result and a control output. FIG. 5 is a diagram showing the relationship between the position of the control operation terminal and the degree of margin;

Embodiments of the present invention will be described in detail below with reference to the drawings, but before that, the findings of the present invention and the circumstances leading up to the present invention will be described by taking shape control of a rolling mill as an example.

First, in order to solve the above problems in the present invention,
1) Instead of setting the reference shape pattern and the control operation for it separately in advance and learning the control operation method, the combination of the shape pattern and the control operation is applied to the same shape, and the control operation is performed. A plurality of control rules are learned according to the control effect, and among the plurality of control operations calculated from the plurality of control rules, the optimum operation is selected according to the state of the control operation end, and the control operation is performed using this. do. As for the selection of the optimum operation, when there is a margin in the operation range of the control operation end, the control operation is performed even if the control effect is small, and when there is no margin, the control operation with a small control effect is not performed.
2) New control rules cannot be predicted in advance, and there are cases where completely unpredictable control rules are optimal. go.
is required.

In order to achieve this, it is necessary to change the combination of the shape pattern used for shape control and the control operation so as to improve the control result. For this purpose, a neural network that can learn combinations of shape patterns and control operations is constructed, and the output of the neural network's control operations for the shape patterns generated by the rolling mill is changed according to the quality of the control results. things are necessary.

If the above is performed while shape control is being performed on the rolling mill in operation, an erroneous control output may be output, resulting in deterioration of the shape and an operational abnormality such as strip breakage. When sheet breakage occurs, it takes time to replace the rolls used in the rolling mill, and the material to be rolled during rolling is wasted, causing great damage. Therefore, it is necessary to avoid outputting an erroneous control output to the rolling mill as much as possible.

Based on the above, in the present invention, in order to realize this, the quality of the control operation output by the neural network is verified using, for example, a simple model of a rolling mill. To prevent shape deterioration by not outputting to a control operation end of a rolling mill. At this time, with respect to the neural network, learning is carried out assuming that the control operation for that shape pattern is erroneous.

Since there is a possibility that the method of verifying the quality of the control operation itself is incorrect, the control operation output of the neural network that is judged to be incorrect with a certain probability can be output to the control operation end of the rolling mill. It is also possible to learn combinations of external shape patterns and control operations.

In order for the neural network to learn control rules, a large amount of learning data, which is a combination of shape results and corresponding control methods, is required. In this specification, a large amount of learning data used for learning a certain neural network is referred to as a learning data group. Control rules, which are learning results, differ depending on what kind of learning data group is used. When changing the control method for the shape pattern by the above method, a method of creating new learning data and adding it to the existing learning data group is used. The learning data in the learning data group only increases, and the time required for learning of the neural network also increases. For this reason, learning data may be deleted based on a period of time, or may be deleted at random. Therefore, it is ideal to add new learning data while leaving the learning data in the current learning data group.

When new learning data is generated, there should be learning data in the learning data group, which is a combination of the shape result judged to have no control effect and the control method for it. is added to the learning data group, the result is that the two conflict with each other, and the control rule that is the learning result does not become the intended one (the control rule equivalent to the new learning data). In order to solve this problem, it is necessary to add new learning data and at the same time delete learning data determined to have no control effect from the learning data group.

This makes it possible to limit the number of learning data included in the learning data group. As learning data increases, the time required for learning increases accordingly. Since the plant control device performs learning according to a predetermined schedule, it is ideal that the learning time is substantially constant. By deleting the learning data as described above, it is possible to limit the increase of the learning data and perform learning using a certain range of learning data.

Also, the control rules of the neural network do not change until learning is performed using a new set of learning data. Therefore, there is a case where the control is executed again using the control rule determined to have no control effect. Since it takes time to learn the neural network, until the learning result using the new learning data group is obtained, control is not performed using the control rule judged to have no control effect, and the corrected learning data The control effect can be enhanced by performing the control method by

FIG. 1 shows an outline of a plant control device according to an embodiment of the present invention. The plant control device of FIG. 1 inputs a controlled plant 1 and performance data Si from the controlled plant 1 and controls a control manipulated variable output SO determined according to a control rule (neural network) as exemplified in FIG. A control execution device 20 to be given to the target plant 1 for control and performance data Si from the control target plant 1 are input for learning, and control method learning for reflecting the learned control rules in the control rules in the control execution device 20. It consists of a device 21, a plurality of databases DB (DB1 to DB3), and a management table TB of the database DB.

The control execution device 20 includes a control input data generation unit 2, a control rule execution unit 10, a control output calculation unit 3, a control output suppression unit 4, a control output determination unit 5, and a control operation disturbance generation unit 16 as main elements. It is

In the control execution device 20 , control input data S<b>1 to be given to the control rule execution unit 10 is first created using the control input data creation unit 2 from the performance data Si of the rolling mill, which is the plant 1 to be controlled. Here, the actual data Si of the rolling mill is the state quantity of the rolling mill, and among the state quantities of the rolling mill, the state quantity given to the control rule execution unit 10 is distinguished as the control input data S1. Therefore, it can be said that the control input data S1 is also the performance data Si of the rolling mill.

The control rule execution unit 10 uses two neural networks (control rules) that express the relationship between the control target performance data Si (control input data S1 of the control rule execution unit 10) and the control operation command S2 to perform control. The operating terminal operation commands of the two neural networks are determined from the target performance data Si, and the optimum operating terminal operating command is created as the control operating terminal operating command S2. The control output calculation unit 3 calculates a control operation amount S3 to the control operation terminal based on the control operation terminal operation command S2. As a result, the control operation amount S3 is created using a neural network according to the performance data Si of the plant 1 to be controlled.

In addition, in the control output determination unit 5 in the control execution device 20, using the actual data Si from the controlled plant 1 and the control operation amount S3 from the control output calculation unit 3, whether or not the control operation amount is output to the control operation terminal Determine data S4. The control output suppression unit 4 determines whether or not to output the control operation amount S3 to the control operation terminal according to the control operation amount output permission data S4, and provides the controlled operation amount S3 to the plant 1 to be controlled. It is output as the control manipulated variable output SO. As a result, the control operation amount S3 determined to be abnormal is no longer output to the plant 1 to be controlled. In addition, the control operation disturbance generator 16 generates disturbance and gives it to the controlled plant 1 for the purpose of verifying the plant control device.

The control execution device 20 configured as described above refers to the control rule database DB1 and the output determination database DB3 in order to execute the processing, as will be described later. The control rule database DB1 is connected so as to be accessible to both the control rule execution unit 10 in the control execution device 20 and the control rule learning unit 11 in the control method learning device 21, which will be described later. A control rule (neural network) as a learning result in the control rule learning unit 11 is stored in the control rule database DB1, and the control rule execution unit 10 refers to the control rule stored in the control rule database DB1. The output determination database DB3 is connected to the control output determination section 5 in the control execution device 20 so as to be accessible.

