JP2022095082A - Plant control system, plant control method and program - Google Patents

Abstract

To efficiently correct a control rule in a state where a risk of disturbing control over a plant is small in a plant control system.SOLUTION: A plant control system includes a control method learning unit 21 for learning a combination of performance data and control operation of an object plant 1, a control execution unit 20 for executing control over the object plant according to the combination of the performance data and the control operation learned by the control method learning part, and a quality determination rule learning unit 22 for learning a combination between the combination of the performance data and control operation of the object plant and control plant qualities. A quality of a control output is determined according to the determined combinations of the performance data and control operation of the object plant and the control result qualities, and a control rule is learned with the quality determination result, the performance data and teacher data as learning data.SELECTED DRAWING: Figure 1

Description

The present invention relates to a plant control system, a plant control method and a program.

Conventionally, in various plants, plant control based on various control theories has been executed in order to obtain appropriate control results by the control.

To explain an example of a plant, for example, in rolling mill control, fuzzy control and neuro-fuzzy control are applied as control theories for shape control for controlling the wavy state of a plate. Fuzzy control is applied to shape control using a coolant, and neuro-fuzzy control is applied to shape control of a Senzimere rolling mill. Of these, the shape control to which the 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 preset reference shape pattern. The similarity ratio is calculated. Then, the control output amount for the operation end is obtained from the obtained similarity ratio according to the control rule expressed by the control operation end operation amount for the preset reference shape pattern.

Hereinafter, a conventional technique for shape control of a Sendimia rolling mill using neuro-fuzzy control will be described.
Neuro-fuzzy control is used in the shape control of the Sendimia rolling mill. As shown in FIG. 25, in the Sendimia rolling mill 50, the pattern recognition unit 51 recognizes the shape pattern from the actual shape detected by the shape detector 52, and the actual shape is any of the preset reference shape patterns. Is calculated to be the closest to. The actual shape data detected by the shape detector 52 is preprocessed for pattern recognition by the shape detection preprocessing unit 54.

Then, the control calculation unit 53 executes control using a control rule composed of a control operation end operation amount for a preset shape pattern.
Here, as shown in FIG. 26, in the pattern recognition unit 51, the difference (Δε) between the actual shape pattern (ε) detected by the shape detector 52 and the target shape (ε _ref ) is different from the pattern 1. Calculate which of the shapes of pattern 8 is closest to. Then, the control calculation unit 53 selects and executes any of the control methods from the pattern 1 to the pattern 8 based on the calculation result.

Patent No. 2804161 Japanese Unexamined Patent Publication No. 2018-180799

In the prior art described in Patent Document 1, a representative shape is set in advance as a reference shape pattern, and control is performed based on a control rule indicating a relationship with a control operation end operation amount with respect to the reference waveform pattern. The learning of the control rule is also related to the control operation end operation amount with respect to the reference waveform pattern, and a predetermined representative reference shape pattern is used as it is. Therefore, there is a problem that the shape control reacts only to a specific shape pattern.

The standard shape pattern is determined by human beings in advance based on their knowledge of the target rolling mill and the experience of accumulating shape results and manual intervention operations, but all shapes generated in the target rolling mill and the material to be rolled are formed. It is difficult to cover. Therefore, when a shape different from the standard shape pattern is generated, the control by the shape control is not executed and the shape deviation remains unsuppressed, or it is erroneously recognized as a similar standard shape pattern and erroneously controlled. In some cases, the shape may be deteriorated by performing an operation.

Therefore, in the conventional shape control, there is a problem that there is a limit to the improvement of the control accuracy because the control rule is learned and the control is executed by using the preset reference shape pattern and the control rule for the reference shape pattern. ..

In order to solve this problem, for example, the technique described in Patent Document 2 has been proposed. Patent Document 2 describes a process of generating a disturbance during control and gradually making the neural network smarter by learning. However, the process of generating a controlled disturbance as described in Patent Document 2 actually operates the controlled plant and causes a controlled disturbance during the operation, which disturbs the actual operation of the controlled plant. Therefore, it cannot be said that it is very preferable in terms of operation. Furthermore, unless the controlled plant is operated to some extent, the neural network will not be appropriate, and there is a high possibility that proper control will not be possible for a while from the initial stage of operation.

An object of the present invention is to provide a plant control system, a plant control method and a program capable of efficiently modifying control rules while reducing the risk of disturbing the control of the plant.

In order to solve the above problems, for example, the configuration described in the claims is adopted.
The present application includes a plurality of means for solving the above problems. For example, the plant control system recognizes the pattern of the combination of the actual data of the controlled target plant with respect to the controlled target plant. Applies to those that perform control.
Then, the plant control system controls the control target plant according to the combination of the control method learning unit that learns the combination of the actual data and the control operation of the controlled target plant and the actual data and the control operation learned by the control method learning unit. It includes a control execution unit to be executed, and a pass / fail judgment rule learning unit that learns a combination of actual data of a controlled plant, a control operation, and a control result pass / fail combination.
Here, the control execution unit is
A control rule execution unit that gives control output according to the specified combination of actual data of the controlled plant and control operation,
A control output pass / fail judgment rule execution unit that judges the pass / fail of the control output according to the combination of the actual data of the controlled plant, the control operation, and the control result pass / fail.
A new search operation amount calculation unit that calculates the operation amount for new operation search based on the quality judgment in the control output pass / fail judgment rule execution unit,
Control output pass / fail judgment Control when it is judged that the actual data of the control target plant deteriorates when the control output is output to the control target plant using the pass / fail judgment in the rule execution unit and the simulation data using the control simulator. It is provided with a control output suppression unit that prevents the output from being output to the controlled plant.
In addition, the pass / fail judgment rule learning unit is
When the control execution unit outputs the control output to the controlled plant, the control result quality judgment unit that determines the quality of the control result after the time delay until the control effect appears in the actual data,
It is provided with a quality judgment rule learning unit that learns the quality of the control result, the actual data, and the control output as learning data in the control result quality determination unit.
In addition, the control method learning unit
Control output quality judgment rule execution unit controls output quality judgment, learning data creation unit that obtains teacher data using control output, and
It is equipped with a control rule learning unit that learns actual data and teacher data as learning data.

According to the present invention, the control rules of the shape pattern and the operation method used in the shape control during control are efficiently and automatically corrected with less risk to the plant, and the environment of the plant changes over time. It will be possible to make it optimal. Therefore, according to the present invention, it is possible to improve the control accuracy, shorten the start-up period of the control unit, and respond to secular variation.
Further, according to the present invention, by evaluating the performance of the control rule in advance, there is an effect that the risk to the plant is reduced by applying the new control rule and the control performance is improved by selecting the optimum control rule.
Issues, configurations and effects other than those described above will be clarified by the following description of the embodiments.

It is a block diagram which shows the outline of the plant control system which concerns on one Embodiment of this invention. It is a figure which shows the specific structural example of the control rule execution part which concerns on one Embodiment of this invention. It is a block diagram which shows the example of the control output quality judgment rule execution part which concerns on one Embodiment of this invention. It is a figure which shows the specific structural example of the control rule learning part which concerns on one Embodiment of this invention. It is a block diagram which shows the example of the quality determination rule learning part which concerns on one Embodiment of this invention. It is a figure which shows the example of the control result quality judgment with respect to the control method in the shape control of a Sendimia rolling mill. It is a figure which shows the neural network composition when it is used for the shape control of the Sendimia rolling mill which concerns on one Embodiment of this invention. It is a figure explaining the shape deviation and the control method which concerns on one Embodiment of this invention. It is a block diagram which shows the example of the control input data creation part which concerns on one Embodiment of this invention. It is a block diagram which shows the example of the control output calculation part which concerns on one Embodiment of this invention. It is a figure which shows the neural network composition when it is used for the quality determination of the Sendimia rolling mill which concerns on one Embodiment of this invention. It is a figure which shows the operation amount calculation method in the new search operation amount calculation unit which concerns on one Embodiment of this invention. It is a block diagram which shows the example of the control output determination part which concerns on one Embodiment of this invention. It is a block diagram which shows the example of the control output calculation part which concerns on one Embodiment of this invention. It is a figure which shows the processing stage and processing contents in the learning data creation part which concerns on one Embodiment of this invention. It is a figure which shows the example of data stored in the learning data database which concerns on one Embodiment of this invention. It is a figure which shows the example of the neural network management table which concerns on one Embodiment of this invention. It is a block diagram which shows the example of the learning data database which concerns on one Embodiment of this invention. It is a block diagram which shows the example of the control quality determination part which concerns on one Embodiment of this invention. It is a figure which shows the example of data stored in the learning data database which concerns on one Embodiment of this invention. It is a figure which shows the example of the neural network management table which concerns on one Embodiment of this invention. It is a figure which shows the example of the learning data database which concerns on one Embodiment of this invention. It is a block diagram which shows the example which provided the control rule evaluation unit in the plant control system which concerns on one Embodiment of this invention. It is a block diagram which shows the hardware configuration example of the plant control system which concerns on one Embodiment of this invention. It is a block diagram which shows the example of the Sendimia rolling mill. It is a figure which shows the example of the list of the control rule in the shape control of a Sendimia rolling mill.

Hereinafter, a plant control system according to an embodiment of the present invention (hereinafter referred to as “this example”) will be described with reference to the accompanying drawings.

First, prior to explaining this example, the background to the present invention and its outline will be described by taking the case where the plant control system is applied to the shape control device of a rolling mill as an example.

