PLANT CONTROL SYSTEM, ROLLING MILL CONTROL DEVICE, PLANT CONTROL METHOD, AND PLANT CONTROL PROGRAM

Abstract

To reduce a computer load by updating a control model at an appropriate frequency and achieve performance maintenance of a control rule.SOLUTION: A plant control system which controls a plant to be controlled comprises: a control method learning device which learns control rules based on a combination of actual data and control operations of the plant to be controlled; a control execution device which controls the plant to be controlled based on the control rules learned by the control method learning device; and a control rule update determination device which calculates a degree of adaptation of the control rules for the plant to be controlled based on the actual data when controlling the plant to be controlled based on the control rules, and updates the control rules based on the degree of adaptation.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a plant control system, a rolling mill control device, a plant control method, and a plant control program which perform real-time feedback control using artificial intelligence technology such as a neural network and have an automatic update function for the learning of the artificial intelligence. [Background technology]

[0002] 2. Description of the Related Art Conventionally, in various plants, plant control based on various control theories has been carried out in order to obtain desired control results by the control.

[0003] In the control of a rolling mill as an example of a plant, fuzzy control and neuro-fuzzy control have been applied as control theories for shape control, which controls the waviness of a strip as an example of control. Fuzzy control is applied to shape control using coolant, and neuro-fuzzy control is applied to shape control of a Sendzimir rolling mill. As shown in Patent Document 1, shape control using neuro-fuzzy control is performed by calculating the similarity ratio between the difference between an actual shape pattern detected by a shape detector and a target shape pattern and a preset reference shape pattern, and calculating the control output amount for the control terminal from the similarity ratio according to a control rule expressed by the control operation amount for the reference shape pattern, which is also set in advance. Hereinafter, the shape control of a Sendzimir rolling mill using neuro-fuzzy control is used as the conventional technology.

[0004] As a management device for the control device, there is known a management system for the hydraulic roll-down control device, which includes a hydraulic roll-down device that controls the piston position of a hydraulic cylinder, as shown in Patent Document 3, and a plurality of management devices for the hydraulic roll-down control device.

[0005] This management system has a function of determining the timing for updating the parameters of the control model of the hydraulic roll-down control device, and adjusting the parameters based on the command values ​​from the control model to the hydraulic roll-down control device and the actual measured values ​​thereof.

[0006] FIG. 5 shows the shape control of the Sendzimir rolling mill described in FIG. 1 of Patent Document 1. Neuro-fuzzy control is used in the shape control of the Sendzimir rolling mill. In this example, the pattern recognition mechanism 51 recognizes the shape pattern from the actual shape detected by the shape detector 52, and calculates which of the preset reference shape patterns the actual shape is closest to. The control calculation mechanism 53 performs control using a control rule consisting of the control operation terminal operation amount for the preset shape pattern as shown in FIG. 6. More specifically, FIG. 6 is described, and the pattern recognition mechanism 51 calculates which of the shape patterns (ε) from 1 to 8 the difference (Δε) between the shape performance detected by the shape detector 52 and the target shape (εref) is closest to, and the control calculation mechanism 53 selects and executes one of the control methods from 1 to 8.

[0007] However, in the method of Patent Document 1, in order to verify the control rule, an operator may be asked to perform manual operations during rolling, but this may result in unexpected changes in shape. In other words, there are cases where the control rule determined in the above manner does not match reality. This is caused by insufficient consideration of mechanical properties or changes in the operating state and machine conditions of the rolling mill, but it is difficult to verify one by one whether the previously set control rule is the best rule because there are many conditions to be considered. Therefore, once a control rule is set, it is often left as it is unless there is a problem.

[0008] When the control rules no longer correspond to reality due to changes in operating conditions, etc., it becomes difficult to achieve a certain level of control precision because the control rules are fixed. Also, once shape control is activated, the operator will no longer perform manual operations (which will become a disturbance to the control), so it is difficult to find new control rules through manual intervention by the operator. Furthermore, when rolling new standards of rolled material, it is difficult to set control rules to suit that material.

[0009] As described above, conventional shape control has a problem in that it is difficult to modify the control rules because the control is performed using preset control rules.

[0010] In order to solve this problem, as shown in Patent Document 2, the control rules are randomly changed while controlling the shape, and the rules that improve the shape are learned. 1) New control rules will be discovered while controlling shape during rolling. 2) Since new control rules cannot be predicted in advance, and a completely unpredictable control rule may turn out to be optimal, the optimal control rule is found by randomly operating the control terminals and observing the control results. This has been achieved. [Prior art documents] [Patent documents]

[0011] [Patent Document 1] Patent No. 2804161 Specification [Patent Document 2] Patent No. 4003733 specification [Patent Document 3] Patent No. 5363380 specification Summary of the Invention [Problem to be solved by the invention]

[0012] In the above-mentioned conventional technology, a representative shape is set in advance as a reference shape pattern, and control is performed based on a control rule that indicates the relationship between the reference waveform pattern and the control manipulated variable. The learning of the control rule also relates to the control manipulated variable for the reference waveform pattern, and the predetermined representative reference shape pattern is used as is. Therefore, there is a problem that the shape control responds only to a specific shape pattern.

[0013] Although the reference shape pattern is determined by a human being based on knowledge of the target rolling mill and experience accumulated from shape records and manual intervention operations, it is difficult to cover all shapes that may occur in the target rolling mill and the material being rolled. Therefore, when a shape different from the reference shape pattern occurs, control by shape control is not performed, and the shape deviation remains unsuppressed, or it may be mistaken for a similar reference shape pattern and an incorrect control operation is performed, which may worsen the shape.

[0014] As described above, in conventional shape control, a control rule is learned using a preset reference shape pattern and its corresponding control rule, and control is then performed. This has the problem that there is a limit to how much improvement can be made in control accuracy.

[0015] In addition, although the control model used in deep learning applied shape control can demonstrate a high control effect immediately after updating, after a certain period of time, as the plant environment ages and the operating conditions change, its suitability to the plant decreases and it is no longer able to demonstrate a sufficient control effect. Therefore, the control model must be successively optimized according to the plant's condition, but it is difficult to constantly update the control model because creating teaching data and executing the control model construction process in real time using rolling performance data that is generated every moment places a large burden on the computer. [Means for solving the problem]

[0016] In view of the above, in the present invention, a plant control system that controls a controlled plant is characterized by comprising: a control method learning device that learns a control rule based on a combination of performance data and control operations of the controlled plant; a control execution device that controls the controlled plant based on the control rule learned by the control method learning device; and a control rule update determination device that calculates a degree of suitability of the control rule with respect to the controlled plant based on the performance data when control of the controlled plant is executed based on the control rule, and updates the control rule based on the degree of suitability. Effect of the Invention

[0017] By using the present invention, it is possible to automatically correct and optimize the control rules for the shape pattern and operation method used in shape control during control, which has the effects of improving control accuracy, shortening the start-up period of the control unit, and making it possible to deal with deterioration over time.