FIG. 2 shows a specific configuration example of the control rule execution unit 10 according to the embodiment of the present invention. The control rule execution unit 10 receives the control input data S1 created by the control input data creation unit 2 and gives the control output calculation unit 3 a control operation terminal operation command S2.

The control rule execution unit 10 has neural networks 101 and 102 . Here, the neural networks 101 and 102 are neural networks learned according to the specification A regarding the control effect described later. The neural network 102 is a neural network trained according to the specification A1 for the control operation, and the neural network 102 is a neural network trained according to the specification A2 for the control effect in which the change in the shape deviation is small but can be corrected as a result of the control operation. . In the neural networks 101 and 102, the neural network operating terminal operation commands N1 and N2 are basically generated by using the control input data S1 created by the control input data creation unit 2 according to the technique of Patent Document 1 as shown in FIG. It has established.

The neural network operation terminal operation commands N1 and N2 are input to the output presence/absence determination unit 105 . An output presence/absence determination unit 105 determines whether or not an operation terminal operation command N1 with a large control effect and an operation terminal operation command N2 with a small control effect are output from the neural network 101, and if they are not output, a flag indicating no output is turned ON. , and if it is output, the flag indicating no output is turned OFF.

When the control effect is high, the output presence/absence determination unit 105 determines that there is an output from a neural network formed by learning a high control effect. A case where there is no output from the neural network formed by learning and a case where the control effect is high is judged to be a case where there is an output from the neural network formed by learning.

The control rule execution unit 10 according to the embodiment of the present invention manipulates the actual position of the control operation end as the state quantity Si of the rolling mill to be controlled, in addition to the control input data S1 created by the control input data creation unit 2. It is input to the end operation margin determination unit 106 . An operation terminal operation margin determination unit 106 determines whether or not there is a sufficient margin for operation by control at the control operation terminal. to

In the control output selection unit 107, using the no output flag from the output presence/absence determination unit 105 and the presence flag from the operation margin determination unit 106, the operation terminal operation command N1 from the neural network 101 and the operation command N1 from the neural network 102 are selected. is used, and a control output S3 is output to the control output calculation unit 3.

Specifically, as shown in FIG. 17, the control output selection unit 107 determines whether the control output selection unit 107 operates the operation terminal when the no-output flag is OFF (there is an operation instruction for the operation terminal with a large control effect). If the command N1 is selected and the no-output flag is ON (there is no operating-terminal operation command with a large control effect) and the margin flag is ON, then the operating-terminal operation command N2 is selected; When the flag is ON (there is no operating terminal operation command with a large control effect) and the margin flag is OFF, the control output is set to 0.

Thereby, when there is an output of a neural network formed by learning when the control effect is high, and when the control effect is high, the control terminal of the plant to be controlled is controlled according to the output, and when the control effect is low, and the control according to the output of the neural network formed by learning the case where the control effect is low when there is only an output of the neural network formed by learning the control When there is only the output of a neural network formed by controlling the operation terminal of the target plant and learning when the control effect is low, and when there is no margin in the operation terminal position at the operation terminal, the plant to be controlled is operated. Do not control the edge.

This is based on the idea that if the control effect is considered to be small and there is no margin in the control action at the control operation end, the life of the control operation end is ensured by not performing the operation. On the other hand, if the control effect is large, it can be said that the control method is determined by prioritizing the control effect over the service life.

The control rule execution unit 10 further includes neural network selection units 103 and 104, and refers to the control rules stored in the control rule database DB1 to select the optimum control rule as the control rule for the neural networks 101 and 102. Select and execute.

In this manner, the control rule execution unit 10 of FIG. 2 selects and uses a necessary neural network from a plurality of neural networks divided according to operator groups and control purposes. The control rule database DB1 preferably includes, as data from the plant 1 to be controlled, performance data (operation team data, etc.) Si that allows selection of a neural network and pass/fail judgment criteria. It should be noted that since execution of a neural network results in a control rule, in this specification the neural network and the control rule are used interchangeably without distinction.

Returning to FIG. 1 , the control method learning device 21 learns the neural networks 101 and 102 used in the control execution device 20 . When the control execution device 20 outputs the control manipulated variable output SO to the plant 1 to be controlled, it takes time for the actual control effect to appear as a change in the performance data Si. For this reason, learning is performed using data delayed by that amount of time. In FIG. 1, Z ⁻¹ represents an appropriate time delay function for each datum.

The control method learning device 21 is mainly composed of a control result quality determination unit 6, a learning data creation unit 7, a control rule learning unit 11, and a quality determination database DB4.

Of these, the control result quality determination unit 6 uses the performance data Si from the controlled plant 1, the previous performance data value Si0, and the quality determination data S5 stored in the quality determination database DB4 to improve the performance data Si. It is determined whether the direction has changed or the direction has changed for the worse, and control result pass/fail data S6 is output.

In the learning data creation unit 7 in the control method learning device 21, the input data such as the control operation terminal operation command S2, the control operation amount S3, and the control operation amount output availability data S4 created by the control execution device 20 are processed at the same time. Using the delayed data and the control result quality data S6 from the control result quality determination unit 6, new teacher data S7a to be used for neural network learning is created and supplied to the control rule learning unit 11. The teacher data S7a corresponds to the control operation terminal operation command S2 output by the control rule execution unit 10, and the learning data creation unit 7 uses the control result quality data S6 given by the control result quality determination unit 6. It can be said that the data obtained by estimating the control operation terminal operation command S2 output by the control rule execution unit 10 is obtained as the new teaching data S7a.

FIG. 3 shows a specific configuration example of the control rule learning unit 11 according to the embodiment of the present invention. The control rule learning unit 11 is composed of an input data creating unit 114, a teacher data creating unit 115, a neural network processing unit 110, and a neural network selecting unit 113 as main components. Further, the control rule learning unit 11 obtains data S8a obtained by delaying the control input data S1 from the input data generating unit 2 and new teacher data S7a from the learning data generating unit 7 as inputs from the outside. The data accumulated in the rule database DB1 and the learning data database DB3 are referred to.

In the control rule learning section 11, the control input data S1 is taken into the neural network processing section 110 via the input data creating section 114 after appropriate time delay compensation.

In the control rule learning unit 11, the new teacher data S7a from the learning data creating unit 7 is the total teacher data S7c including the past teacher data S7b stored in the learning data database DB2 in the teacher data creating unit 115. , is given to the neural network processing unit 110 . These teacher data S7a and S7b are appropriately stored in the learning data database DB2 and used.

Similarly, the input data S8a from the control input data generation unit 2 is used as the total input data S8c including the past input data S8b stored in the learning data database DB2 in the input data generation unit 114, and the neural network processing unit 110. These input data S8a and S8b are appropriately stored in the learning data database DB2 and used.

The neural network processing unit 110 includes a neural network 111 and a neural network learning control unit 112. The neural network 111 receives input data S8c from the input data generation unit 114, teacher data S7c from the teacher data generation unit 115, The control rule (neural network) selected by the neural network selection unit 113 is taken in, and the finally determined neural network is stored in the control rule database DB1.