First, in order to obtain a plant control system capable of efficiently modifying control rules with less risk to the plant, which is one of the objects of the present invention, the following requirements (1), (2), ( 3) and (4) are required.
-Requirement (1): In order to improve the control rule, if the control operation with good control result cannot be learned, the control operation is changed significantly, and if the control result is good, as a new control operation method. If you can learn the control operation with good control result, do not change the control operation or make a small change, and if the control result for it is good, adopt it as a new control operation method. ..
-Requirement (2): A model that can estimate the quality of the control result more accurately than the simulator using the machine model by learning the combination of the shape pattern, the control operation, and the quality of the control result based on the actual machine data. And always build a model that fits the latest plant conditions by regular automatic learning.
-Requirement (3): By using a model that estimates the quality of the control result, the reliability of the control output suppression function for the plant, which was performed only by the simple mechanical model in the conventional technique, is enhanced.
-Requirement (4): Noise included in the plant data by using a model that estimates the quality judgment of the control result in the control rule learning data generation function that was performed by the quality judgment of the control result once in the conventional technique. By suppressing the influence of the training data and making it possible to target the training data even for fine adjustments with a small effect, and at the same time, by preventing erroneous judgment of the control effect, fluctuations in the training data are suppressed and the control performance is stabilized.

In order to realize these requirements (1) to (4), it is necessary to construct a neural network that can learn the combination of the shape pattern used for shape control, the control operation, and the quality of the control result in the control device. preferable. Then, the control device estimates the quality of the control result by the output of the control operation for the shape pattern generated by the rolling mill by the value obtained by inputting the shape pattern generated by the rolling mill and the output of the control operation into the neural network. Is needed. Further, the control device selects a control operation amount calculation method for searching for a new control operation by using the estimated value of the quality of the control result.

Verification is performed using a simple model of the rolling mill, and the control device prevents the shape deterioration by not outputting to the control operation end of the rolling mill for the output whose shape is clearly considered to be deteriorated. At this time, the control device uses the estimated value of the quality of the control result to judge the output suppression, thereby increasing the reliability of protection and optimizing the range of suppression, thereby increasing the range that can be handled by the control function. Can be expanded.

Insufficient accuracy in estimating the quality of control results In the initial stage of application, even control operation outputs estimated to be bad can be output to the plant to determine the quality of unlearned shape patterns, control operations, and control results. It is necessary to expand the range of learning about combinations.

At the stage where the estimation accuracy of the quality judgment of the control result is sufficiently high, the quality of the control result can be estimated without outputting the operation amount to the plant, so that the control device can generate the learning data of the control rule. ..

By estimating the quality of the control result using a neural network that can estimate the quality of the control result, the control device can reduce the influence of noise on the plant data and can judge the quality of fine-tuned data with a small effect. It becomes. This allows the control device to generate learning data. Further, the control device can improve the accuracy of the learning data by preventing erroneous quality determination due to noise.

In addition, when the estimation accuracy of the quality judgment of the control result deteriorates due to changes in the plant environment due to aging, the control device adapts to the latest plant state by re-learning using the actual data of the latest plant. It is possible to estimate the quality of the control result.

In order to confirm the estimation accuracy of the quality judgment of the control result, test data is prepared for accuracy verification in addition to the data used for learning the neural network. Then, the control device determines whether the shape pattern included in the test data for accuracy verification and the output of the control operation are good or bad between the value output by inputting the output into the neural network and the control result included in the test data. Based on the error, it becomes possible to confirm the estimation accuracy of the pass / fail judgment.

FIG. 1 shows the configuration of the plant control system of this example.
The plant control system of FIG. 1 includes a control execution unit 20, a control method learning unit 21, a pass / fail judgment rule learning unit 22, a plurality of database DBs (DB1 to DB8), and a management table TB of each database DB.

The control execution unit 20 inputs the actual data Si from the controlled target plant 1 and gives the controlled operation amount output SO determined according to the control rule (neural network) to the controlled target plant 1 to control the controlled target plant 1. The plant 1 to be controlled here is the Sendimia rolling mill 50 shown in FIG. 25, which has already been described.
Here, as described with reference to FIG. 26, the control rule is a plurality of patterns in which, for example, the difference a (Δε) between the detected actual shape pattern A (ε) and the target shape (ε _ref ) is prepared. It is a rule that calculates which of the shapes is closest to. The control execution unit 20 selects and executes one of the pattern control methods based on the calculation result according to this control rule.

The control method learning unit 21 inputs the control input data S1 and the like created by the control execution unit 20 to perform learning, and reflects the learned control rules in the control rules in the control execution unit 20.
The pass / fail determination rule learning unit 22 inputs and learns the actual data Si before and after the control of the controlled target plant 1, and reflects the learned pass / fail determination rule in the pass / fail determination rule in the control execution unit 20.

The control execution unit 20 includes a control input data creation 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, a control output quality determination rule execution unit 17, and a new search operation amount. A calculation unit 33 and a control output operation method selection unit 18 are provided.

The control execution unit 20 creates the input data S1 of the control rule execution unit 10 by using the control input data creation unit 2 from the actual data Si of the rolling mill which is the control target plant 1.
The control rule execution unit 10 creates a control operation end operation command S2 from the control target actual data Si using a neural network (control rule) expressing the relationship between the control target actual data Si and the control operation end operation command S2. Control rule execution processing is performed. The control output calculation unit 3 calculates the control operation amount S3 to the control operation end based on the control operation end operation command S2. As a result, the control execution unit 20 creates the control operation amount S3 using the neural network according to the actual data Si of the controlled target plant 1.

Further, the control output pass / fail determination rule execution unit 17 controls using a neural network (pass / fail determination rule) that expresses the relationship between the actual data Si to be controlled, the control operation amount S3, and the control result pass / fail data S6 of the control result. The control output pass / fail judgment rule execution process for creating the control output pass / fail judgment estimated value S9 from the target actual data Si and the control operation amount S3 is performed. Further, the control output pass / fail determination rule execution unit 17 creates a control result pass / fail determination estimated value S11 from the actual data Si to be controlled and the selection control operation amount S8 described later.

The new search operation amount calculation unit 33 performs a new search operation amount calculation process for calculating the new search operation amount S12 based on the control output pass / fail determination estimated value S9.
The control output operation method selection unit 18 creates a selection control operation amount S8 and a control method selection flag S14 based on the control operation amount S3 or the new search control operation amount S12.

Further, the control output determination unit 5 in the control execution unit 20 can output the control operation amount to the control operation end by using the actual data Si from the controlled target plant 1 and the control operation amount S3 from the control output calculation unit 3. The data S4 is determined. The control output suppression unit 4 determines whether or not the selection control operation amount S8 can be output to the control operation end based on the control operation amount output availability data S4 and the control result pass / fail judgment estimated value S11, and the selection control operation is permitted. The quantity S8 is output as a control operation quantity output SO given to the controlled target plant 1. As a result, the selection control operation amount S8 determined to be abnormal is not output from the control execution unit 20 to the controlled target plant 1.

The control execution unit 20 configured as described above refers to the control rule database DB1, the output judgment database DB3, and the pass / fail judgment rule database DB5 in order to execute the processing.
The control rule database DB 1 is accessiblely connected to both the control rule execution unit 10 in the control execution unit 20 and the control rule learning unit 802 in the control method learning unit 21 described later.

The control rule database DB 1 stores a control rule (neural network) as a learning result in the control rule learning unit 802. The control rule execution unit 10 refers to the control rule stored in the control rule database DB1.
The learning data database DB2 stores the learning data obtained by the control rule learning unit 802.
The output determination database DB 3 is accessiblely connected to the control output determination unit 5 in the control execution unit 20, and the output determination result is stored in the output determination database DB 3.

Data for pass / fail judgment is stored in the pass / fail judgment database DB4.
The pass / fail determination rule database DB 5 stores a pass / fail determination rule (neural network) as a learning result in the pass / fail determination rule learning unit 31. The pass / fail judgment rule database DB 5 is accessiblely connected to both the control output pass / fail judgment rule execution unit 17 in the control execution unit 20 and the pass / fail judgment rule learning unit 31 in the pass / fail judgment rule learning unit 22, which will be described later. The control output pass / fail determination rule execution unit 17 refers to the pass / fail determination rule stored in the pass / fail determination rule database DB 5.
The learning data database DB 6 stores the learning data learned by the control method learning unit 21.
The verification data database DB 7 stores verification data necessary for making a pass / fail judgment.

FIG. 2 shows a specific configuration example of the control rule execution unit 10 of this example.
The input data S1 created by the control input data creation unit 2 is input to the control rule execution unit 10. The control rule execution unit 10 processes the input data S1 and gives the control operation end operation command S2 to the control output calculation unit 3. The control rule execution unit 10 includes a neural network 101, and the neural network 101 outputs a control operation end operation command S2 according to a shape control rule as shown in FIG. 26.

The control rule execution unit 10 further includes a neural network selection unit 102, and by referring to the control rules stored in the control rule database DB 1, the optimum control rule is selected as the control rule in the neural network 101, and the neural network is selected. Let the net 101 execute.

In this way, the control rule execution unit 10 selects and uses a necessary neural network from a plurality of neural networks divided by an operator group or a control purpose. It is preferable that the control rule database DB 1 also includes the actual data (data of the operation group, etc.) Si that allows the selection of the neural network and the quality determination criteria as the data from the controlled plant 1.
In addition, since the execution of the neural network results in a control rule, the neural network and the control rule are used interchangeably in the present specification.

FIG. 3 shows a specific configuration of the control output quality determination rule execution unit 17.
The control operation amount S3 created by the control input data creation unit 2 and the input data S1 and the control output calculation unit 3 is input to the control output pass / fail determination rule execution unit 17. The control output pass / fail determination rule execution unit 17 generates a control output pass / fail determination estimated value S9 based on these input data and gives it to the new search operation amount calculation unit 33.

Further, the input data S1 created by the control input data creation unit 2 and the selection control operation amount S8 created by the control output operation method selection unit 18 are input to the control output pass / fail determination rule execution unit 17. The control output pass / fail determination rule execution unit 17 generates a control result pass / fail determination estimated value S11 based on these input data and gives it to the control output suppression unit 4.

The control output quality determination rule execution unit 17 includes a neural network 171 and a neural network selection unit 172.
The neural network 171 estimates the quality judgment value of the control result when the control operation amount S3 (control pattern) is output with respect to the input data S1 (shape pattern) based on the past control results.
The neural network selection unit 172 selects the optimum pass / fail determination rule as the pass / fail determination rule in the neural network 171 by referring to the pass / fail determination rule stored in the pass / fail determination rule database DB 5.