[0018] Furthermore, according to the present invention, the degree of compatibility between the control rules and performance data is evaluated, the timing for updating the control rules is determined, and automatic re-learning is performed, thereby updating the control model at an appropriate frequency, thereby reducing the load on the computer and maintaining the performance of the control rules. [Brief description of the drawings]

[0019] [Figure 1] 1 is a diagram showing an overview of a plant control system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram showing a specific example of the configuration of the control rule execution unit 10 according to the embodiment of the present invention. [Diagram 3] FIG. 2 is a diagram showing a specific configuration example of a control rule learning unit 11 according to an embodiment of the present invention. [Figure 4] FIG. 13 is a diagram showing a neural network configuration when the present invention is used for shape control of a Sendzimir rolling mill. [Diagram 5] FIG. 2 is a diagram showing shape control of the Sendzimir rolling mill described in FIG. 1 of Patent Document 1. [Figure 6] FIG. 2 is a diagram showing a control rule for shape control in the Sendzimir rolling mill described in FIG. 1 of Patent Document 1. [Figure 7] FIG. 2 is a diagram showing an overview of a control input data creation unit 2. [Figure 8] FIG. 2 is a diagram showing an overview of a control output calculation unit 3. [Figure 9] FIG. 2 is a diagram showing an overview of a control output determination unit 5. [Figure 10] 13A and 13B are diagrams showing shape deviation and a control method. [Figure 11] FIG. 2 is a diagram showing an overview of a control result pass / fail determination unit 6. [Figure 12]FIG. 2 is a diagram showing the relationship between data and symbols in the control output calculation unit 3. [Figure 13] FIG. 4 is a diagram showing processing steps and processing contents in a learning data creation unit 7. [Figure 14] FIG. 13 shows an example of data stored in the learning data database DB2. [Figure 15] FIG. 4 is a diagram showing an example of a database management table TB. [Figure 16] FIG. 2 shows an example of a learning data database DB2. [Figure 17] FIG. 2 is a diagram showing an overview of a control rule conformance evaluation unit 25. [Figure 18] FIG. 2 is a diagram showing an overview of a control rule update evaluation unit 26. [Figure 19] FIG. 2 is a diagram showing an overview of a control rule update processing management unit 24. [Figure 20] FIG. 4 is a diagram showing an overview of a control rule evaluation value database DB5. [Figure 21] FIG. 5 is a diagram showing an overview of the hardware of a computer 500. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Before that, however, the findings of the present invention and the process by which the present invention was arrived at will be described with reference to shape control of a rolling mill as an example.

[0021] First, in order to solve the above problems in the present invention, 1) Instead of separately setting a reference shape pattern and a control operation for that pattern and learning the control operation method, a combination of a shape pattern and a control operation is learned and used to carry out the control operation. 2) Since new control rules cannot be predicted in advance, and a completely unpredictable control rule may turn out to be optimal, the optimal control rule is found by randomly operating the control terminals and observing the control results. It becomes necessary to do so.

[0022] To achieve this, it is necessary to change the combination of the shape pattern and the control operation used for shape control, and to change the control operation so as to improve the control result.To do this, it is necessary to configure a neural network that can learn the combination of the shape pattern and the control operation, and to change the output of the neural network's control operation for the shape pattern generated by the rolling mill according to the quality of the control result.

[0023] If the above is carried out while performing shape control on an operating rolling mill, erroneous control output may be output, causing deterioration of the shape and operational abnormalities such as strip breakage. If strip breakage occurs, it will take time to replace the rolls used in the rolling mill and the material being rolled will be wasted, causing great damage. Therefore, it is necessary to avoid outputting erroneous control output to the rolling mill as much as possible.

[0024] In view of the above, in order to realize this, in the present invention, the quality of the control operation output by the neural network is verified using, for example, a simple model of a rolling mill, and an output that is considered to clearly deteriorate the shape is not output to the control operation terminal of the rolling mill, thereby preventing deterioration of the shape. At this time, as for the neural network, learning is carried out assuming that the control operation for that shape pattern is incorrect.

[0025] Since the method of verifying whether the control operation is correct or not may itself be incorrect, even if the neural network's control operation output is judged to be incorrect with a certain probability, by outputting it to the control operation terminal of the rolling mill, it becomes possible to learn about combinations of unexpected shape patterns and control operations.

[0026] In addition, if the control rules are no longer optimal for the plant due to changes in the plant's environment or operating conditions over time, it is possible to avoid a deterioration in control performance by monitoring the computer load and successively optimizing the control rules at the appropriate time based on the plant's condition.

[0027] The suitability of the control rules for the plant can be evaluated based on the error between the value output by inputting the shapes contained in the performance data generated sequentially during plant operation into a neural network and the control output contained in the performance data. EXAMPLES

[0028] An outline of a plant control system according to an embodiment of the present invention is shown in Fig. 1. The plant control system in Fig. 1 is composed of a controlled plant 1, a control execution device 20 that inputs performance data Si from the controlled plant 1 and controls the controlled plant 1 by providing a control operation output SO determined according to a control rule (neural network) as exemplified in Fig. 6, a control method learning device 21 that inputs performance data Si from the controlled plant 1 and performs learning and reflects the learned control rule in the control rule in the control execution device 20, a control rule update determination device 22 that evaluates the compatibility of the performance data Si created successively during plant operation with the control rule and provides a control rule learning instruction to the control rule learning unit 11 at an appropriate timing, a plurality of databases DB (DB1 to DB5), and a database management table TB of the databases DB.

[0029] The control execution device 20 is composed of a control input data creation section 2, a control rule execution section 10, a control output calculation section 3, a control output suppression section 4, a control output determination section 5, and a control operation disturbance generation section 16 as main elements.

[0030] In the control execution device 20, first, input data S1 for the control rule execution unit 10 is created using the control input data creation unit 2 from the performance data Si of the rolling mill, which is the controlled plant 1. The control rule execution unit 10 creates a control operation terminal operation command S2 from the performance data Si of the controlled object, using a neural network (control rule) that expresses the relationship between the performance data Si of the controlled object and the control operation terminal operation command S2. The control output calculation unit 3 calculates a control operation amount S3 for the control operation terminal based on the control operation terminal operation command S2. As a result, the control operation amount S3 is created using the neural network according to the performance data Si of the controlled plant 1.