The neural network learning control unit 112 controls the input data generation unit 114, the teacher data generation unit 115, and the neural network selection unit 113 at appropriate timings, obtains input to the neural network 111, and outputs processing results. It is controlled to be stored in the control rule database DB1.

Here, the neural networks 101 and 102 in the control execution device 20 in FIG. 2 and the neural network 111 in the control method learning device 21 in FIG. The above differences are explained as follows. First, the neural networks 101 and 102 in the control execution device 20 are neural networks with predetermined contents, and obtain a control operation terminal operation command S2 as a corresponding output when control input data S1 is given, It is, so to speak, a neural network used for one-way processing. On the other hand, the neural network 111 in the control method learning device 21 satisfies this input/output relationship when the control input data S1 and the input data S8c and teacher data S7c for the control input data S1 and the control operation terminal operation command S2 are set as learning data. It is for finding a neural network that

The concept of basic processing in the control method learning device 21 configured as described above is as follows. First, when the content of the control operation amount output availability data S4 is "permissible", the control operation amount output SO is output to the controlled plant 1, and the content of the control result quality data S6 is "good" (actual data Si improves). direction change), the control operation terminal operation command S2 output by the control rule execution unit 10 is determined to be correct, and learning data is created so that the output of the neural network becomes the control operation terminal operation command S2.

On the other hand, the content of the control operation amount output propriety data S4 is "no", or the control operation amount output SO is output to the controlled plant 1, and the content of the control result quality data S6 is "no" (actual data Si deteriorates). direction change), it is determined that the control operation terminal operation command S2 output by the control rule execution unit 10 is erroneous, and learning data is created so as not to output the output of the neural network. At this time, as the control output, the neural network output is configured so that two types of output, + direction and - direction, are output to the same control operation terminal, and the control operation terminal operation command S2 on the output side is not output. Create training data as follows.

Further, in the control rule learning unit 11 illustrated in FIG. 3, the results of data processing by the neural network learning control unit 112 are processed as follows. Here, first, using learning data that is a combination of S8c obtained by delaying the control input data S1 to the control execution device 20 and teacher data S7c created by the teacher data creation unit 115, the control rule execution unit 10 The neural networks 101 and 102 used are trained. Actually, a neural network 111 which is the same as the neural networks 101 and 102 of the control rule execution unit 10 is provided in the control rule learning unit 11, and operation tests are performed under various conditions to learn the response at that time. One obtains control rules that have been verified to produce good results. Since learning needs to be performed using a plurality of pieces of learning data, a plurality of pieces of past learning data are taken out from the learning data database DB2 that stores learning data created in the past, and learning is performed. While executing, the learning data of this time is stored in the learning data database DB2. Also, the learned neural network is stored in the control rule database DB1 for use by the control rule execution unit 10. FIG.

At this time, the past learning data should include the learning data that causes the output of the control operation terminal operation command S2, which is the basis of the learning data updated this time, and the learning data updated this time is added as it is. Even if it learns by doing so, it will result in learning with conflicting learning data, which will hinder the neural network from learning a new control method. Therefore, it is preferable to delete the past learning data most similar to the combination of the original control input data S1 and the control operation terminal operation command S2 of the learning data updated and added this time.

Neural network learning may be performed using past learning data each time new learning data is created. You may learn using data together.

In addition, the control result pass/fail determination unit 6 performs pass/fail determination based on pass/fail determination criteria from the pass/fail determination database DB4. Since the judgment result of the control result differs depending on the control purpose, it is possible to create multiple neural networks according to multiple control purposes, and even if the input data is the same, create training data for each control purpose and learn. By creating multiple teacher data for one input data and using it for training the neural network corresponding to each teacher data, the neural network corresponding to multiple control purposes can be learned at the same time. is possible. Here, the plurality of control purposes are, for example, in the case of shape control, which part in the strip width direction (strip edge part, center part, asymmetric part, etc.) is to be preferentially controlled, and a plurality of control target items (for example, which of the plate thickness, tension, rolling load, etc.) should be preferentially controlled.

In the above configuration, once the neural network 101 used in the control rule execution unit 10 has learned, no new control operation is performed. Therefore, the control operation disturbance generator 16 generates a new operation method in a random manner at appropriate times, and by executing the control operation in addition to the control operation amount S3, the new control method is learned.

The details of the present plant control method will be described below for shape control in a Sendzimir rolling mill as disclosed in Patent Document 1. Regarding the shape control, it is assumed that the following specifications A and B are adopted.

The specification A is a specification relating to the control effect. As a result of the control operation, A1 indicates a case where the shape deviation is greatly corrected, and A2 indicates a case where the change in the shape deviation is small but can be corrected.

The specification B is a specification for dealing with previously known conditions. To give an example, since the relationship between the shape pattern and the control method changes depending on various conditions, it is conceivable that it is necessary to classify the specifications B1 by sheet width and the specifications B2 by steel grade, for example. As each of the above changes, the degree of influence on the shape of the shape manipulation end changes.

In this example, the plant 1 to be controlled is a Sendzimir rolling mill, and the performance data is the shape performance. The Sendzimir rolling mill is a rolling mill with cluster rolls for cold rolling hard materials such as stainless steel. A Sendzimir mill uses small diameter work rolls to apply a high reduction to hard materials. Therefore, it is difficult to obtain a flat steel plate. As a countermeasure, a cluster roll structure and various shape control parts are adopted. The Sendzimir rolling mill is generally equipped with upper and lower first intermediate rolls having a single taper so that they can be shifted, as well as upper and lower 6 split rolls and 2 rolls called AS-U. In the example described below, detection data of a shape detector is used as actual shape data Si, and shape deviation, which is a difference from a target shape, is used as control input data S1. As the control operation amount S3, the AS-U of #1 to #n and the roll shift amounts of the upper and lower first intermediate rolls are used.

FIG. 4 shows the configuration of a neural network used for shape control of a Sendzimir rolling mill. Here, the neural networks are the neural networks 101 and 102 for the control rule execution section 10, and the neural network shown in the neural network 111 for the control rule learning section 11. are the same.

In the example of the shape control of the Sendzimir rolling mill shown in FIG. 4, the actual data Si from the plant 1 to be controlled is the data of the shape detector (here, the shape deviation, which is the difference between the actual shape and the target shape, is output). ), and the control input data generator 2 obtains a normalized shape deviation 201 and a shape deviation stage 202 as the control input data S1. Thus, the input layers of the neural networks 101 , 102 and 111 are composed of the normalized shape deviation 201 and the shape deviation stage 202 . In FIG. 4, the shape deviation stage 202 is used as an input to the neural network input layer, but the neural network may be switched according to the stage.