In this way, the control output quality determination rule execution unit 17 selects a necessary neural network from a plurality of neural networks divided according to the difference in the material properties to be controlled and the difference in the quality determination criteria.
The pass / fail judgment rule database DB 5 should include, as data from the controlled target plant 1, actual data (data of the operation group, etc.) Si that allows selection of the material properties to be controlled and the pass / fail judgment criteria. Since the execution of the neural network results in a pass / fail judgment rule, the neural network and the pass / fail judgment rule are used interchangeably in the present specification.

Returning to the description of FIG. 1, the control method learning unit 21 executes learning of the neural network 101 used by the control execution unit 20.
The control method learning unit 21 includes a learning data creation unit 801 and a control rule learning unit 802.

The learning data creation unit 801 in the control method learning unit 21 determines whether the control result is good or bad, which is created by the selection control operation amount S8 created by the control execution unit 20, the control method selection flag S14, and the control output quality determination rule execution unit 17. Using the estimated value S11, a learning data creation process for creating new teacher data S7a used for learning the neural network is performed. The learning data S7a created by the learning data creation unit 801 is given to the control rule learning unit 802.
The teacher data S7a corresponds to the control operation end operation command S2 output by the control rule execution unit 10.

The learning data creation unit 801 uses the control result pass / fail judgment estimated value S11 created by the control output pass / fail judgment rule execution unit 17 to estimate the control operation end operation command S2 output by the control rule execution unit 10. Is obtained as new teacher data S7a.

FIG. 4 shows a specific configuration example of the control rule learning unit 802.
The control rule learning unit 802 includes an input data creation unit 114, a teacher data creation unit 115, a neural network processing unit 110, and a neural network selection unit 113.
Input data S1 from the control input data creation unit 2 and new teacher data S7a from the learning data creation unit 801 are input to the control rule learning unit 802 as external inputs. Further, the control rule learning unit 802 refers to the data stored in the control rule database DB1 and the learning data database DB2.

In the control rule learning unit 802, the input data S1 is taken into the neural network processing unit 110 via the input data creation unit 114.

Further, in the control rule learning unit 802, the new teacher data S7a from the learning data creation unit 801 is the total teacher data including the past teacher data S7b stored in the learning data database DB2 in the teacher data creation unit 115. As S7c, it is given to the neural net 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 creation unit 2 is the neural net processing unit as the total input data S8c including the past input data S8b stored in the learning data database DB2 in the input data creation unit 114. Given to 110. These input data S8a and S8b are also appropriately stored in the learning data database DB2 and used.

The neural network processing unit 110 is composed of a neural network 111 and a neural network learning control unit 112.
The neural network 111 takes in the input data S8c from the input data creation unit 114, the teacher data S7c from the teacher data creation unit 115, and the control rule (neural network) selected by the neural network selection unit 113, and finally determines the neural network. The net is stored in the control rule database DB1.

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

Here, the neural network 101 in the control rule execution unit 10 of FIG. 2 and the neural network 111 in the control method learning unit 21 of FIG. 4 are both neural networks of the same concept, but they differ as follows. ..
That is, the neural network 101 in the control rule execution unit 10 is a neural network having predetermined contents, and is a neural network that obtains a control operation end operation command S2 as a corresponding output when input data S1 is given.

On the other hand, the neural net 111 in the control method learning unit 21 obtains this input / output relationship by learning when the input data S8c and the teacher data S7c for the input data S1 and the control operation end operation command S2 are set as learning data. It is the net.

The concept of basic processing in the control method learning unit 21 is as follows.
First, when the content of the control operation quantity output availability data S4 is “possible”, the control execution unit 20 outputs the control operation quantity output SO to the controlled target plant 1. Here, when the content of the control result pass / fail determination estimated value S11 is “good” (changes in the direction in which the actual data Si improves), the learning data creation unit 801 performs the selection control operation output by the control output operation method selection unit 18. It is determined that the quantity S8 is correct, and training data is created so that the output of the neural network becomes the selection control operation quantity S8.

On the other hand, the content of the control operation quantity output availability data S4 is "No", or the content of the control result pass / fail judgment estimated value S11 that outputs the control operation quantity output SO to the controlled target plant 1 is "No" (actual data Si is bad). In the case of (change in the direction of)), the learning data creation unit 801 determines that the selection control operation amount S8 output by the control output operation method selection unit 18 is incorrect.

In this case, the learning data creation unit 801 confirms whether or not the control operation amount S3 has been selected in the control output operation method selection unit 18 from the control method selection flag S14. When the control operation amount S3 is selected in this confirmation, the learning data creation unit 801 determines that the control operation end operation command S2 output by the control rule execution unit 10 is incorrect, and the neural network is not output. Create training data like this. At this time, the neural network output is configured so that two types of outputs, + direction and-direction, are output to the same control operation end as the control output, and the control operation end operation command S2 on the output side is not output. Create training data as in.

Further, the control rule learning unit 802 shown in FIG. 4 performs processing as follows as a result of data processing by the neural network learning control unit 112.
First, the control rule learning unit 802 uses learning data that is a combination of the data S8c obtained from the input data S1 to the control execution unit 20 and the teacher data S7c created by the teacher data creation unit 115 to execute the control rule. The learning of the neural network 101 used in the part 10 is executed.

Actually, the control rule learning unit 802 prepares the same neural network 111 as the neural network 101 of the control rule execution unit 10 in the control rule learning unit 802, performs an operation test under various conditions, and learns the response at that time. , Get control rules that have been confirmed to produce better results as a result of learning.

Since the learning here needs to be performed using a plurality of learning data, a plurality of past learning data are taken out from the learning data database DB2 that stores the learning data created in the past, and learning is performed. And execute the process. Then, the learning data this time is stored in the learning data database DB2. Further, the learned neural network is stored in the control rule database DB 1 for use by the control rule execution unit 10.

In the learning of the neural net, each time new learning data is created, the past learning data may be used together for learning, or after the learning data is accumulated to some extent (for example, 100 pieces), the past learning is performed. You may learn by using the data together.

With such a configuration, the control output operation method selection unit 18 selects a new search operation amount, outputs the new search operation amount to the target plant, and creates learning data according to the control result. It becomes possible to learn new control methods.

Returning to the explanation of FIG. 1, the pass / fail determination rule learning unit 22 executes learning of the neural network 171 (FIG. 3) used by the control execution unit 20. When the control execution unit 20 outputs the control operation amount output SO to the controlled target plant 1, it takes time for the control effect to actually appear as a change in the actual data Si. Therefore, learning is executed using the data delayed by that time. In addition, in FIG. 1 and the like, the processing unit DL described as “Z ^-1 ” indicates that there is an appropriate time delay when each data is transmitted.

The pass / fail determination rule learning unit 22 includes a control result pass / fail determination unit 6, a pass / fail determination rule learning unit 31, and a pass / fail determination database DB 4.

The control result pass / fail judgment unit 6 changes in the direction of improving the actual data Si by using the actual data Si from the controlled plant 1, the actual data previous value Si0, and the pass / fail judgment data S5 stored in the pass / fail judgment database DB4. A control result pass / fail judgment process is performed to determine whether the data has changed or the data has changed in the worse direction. Then, the control result pass / fail determination unit 6 outputs the control result pass / fail data S6 indicating the determination result.

FIG. 5 shows a specific configuration of the pass / fail determination rule learning unit 31.
The pass / fail determination rule learning unit 31 includes an input data creation unit 314, a teacher data creation unit 315, a neural network processing unit 310, and a neural network selection unit 313.
The pass / fail determination rule learning unit 31 delays the time of the data S12a1 in which the input data S1 from the control input data creation unit 2 is delayed and the control output amount S0 from the control output suppression unit 4 as inputs from the outside. Obtain data S12a2. Further, the pass / fail determination rule learning unit 31 obtains the control result pass / fail data S6 (S13a) from the control result pass / fail determination unit 6.
Further, the pass / fail determination rule learning unit 31 refers to the data accumulated in the pass / fail determination rule database DB 5 and the learning data database DB 6.

The input data S1 and the control output amount S0 are taken into the neural network processing unit 310 via the input data creation unit 314 after appropriate time delay compensation.

Further, the control result pass / fail data S6 (S13a) from the control result pass / fail determination unit 6 is used as the total teacher data S13c including the past teacher data S13b stored in the learning data database DB 6 in the teacher data creation unit 315. It is given to the neural network processing unit 310. These teacher data S13a and S13b are appropriately stored in the learning data database DB6 and used.

Similarly, the input data S12a1 and S12a2 from the control input data creation unit 2 and the control output suppression unit 4 are total inputs including the past input data S12b stored in the learning data database DB 6 in the input data creation unit 314. The data S12c is given to the neural net processing unit 310. These input data S12a1, S12a2, and S12b are appropriately stored in the learning data database DB6 and used.

The neural network processing unit 310 is composed of a neural network 311 and a neural network learning control unit 312.
The neural network 311 takes in the input data S12c from the input data creation unit 314, the teacher data S13c from the teacher data creation unit 315, and the control rule (neural net) selected by the neural network selection unit 313. Then, the neural network 311 stores the finally determined neural network in the quality determination rule database DB5.

The neural network learning control unit 312 controls the input data creation unit 314, the teacher data creation unit 315, and the neural network selection unit 313 at appropriate timings to obtain the input of the neural network 311. Further, the neural network learning control unit 312 stores the processing result in the quality determination rule database DB 5 via the neural network selection unit 313.

FIG. 6 is a diagram showing a specific example of control result quality determination for a control method in shape control of a Sendimia rolling mill. FIG. 6 shows the control result pass / fail determination result for each shape control rule shown in FIG. 26.