[0031] Furthermore, the control output determination unit 5 in the control execution device 20 uses the performance data Si from the controlled plant 1 and the control operation amount S3 from the control output calculation unit 3 to determine data S4 on whether or not to output the control operation amount to the control operation end. The control output suppression unit 4 determines whether or not to output the control operation amount S3 to the control operation end in accordance with the control operation amount output permission data S4, and outputs the control operation amount S3 that has been determined to be permitted as the control operation amount output SO to be given to the controlled plant 1. As a result, the control operation amount S3 that is determined to be abnormal is not output to the controlled plant 1. The control operation disturbance generation unit 16 generates a disturbance and gives it to the controlled plant 1 for the purpose of verifying the plant control system.

[0032] The control execution device 20 configured as above refers to a control rule evaluation value database DB1 and an output judgment database DB3 for executing its processing, as will be described further below. The control rule evaluation value database DB1 is connected to both a control rule execution unit 10 in the control execution device 20 and a control rule learning unit 11 in a control method learning device 21, which will be described later, so as to be accessible to each other. Control rules (neural networks) as learning results in the control rule learning unit 11 are stored in the control rule evaluation value database DB1, and the control rule execution unit 10 refers to the control rules stored in the control rule evaluation value database DB1. The output judgment database DB3 is connected to a control output judgment unit 5 in the control execution device 20 so as to be accessible to each other.

[0033] FIG. 2 shows a specific example of the configuration of the control rule execution unit 10 according to the embodiment of the present invention. The control rule execution unit 10 inputs the input data S1 created by the control input data creation unit 2, and gives the control operation terminal operation command S2 to the control output calculation unit 3. The control rule execution unit 10 includes a neural network 101, which basically determines the control operation terminal operation command S2 by the method of Patent Document 1 as illustrated in FIG. 6. In the present invention, the control rule execution unit 10 further includes a neural network selection unit 102, which refers to the control rules stored in the control rule evaluation value database DB1, and selects and executes the optimal control rule as the control rule in the neural network 101. In this way, the control rule execution unit 10 of FIG. 2 selects and uses a necessary neural network from a plurality of neural networks divided by operator teams or control purposes. The control rule evaluation value database DB1 preferably includes performance data (such as data from the operation team) Si from which a neural network and a pass / fail judgment criterion can be selected as data from the controlled plant 1. Since the neural network becomes a control rule when executed, the neural network and the control rule are not differentiated in this specification and are used synonymously.

[0034] Returning to FIG. 1, in the control method learning device 21, learning is performed on the neural network 101 used in the control execution device 20. When the control execution device 20 outputs a control operation amount output SO to the controlled plant 1, it takes time for the control effect to actually appear as a change in the performance data Si. For this reason, learning is performed using data that is delayed by that time. In FIG. 1, Z -1 represents an appropriate time delay function for each data.

[0035] The control method learning device 21 is configured with a control result pass / fail judgement section 6, a learning data creation section 7, a control rule learning section 11, and a pass / fail judgement database DB4 as main elements.

[0036] Of these, the control result pass / fail judgment unit 6 judges whether the performance data Si has changed for the better or for the worse, using the performance data Si and previous performance data value Si0 from the controlled plant 1, and the pass / fail judgment data S5 stored in the pass / fail judgment database DB4, and outputs the control result pass / fail data S6.

[0037] The learning data creation section 7 in the control method learning device 21 creates new teacher data S7a to be used for learning the neural network using data, such as the control operation terminal operation command S2, the control operation amount S3, and the control operation amount output propriety data S4, created by the control execution device 20, delayed by the same time, and the control result pass / fail data S6 from the control result pass / fail judging section 6, and provides this to the control rule learning section 11. Note that the teacher data S7a corresponds to the control operation terminal operation command S2 output by the control rule executing section 10, and it can be said that the learning data creation section 7 has obtained data, as the new teacher data S7a, by estimating the control operation terminal operation command S2 output by the control rule executing section 10 using the control result pass / fail data S6 provided by the control result pass / fail judging section 6.

[0038] The control rule update determination device 22 evaluates the suitability of the control rules to the plant from the performance data Si and the control rules, and learns the control rules from new teacher data S7a at a timing when the control processing computer 23 can process the control rules without placing a high load on the control processing computer 23, and updates the control rules. The control processing computer 23 is a computer that realizes the control execution device 20 and the control rule update determination device 22.

[0039] Returning to the explanation of Fig. 1, the control rule update determination device 22 executes updates of the neural network 101 used in the control execution device 20. The control rule update determination device 22 includes a control rule update processing management unit 24, a control rule conformance evaluation unit 25, a control rule update evaluation unit 26, and a control rule evaluation value database DB5.

[0040] The control rule conformance evaluation unit 25 obtains the neural network (control rule) No. S9 corresponding to the performance data Si, which is sequentially created during the operation of the plant, from the database management table TB, and selects the corresponding control rule S10 from the control rule evaluation value database DB1. The control rule conformance evaluation unit 25 calculates the difference (error) between the output when the shape of the performance data Si is input to the selected control rule and the control output included in the performance data Si, or an index based on the difference, as the conformance of the control rule to the plant, and stores it in the control rule evaluation value database DB5. The control rule conformance evaluation unit 25 outputs the No. of the evaluated neural network (control rule) and an instruction S12 to execute a control rule update evaluation to the control rule update evaluation unit 26.

[0041] After receiving an instruction to execute a control rule update evaluation from the control rule compatibility evaluation unit 25, the control rule update evaluation unit 26 acquires the control rule of the neural net (control rule) No. evaluated by the control rule compatibility evaluation unit 25 (S13), evaluates whether or not it needs to be updated, registers the evaluation result in the control rule evaluation value database DB5, and updates the update priority of the control rule registered in the control rule evaluation value database DB5 (S14). For each neural net No., the control rule is assigned a higher update priority in ascending order of the average of the latest predetermined number of control rule compatibility evaluation values.