The output layer is composed of an AS-U operation degree 301 and a first intermediate operation degree 302 in accordance with the AS-U and the first intermediate roll, which are shape control operation ends of the Sendzimir rolling mill. For AS-U, the AS-U opening direction (the direction in which the roll gap (the space between the upper and lower work rolls of the rolling mill) opens) and the AS-U closing direction (the direction in which the roll gap closes) Have about AS-U. In addition, for the first intermediate roll, the first intermediate roll opening direction (the direction in which the first intermediate roll moves outward from the center of the rolling mill), the first intermediate roll closing direction (the first intermediate roll moves toward the center of the rolling mill) direction) for the upper and lower first intermediate rolls. For example, if the shape detector has 20 zones and the shape deviation stage 202 has three stages (large, medium, and small), the input layer will have 23 inputs. If the number of AS-U saddles is 7 and the upper and lower first intermediate rolls are shiftable in the strip width direction, the output layer has 14 AS-U operation degrees 301 and 4 intermediate operation degrees, for a total of 18 pieces. becomes. The number of intermediate layers and the number of neurons in each layer are set as appropriate. As will be described later with reference to FIG. 8, the shape control operation terminal of the Sendzimir rolling mill, which is the output layer, is output from a neural network so that two types of outputs, + direction and - direction, are output for each control operation terminal. constitutes

FIG. 10 shows the shape deviation and the control method. Here, the upper part of FIG. 10 shows the control method when the shape deviation is large, and the lower part of FIG. 10 shows the control method when the shape deviation is small. The height direction is the size of the shape deviation, and the horizontal axis direction is the width direction of the plate. As shown in the upper part of FIG. 10, when the shape deviation is large, priority is given to correcting the overall shape over local shape deviation in the strip width direction. On the other hand, as shown in the lower part of FIG. 10, when the shape deviation is small, priority is given to reducing the local shape deviation.

Since it is necessary to change the control method according to the magnitude of the shape deviation, a shape deviation stage 202 is provided as shown in FIG. judge. As for the shape deviation, it is preferable to use the one normalized to, for example, 0 to 1 regardless of the magnitude of the shape deviation. This is just an example, and it is possible to input the shape deviation as it is to the input layer of the neural network without normalization, or change the neural network itself according to the size of the shape deviation (for example, two neural It is also conceivable to prepare nets and separate the neural net to be used when the shape deviation is large and the neural net to be used when the shape deviation is small).

The neural networks 101, 102, and 111 configured as shown in FIG. 4 described above are made to learn how to operate shape patterns, and shape control is performed using the learned neural networks. Even neural networks with the same configuration can have different characteristics depending on learning conditions, and can output different control outputs for the same shape pattern.

Therefore, by selectively using a plurality of neural networks according to other conditions of the actual shape, optimum control can be configured for various conditions. This corresponds to specification B. The configuration of FIG. 2 described above shows a specific example of such specifications. In the configuration example of FIG. 2, the neural networks 101 and 102 used in the control rule execution unit 10 are prepared according to the rolling performance, the name of the rolling mill operator, the steel type of the material to be rolled, the strip width, etc., and controlled. It is registered in the rule database DB1. Neural network selection units 103 and 104 select neural networks that match the conditions at that time, and set them to neural networks 101 and 102 of control rule execution unit 10 . As a condition at that time in the neural network selection units 103 and 104, it is preferable to take in strip width data from the performance data Si of the plant 1 to be controlled and select a neural network accordingly. Also, the plurality of neural networks used here may have different numbers of intermediate layers and different numbers of units in each layer as long as they have an input layer and an output layer as shown in FIG.

FIG. 7 shows an outline of the control input data creating section 2 that creates the control input data S1 (normalized shape deviation 201, shape deviation stage 202) to be input to the input layers of the neural networks 101 and 102111. As shown in FIG. Here, as the performance data Si, the shape detector data of the shape detector that detects the strip shape during rolling in the Sendzimir rolling mill, which is the plant 1 to be controlled, is input. A shape deviation PP value (Peak To Peak value) S _PP , which is the difference between the maximum value and the minimum value of the detection result of the shape detector zone, is obtained. The shape deviation stage calculator 211 classifies the shape deviation into three stages of large, medium, and small according to the shape deviation PP value _SPP . The shape is the widthwise distribution of the elongation of the material to be rolled, and the unit is I-UNIT, which expresses the elongation in units of 10-5. For example, they are classified as in the following formula.

Here, the shape deviation stage is set to (large = 1, medium = 0, small = 0) by formula (1), and the shape deviation stage is set to (large = 0, medium = 1, small by formula (2). = 0), and according to the formula (3), the shape deviation stages are classified as (large = 0, medium = 0, small = 1). Here, the shape deviation of each zone is normalized using S _PM , where S _PM =S _PP .

As described above, the normalized shape deviation 201 and the shape deviation stage 202, which are input data to the neural networks 101 and 102, are created. The normalized shape deviation 201 and the shape deviation step 202 are the control input data S1 of the control rule execution unit 10. FIG.

FIG. 8 shows an outline of the control output calculator 3. As shown in FIG. The control output calculation unit 3 receives the control operation terminal operation command S2 output from the neural network 101 in the control rule execution unit 10 (in the case of shape control of a Sendzimir rolling mill, AS-U operation degree 301, first intermediate The operation degree 302 corresponds to this), a control operation amount S3, which is an operation command to each shape control operation end, is created. Here, one data example is shown for each of the AS-U operation degree 301 and the first intermediate operation degree 302, which have a plurality of numbers, and each data is composed of a pair of data of the opening direction degree and the closing direction degree. It is

In the control output calculation unit 3, the input AS-U operation degree 301 has _outputs for each AS-U opening direction and closing direction. Outputs an operation command to U. Since the control output to each AS-U is the AS-U position change amount (unit: length), the conversion gain _GASU is the conversion gain from the degree to the position change amount.

Also, since the input first intermediate operation degree 302 has first intermediate outer and inner outputs, by multiplying the difference between them by the conversion gain G _1ST , an operation command for each first intermediate roll shift is output. do. The conversion gain G _1ST is a conversion gain from the degree to the position change amount because the control output to each first intermediate roll is the first intermediate roll shift position change amount (unit: length).

As described above, the control operation amount S3 can be calculated. The control operation amount S3 is composed of #1 to #n AS-U position change amount (where n is the number of AS-U roll saddles), upper first intermediate shift position change amount, and lower first intermediate shift position change amount. ing. FIG. 8 shows a system for adding disturbance data from the control operation disturbance generator 16 to the control operation terminal operation command S2.

FIG. 9 shows an outline of the control output determination section 5. As shown in FIG. The control output determination unit 5 is composed of a rolling phenomenon model 501 and a shape correction quality determination unit 502, and includes performance data Si from the controlled plant 1, control operation amount S3 from the control output calculation unit 3, and an output determination database. The information of DB3 is obtained, and the control operation amount output propriety data S4 to the control operation terminal is given. With such a configuration, in the control output determination unit 5, the change in shape when the control operation amount S3 calculated by the control output calculation unit 3 is output to the rolling mill, which is the plant 1 to be controlled, is determined as the known plant 1 to be controlled. Prediction is made by inputting it into the model (rolling phenomenon model 501 in the case of the embodiment of FIG. 9), and if the shape is expected to deteriorate, the control operation amount output SO is suppressed to prevent the shape from greatly deteriorating. do.