Here, the neural network 171 of the control execution unit 20 shown in FIG. 3 and the neural network 311 in the pass / fail judgment rule learning unit 22 shown in FIG. 5 are both neural networks of the same concept, but differ in the following points. ing.
The neural network 171 in the control execution unit 20 is a neural network having predetermined contents. That is, the neural network 171 obtains the control output pass / fail judgment estimated value S9 or S11 as the output corresponding to the input data S1 and the selection control operation amount S8 or the control operation amount S3, so to speak, in one direction. It is a neural network used for processing.

On the other hand, the neural network 311 in the pass / fail determination rule learning unit 22 is a neural network that satisfies these input / output relationships when the input data S1, the input data S12c of the control output amount S0, and the teacher data S13c are set as learning data. It is the net.

Next, a specific example of the plant control method will be described for shape control in the Sendimia rolling mill. The shape control will be described assuming that the following specifications A and B are adopted.

Specification A is a specification regarding priority, and has information on priority in the plate width direction. For example, in shape control, it is often difficult to control the target value over the entire plate width direction due to mechanical characteristics. Therefore, specifications A1 and A2 for the following two priorities are provided in the plate width direction. Of these, the specification A1 regarding the priority is "priority is given to the plate end portion". Further, the specification A2 regarding the priority is to "prioritize the central portion".
Control is executed according to the two priorities of the specifications A1 and A2. That is, when the plant control system executes the control, either the specification A1 or A2 regarding the priority is taken into consideration.

Specification B is a specification for dealing with a condition that is known in advance. As an example, since the relationship between the shape pattern and the control method changes under various conditions, for example, it is necessary to classify the specification B1 into a plate width and the specification B2 into a steel type. As each specification changes, the degree of influence on the shape of the shape operation end changes.

The controlled target plant 1 in this example is a Sendimia rolling mill, and the actual data is the actual shape. The Sendimia rolling mill is a rolling mill having cluster rolls for cold rolling hard materials such as stainless steel. Zenjimia rolling mills use small diameter work rolls for the purpose of applying strong rolling pressure to hard materials. For this reason, it is difficult for the Zendimia rolling mill to control to obtain a flat steel sheet. As a countermeasure, the Zenjimia rolling mill employs a cluster roll structure and various shape control units.

In a Sendimia rolling mill, the upper and lower first intermediate rolls generally have a one-sided taper so that they can be shifted, and the upper and lower first intermediate rolls are equipped with six split rolls and two rolls called AS-U. In the example described below, the actual shape data Si uses the detection data of the shape detector, and the input data S1 uses the shape deviation, which is the difference from the target shape. Further, the control operation amount S3 is the AS-U of # 1 to #n and the roll shift amount of the upper and lower first intermediate rolls.

FIG. 7 shows a neural network configuration when used for shape control of a Sendimia rolling mill. Here, the neural network indicates the neural network 101 for the control rule execution unit 10. Further, for the control rule learning unit 802, the neural network 111 is shown. Both the neural network 101 and the neural network 111 have the same structure.

In the case of shape control of the Sendimia rolling mill, the actual data Si from the controlled plant 1 is the data of the shape detector (here, it is assumed that the shape deviation which is the difference between the actual shape and the target shape is output). It is the actual data of the Sendimia rolling mill including. The control input data creation unit 2 obtains the normalized shape deviation 201 and the shape deviation step 202 as the input data S1. As a result, the input layer of the neural networks 101 and 111 is composed of the normalized shape deviation 201 and the shape deviation step 202. In FIG. 7, the shape deviation step 202 is used as the input to the neural network input layer, but the neural network may be switched according to the step.

The output layers of the neural networks 101 and 111 are 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 the shape control operation ends of the Sendimia rolling mill. .. For AS-U, the degree of operation is the AS-U opening direction (the direction in which the roll gap (distance between the upper and lower working rolls of the rolling mill) opens) and the AS-U closing direction (the direction in which the roll gap closes). I have about AS-U.

Regarding the first intermediate roll, the first intermediate roll opening direction (the direction in which the first intermediate roll operates outward from the center of the rolling mill) and the first intermediate roll closing direction (the first intermediate roll faces the center side of the rolling mill). The direction of operation) is provided for the upper and lower first intermediate rolls.
For example, if the shape detector has 20 zones and the shape deviation stage 202 is set to 3 stages (large, medium, small), the input layer has 23 inputs. If the AS-U saddles are 7 and the upper and lower first intermediate rolls can be shifted in the plate width direction, the output layer has 14 AS-U operation degrees 301 and 4 intermediate operation degrees, for a total of 18 pieces. Will be. The number of layers in the middle layer and the number of neurons in each layer are set in a timely manner.
The shape control operation end of the Sendimia rolling mill, which is the output layer, is configured to output a neural network so that two types of outputs, a + direction and a-direction, are output to each control operation end.

FIG. 8 shows the shape deviation and the control method in this example.
FIG. 8A shows a control method when the shape deviation is large, and FIG. 8B shows a control method when the shape deviation is small. The height direction (vertical axis direction) of FIGS. 8A and 8B is the magnitude of the shape deviation, the horizontal axis direction is the plate width direction, both sides of the plate width are the plate end portions, and the center is the plate center portion. show.

As shown in FIG. 8A, when the shape deviation is large, the correction of the overall shape is prioritized over the local shape deviation in the plate width direction.
On the other hand, as shown in FIG. 8B, when the shape deviation is small, priority is given to reducing the local shape deviation.

As described above, since it is necessary to change the control method according to the magnitude of the shape deviation, the shape deviation step 202 is provided to the neural networks 101 and 111 as shown in FIG. 7, and the magnitude of the shape deviation is determined. .. As for the shape deviation, regardless of the magnitude of the shape deviation, it is preferable to use one standardized to, for example, 0 to 1. This is an example, and the shape deviation may be input to the input layer of the neural network as it is without standardization, or the neural network itself may be changed according to the magnitude of the shape deviation. For example, two neural networks may be prepared and the neural network used when the shape deviation is large and the neural network used when the shape deviation is small may be separated.

In the plant control of this example, the neural networks 101 and 111 having the configuration as shown in FIG. 6 described above are trained to learn the operation method for the shape pattern, and the shape control is executed using the trained neural network. Even neural networks with the same configuration have different characteristics depending on the learning conditions, and different control outputs can be output for the same shape pattern.

Therefore, by properly using a plurality of neural networks according to other conditions of the shape record, it is possible to configure the optimum control for various conditions. This is a response to specification B. The configuration of FIG. 2 described above shows a specific example in the case of making such a specification.
That is, in the configuration example of FIG. 2, the neural network 101 used in the control rule execution unit 10 is controlled by preparing a separate neural network according to the rolling results, the rolling mill operator name, the steel type of the material to be rolled, the plate width, and the like. Registered in the rule database DB1. The neural network selection unit 102 selects a neural network that matches the conditions at that time, and sets it in the neural network 101 of the control rule execution unit 10.

As a condition at that time in the neural network selection unit 102, it is preferable to take in the plate width data from the actual data Si in the controlled target plant 1 and select the neural network accordingly. Further, if the plurality of neural networks used here have an input layer and an output layer as shown in FIG. 6, the number of layers of the intermediate layer and the number of units of each layer may be different.

FIG. 9 shows the configuration of the control input data creating unit 2 that creates the data S1 (normalized shape deviation 201, shape deviation step 202) for inputting to the input layer of the neural networks 101 and 111.
The control input data creation unit 2 inputs the shape detector data of the shape detector, which detects the plate shape at the time of rolling in the Sendimia rolling mill, which is the controlled target plant 1, as the actual data Si. Then, the control input data creation unit 2 obtains the shape deviation PP value (Peak To Peak value) SPP, which is the difference between the maximum value and the minimum value of the detection result of each shape detector zone, by the shape deviation PP value calculation unit 210. ..

The shape deviation step calculation unit 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 distribution in the plate width direction of the elongation rate of the material to be rolled, and I-UNIT, which expresses the elongation rate in 10-5 units, is used as a unit. For example, classify as shown by the following formula.

Here, the shape deviation stage is set to (large = 1, medium = 0, small = 0) by the establishment of the [Equation 1] equation, and the shape deviation stage is (Large = 0, medium = 1, by the establishment of the [Equation 2] equation. Small = 0), and the shape deviation stage is classified as (Large = 0, Medium = 0, Small = 1) by the establishment of the equation [Equation 3]. For the shape deviation of each zone, normalization is performed using SPM with SPM = SPP.

As described above, the control input data creation unit 2 creates the normalized shape deviation 201 and the shape deviation step 202, which are the input data to the neural network 101. The standardized shape deviation 201 and the shape deviation step 202 are input data S1 of the control rule execution unit 10.

FIG. 10 shows the configuration of the control output calculation unit 3.
The control output calculation unit 3 creates a control operation amount S3 which is an operation command to each shape control operation end from the control operation end operation command S2 which is an output from the neural network 101 in the control rule execution unit 10. The control operation end operation command S2 corresponds to the AS-U operation degree 301 and the first intermediate operation degree 302 in the case of shape control of the Sendimia rolling mill.
FIG. 10 shows one data example for each of the AS-U operation degree 301 and the first intermediate operation degree 302 in which a plurality of numbers exist, and each data is composed of a pair of data of the open direction degree and the closed direction degree. Has been done.

In the control output calculation unit 3, the input AS-U operation degree 301 has outputs in the open direction and the closed direction of each AS-U, so the difference between them is calculated by the subtractor 303. Then, by multiplying the output of the subtractor 303 by the conversion gain G _ASU with the multiplier 304, an operation command to each AS-U is generated and output. Since the control output to each AS-U is the AS-U position change amount (unit is length), the conversion gain G _ASU is the conversion gain from the degree to the position change amount.

Since the first intermediate operation degree 302 similarly input has the outputs of the first intermediate outer side and the inner side, the difference between them is calculated by the subtractor 305. Then, by multiplying the output of the subtractor 305 by the conversion gain G _1ST with the multiplier 306, an operation command for each first intermediate roll shift is generated and output. Since the control output to each first intermediate roll is the first intermediate roll shift position change amount (unit is length), the conversion gain G _1ST is the conversion gain from the degree to the position change amount.