[0042] The control rule update processing management unit 24 selects the control rule with the highest update priority from the control rule evaluation value database DB5 (S15), and issues a processing execution instruction S16 to the control rule learning unit 11 to instruct learning of the control rule, thereby causing the control rule to be updated. Since the execution of control rule learning places a large load on the computer, the control rule update processing management unit 24 monitors the resource usage status such as the CPU load and memory usage rate of the control processing computer 23 (S17), monitors whether or not the computer is in operation from the performance data Si, and outputs the processing execution instruction S16 when the control model to be updated is not used for rolling in the control plant and / or the processing of the control execution device 20 is not delayed even if the control rule learning processing is executed. Thereafter, the control rule update processing management unit 24 outputs a deletion instruction S18 to delete information on the control rule for which the processing execution instruction S16 was output from the control rule evaluation value database DB5.

[0043] 3 shows a specific example of the configuration of the control rule learning unit 11 according to the embodiment of the present invention. The control rule learning unit 11 is configured with 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 as main components. The control rule learning unit 11 also obtains data S8a, which is a time-delayed version of the input data S1 from the control input data creation unit 2 as an external input, and new teacher data S7a from the learning data creation unit 7, and also refers to data stored in the control rule evaluation value database DB1 and the learning data database DB2.

[0044] In the control rule learning section 11, the input data S1 is taken into the neural network processing section 110 via the input data creation section 114 after appropriate time delay compensation.

[0045] In the control rule learning unit 11, new teacher data S7a from the learning data creation unit 7 is provided to the neural network processing unit 110 as total teacher data S7c including past teacher data S7b stored in the learning data database DB2 in the teacher data creation unit 115. These teacher data S7a and S7b are appropriately stored in the learning data database DB2 and used.

[0046] Similarly, input data S8a from the control input data creation unit 2 is provided to the neural network processing unit 110 as total input data S8c including past input data S8b stored in the learning data database DB2 in the input data creation unit 114. These input data S8a and S8b are appropriately stored in the learning data database DB2 and used.

[0047] 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 input data S8c from an input data creation unit 114, teacher data S7c from a teacher data creation unit 115, and a control rule (neural network) selected by a neural network selection unit 113, and stores the finally determined neural network in a control rule evaluation value database DB1.

[0048] 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 timing to obtain input for the neural network 111, and also controls the processing results to be stored in the control rule evaluation value database DB1.

[0049] Here, the neural network 101 in the control execution device 20 in Fig. 2 and the neural network 111 in the control method learning device 21 in Fig. 3 are both neural networks of the same concept, but the basic conceptual differences in their use will be explained as follows. First, the neural network 101 in the control execution device 20 is a neural network with predetermined contents, which obtains a control operation terminal operation command S2 as a corresponding output when input data S1 is given, and is a neural network used for so-called one-way processing. In contrast, the neural network 111 in the control method learning device 21 is a neural network for obtaining, by learning, a neural network that satisfies the input / output relationship when input data S1 and input data S8c and teacher data S7c regarding the control operation terminal operation command S2 are set as learning data.

[0050] The basic processing concept of the control method learning device 21 configured as above is as follows: First, if the content of the control manipulated variable output feasibility data S4 is "possible", the control manipulated variable output SO is output to the controlled plant 1, and if the content of the control result pass / fail data S6 is "good" (the performance data Si changes in an improving direction), it is determined that the control manipulated variable operation command S2 output by the control rule execution unit 10 is correct, and learning data is created so that the output of the neural network becomes the control manipulated variable operation command S2.

[0051] On the other hand, if the content of the control operation amount output propriety data S4 is "prohibited", or if the control operation amount output SO is output to the controlled plant 1 and the content of the control result pass / fail data S6 is "prohibited" (the performance data Si changes in a worsening direction), it is determined that the control operation terminal operation command S2 output by the control rule execution unit 10 is incorrect, and learning data is created so that no neural network output is generated. At this time, the neural network output is configured so that two types of output, a + direction output and a - direction output, are generated for the same control operation terminal as the control output, and learning data is created so that the control operation terminal operation command S2 on the output side is not output.

[0052] In the control rule learning unit 11 illustrated in FIG. 3, the following processing is performed as a result of data processing by the neural network learning control unit 112. Here, learning data that is a combination of S8c, which is input data S1 to the control execution device 20 delayed in time, and teacher data S7c created by the teacher data creation unit 115, is used to learn the neural network 101 used in the control rule execution unit 10. In practice, the same neural network 111 as the neural network 101 of the control rule execution unit 10 is provided in the control rule learning unit 11, and operation tests are performed under various conditions to learn the responses at that time, and a control rule that has been confirmed to produce better results as a result of learning is obtained. Since learning needs to be performed using multiple learning data, multiple past learning data are taken out from the learning data database DB2 that accumulates learning data created in the past, and learning and processing are performed, and the current learning data is stored in the learning data database DB2. In addition, the learned neural network is stored in the control rule evaluation value database DB1 to be used by the control rule execution unit 10.

[0053] When training a neural network, new training data may be created each time new training data is generated, and the new training data may be used together with past training data. Alternatively, after a certain amount of training data (e.g., 100 pieces of data) has been accumulated, the new training data may be used together with past training data.

[0054] In addition, the control result pass / fail judgment unit 6 performs pass / fail judgment based on pass / fail judgment criteria from the pass / fail judgment database DB4. Since the judgment result of the pass / fail judgment of the control result differs depending on the control purpose, multiple neural networks corresponding to multiple control purposes are created, and even if the input data is the same, teacher data is created for each control purpose and learned. By creating multiple teacher data for one input data and using it to learn the neural network corresponding to each teacher data, it is possible to simultaneously learn neural networks corresponding to multiple control purposes. Here, multiple control purposes refer to, for example, in the case of shape control, which part in the strip width direction (strip end, center part, asymmetric part, etc.) is to be controlled with priority, which of multiple control target items (for example, strip thickness and tension, rolling load, etc.) is to be controlled with priority, etc.

[0055] In the above-mentioned configuration, once the neural network 101 used in the control rule execution unit 10 has learned, new control operations will no longer be executed. Therefore, the control operation disturbance generation unit 16 randomly generates new operation methods at appropriate times, and by executing the control operation in addition to the control operation amount S3, the new control method is learned.

[0056] Hereinafter, the details of this plant control method will be described with respect to shape control in a Sendzimir rolling mill as shown in Patent Document 1. Note that the shape control will be described assuming that the following specifications A and B are adopted.

[0057] Specification A is a specification for priority, and has priority information in the strip width direction. For example, in shape control, it is often difficult to control to the target value over the entire strip width direction due to machine characteristics. For this reason, the following two priority specifications A1 and A2 are provided in the strip width direction. Of these, priority specification A1 is "give priority to strip ends," and priority specification A2 is "give priority to the center," and control is implemented according to the two priorities A1 and A2. When control is implemented, either priority specification A1 or A2 is taken into consideration.