More specifically, the control operation amount S3 is input to the rolling phenomenon model 501, the change in shape due to the control operation amount S3 is predicted, and the shape deviation correction amount prediction data 503 is calculated. On the other hand, the shape deviation correction amount prediction data 503 is added to the shape detector data Si (current shape deviation actual data 504) from the controlled plant 1 to obtain the shape deviation prediction data 505. By evaluating , it is possible to predict how the shape will change when the control operation amount S3 is output to the plant 1 to be controlled. Based on the actual shape deviation actual data 504 and the shape deviation prediction data 505, the shape correction quality determination unit 502 determines whether the shape will change in the direction of improvement or deterioration. Get S4.

Specifically, the shape correction quality determination unit 502 determines the quality of the shape correction as follows. First, in order to consider the control priority in the strip width direction, a weighting factor w(i) in the strip width direction is set in the output determination database DB3. Using this, the quality of the shape change is determined using, for example, an evaluation function J such as the following equation (4). In equation (4), w(i) is the weighting factor, εfb(i) is the actual shape deviation 504, εest(i) is the predicted shape deviation 505, i is the shape detector zone, and ran is a random number term.

When the evaluation function J of the formula (4) is used, the evaluation function J is positive when the shape is good, and negative when the shape is bad. Also, rand is a random number term that changes the evaluation result of the evaluation function J in a random number manner. As a result, even if the shape deteriorates, the evaluation function J may become positive. Therefore, even if the rolling phenomenon model 501 is not correct, the relationship between the shape pattern and the control method can be learned. is possible. Here, rand increases the maximum value when the model of the controlled plant 1 is uncertain, such as at the beginning of trial operation, and sets it to 0 when it is desired to learn the control method to some extent and implement stable control. change.

In the shape correction quality determination unit 502, an evaluation function J is calculated. No), the control operation amount output enable/disable data S4 is output.

The control output suppression unit 4 determines whether or not to output the control operation amount output SO to the plant 1 to be controlled according to the control operation amount output propriety data S<b>4 that is the determination result of the control output determination unit 5 . The control operation amount output enable/disable data S4 is #1 to #nAS-U position change amount output, upper first intermediate shift position change amount output, lower first intermediate shift position change amount output,
IF (control operation amount output enable/disable data S4=0) THEN
#1 to #nAS-U position change amount output = 0
Upper first intermediate shift position change amount output=0
Lower first intermediate shift position change amount output=0
ELSE
#1 to #nAS-U position change amount output = #1 to #nAS-U position change amount Upper 1st intermediate shift position change amount output = Upper 1st intermediate shift position change amount Lower 1st intermediate shift position change amount output = Lower first intermediate shift position change amount ENDIF
determined by

In the control execution device 20, the above calculation is executed from the performance data Si from the control target plant 1 (rolling mill), and the shape control is performed by outputting the control operation amount output SO to the control target plant 1 (rolling mill). to implement.

Next, an overview of the operation of the control method learning device 21 will be described. In the control method learning device 21, the time delay data of the data used in the control execution device 20 is used. The time delay Z ^-1 means e ^-TS and indicates that it is delayed by a preset time T. Since the controlled plant 1 has a time response, there is a time delay until the actual data changes due to the control manipulated variable output SO. Therefore, the learning is performed using the performance data when the delay time T has passed after the execution of the control operation. In shape control, it takes several seconds for the shape meter to detect a change in shape after outputting an operation command to the AS-U or the first intermediate roll. Since the delay time changes depending on the type and the rolling speed, the optimum time until the change of the control end changes the shape may be set as T).

FIG. 11 shows an outline of the operation of the control quality determination section 6. As shown in FIG. The shape change quality determination unit 602 uses a quality determination evaluation function J _C for the control effect as expressed by the following equation.

In the equation (5), εfb(i) is the shape deviation actual data included in the actual data Si, εlast(i) is the previous value of the shape deviation actual data, and wC(i) is the strip width direction weight for quality judgment. is the coefficient. Here, the weighting factor wC(i) for quality determination is set according to the specification of the priority of control in the strip width direction from the quality determination database DB4.

Also, in the equation (5), a(a1, a2) is set according to the specifications A1, A2 regarding the control effect. The a1 is the specification A1 for correcting the shape deviation to a large extent, and the a2 is the specification A2 for correcting the shape deviation small. S ₂ (j) is the control final command for control device j, and max|S ₂ (j)| is the maximum absolute value of the control final command.

The quality of the control result is determined by the quality determination evaluation function Jc. In addition, when the control operation amount output availability data S4, which is the determination result of the control output determination unit 5, is 0 (control output is not possible), the control operation amount output to the controlled plant 1 is actually 0, but the shape is judge it to be bad.

Here, when the control operation amount output propriety data S4=0, the control result pass/fail data S6=-1. Also, the threshold upper limit LCU and the threshold lower limit LCL are set in advance under the threshold condition (LCU≧0≧LCL). At this time, if the result of comparison with the quality judgment evaluation function Jc is Jc>LCU, the control result quality data S6 is set to -1 (the shape has deteriorated),
If LCU≧Jc≧0, control result good/bad data S6=0 (changes to worse shape)
year,
If 0>Jc≧LCL, control result good/bad data S6=1 (changes in the direction of better shape)
year,
If Jc<LCL, the control result quality data S6 is set to 0 (the shape is improved).

Here, the control result good/bad data S6=-1 indicates that the output control output is suppressed because the shape has deteriorated. When the control output is held, the control result good/bad data S6=1 has changed in the direction of improving the shape, but since there is a possibility that the shape will be improved further, the output control amount is increased.

As described above, the weighting factor wC(i) in the sheet width direction changes according to the specifications A1 and A2 regarding the control effect, so the pass/fail judgment evaluation function Jc differs. Therefore, it is conceivable that the determination result of the control result pass/fail data S6 is also different. Therefore, in the control method learning device 21, the control result pass/fail data S6 is determined for two types of specifications A1 and A2 regarding the control effect.

Next, an overview of the learning data creation unit 7 will be described. As shown in FIG. 1, in the learning data generation unit 7, based on the determination result (control result quality data S6) from the control result quality determination unit 6, the control operation terminal operation command S2, the control operation amount S3, Teacher data S7a for the neural network 111 used in the control rule learning unit 11 is created from the determination result of the control output suppressing unit (control operation amount output propriety data S4).

The teacher data S7a in this case are the AS-U manipulation degree 301 and the 1 intermediate manipulation degree 302, which are outputs from the output layer of the neural network 111 shown in FIG. The learning data generation unit 7 generates the control operation terminal operation command S2 (AS-U operation degree 301, 1 intermediate operation degree 302) which are the outputs of the neural networks 101 and 102, and #1 to #nAS which are the control operation amount outputs SO. -Using the U position change amount output, the upper first intermediate shift position change amount output, and the lower first intermediate shift position change amount output, teacher data S7a for the neural network 111 used in the control rule learning unit 11 is created.