As described above, the control output calculation unit 3 can calculate the control operation amount S3. The control operation amount S3 is composed of # 1 to # nAS-U position change amount (n depends on the number of saddles of the AS-U roll), the upper first intermediate shift position change amount, and the lower first intermediate shift position change amount. ing.

FIG. 11 shows a neural network configuration when used for quality determination of the shape control result of the Sendimia rolling mill used in the control output quality determination rule execution unit 17 and the quality determination rule learning unit 31. The neural network here is a neural network 171 for the control output pass / fail determination rule execution unit 17, and a neural network 311 for the pass / fail determination rule learning unit 31, but both have the same structure.

For the standardized shape deviation 201 and the shape deviation step 202 as the input data S1, the same signal as the input to the neural network input layer described with reference to FIG. 7 is used. Further, the control operation amount S3 or the selection control operation amount S8 described later is used as an input to the input layer. The control operation amount S3 or the selection control operation amount S8 is composed of the position change amount of each control operation device.

Further, the output layer outputs a value estimated to determine the quality of the control result when the control operation amount S3 or the selection control operation amount S8 is output to the input data S1. The number of layers in the middle layer and the number of neurons in each layer are set in a timely manner.

FIG. 12 shows an operation amount calculation method in the new search operation amount calculation unit 33.
The new search operation amount calculation unit 33 calculates the new search operation amount S12 according to the following policy using the control output quality determination estimated value S9 output by the control output quality determination rule execution unit 17.
That is, when the value of the control output quality determination estimated value S9 is large, the quality determination of the control operation is well estimated, so that the new search operation amount calculation unit 33 makes fine adjustments as the new search operation amount.
When the value of the control output good / bad judgment estimated value S9 is small, the good / bad judgment of the control operation is badly estimated. Therefore, the new search operation amount calculation unit 33 can make a new appropriate operation method by significantly changing the control operation. Search for.

Based on the above policy, the formula for obtaining the new search operation amount Crand is set as follows.
IF (S9> th) THEN Crand = Cref ^* (1 + β ^* th1)
IF (th> = S9) THEN Crand = Cref + γ ^* th2 ^* G
Here, β and γ indicate random values generated between -1 and 1. th1 indicates the degree of fine adjustment. For example, when the range of ± 10% of the original command is to be fine adjustment, th1 is set to 0.1.

th2 is a setting to the extent that the operation method is significantly changed. For example, when th2 is set to 0.1, an offset of 10% is added to the original command, and there is a possibility that the operation polarity may change or originally. It occurs that the command of the device that was not operated is output.
As the values of β and γ, different values are used for each operating device, and the operating amount of each device is changed independently. G indicates the maximum operation position control command of each control operation device, and by multiplying the above-mentioned command%, the value of% is converted into the operation position control command.

The control output operation method selection unit 18 selects the control operation amount S3 or the new search control operation amount S12, and outputs the selection control operation amount S8. Whether to select the control operation amount S3 or the new search control operation amount S12 is stochastically determined, and the probability Grand of using the new search control operation amount S12 can be set by the user from 0 to 1. To. It is determined by the following equation using a value δ that randomly takes a value from 0 to 1.
IF (δ> Grand) THEN C "ref = Cref, α = 1
ELSE C "ref = Crand, α = 0

Here, C "ref indicates a selection control operation amount S8 output by the control output operation method selection unit 18 to the subsequent calculation units. δ uses a shared value for the calculation of the operation amount of all devices, and is the same for all devices. The operation amount on the side is used. α is a control method selection flag S14, which is 1 when the control operation amount S3 is selected and 0 when the new search control operation amount S12 is selected. This control method selection flag S14. Is output to the subsequent calculation unit together with the selection control operation amount S8. As a method of setting the Prad, if you do not want to give a risk to the plant by a random operation in the control of the actual machine, set it to 0. To improve the control rule, set a ratio other than 0 when you want to output the operation amount for new search.

FIG. 13 shows the configuration of the control output determination unit 5.
The control output determination unit 5 includes a rolling phenomenon model 501 and a shape correction pass / fail determination unit 502. Then, the control output determination unit 5 obtains the actual data Si from the controlled target plant 1, the control operation amount S3 from the control output calculation unit 3, and the information of the output determination database DB3, and controls the control operation amount to the control operation end. Output availability data S4 is given.

The control output determination unit 5 having such a configuration controls known changes in shape when the selection control operation amount S8 calculated by the control output operation method selection unit 18 is output to the rolling mill, which is the control target plant 1. Predict by inputting to the model of the target plant 1. The known model of the controlled plant 1 is here the rolling phenomenon model 501. If the shape is expected to deteriorate in this prediction, the control output determination unit 5 suppresses the control operation amount output SO and prevents the shape from being significantly deteriorated.

More specifically, the control output determination unit 5 inputs the selection control operation amount S8 into the rolling phenomenon model 501, predicts the shape change due to the selection control operation amount S8, and calculates the shape deviation correction amount prediction data 503.
On the other hand, the control output determination unit 5 obtains the shape deviation prediction data 505 by adding the shape deviation correction amount prediction data 503 to the shape detector data Si from the controlled target plant 1, and evaluates the shape deviation prediction data 505. .. As a result, the control output determination unit 5 can predict how the shape will change when the control operation amount S3 is output to the controlled target plant 1. The shape detector data Si here is the shape deviation actual data 504 at the present time.
The control output determination unit 5 determines whether the shape is changed in the direction of improvement or deterioration in the shape correction quality determination unit 502 based on the current shape deviation actual data 504 and the shape deviation prediction data 505. Then, the control operation amount output availability data S4 is obtained.

Specifically, the shape modification pass / fail determination unit 502 determines the pass / fail of the shape modification as follows. First, as shown in the specifications A and B regarding the priority of shape control, in order to consider the control priority in the plate width direction, the weighting coefficient w (i) in the plate width direction is specified in the output determination database DB3. Set for each specification of A1 and specification A2. Using it, for example, the quality of the shape change is determined by using the evaluation function J as shown in the following equation [Equation 4]. In the equation [Equation 4], w (i) is the weight coefficient, εfb (i) is the shape deviation actual data 504, εest (i) is the shape deviation prediction data 505, i is the shape detector zone, and rand is a random number term. Is.

When the evaluation function J of the equation [Equation 4] is used, the evaluation function J is positive when the shape is good, and the evaluation function J is negative when the shape is bad. Further, rand is a random number term, and the evaluation result of the evaluation function J is randomly changed. As a result, even if the shape deteriorates, the evaluation function J may be positive. Therefore, even if the rolling phenomenon model 501 is incorrect, the relationship between the shape pattern and the control method should be learned. Is possible.

Here, the random number term rand should be increased to a maximum value when the model of the controlled plant 1 is uncertain, as in the beginning of the test run, and set to 0 when it is desired to learn the control method to some extent and execute stable control. , Change in a timely manner.

The shape correction pass / fail judgment unit 502 calculates the evaluation function J, and when J ≧ 0, the control operation amount output enable / fail data S4 = 1 (possible), and when J <0, the control operation amount output enable / fail data S4 = 0 (no). ), The control operation amount output availability data S4 is output.

As described above, the standardized shape deviation 201, the shape deviation step 202, and the selection control operation amount S8 are input to the control output quality determination rule execution unit 17, and the control result quality determination estimated value S11 is output. The control result pass / fail determination estimated value S11 takes a value of 1 when it is estimated that the control result is improved, and 0 in other cases.

The control output suppression unit 4 outputs the control operation amount output SO to the controlled plant 1 according to the control operation amount output availability data S4 and the control result pass / fail determination estimated value S11, which are the determination results of the control output determination unit 5. To decide. The control operation amount output availability data S4 is # 1 to # nAS-U position change amount output, upper first intermediate shift position change amount output, and lower first intermediate shift position change amount output, and is determined under the following conditions. ..

IF (control method selection flag = 1) THEN
IF (control operation amount output availability data S4 = 0 OR control result pass / fail judgment estimated value S11 <= thprot) THEN
# 1 to # nAS-U Position change amount output = 0
Upper 1st intermediate shift position change amount output = 0
Lower 1st 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 1st intermediate shift position change amount ENDIF
ELSE
IF ((control operation amount output availability data S4 = 0 OR control result pass / fail judgment estimated value S11 <= thprot) AND (PTRIAL <η)) THEN
# 1 to # nAS-U Position change amount output = 0
Upper 1st intermediate shift position change amount output = 0
Lower 1st 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 1st intermediate shift position change amount ENDIF
ENDIF

Here, thprot sets a reference value for suppressing output based on the estimated value of the control result pass / fail judgment. Specifically, it is considered that the estimation accuracy of the quality judgment is low at the initial stage of startup when the operation data of the plant is insufficient, so keep the reference value low and do not suppress the output by estimating the quality judgment too much. To.
On the other hand, after the actual operation data is sufficiently accumulated and the accuracy of the pass / fail judgment is improved, the reference value is raised to enhance the effect of output suppression by the pass / fail judgment estimation of the control result. The accuracy of the pass / fail judgment is determined based on the verification result of the estimation accuracy of the pass / fail judgment rule currently used by receiving the pass / fail judgment rule accuracy S15 from the pass / fail judgment rule accuracy verification unit 34 in the pass / fail judgment rule learning unit. ..

Further, η is a variable that takes a random value from 0 to 1, and PTRIAL indicates the probability that output suppression is invalidated and a new search operation is output to the plant. When the control method selection flag S14 is 0, the effect of the control method in an unknown region is verified. Therefore, the output to the plant is ignored with a certain probability and the output is performed to the plant.

In the control execution unit 20, the above-mentioned calculation is executed from the actual data Si from the controlled target plant 1 (rolling mill), and the control operation amount output SO is output to the controlled target plant 1 (rolling mill) to control the shape. To execute. Further, the control method learning unit 21 uses the data used in the control execution unit 20.