[0058] Specification B is a specification for dealing with conditions that are known in advance. For example, since the relationship between the shape pattern and the control method changes under various conditions, it may be necessary to classify them into categories, for example, specification B1 for plate width and specification B2 for steel type. By changing each of the above, the degree of influence on the shape of the shape control terminal changes.

[0059] In this example, the controlled plant 1 is a Sendzimir rolling mill, and the actual data is the shape actual data. The Sendzimir rolling mill is a rolling mill with cluster rolls for cold rolling hard materials such as stainless steel. In the Sendzimir rolling mill, small diameter work rolls are used to apply strong reduction to hard materials. For this reason, it is difficult to obtain flat steel sheets. To address this issue, cluster roll structures and various shape control devices are adopted. In general, the upper and lower first intermediate rolls have a single taper and can be shifted, and the Sendzimir rolling mill is equipped with six split rolls and two rolls called AS-U on the upper and lower sides. In the example described below, the detection data of the shape detector is used as the actual shape data Si, and the shape deviation, which is the difference from the target shape, is used as the input data S1. 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.

[0060] Fig. 4 shows the neural network configuration when used for shape control of a Sendzimir rolling mill. Here, the neural network refers to neural network 101 for the control rule execution unit 10, and neural network 111 for the control rule learning unit 11, but both have the same structure.

[0061] In the example of shape control of a Sendzimir rolling mill shown in Fig. 4, actual data Si from the controlled plant 1 is actual data of the Sendzimir rolling mill including data from a shape detector (here, a shape deviation, which is the difference between the actual shape and the target shape, is assumed to be output), and the control input data creation unit 2 obtains a standardized shape deviation 201 and a shape deviation stage 202 as input data S1. As a result, the input layer of the neural networks 101, 111 is composed of the standardized shape deviation 201 and the shape deviation stage 202. Note that, although the shape deviation stage 202 is input to the neural network input layer in Fig. 4, the neural network may be switched depending on the stage.

[0062] The output layer is composed of an AS-U operation degree 301 and a first intermediate operation degree 302 according to the AS-U and the first intermediate roll, which are shape control operation terminals of the Sendzimir rolling mill. The operation degrees for the AS-U include the AS-U opening direction (the direction in which the roll gap (the gap between the upper and lower work rolls of the rolling mill) opens) and the AS-U closing direction (the direction in which the roll gap closes) for each AS-U. For the first intermediate roll, the operation degrees include the first intermediate roll opening direction (the direction in which the first intermediate roll moves toward the outside from the center of the rolling mill) and the first intermediate roll closing direction (the direction in which the first intermediate roll moves toward the center of the rolling mill) for the upper and lower first intermediate rolls. For example, if the shape detector has 20 zones and the shape deviation stage 202 has three stages (large, medium, small), the input layer will have 23 inputs. In addition, if the AS-U has seven saddles and the upper and lower first intermediate rolls can be shifted in the strip width direction, the output layer will have 14 AS-U operation degrees 301 and 4 first intermediate operation degrees, totaling 18. The number of layers in the intermediate layer and the number of neurons in each layer are set appropriately. As will be described later with reference to Fig. 8, the neural network output is configured so that two types of output, a + direction and a - direction, are output for each shape control operation end of the Sendzimir rolling mill, which is the output layer.

[0063] Figure 10 shows shape deviation and control methods. The upper part of Fig. 10 shows the control method when the shape deviation is large, and the lower part of Fig. 10 shows the control method when the shape deviation is small. Note that the height direction is the magnitude of the shape deviation, the horizontal axis direction is the strip width direction, with both sides of the strip width representing the strip ends and the centre representing the strip centre. As shown in the upper part of Fig. 10, when the shape deviation is large, priority is given to correcting the overall shape over the local shape deviation in the strip width direction. On the other hand, as shown in the lower part of Fig. 10, when the shape deviation is small, priority is given to reducing the local shape deviation.

[0064] As described above, since it is necessary to change the control method according to the magnitude of the shape deviation, a shape deviation stage 202 is provided as shown in Fig. 4, and the shape deviation is provided to neural networks 101, 111 to determine the magnitude of the shape deviation. It is preferable to use a shape deviation that is normalized, for example, to 0 to 1, regardless of the magnitude of the shape deviation. This is just one example, and it is also possible to input the shape deviation directly to the input layer of the neural network without normalizing it, or to change the neural network itself according to the magnitude of the shape deviation (for example, to prepare two neural networks, and separate the neural network used when the shape deviation is large and the neural network used when it is small).

[0065] The neural networks 101 and 111 configured as shown in Fig. 4 described above are trained to learn the operation method for the shape pattern, and the trained neural network is used to perform shape control. Even neural networks with the same configuration can have different characteristics depending on the learning conditions, and can be made to output different control outputs for the same shape pattern.

[0066] Therefore, by using a plurality of neural networks according to other conditions of the shape record, it is possible to configure optimal control for various conditions. This corresponds to the specification B. The configuration of FIG. 2 described above shows a specific example of the case where such a specification is performed. In the configuration example of FIG. 2, the neural network 101 used in the control rule execution unit 10 is prepared separately according to the rolling record, the name of the rolling mill operator, the steel type of the material to be rolled, the plate width, etc., and registered in the control rule evaluation value database DB1. The neural network selection unit 102 selects a neural network that matches the condition at that time and sets it as the neural network 101 of the control rule execution unit 10. Note that, as the condition at that time in the neural network selection unit 102, it is preferable to take in the plate width data from the record data Si in the controlled plant 1 and select a neural network according to this. In addition, the plurality of neural networks used here may have different numbers of layers of the intermediate layer and the number of units in each layer as long as they have the input layer and the output layer as shown in FIG. 4.

[0067] 7 shows an overview of the control input data creation unit 2 that creates data S1 (normalized shape deviation 201, shape deviation stage 202) to be input to the input layers of the neural networks 101, 111. Here, shape detector data from a shape detector that detects the strip shape during rolling in a Sendzimir rolling mill, which is the controlled plant 1, is input as actual data Si. First, a shape deviation PP value calculation device 210 calculates a shape deviation PP value (Peak To Peak value) S, which is the difference between the maximum and minimum values ​​of the detection results of each shape detector zone. PP The shape deviation step calculation device 211 calculates the shape deviation PP value S PP The shape deviation is classified into three levels: large, medium, and small. The shape is the distribution of the elongation rate of the rolled material in the plate width direction, and I-UNIT, which expresses the elongation rate in 10-5 units, is used as the unit. For example, it is classified as follows.