In order to explain the outline of the operation of the learning data generating section 7, FIG. 12 summarizes the relationship between data and symbols in the control output computing section 3 of FIG. Here, the AS-U operation degree 301 is representatively shown for the control operation terminal operation command S2, which is the output of the neural network 101. OPref is the data on the positive side of the operation degree, OMref is the data on the negative side of the operation degree, and OMref is the data on the negative side of the operation degree. The operation degree randomly generated from the operation disturbance generator 16 is assumed to be an operation degree random number Oref, the conversion gain is G, and the control operation amount output SO is Cref. Thus, here, for the sake of simplicity, the output from the output layer of the neural network 101 of the control rule execution unit 10 is the positive side of the operation degree and the negative side of the operation degree, and the random number generated from the control operation disturbance generation unit 16. The degree of manipulation is used as the degree of manipulation random number. Also, the control operation amount output SO for the control operation terminal is used as the operation command value.

FIG. 13 shows the processing steps and processing contents in the learning data creation unit 7. As shown in FIG. 12, in the first processing step 71, the operation command value Cref is obtained by the formula (6).

In the next processing step 72, the operation command value Cref is corrected according to the control result good/bad data S6 to be C'ref. Specifically, when the control result pass/fail data S6=-1, the operation command is given by the formula (7), when the control result pass/fail data S6=0, by the formula (8), and when the control result pass/fail data S6=1, by the formula (9). Let C'ref be the modified value of Cref.

At the processing step 73, the operation degree correction amount ΔOref is obtained from the corrected operation command value C'ref by the equations (10) and (11).

At the processing stage 74, teacher data OP'ref and OM'ref to the neural network 111 are obtained from the equation (12).

As shown in FIG. 12, the learning data creation unit 7 converts the operation command value Cref actually output to the plant 1 to be controlled into the control result quality data S6, which is the judgment result of the control result judgment unit 6. , the operation command value correction value C'ref is calculated. Specifically, when the control result data S6=1, the control direction is OK, but the control output is insufficient, and the operation command value is increased by ΔCref in the same direction. to Conversely, when the control result quality data S6=-1, it is determined that the control direction is wrong, and the operation command value is decreased by ΔCref in the opposite direction. Since the conversion gain G is set in advance and is already known, if the values of the positive side and the negative side of the degree of operation are known, it is possible to obtain the correction amount ΔOref. Here, ΔCref is set by finding an appropriate value in advance by simulation or the like. Through the above procedure, OP'ref and OM'ref can be obtained by the above equation (12) in the control rule learning unit 11. FIG.

Note that FIG. 13 illustrates a simple example, but in reality, the AS-U operation degree 301 for #1 to #n AS-Us and the first upper intermediate roll shift and the first lower intermediate roll shift. All of them are performed for 1 intermediate operation degree 302 and used as teacher data (AS-U operation degree teacher data, 1 intermediate operation degree teacher data) of neural network 111 used in control rule learning unit 11 .

FIG. 14 shows an example of data stored in the learning data database DB2. In order to learn neural network 111, a large number of combinations of input data S8a and teacher data S7a are required. Therefore, the teacher data S7a (AS-U manipulation degree teacher data, first intermediate manipulation degree) created by the learning data creation unit 7 is the input data S1 (standard It is stored in the learning data database DB2 as a set of learning data S11 in combination with the shape deviation 201 and the time delay data S8a of the shape deviation stage 202).

1 uses various databases DB1, DB2, DB3, DB4, and DB5. 2 shows the configuration of the neural network management table TB of . The management table TB includes a specification management table. Specifically, the management table TB is divided according to specifications A1 and A2 regarding (B1) strip width, (B2) steel type, and control priority. (B1) Sheet widths are, for example, 3 foot width, meter width, 4 foot width, and 5 foot width, and steel grades are about 10 grades (1) to (10). In addition, two specifications A1 and A2 are used for the specification A regarding the priority of control. In this case, there are 80 divisions, and 80 neural networks are selectively used according to the rolling conditions.

The neural network learning control unit 112 assigns learning data, which is a combination of input data and teacher data as shown in FIG. and the used neural network are stored in the learning data database DB2 as shown in FIG.

Each time the control execution device 20 executes shape control on the plant 1 to be controlled, two sets of learning data are created. This is because, for the same input data and control output, the quality of the control result is determined using two evaluation criteria, specification A1 and specification A2, regarding the priority of control, and two types of teacher data are created. is. When a certain amount of training data (for example, 200 sets) is accumulated, or when the learning data database DB2 is newly accumulated, the neural network learning control unit 112 instructs the neural network 111 to learn.

A plurality of neural networks are stored in the control rule database DB1 according to a management table TB as shown in FIG. is designated, and the neural network selection unit 113 extracts the neural network from the control rule database DB1 and sets it in the neural network 111 . The neural network learning control unit 112 instructs the input data creating unit 114 and the teaching data creating unit 115 to extract the input data and the teacher data corresponding to the neural network from the learning data database DB2, and uses them to perform neural training. The net 111 is trained. Various methods have been proposed for the neural network learning method, and any of these methods may be used.

When the learning of the neural network 111 is completed, the neural network learning control unit 112 assigns the learning result of the neural network 111 to the neural network No. of the control rule database DB1. Learning is completed by writing back to the position of .

Learning may be performed at regular time intervals (for example, every day) for all the neural networks defined in FIG. No. Only the neural network of is allowed to learn at that time.

As described above, without greatly disturbing the shape of the rolling mill, which is the controlled plant 1,
1) Instead of setting reference shape patterns and control operations for them separately in advance and learning control operation methods, learn combinations of shape patterns and control operations and use them to perform control operations. .
2) New control rules cannot be predicted in advance, and there are cases where completely unpredictable control rules are optimal. go.

The neural network used by the control execution device 20 is stored in the control rule database DB1, but if the stored neural network only performs initial processing with random numbers, the learning of the neural network progresses. However, it will take some time before a reasonable degree of control becomes possible. Therefore, when the control unit is constructed for the controlled plant 1, based on the control model of the controlled plant 1 that is known at that time, the control rules are learned in advance by simulation, and the simulator By storing the trained neural network in the database, it is possible to control the performance to some extent from the beginning of the start-up of the plant to be controlled.

As is clear from the above description, the contents of the control rule database DB1 formed as a result of the learning process in the neural network learning control unit 112 include the neural network learned with respect to the specification A1 with high control effect and the specification with low control effect. and a neural network trained on A2. The former is reflected in the neural network 101 of the control rule execution section via the control rule database DB1, and the latter is reflected in the neural network 102 of the control rule execution section via the control rule database DB1.

The plant control apparatus of the present invention is actually implemented as a computer system, and in this case, a plurality of program groups are formed within the computer system.