Next, the operation performed by the learning data creation unit 801 will be described.
As shown in FIG. 1, the learning data creation unit 801 uses the control operation end operation command S2, the selection control operation amount S8, based on the control result quality determination estimated value S11 from the control output quality determination rule execution unit 17. From the control method selection flag S14 and the determination result of the control output suppression unit (control operation amount output availability data S4), the teacher data S7a for the neural network 111 used in the control rule learning unit 802 is created.

The teacher data S7a in this case has an AS-U operation degree 301 and an intermediate operation degree 302, which are outputs from the output layer of the neural network 111 shown in FIG. 7. The learning data creation unit 7 has a control operation end operation command S2 (AS-U operation degree 301, 1 intermediate operation degree 301) which is an output of the neural network 101, and # 1 to # nAS-U which are selection control operation amounts S8. Using the position change amount output, the upper first intermediate shift position change amount output, and the lower first intermediate shift position change amount output, the teacher data S7a for the neural network 111 used in the control rule learning unit 802 is created.

In explaining the operation of the learning data creation unit 801, FIG. 14 shows the relationship between the data and symbols of each unit in the control output calculation unit 3 shown in FIG. Here, the AS-U operation degree 301 is typically shown for the control operation end operation command S2 which is the output of the neural network 101, the data on the operation degree positive side is OPref, the data on the operation degree negative side is OMref, and conversion is performed. Let G be the gain and Cref be the control operation amount S3.

The difference between the operation degree correct data OPref and the operation degree control data OMref is taken by the subtractor 701, and the conversion gain G is multiplied by the multiplier 702 to obtain the control operation amount output Cref. This control operation amount output Cref is supplied to the control output operation method selection unit 18, and the selected operation command value C "ref is obtained.
Here, for the sake of simplicity, as the output from the output layer of the neural network 101 of the control rule execution unit 10, the operation degree that is randomly generated from the operation degree positive side and the operation degree negative side, and the control operation disturbance generation unit 16 is used. The operation degree is a random number. Further, the control operation amount output SO for the control operation end is set as the operation command value.

FIG. 15 shows the processing stage and the processing content in the learning data creation unit 7.
In the first processing step 71, the operation command value C "ref refers to the selection control operation amount S8 which is the output value of the control output operation method selection unit 18.

In the next processing step 72, the operation command value Cref is modified to be C'ref according to the control result pass / fail judgment estimated value S11 and the control operation amount output availability data S4. Specifically, when the control result pass / fail judgment estimated value S11 = 0 or the control operation quantity output availability data S4 = 0, the following [Equation 5] formula, the control result pass / fail judgment estimated value S11 = 1 and the control operation quantity output availability data When S4 = 1, the correction value C'ref of the operation command value C "ref is set by the following equation [Equation 6].

In the processing step 73, the operation degree correction amount ΔOref is obtained from the modified operation command value C'ref by the [Equation 7] equation and the [Equation 8] equation.

In the processing step 74, the teacher data OP'ref and OM'ref to the neural network 111 are obtained by the equation [Equation 9].

In this way, as shown in FIG. 14, in the learning data creation unit 7, the operation command value C ”ref actually output to the controlled target plant 1 is estimated by the control result quality judgment rule execution unit 17 of the control output quality judgment rule. The operation command value correction value C'ref is calculated according to the value S11 and the control operation amount output availability data S4 of the control output suppression unit 4.
Specifically, when the control result pass / fail judgment estimated value S11 = 1 and the control operation amount output availability data S4 = 1, the operation command value is increased by ΔCref in the same direction when it is judged to be a good operation. To do so.

On the contrary, when the control result pass / fail judgment estimated value S11 = 0 or the control operation amount output availability data S4 = 0, the operation command value is reduced by ΔClef in the opposite direction when it is judged that the operation is not good. To. Since the conversion gain G is known because it is set in advance, it is possible to obtain the correction amount ΔOref if the values on the positive operation degree side and the negative operation degree side are known. Here, ΔCref is set by obtaining an appropriate value in advance by simulation or the like. By the above procedure, the teacher data OP'ref and OM'ref used in the control rule learning unit 802 can be obtained by the equation [Equation 9].

Although the description is given with a simple example in FIG. 14, in reality, the AS-U operation degree 301 for # 1 to # nAS-U, the upper first intermediate roll shift, and the lower first intermediate roll shift are described. All of the first intermediate operation degree 302 is executed, and the teacher data (AS-U operation degree teacher data, 1 intermediate operation degree teacher data) of the neural network 111 used in the control rule learning unit 802 is used.

FIG. 16 shows an example of data stored in the learning data database DB2.
In order to learn the neural network 111, a combination of a large number of input data S8a and teacher data S7a is required. Therefore, the teacher data S7a created by the training data creation unit 7 is combined with the input data S1 (S8a) input to the control rule execution unit 10 by the control execution unit 20 to form a set of training data, the training data database DB2. It is saved in. The teacher data S7a here is the AS—U operation degree teacher data and the first intermediate operation degree. Further, the input data S1 (S8a) is a standardized shape deviation 201 and a shape deviation stage.

The plant control system of FIG. 1 uses various databases DB1, DB2, DB3, and DB4, but each database DB1, DB2, DB3, and DB4 are managed and operated interconnectedly by the neural net management table TB. Will be done.

FIG. 17 shows the configuration of the neural network management table TB.
The neural network management table TB is classified according to specifications A1 and A2 regarding (B1) plate width, (B2) steel grade, and control priority. (B1) As the plate width, for example, 4 divisions of 3 feet width, meter width, 4 feet width, and 5 feet width are used, and as the steel grade, about 10 divisions of steel grade (1) to steel grade (10) are used. Further, regarding the specification A regarding the priority of control, there are two types, A1 and A2. In this case, there are 80 divisions, and 80 neural networks are used properly according to the rolling conditions.

The neural network learning control unit 112 applies learning data, which is a combination of input data and teacher data, as shown in FIG. 16, according to the neural network management table TB shown in FIG. Is stored in the learning data database DB2 as shown in FIG.

The control execution unit 20 creates two sets of learning data each time the shape control is executed for the controlled target plant 1. This is because the control result pass / fail judgment is performed for the same input data and control output using the two evaluation criteria of the specification A1 and the specification A2 regarding the control priority, so that two types of teacher data are created. Is. When the teacher data is accumulated to some extent (for example, 200 sets) or newly accumulated in the learning data database DB2, the neural network learning control unit 112 instructs the learning of the neural network 111.

In the control rule database DB1, a plurality of neural networks are stored according to the management table TB as shown in FIG. The neural network learning control unit 112 has a neural network No. that requires learning. Is specified, the neural network selection unit 113 takes out 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 creation unit 114 and the teacher data creation unit 115 to take out the input data and the teacher data corresponding to the corresponding neural network from the training data database DB2, and uses them. The training of the neural network 111 is executed. Various methods have been proposed for learning the neural network, and any method may be used.

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

The training may be performed simultaneously for all the neural networks defined as shown in FIG. 17 at regular time intervals (for example, every day), or new training data is accumulated to some extent (for example, 100 sets). Neural net No. Only the neural network of may be trained at that time.

Next, the operation of the pass / fail determination rule learning unit 22 will be described.
The pass / fail determination rule learning unit 22 uses the time delay data of the data used in the control execution unit 20. Here, the time delay Z ^-1 means e-TS, and indicates that the time is delayed by a preset time T.
Since the controlled target plant 1 has a time response, there is a time delay until the actual data changes due to the control operation amount output SO. Therefore, the learning is executed using the actual data at the time when the delay time T has elapsed after the control operation is executed.

In the shape control, it takes several seconds for the shape meter to detect the shape change after the operation command is output to the AS-U or the first intermediate roll, so it is preferable to set T = 2 seconds to about 3 seconds. Since the delay time changes depending on the type of the shape detector and the rolling speed, it is preferable to set the optimum time until the change of the control operation end becomes the shape change as T.

FIG. 19 shows the operation of the control result quality determination unit 6. In the shape change quality determination unit 602, the quality change evaluation function Jc shown in the equation [Equation 10] is used.

In this equation [Equation 10], 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 plate width direction for pass / fail determination. It is a weighting coefficient. Here, the weighting coefficient wC (i) for pass / fail determination is set from the pass / fail determination database DB4 according to the specifications A1 and A2 regarding the priority of control. The quality of the control result is determined by the quality determination evaluation function Jc.

The threshold upper limit LCU and the threshold addition / subtraction LCL are set in advance under the threshold condition (LCU ≧ 0 ≧ LCL). At this time, if the result of comparison with the pass / fail judgment evaluation function Jc is Jc> LCU, the control result pass / fail data S6 = 0 (the shape has deteriorated), and if Jc <LCL, the control result pass / fail data. It is assumed that S6 = 1 (the shape has improved).

As described above, since the weighting coefficient wC (i) in the plate width direction changes according to the specifications A1 and A2 regarding the priority of control, the quality determination evaluation function Jc is different. Therefore, it is conceivable that the determination result of the control result pass / fail data S6 is also different. Therefore, the pass / fail determination rule learning unit 22 executes the determination of the control result pass / fail data S6 for the two types of specifications A1 and A2 regarding the priority of control.
This control result pass / fail data S6 is used as it is as the teacher data S13a for the neural network 311 used in the pass / fail determination rule learning unit 31.

FIG. 20 shows an example of data stored in the learning data database DB6.
In order to learn the neural network 311, a combination of a large number of input data S12a and teacher data S13a is required. Therefore, the teacher data S13a (control result pass / fail data) created by the control result pass / fail determination unit 6 is the input data S1 (standardized shape deviation 201 and shape deviation step) input to the control rule execution unit 10 in the control execution unit 20. ) Is combined with the time delay data S12a and stored in the training data database DB 6 as a set of training data.
At this time, the learning data is stored in the verification data database DB 7 at a constant ratio so that it can be used for the pass / fail judgment rule verification in the pass / fail judgment rule accuracy verification unit 34.