[0068] Here, the shape deviation stages are classified as follows: (large=1, medium=0, small=0) when formula (1) is satisfied, (large=0, medium=1, small=0) when formula (2) is satisfied, and (large=0, medium=0, small=1) when formula (3) is satisfied. Note that the shape deviations of each zone are classified as follows: PM =S PP And, S PM Normalization is performed using

[0069]

number

[0070]

number

[0071]

number

[0072] In this manner, the normalized shape deviation 201 and the shape deviation stage 202 are generated as input data to the neural network 101. The normalized shape deviation 201 and the shape deviation stage 202 are input data S1 to the control rule executing section .

[0073] An overview of the control output calculation unit 3 is shown in Figure 8. The control output calculation unit 3 creates a control operation amount S3, which is an operation command to each shape control operation end, from a control operation end operation command S2 (in the case of shape control of a Sendzimir rolling mill, this corresponds to the AS-U operation degree 301 and the first intermediate operation degree 302) which is an output from the neural network 101 in the control rule execution unit 10. Note that here, one data example is shown for each of the AS-U operation degree 301 and the first intermediate operation degree 302, of which there are multiple values, and each data is composed of a pair of data of an opening direction degree and a closing direction degree.

[0074] In the control output calculation unit 3, the input AS-U operation degree 301 has outputs for the opening and closing directions of each AS-U, so the difference between them is the conversion gain G ASU By multiplying this by the conversion gain G, an operation command is output to each AS-U. ASU Since the control output to each AS-U is the AS-U position change amount (unit is length), this is the conversion gain from degree to position change amount.

[0075] Similarly, the first intermediate operation degree 302 inputted has a first intermediate outer and inner output, and the difference between them is the conversion gain G 1ST By multiplying this by the conversion gain G, an operation command for each first intermediate roll shift is output. 1ST Since the control output to each first intermediate roll is the first intermediate roll shift position change amount (unit is length), this is a conversion gain from degree to position change amount.

[0076] In this way, the control operation amount S3 can be calculated. The control operation amount S3 is composed of #1 to #n AS-U position change amounts (n is 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. Note that Fig. 8 shows a system for adding disturbance data from the control operation disturbance generating unit 16 to the control operation end operation command S2.

[0077] 9 shows an overview of the control output judgment unit 5. The control output judgment unit 5 is composed of a rolling phenomenon model 501 and a shape correction pass / fail judgment unit 502, and obtains the performance data Si from the controlled plant 1, the control operation amount S3 from the control output calculation unit 3, and information from the output judgment database DB3, and gives the control operation amount output availability data S4 to the control operation end. With this configuration, the control output judgment unit 5 predicts the change in shape when the control operation amount S3 calculated by the control output calculation unit 3 is output to the rolling mill, which is the controlled plant 1, by inputting it into a known model of the controlled plant 1 (rolling phenomenon model 501 in the case of the embodiment of FIG. 9), and when it is predicted that the shape will deteriorate, the control operation amount output SO is suppressed to prevent the shape from deteriorating significantly.

[0078] More specifically, the control operation amount S3 is input to a rolling phenomenon model 501 to predict a shape change due to the control operation amount S3, and shape deviation correction amount prediction data 503 is calculated. On the other hand, the shape deviation correction amount prediction data 503 is added to the shape detector data Si (current shape deviation actual data 504) from the controlled plant 1 to obtain shape deviation prediction data 505, and by evaluating the shape deviation prediction data 505, it is possible to predict how the shape will change when the control operation amount S3 is output to the controlled plant 1. From the current shape deviation actual data 504 and the shape deviation prediction data 505, the shape correction pass / fail judgment unit 502 judges whether the shape will change in a better or worse direction, and obtains control operation amount output pass / fail data S4.

[0079] Specifically, the shape modification pass / fail judgment unit 502 judges the pass / fail of shape modification as follows. First, as shown in the specifications A1 and A2 regarding the priority of shape control, in order to take into account the control priority in the strip width direction, a weighting coefficient w(i) in the strip width direction is set in the output judgment database DB3 for each of the specifications A1 and A2. Using this, the pass / fail of the shape change is judged using an evaluation function J such as the following equation (4). In equation (4), w(i) is a weighting coefficient, εfb(i) is the shape deviation actual data 504, εest(i) is the shape deviation predicted data 505, i is the shape detector zone, and rand is a random number term.

[0080]

number

[0081] When the evaluation function J in the formula (4) is used, the evaluation function J is positive when the shape improves, and negative when the shape deteriorates. Moreover, rand is a random term, and randomly changes the evaluation result of the evaluation function J. As a result, even if the shape deteriorates, the evaluation function J may become positive, so that it is possible to learn the relationship between the shape pattern and the control method even when the rolling phenomenon model 501 is incorrect. Here, rand is changed as appropriate so that the maximum value is increased when the model of the controlled plant 1 is uncertain, such as at the beginning of a trial run, and is set to 0 when it is desired to learn the control method to a certain extent and perform stable control.

[0082] The shape modification pass / fail judgment unit 502 calculates an evaluation function J, and outputs the control operation amount output feasibility data S4 as S4 = 1 (passive) when J ≥ 0, and as S4 = 0 (failed) when J < 0.

[0083] The control output suppression unit 4 determines whether or not to output the control operation amount output SO to the controlled plant 1 according to the control operation amount output feasibility data S4, which is the result of the judgment by the control output judgment unit 5. The control operation amount output feasibility data S4 is the #1 to #nAS-U position change amount output, the upper first intermediate shift position change amount output, and the lower first intermediate shift position change amount output, IF (control operation amount output availability data S4 = 0) THEN #1~#nAS-U position change output = 0 Upper 1st intermediate shift position change amount output = 0 Lower 1st intermediate shift position change output = 0 ELSE #1~#nAS-U position change amount output=#1~#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 is determined by.

[0084] The control execution device 20 executes the above calculations based on the performance data Si from the controlled plant 1 (rolling mill), and performs shape control by outputting the control operation amount output SO to the controlled plant 1 (rolling mill).

[0085] Next, an outline of the operation of the control method learning device 21 will be described. In the control method learning device 21, time-delayed data of the data used in the control execution device 20 is used. Time delay Z -1 e -TS and indicates a delay of a preset time T. The controlled plant 1 has a time response, so there is a time delay before the performance data changes due to the control operation output SO. Therefore, learning is performed using the performance data at the point in time when the delay time T has elapsed after the control operation is executed. In shape control, since it takes several seconds for the shape meter to detect a change in shape after an operation command is output to the AS-U or the first intermediate roll, it is best to set T to around 2 to 3 seconds (since the delay time changes depending on the type of shape detector and the rolling speed, it is best to set T to the optimal time until a change in the control operation terminal results in a change in shape).