These programs are, for example,
A control rule execution program that provides a control output in accordance with a predetermined combination of actual data of a plant to be controlled and a control operation in order to achieve the processing of the control execution device, and determines whether or not the control output output by the control rule execution program is permitted. , a control output determination program for notifying the control method learning device that the actual data and the control operation are erroneous; and when the control output determination program outputs the control output to the plant to be controlled, the actual data of the plant to be controlled is a control output suppression program that prevents the control output from being output to the controlled plant when it is determined that the
In order to achieve the processing of the control method learning device, when the control execution device actually outputs the control output to the plant to be controlled, after the time delay until the control effect appears in the actual data, the actual data is A control result quality determination program for achieving a control result quality determination process that determines whether the control result is good or bad as compared to , the quality of the control result in the control result quality determination program, They are a learning data creation program that obtains teacher data using control output, and a control rule learning program that learns the performance data and the teacher data as learning data.
Then, by learning by the control method learning device, individual combinations of actual data and control operations are obtained for a plurality of control targets according to the state of the plant to be controlled, and the combinations of the obtained actual data and control operations are obtained. is used as a prescribed combination of actual data of the plant to be controlled and the control operation in the control rule execution program.

In applying the device of the present invention to an actual plant, it is necessary to determine the initial values of the neural network. It is preferable to shorten the learning period for combinations of actual data and control operations in the plant to be controlled by creating a simulation using a control model of the plant to be controlled.

The effects of the present invention described above will be described in detail with reference to FIGS. 18 and 19. FIG. First, the neural networks 101 and 102 in FIG. 2 each obtain the same control input data S1 from the control input data generator 2, but are neural networks reflecting results of learning from different viewpoints of control effects. , give different outputs N1 and N2 of the operation terminal operation commands from each other. The outputs N1 and N2 of the operation terminal operation commands may be obtained from both at the same timing, or may be obtained from only one of them.

FIG. 18 is a diagram showing the relationship between shape evaluation results and control outputs. Here, the shape evaluation result is plotted on the vertical axis and the time on the horizontal axis to illustrate that the shape evaluation result has decreased over time.

In this example, in the first section T1, the neural network 101 outputs the operated terminal operation command N1 and the neural network 102 outputs the operated terminal operation command N2. The operation command N1 is selected, and in this case, the presence or absence of a control operation margin at the control operation end is not taken into consideration. Whether or not there is a margin, the operation terminal operation command N1 having a large control effect is selected and reflected in the control. Note that even when the neural network 102 does not output the operation terminal operation command N2, the control output selection unit 107 selects the operation terminal operation command N1 having a large control effect, and determines whether or not there is a margin for the control operation at the control operation terminal. do not consider.

Next, in the second section T2, the neural network 101 does not output the operation terminal operation command N1, and the neural network 102 outputs only the operation terminal operation command N2. An interval T21 in the first half of this interval T2 represents a state in which there is a margin in the control operation at the control operation terminal, and the control output selection unit 107 selects the operation terminal operation command N2 with a small control effect, and performs control using this. to execute. On the other hand, the latter half of the interval T22 in the interval T2 represents a state in which there is no margin for the control operation at the control operation terminal, and the control output selection unit 107 does not select the operation terminal operation command N2 with a small control effect. . As a result, this section will be in an uncontrolled state.

FIG. 19 shows the relationship between the operating end position and margin at the control operating end. The vertical axis in FIG. 19 represents the operating end position and margin, and the vertical axis represents time. In this figure, if the operating end position is, for example, the position of a valve, the valve is movable in the range from 0 to the control limit LL. In the present invention, a margin level LM (0<margin level LM<control limit LL) is set. It is movable.

On the other hand, in the case of the operation terminal operation command N2 having a small control effect, the movable range of the valve is from 0 to the margin level LM. In addition, during the period T0 in which the valve position reaches the margin level LM when the operating-end operation command N2 has a small control effect, control by the operating-end operation command N2 having a small control effect is blocked. Note that this diagram only shows the movable range, and does not show that two operation terminal operation commands exist at the same time. Moreover, although the example in which the margin level is set on the upper limit side for the valve position is shown, it is also possible to set it on the lower limit side in the same way.

As is clear from FIGS. 18 and 19, the present invention has the following effects. First, for example, in the case of a rolling mill, there are a plurality of valves serving as operating ends. In such a case, if one of the plurality of valves reaches the margin level LM, no control output is output. . As a result, it is possible to suppress the wear and tear of the valve mechanism due to the extreme movement of the valve when the control effect is small.

It may also indicate the event that some of the valves continue to move in the open direction and some other valves continue to move in the closed direction. In general, it is desirable to operate the valve in the vicinity of the central position, and it is not desirable to operate it in the end positions. It is useful in the sense of increasing or enlarging the operation allowance. In particular, it is possible to leave an operation allowance for when the operation terminal operation command N1 having a large control effect starts to be output after this, and it is possible to give priority to the operation having a high control effect.

The present invention relates to a control method and unit of a rolling mill, which is one of rolling equipment, for example,
There are no particular problems in actual application.

1: controlled plant 2: control input data creation unit 3: control output calculation unit 4: control output suppression unit 5: control output determination unit 6: control result quality determination unit 7: learning data creation unit 10: control rule execution unit 11 : Control rule learning unit 20: Control execution device 21: Control method learning device DB1: Control rule database DB2: Output determination database DB3: Learning data database Si: Actual data SO: Control operation amount output S1: Input data S2: Control operation terminal Operation command S3: control operation amount S4: control operation amount output propriety data S5: pass/fail judgment data S6: control result pass/fail data S7a, S7b, S7c: teacher data S8a, S8b, S8c: input data (for control rule learning section)

Claims (16)