The pass / fail determination rule accuracy verification unit 34 is provided with a neural network that performs only one-way calculation like the control output pass / fail determination rule execution unit 17. Then, the pass / fail judgment rule accuracy verification unit 34 takes out the test data from the verification data database DB7, inputs the input data of the data to the neural net, and calculates the error between the output data obtained and the output data of the test data. .. For example, the pass / fail determination rule accuracy verification unit 34 calculates the average value of the errors of all the test data as the pass / fail determination rule accuracy S15 of the pass / fail determination rule.

The plant control system of FIG. 1 uses various databases DB5 and DB6, but FIG. 21 shows the configuration of a neural network management table TB for interconnectedly managing and operating each of the databases DB5 and DB6. .. That is, the management table TB includes a specification management table.

Specifically, as shown in FIG. 21, the management table TB is classified according to (B1) plate width, (B2) steel grade, and specifications A1 and A2 regarding control priorities. (B1) As the plate width, for example, 4 divisions of 3 feet width, meter width, 4 feet width, and 5 feet width are used, and as the steel grade, about 10 divisions of steel grade (1) to steel grade (10) are used. Further, regarding the specification A regarding the priority of control, there are two types, A1 and A2. In this case, there are 80 divisions, and 80 neural networks are used properly according to the rolling conditions.

The neural network learning control unit 312 applies the training data, which is a combination of the input data and the teacher data, as shown in FIG. 20, according to the neural network management table TB of FIG. Is stored in the learning data database DB 6 as shown in FIG. 22.

Each time the control execution unit 20 executes shape control for the controlled target plant 1, two sets of learning data are created. This is because the control result pass / fail judgment is performed for the same input data and control output using the two evaluation criteria of the specification A1 and the specification A2 regarding the control priority, so that two types of teacher data are created. Is. When the teacher data is accumulated to some extent (for example, 200 sets) or newly accumulated in the learning data database DB 6, the neural network learning control unit 312 instructs the learning of the neural network 311.

The pass / fail judgment rule database DB 5 stores a plurality of neural networks according to the management table TB as shown in FIG. Then, the neural network learning control unit 312 determines the neural network No. that requires learning. Is specified, the neural network selection unit 313 extracts the corresponding neural network from the pass / fail determination rule database DB5, and sets it in the neural network 311. The neural network learning control unit 312 extracts input data and teacher data corresponding to the corresponding neural network from the training data database DB6, instructs the input data creation unit 314 and the teacher data creation unit 315, and uses them to make a neural network. Perform the learning of the net 311. Various methods have been proposed for learning the neural network, and any method may be used.

When the learning of the neural network 311 is completed, the neural network learning control unit 312 applies the neural network 311 which is the learning result to the neural network No. 3 of the control rule database DB6. Learning is completed by writing back to the position of.

The learning is executed all at once for all the neural networks defined in the management table TB shown in FIG. 21 at regular time intervals (for example, every day). Alternatively, the neural network No. in which new learning data is accumulated to some extent (for example, 100 sets). You may learn only the neural network of at that time.

In addition, by including the rolling record, steel type, and plate width in the input data of the pass / fail judgment rule, it is possible to learn with one neural network including the difference in the pass / fail judgment criteria. In this case, it is not necessary to switch the pass / fail judgment rule according to the rolling conditions when the pass / fail judgment rule is executed.

As described above, in order to improve the control rule of the controlled plant 1, if the control operation with a good control result cannot be learned, the control operation is significantly changed. If the control result is good, it will be adopted as a new control operation method. Furthermore, if the control operation with good control results can be learned, the control operation should not be changed or only a small change should be made. Then, when the control results for these are good, it is efficient to incorporate them as a new control operation method.

Then, by learning the combination of the shape pattern, the control operation, and the quality of the control result based on the actual machine data, a model that can estimate the quality of the control result more accurately than the simulator using the machine model is constructed. With regular automatic learning, it is possible to always build a model that matches the latest plant conditions.

In addition, by using a model that estimates the quality of the control result, it is possible to improve the reliability of the control output suppression function for the plant, which was performed only by the simple mechanical model in the conventional technique.
In addition, in the case of this example, the generation of control rule learning data, which was conventionally performed by determining the quality of the control result once, is generated by using a model that estimates the quality of the control result. It is possible to suppress the influence and target the training data even for fine adjustments with small effects. Further, according to this example, by preventing erroneous determination of the control effect, it is possible to suppress fluctuations in the learning data and stabilize the control performance.

The control rule database DB 1 stores a neural network used by the control execution unit 20. Here, when the stored neural network only executes the initial processing with random numbers, it takes time for the learning of the neural network to proceed and the control to be possible to some extent. Therefore, when the control unit is constructed for the controlled target plant 1, the learning of the control rule is executed in advance by simulation based on the control model of the controlled target plant 1 known at that time. Then, by storing the neural network that has been learned by the simulator in the database, it becomes possible to control the performance to some extent from the beginning of the controlled plant.

Alternatively, by learning the pass / fail judgment rule in the pass / fail judgment rule learning unit 22 based on the actual data of the operation data of the actual machine, it is possible to learn the control rule without controlling the actual machine. It is possible to perform some performance control before application to the controlled plant.

FIG. 23 shows a configuration in the case where the plant control system of this example includes a control rule evaluation unit 23 that performs control rule evaluation processing.
The control rule evaluation unit 23 includes a control rule pass / fail judgment data collection unit 35, a control rule evaluation data calculation unit 36, a control rule database update unit 37, a control rule evaluation data database DB 8, and a control rule evaluation value database DB 9.

The control rule pass / fail judgment data collection unit 35 receives the control output pass / fail judgment estimated value S9 from the control output pass / fail judgment rule execution unit 17, and also receives the pass / fail judgment rule accuracy S15 from the pass / fail judgment rule accuracy verification unit 34. Then, the control rule pass / fail determination data collection unit 35 stores the control rule pass / fail determination data S16 together with the control rule number used in the control execution unit 20 in the control rule evaluation data database DB 8. The control rule pass / fail determination data S16 is a control output pass / fail determination estimated value S9. However, when the quality determination rule accuracy S15 is equal to or less than a certain level, the database DB 8 is not saved.

The control rule pass / fail judgment data S16 obtains new data each time the control output calculation using the control rule is performed in the control execution unit 20, and the obtained control rule pass / fail judgment data S16 is the control rule evaluation data database. It is saved in DB8. In this case, a large amount of data is stored for each control rule. Therefore, the control rule evaluation data database DB8 sets an upper limit of the data to be stored in each control rule, and if it exceeds a certain level, it is old. Delete the data and save the new data.

The control rule evaluation data calculation unit 36 collectively takes out the control rule pass / fail judgment data S17 accumulated for each control rule from the control rule evaluation data database DB8, and obtains the average value as the control rule evaluation data S18 by calculation. The calculated average value corresponds to the evaluation value.
The control rule evaluation data S18 calculated by the control rule evaluation data calculation unit 36 is stored in the control rule evaluation value database DB 9. However, if the number of control rule pass / fail judgment data is less than a certain number, the evaluation value is not saved because the reliability of the evaluation value is low.

In the database management table TB, the neural network No. used according to the conditions. (Control rules) are registered one by one. On the other hand, the control rule evaluation value database DB 9 manages the evaluation values of a plurality of control rules. The control rule database update unit 37 refers to the control rule evaluation value database DB9 and refers to the neural network No. 1 registered in the database management table TB. The control rule evaluation value of (control rule) is compared with the control rule evaluation value of other control rules applicable to the condition, and the control rule having the highest evaluation value is referred to as the neural network No. of the database management table TB. Update to (control rule).

Other parts of the plant control system shown in FIG. 23 are configured in the same manner as the plant control system shown in FIG. However, in the case of the plant control system shown in FIG. 23, since the control rule evaluation unit 23 evaluates based on the past performance as the past performance data of the control target plant 1, the control execution unit 20 is actually the control target. There is no need to run plant 1. Specifically, it is not necessary to supply the control output amount S0 from the control output suppression unit 4 to the controlled target plant 1.

According to the plant control system shown in FIG. 23, by setting the control rule to be evaluated in the control rule execution unit 10 and giving the past actual data as Si, the control output is not actually output to the controlled plant 1. However, the control rule evaluation value database DB 9 can be updated.

<Modification example>
The present invention is not limited to the above-described embodiments, but includes various modifications. For example, the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one including all the described configurations.

For example, the plant control system shown in FIGS. 1 and 23 is configured to include a processing unit that performs processing such as data creation, learning, and control. The control execution unit 20, the control method learning unit 21, the pass / fail judgment rule learning unit 22, and the control rule evaluation unit 23 shown in FIGS. 1 and 23 are composed of a program (software) in which the processor realizes each function. You may let the computer run the program. FIG. 24 shows an example of a computer configuration in this case.

That is, as shown in FIG. 24, the computers constituting the units 20 to 23 have a CPU (Central Processing Unit) a, a ROM (Read Only Memory) b, and a RAM (RAM (Read Only Memory) b) connected to the bus, respectively. Random Access Memory) c is provided. Further, the computer includes a non-volatile storage d and a network interface e.

The CPUa is an arithmetic processing unit that reads out the program code of the software that executes the processing in each unit 20 to 23 from the ROM b and executes the processing. Variables, parameters, etc. generated during the arithmetic processing are temporarily written in the RAMc. For the non-volatile storage d, for example, a large-capacity information storage unit such as an HDD (Hard Disk Drive) or SSD (Solid State Drive) is used, and programs executed by each unit 20 to 23, data of each database, etc. Is stored.
Each unit 20 to 23 may be configured by a different computer, or each program may be mounted on a small number of computers such as one and executed at the same time.

For the network interface e, for example, a NIC (Network Interface Card) or the like is used, and data is transmitted / received to / from another unit or the controlled target plant 1.
Information such as a program that realizes each processing function in this case can be placed in a recording medium such as a memory, an IC card, an SD card, or an optical disk, in addition to the non-volatile storage d such as an HDD or SSD.