[0086] 11 shows an outline of the operation of the control result pass / fail judging section 6. The shape change pass / fail judging section 602 uses a pass / fail judgment evaluation function Jc as shown in the following equation.

[0087]

number

[0088] In equation (5), εfb(i) is the shape deviation actual data included in the actual data Si, εlast(i) is the previous value of the shape deviation actual data, and wC(i) is the weighting coefficient in the plate width direction for pass / fail determination. Here, the weighting coefficient wC(i) for pass / fail determination is set according to the specifications A1 and A2 regarding the control priority from the pass / fail determination database DB4. The pass / fail of the control result is determined by the pass / fail evaluation function Jc. Also, even when the control operation amount output permission data S4, which is the determination result of the control output determination unit 5, is 0 (control output not allowed), although the control operation amount output to the actual control target plant 1 is 0, it is determined that the shape has deteriorated.

[0089] Here, when the control operation amount output permission data S4 = 0, the control result pass / fail data S6 is set to -1. Also, 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 evaluation function Jc is Jc > LCU, the control result pass / fail data S6 is set to -1 (the shape has deteriorated); if LCU ≥ Jc ≥ 0, the control result pass / fail data S6 is set to 0 (changing in the direction of shape deterioration); if 0 > Jc ≥ LCL, the control result pass / fail data S6 is set to 1 (changing in the direction of shape improvement); if Jc < LCL, the control result pass / fail data S6 is set to 0 (the shape has improved).

[0090] Here, when the control result pass / fail data S6 = -1, since the shape has deteriorated, when suppressing the output control output, when the control result pass / fail data S6 = 0, since there is no shape change or the shape has improved, when holding the output control output, when the control result pass / fail data S6 = 1, since the shape has changed in the direction of improvement but there is still a possibility of further improvement, it is the case of increasing the output control amount.

[0091] In this way, since the weighting coefficient wC(i) in the plate width direction changes according to the specifications A1 and A2 regarding the control priority, the pass / fail 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, in the control method learning device 21, the determination of the control result pass / fail data S6 is carried out for two types of specifications A1 and A2 regarding the control priority.

[0092] Next, an overview of the learning data creation unit 7 will be described. As shown in Fig. 1, the learning data creation unit 7 creates teacher data S7a for the neural network 111 used in the control rule learning unit 11 from the control operation terminal operation command S2, the control operation amount S3, and the judgment result of the control output suppression unit (control operation amount output propriety data S4) based on the judgment result from the control result propriety judgment unit 6 (control result propriety data S6),

[0093] The teacher data S7a in this case is the AS-U operation degree 301 and the first intermediate operation degree 302, which are outputs from the output layer of the neural network 111, as shown in Fig. 4. The learning data creation unit 7 creates teacher data S7a for the neural network 111 used in the control rule learning unit 11, using the control operation end operation command S2 (AS-U operation degree 301, first intermediate operation degree 302) which is the output of the neural network 101, and the #1 to #n AS-U position change amount output, the upper first intermediate shift position change amount output, and the lower first intermediate shift position change amount output, which are the control operation amount output SO.

[0094] In order to explain the operation overview of the learning data creation unit 7, the relationship between the data of each unit and symbols in the control output calculation unit 3 in FIG. 8 is summarized in FIG. 12. Here, the AS-U operation degree 301 is representatively shown for the control operation terminal operation command S2 which is the output of the neural network 101, and the operation degree positive side data is OPref, the operation degree negative side data is OMref, the operation degree randomly generated from the control operation disturbance generation unit 16 is operation degree random number Oref, the conversion gain is G, and the control operation amount output SO is Cref. Thus, for simplicity, the operation degree positive side and operation degree negative side, and the operation degree randomly generated from the control operation disturbance generation unit 16 are the operation degree random number as outputs from the output layer of the neural network 101 of the control rule execution unit 10. Also, the control operation amount output SO for the control operation terminal is the operation command value.

[0095] Fig. 13 shows the processing steps and processing contents in learning data creation unit 7. Here, to explain in accordance with the symbol conventions in Fig. 12, in the first processing step 71, the operation command value Cref is calculated by equation (6).

[0096]

number

[0097] In the next process step 72, the operation command value Cref is corrected to C'ref in accordance with the control result pass / fail data S6. Specifically, the corrected value C'ref of the operation command value Cref is determined according to equation (7) when the control result pass / fail data S6=-1, equation (8) when the control result pass / fail data S6=0, and equation (9) when the control result pass / fail data S6=1.

[0098]

number

[0099]

number

[0100]

number

[0101] In a processing step 73, the operation degree correction amount ΔOref is calculated from the corrected operation command value C′ref by using equations (10) and (11).

[0102]

number

[0103]

number

[0104] In the processing step 74, the teaching data OP'ref and OM'ref for the neural network 111 are obtained by the formula (12).

[0105]

number

[0106] In this manner, as shown in FIG. 12, the learning data creation unit 7 calculates the operation command value correction value C'ref of the operation command value Cref actually output to the controlled plant 1 according to the control result pass / fail data S6, which is the judgment result in the control result pass / fail judgment unit 6. Specifically, when the control result pass / fail data S6=1, the control direction is OK, but the control output is judged to be insufficient, and the operation command value is increased in the same direction by ΔCref. Conversely, when the control result pass / fail data S6=-1, the control direction is judged to be incorrect, and the operation command value is decreased in the opposite direction by ΔCref. Since the conversion gain G is previously set and therefore known, if the values ​​of the operation degree positive side and the operation degree negative side are known, it is possible to find the correction amount ΔOref. Here, ΔCref is set by finding 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 11 can be found by the above formula (12).

[0107] Note that Figure 13 uses a simple example for explanation, but in reality, all of the AS-U operation degrees 301 for #1 to #n AS-U and the first intermediate operation degree 302 for the first upper intermediate roll shift and the first lower intermediate roll shift are performed, and the data are used as teacher data for the neural network 111 used in the control rule learning unit 11 (AS-U operation degree teacher data, 1 intermediate operation degree teacher data).