A control method learning device that learns a combination of actual data of a plant to be controlled and a control operation according to control effects, and forms a plurality of neural networks with different control effects. A plant control device comprising a control execution device that provides a control output for controlling an operation terminal,
When there is an output of a neural network formed by learning when the control effect is high and the control effect is high, the control execution device controls the operation terminal of the plant to be controlled according to the output, and controls the control effect When there is only the output of the neural network formed by learning when the control effect is low, and when there is a margin in the operating terminal position at the operating terminal, the output of the neural network formed by learning when the control effect is low is used to control the operating terminal of the plant to be controlled, and only the output of the neural network formed by learning when the control effect is low exists, and when there is no margin in the operating terminal position at the operating terminal, the controlled object A plant control device that does not control an operating end of a plant. The plant control device according to claim 1,
The control execution device selects an optimum control output in consideration of outputs from a plurality of neural networks according to control effects and the control terminal position according to a predetermined combination of actual data of the plant to be controlled and control operation. , a control rule execution unit that determines whether or not the control output output by the control rule execution unit is permitted, and a control output determination unit that notifies the control method learning device that the actual data and the control operation are erroneous; and, if the control output determining unit determines that the performance data of the controlled plant deteriorates when the control output is output to the controlled plant, the control output is prevented from being output to the controlled plant. and a control output suppression unit,
In the control method learning device, when the control execution device actually outputs the control output to the plant to be controlled, after the time delay until the control effect appears in the actual data, the actual data is better than before the control. Control rule learning that learns the quality of the control result in the control result quality determination unit, the actual data and the teacher data as learning data. and the control method learning device learns to obtain a combination of separate performance data and control operations for a plurality of control targets according to the state of the plant to be controlled, and the obtained performance data and When the combination of control operations is used as a predetermined combination of the actual data of the controlled plant and the control operations in the control rule execution unit, and the state of the controlled plant is similar to the combination of the actual data and the control operations before correction, A plant control device that performs control using corrected learning data. The plant control device according to claim 1,
In order to change the combination of the actual data and the control operation according to the size of the actual data of the plant to be controlled, we use the information on the size of the actual data and the information that standardizes the actual data to facilitate pattern recognition. A plant control device characterized by learning and controlling a combination of a control operation and a control operation. The plant control device according to claim 1,
The control execution device holds, as a first neural network, a predetermined combination of performance data and control operations of a plant to be controlled, and the control method learning device stores a combination of performance data and control operations in a second neural network. and uses a second neural network obtained as a result of learning in the control method learning device as the first neural network in the control execution device. The plant control device according to claim 1,
The plant control device, wherein the control execution device includes a control operation disturbance generation unit that applies a disturbance to the control output, and the control method learning device learns even when the disturbance is applied. The plant control device according to claim 1,
The control method learning device obtains a plurality of combinations of performance data and control operations by learning under a plurality of predetermined specifications, and the control execution device acquires a plurality of combinations of performance data and control operations. 1. A plant control device, which selects a plurality of combinations of one performance data and control operations from among the combinations according to the operating state of a plant to be controlled, and provides the control output. The plant control device according to claim 4,
A plant control device that changes a neural network that learns a combination of actual data to be used and an operation method according to the size of the actual data. The plant control device according to claim 1,
Based on the state of the plant to be controlled or the experience of the operator of the plant to be controlled, criteria for judging the quality of control results are selected , and the relationship between actual data and operation methods for the plant to be controlled is obtained and stored in a database. A plant control apparatus, wherein control is performed by different control methods according to the state of the plant to be controlled or the experience of an operator of the plant to be controlled. The plant control device according to claim 1,
The combination of the actual data and the control operation is created by simulation using the control model of the controlled plant before the controlled plant is controlled, and the combination of the actual data and the control operation in the controlled plant is learned. A plant control device characterized by shortening a period. The plant control device according to claim 1,
The control execution device includes an output presence/absence determination unit that determines presence/absence of outputs from a plurality of neural networks. When the output of the net exists and the control effect is low, it is formed by learning when the control effect is high. 1. A plant control device, characterized in that it determines that there is an output of a neural network. A rolling mill control device to which the plant control device according to claim 1 is applied,
The rolling mill control device, wherein the plant to be controlled is a rolling mill, and the performance data is a delivery side shape of the rolling mill. The operation terminal of the controlled plant according to the output of the neural network learned by the learning unit that learns the combination of the actual data of the controlled plant and the control operation according to the control effect, and forms a plurality of neural networks with different control effects. A plant control method comprising a control unit that provides a control output that controls
When there is an output of a neural network formed by learning when the control effect is high and the control effect is high, the control unit controls the operating end of the plant to be controlled according to the output, and the control effect is high. According to the output of the neural network formed by learning the case where the control effect is low when there is only the output of the neural network formed by learning the low case and there is a margin in the manipulating end position at the manipulating end When there is only the output of a neural network formed by controlling the operation terminal of the controlled plant and learning when the control effect is low, and when there is no margin in the operation terminal position at the operation terminal, the controlled plant A plant control method characterized by not controlling the operating end of The plant control method according to claim 12,
The control unit selects an optimum control output in consideration of outputs from a plurality of neural networks according to control effects and the control terminal position according to a predetermined combination of actual data of the plant to be controlled and the control operation. and, if it is determined that the performance data of the controlled plant deteriorates when the control output is output to the controlled plant, preventing the control output from being output to the controlled plant,
When the control unit actually outputs the control output to the plant to be controlled, the learning unit determines whether the actual data has improved compared to before the control after a time delay until the control effect appears in the actual data. It determines whether the control result is good or bad, learns the good or bad of the control result, the actual data and the teacher data as learning data, and by learning, sets a plurality of control targets according to the state of the plant to be controlled. A separate combination of performance data and control operation is obtained for the plant, and the obtained combination of performance data and control operation is used as a predetermined combination of performance data and control operation of the controlled plant in the control unit, and the controlled object A plant control method, wherein when a plant state is similar to a combination of actual data and control operation before correction, control is performed using learned data after correction. A rolling mill control method to which the plant control method according to claim 12 is applied,
A rolling mill control method, wherein the plant to be controlled is a rolling mill, and the performance data is a delivery side shape of the rolling mill. A program for realizing a plant control device that recognizes patterns of combination of performance data of a plant to be controlled and implements control by a computer system,
The computer system comprises a control method learning program that learns combinations of actual data and control operations of the plant to be controlled, forms a plurality of neural networks with different control effects, and controls according to the learned combinations of the actual data and control operations. Equipped with a control execution program that controls the target plant,
When there is an output of a neural network formed by learning when the control effect is high and the control effect is high, the control execution program controls the operation end of the plant to be controlled according to the output, and controls the control effect When there is only the output of the neural network formed by learning when the control effect is low, and when there is a margin in the operating terminal position at the operating terminal, the output of the neural network formed by learning when the control effect is low is used to control the operating terminal of the plant to be controlled, and only the output of the neural network formed by learning when the control effect is low exists, and when there is no margin in the operating terminal position at the operating terminal, the controlled object A program characterized in that it does not control the operating end of the plant. 16. The program according to claim 15,
The control execution program includes a control rule execution program that provides a control output in accordance with a predetermined combination of performance data of a plant to be controlled and a control operation; A control output determination program for notifying the control method learning program that the data and control operation are erroneous, and when the control output determination program outputs the control output to the controlled plant, the actual data of the controlled plant If it is determined that it will deteriorate, it comprises a control output suppression program that prevents output of the control output to the controlled plant,
In the control method learning program, when the control execution device actually outputs the control output to the controlled plant, after the time delay until the control effect appears in the actual data, the actual data becomes better than before the control. A control result quality determination program for achieving a control result quality determination process for determining whether the control result is good or bad, the quality of the control result in the control result quality determination program, and the control output. A learning data creation program for obtaining teacher data, and a control rule learning program for learning the performance data and the teacher data as learning data,
Through learning by the control method learning device, individual combinations of performance data and control operations are obtained for a plurality of control targets according to the state of the plant to be controlled, and the combinations of the obtained performance data and control operations are obtained as described above. A program characterized by being used as a prescribed combination of performance data of a plant to be controlled and control operations in a control rule execution program.

Images