Further, a part or all of the functions performed by each unit 20 to 23 may be realized by hardware such as FPGA (Field Programmable Gate Array) or ASIC (Application Specific Integrated Circuit).

Further, in the block diagrams shown in FIGS. 1 and 23, only the control lines and information lines considered to be necessary for explanation are shown, and not all the control lines and information lines are necessarily shown in the product. do not have. In practice, it can be considered that almost all configurations are interconnected.
Further, in the above-described embodiment, the control target plant 1 is applied to a Sendimia rolling mill, but the present invention can be applied to the control of various other plants. The control rule when applied to the Sendimia rolling mill is also shown as an example, and the present invention is not limited to the above-described embodiment.

1 ... Control target plant, 2 ... Control input data creation unit, 3 ... Control output calculation unit, 4 ... Control output suppression unit, 5 ... Control output judgment unit, 6 ... Control result pass / fail judgment unit, 7 ... Learning data creation unit, 10 ... Control rule execution unit, 16 ... Control operation disturbance generation unit, 17 ... Control output quality judgment rule execution unit, 18 ... Control output operation method selection unit, 20 ... Control execution unit, 21 ... Control method learning unit, 22 ... Good or bad Judgment rule learning unit, 23 ... Control rule evaluation unit, 31 ... Good / bad judgment rule learning unit, 33 ... New search operation amount calculation unit, 34 ... Good / bad judgment rule accuracy verification unit, 35 ... Control rule good / bad judgment data collection unit, 36 ... Control rule evaluation data calculation unit, 37 ... Control rule database update unit, 50 ... Sendimia rolling mill, 51 ... Pattern recognition unit, 52 ... Shape detector, 53 ... Control calculation unit, 54 ... Shape detection preprocessing unit, 101 ... Neural Net, 102 ... Neural net selection unit, 110 ... Neural net processing unit, 111 ... Neural net, 112 ... Neural net learning control unit, 113 ... Neural net selection unit, 114 ... Input data creation unit, 115 ... Teacher data creation unit, 171 ... Neural net, 172 ... Neural net selection unit, 201 ... Standardized shape deviation, 202 ... Shape deviation stage, 210 ... Shape deviation PP value calculation unit, 211 ... Shape deviation stage calculation unit, 310 ... Neural net processing unit, 311 ... Neural net, 312 ... Neural net learning control unit, 313 ... Neural net selection unit, 314 ... Input data creation unit, 315 ... Teacher data creation unit, 501 ... Rolling phenomenon model, 502 ... Shape correction pass / fail judgment unit, 503 ... Shape Deviation correction amount prediction data, 504 ... Shape deviation actual data, 505 ... Shape deviation prediction data, 602 ... Shape change quality judgment unit, 801 ... Learning data creation unit, 802 ... Control rule learning unit, DB1 ... Control rule database, DB2 ... Learning data database, DB3 ... Output judgment database, DB4 ... Good / bad judgment database, DB5 ... Good / bad judgment rule database, DB6 ... Learning data database, DB7 ... Verification data database, DB8 ... Control rule evaluation data database, DB9 ... Control rule evaluation value database

Claims (7)

A plant control system that recognizes a pattern of combinations of actual data of the controlled plant for the controlled plant and executes control.
A control method learning unit that learns a combination of actual data and control operations of the controlled plant, and a control execution unit that executes control of the controlled plant according to the combination of actual data and control operations learned by the control method learning unit. And a pass / fail judgment rule learning unit that learns the combination of the actual data of the controlled plant, the control operation, and the control result pass / fail.
The control execution unit is
A control rule execution unit that gives control output according to a specified combination of actual data of the controlled plant and control operations.
A control output quality judgment rule execution unit that determines the quality of the control output according to a combination of actual data of the controlled plant, control operation, and control result quality.
A new search operation amount calculation unit that calculates an operation amount for new operation search based on the quality judgment in the control output pass / fail judgment rule execution unit, and a new search operation amount calculation unit.
When it is determined that the actual data of the controlled target plant deteriorates when the control output is output to the controlled target plant by using the pass / fail judgment in the control output pass / fail judgment rule execution unit and the simulation data using the control simulator. Also equipped with a control output suppression unit that prevents the control output from being output to the controlled plant.
The pass / fail judgment rule learning unit is
When the control execution unit outputs the control output to the controlled target plant, the control result quality determination unit that determines the quality of the control result after a time delay until the control effect appears in the actual data, and the control result quality determination unit.
It is provided with a quality determination rule learning unit that learns the quality of the control result in the control result quality determination unit, the actual data, and the control output as learning data.
The control method learning unit is
The control output quality determination rule execution unit determines the quality of the control output, and the learning data creation unit that obtains teacher data using the control output.
A plant control system including a control rule learning unit that learns the actual data and the teacher data as learning data. By learning by the control method learning unit, it is possible to obtain different combinations of actual data and control operations for a plurality of control targets according to the state of the controlled plant.
The plant control system according to claim 1, wherein the combination of the obtained actual data and the control operation is used as a defined combination of the actual data of the controlled plant and the control operation in the control rule execution unit. The control output pass / fail judgment rule execution unit holds the combination of the actual data of the controlled plant, the control operation, and the control result pass / fail as the first neural network.
The pass / fail judgment rule learning unit holds a combination of actual data, control operation, and control result pass / fail as a second neural network.
The plant control system according to claim 1, wherein the second neural network obtained as a result of learning in the pass / fail determination rule learning unit is used as the first neural network in the pass / fail determination rule execution unit. The pass / fail judgment rule learning unit includes a pass / fail judgment rule accuracy verification unit.
The plant control system according to claim 1, wherein the quality determination rule accuracy generated by the quality determination rule verification unit is used to change the output suppression standard using the control result quality in the control output suppression unit. In addition, it is equipped with a control rule evaluation unit.
The control rule evaluation unit is
A control rule pass / fail judgment data collection unit that stores the pass / fail judgment data of the pass / fail judgment rule execution unit of the control execution unit and the accuracy verification result of the pass / fail judgment rule obtained by the pass / fail judgment rule learning unit in a database.
It has a control rule evaluation data calculation unit that calculates control rule evaluation data based on the pass / fail judgment data stored in the database and the accuracy verification result of the pass / fail judgment rule.
The plant control system according to any one of claims 1 to 4, wherein the evaluation of the control rule used for the control execution unit is executed without being output to the controlled plant. It is a plant control method that recognizes a pattern of combination of actual data of the controlled target plant for the controlled target plant and executes control of the controlled target plant by a computer.
As a process executed by the computer,
A control method learning process that learns a combination of actual data and control operations of the controlled plant, and a control execution process that executes control of the controlled plant according to the combination of actual data and control operations learned by the control method learning process. And a pass / fail judgment rule learning process for learning a combination of the actual data of the controlled plant, a control operation, and a control result pass / fail combination.
The control execution process is
Control rule execution processing that gives control output according to the specified combination of the actual data of the controlled plant and the control operation,
The control output quality judgment rule execution process that determines the quality of the control output according to the combination of the actual data of the controlled plant, the control operation, and the control result quality.
A new search operation amount calculation process that calculates an operation amount for a new operation search based on the pass / fail judgment by the control output pass / fail judgment rule execution process, and
When it is determined that the actual data of the controlled target plant deteriorates when the control output is output to the controlled target plant by using the quality judgment by the control output pass / fail judgment rule execution process and the simulation data using the control simulator. , A control output suppression process that prevents the control output from being output to the controlled plant.
The pass / fail judgment rule learning process is
When the control output is output to the controlled plant by the control execution process, the control result pass / fail determination process for determining the quality of the control result after a time delay until the control effect appears in the actual data, and the control result pass / fail determination process.
Includes the quality of the control result in the control result quality determination process, the quality determination rule learning process for learning the actual data and the control output as learning data, and the like.
The control method learning process is
The control output pass / fail judgment by the control output pass / fail judgment rule execution process, the learning data creation process for obtaining teacher data using the control output, and the control output pass / fail determination rule execution process.
A plant control method including a control rule learning process for learning the actual data and the teacher data as learning data. It is a program that recognizes the pattern of the combination of the actual data of the controlled target plant for the controlled target plant and causes the computer to execute the plant control.
The program executes control of the controlled plant according to the control method learning procedure for learning the combination of the actual data and the control operation of the controlled plant and the combination of the actual data and the control operation learned by the control method learning procedure. The computer is made to execute the control execution procedure to be performed, the pass / fail judgment rule learning procedure for learning the combination of the actual data of the controlled target plant and the control operation, and the combination of the control result pass / fail.
The control execution procedure is
A control rule execution procedure that gives control output according to a defined combination of actual data of the controlled plant and control operations, and
The control output quality judgment rule execution procedure for determining the quality of the control output according to the combination of the actual data of the controlled plant, the control operation, and the control result quality,
A new search operation amount calculation procedure for calculating a new operation search operation amount based on the quality judgment by the control output pass / fail judgment rule execution procedure, and a new operation search operation amount calculation procedure.
When it is determined that the actual data of the controlled target plant deteriorates when the control output is output to the controlled target plant by using the quality judgment by the control output pass / fail judgment rule execution procedure and the simulation data using the control simulator. , A control output suppression procedure that prevents the control output from being output to the controlled plant.
The pass / fail judgment rule learning procedure is as follows.
When the control output is output to the controlled plant by the control execution procedure, the control result quality determination procedure for determining the quality of the control result after the time delay until the control effect appears in the actual data, and the control result quality determination procedure.
Includes the good / bad of the control result in the control result good / bad judgment procedure, and the good / bad judgment rule learning procedure for learning the actual data and the control output as learning data.
The control method learning procedure is
The pass / fail judgment of the control output by the control output pass / fail judgment rule execution procedure, the learning data creation procedure of obtaining the teacher data using the control output, and the procedure.
A program including a control rule learning procedure for learning the actual data and the teacher data as learning data.

Images