[0108] 14 shows an example of data stored in the learning data database DB2. In order to train the neural network 111, a large number of combinations of input data S8a and teacher data S7a are required. Therefore, the teacher data S7a (AS-U operation degree teacher data, first intermediate operation degree) created by the learning data creation unit 7 is combined with the time-delay data S8a of the input data S1 (normalized shape deviation 201 and shape deviation stage) input to the control rule execution unit 10 by the control execution device 20, and is stored in the learning data database DB2 as a set of learning data.

[0109] In the plant control system of FIG. 1, various databases DB1, DB2, DB3, and DB4 are used. FIG. 14 shows the configuration of a database management table TB for managing and operating each database DB1, DB2, DB3, and DB4 in an interconnected manner. The database management table TB has a management table for specifications. Specifically, the database management table TB is divided according to (B1) plate width, (B2) steel type, and specifications A1 and A2 for control priority. For (B1) plate width, for example, four divisions are used, 3 feet width, meter width, 4 feet width, and 5 feet width, and for steel type, about 10 divisions are used, from steel type (1) to steel type (10). For specification A for control priority, two types are used, A1 and A2. In this case, there are 80 divisions, and 80 neural networks are used according to the rolling conditions.

[0110] The neural network learning control unit 112 associates the learning data, which is a combination of input data and teacher data as shown in FIG. 14, with the corresponding neural network No. in accordance with the database management table TB in FIG. 15, and stores it in a learning data database DB2 as shown in FIG. 16.

[0111] Each time the control execution device 20 executes shape control on the controlled plant 1, two sets of learning data are created. This is because two types of teacher data are created because the control result pass / fail judgment is performed using two evaluation criteria for the control priority, specifications A1 and A2, for the same input data and control output. When a certain amount of teacher data (e.g., 200 sets) has been accumulated, or when new teacher data has been accumulated in the learning data database DB2, the neural network learning control unit 112 instructs the neural network 111 to learn.

[0112] A plurality of neural nets are stored in the control rule database DB1 according to a database management table TB as shown in Fig. 15, and in the neural net learning control unit 112, the neural net number that needs to be learned is designated, and the neural net selection unit 113 retrieves the neural net from the control rule evaluation value database DB1 and sets it as the neural net 111. The neural net learning control unit 112 instructs the input data creation unit 114 and the teacher data creation unit 115 to retrieve input data and teacher data corresponding to the neural net from the learning data database DB2, and uses them to carry out learning of the neural net 111. Note that various methods have been proposed for learning neural nets, and any of these methods may be used.

[0113] When the learning of the neural net 111 is completed, the neural net learning control unit 112 writes the neural net 111, which is the learning result, back to the position of the neural net number in the control rule evaluation value database DB1, thereby completing the learning.

[0114] The learning may be performed simultaneously for all neural networks defined in FIG. 15 at regular intervals (e.g., once a day), or only the neural network with the neural net number for which a certain amount of new learning data (e.g., 100 sets) has been accumulated may be trained at that point.

[0115] FIG. 17 shows the configuration of the control rule conformity evaluation unit 25. The control rule conformity evaluation unit 25 inputs the performance data Si and the control result conformity data S6 from the control result conformity judgment unit 6 to the process execution judgment unit 254. When the control rule conformity evaluation unit 25 confirms that an effective operation that improves the shape has been performed, it acquires the control rule S10 used in the performance data Si from the control rule evaluation value database DB5 and the neural net No. S9 of this control rule S10 from the database management table TB. The process execution judgment unit 254 judges whether or not the process of the control output acquisition unit 251 needs to be executed from the performance data Si and the control result conformity data S6, and outputs a process execution instruction S2505 when the process of the control output acquisition unit 251 is to be executed. The control output acquisition unit 251 receives the control rule S10, the performance data Si, and the process execution instruction S2505, inputs the actual shape included in the performance data Si to the control rule S10, and obtains an output S2501 for the shape control device. The control output error calculation unit 252 calculates the difference between the output S2501 and the output of the shape control included in the performance data Si using equation (13).

[0116]

number

[0117] Here, the output S2501 of the control rule S10 is r, the shape control output included in the performance data Si is g, the total number of devices for shape control is N, and the output error between the control rule S10 and the performance data Si is SUV.

[0118] If the output error SUV is equal to or less than a certain threshold value TS, the control output error calculation unit 252 determines that the control rule S10 is suitable for the controlled plant and sets the degree of suitability DSU to 1, and if it exceeds the threshold value TS, it determines that the control rule is unsuitable for the controlled plant and sets the degree of suitability DSU to 0. The control output error calculation unit 252 registers the neural network number and the value of the degree of suitability DSU corresponding to this control rule S10 in the control rule evaluation value database DB5. The degree of suitability DSU is calculated using equation (14).

[0119]

number

[0120] The larger the threshold TS, the higher the proportion of fitness scores that are 1, and the smaller the threshold TS, the higher the proportion of fitness scores that are 0, so the control rule update decision results described below will vary greatly depending on the threshold TS. If you want to update the control rules in short intervals, you need to set the threshold TS to a small value, for example, and flexibly according to the operation of the plant.

[0121] Thereafter, the control output error calculation unit 252 outputs a processing execution instruction S2504 to the control rule update evaluation instruction unit 253. Upon receiving the processing execution instruction from the control output error calculation unit 252, the control rule update evaluation instruction unit 253 outputs a processing execution instruction S12 including the neural net number to the control rule update evaluation unit 26.

[0122] 18 shows the configuration of the control rule update evaluation unit 26. The control rule update evaluation unit 26 executes processing at the timing when it receives a process execution instruction S12 including a neural net number from the control rule compatibility evaluation unit 25. The control rule update necessity determination unit 261 acquires the compatibility DSU for a certain period (a period in which an operator's shift goes round, for example, one week, in order to suppress variation in operation for each operator) for the neural net number included in the process execution instruction S12 from the control rule evaluation value database DB5 (S13a), calculates the average value UEV, and registers it in the control rule evaluation value database DB5 (S14a). The calculation formula is shown in formula (15).

[0123]

number

[0124] Here, the amount of data for a certain period is assumed to be W.

[0125] If the average value UEV becomes equal to or less than the threshold value TU, the control rule update necessity flag RUF is set to 1, otherwise the control rule update necessity flag RUF is set to 0 and registered in the control rule evaluation value database DB5. However, if the suitability DSU has not been accumulated for a certain period of time, the calculation of the control rule update necessity flag RUF is not executed. The calculation formula is shown in formula (16).

[0126]

number

[0127] The calculation result of the control rule update necessity flag changes depending on the setting of the threshold TU. For this reason, it is necessary to flexibly set the threshold TU according to the operation of the plant, for example, to set the threshold TU to a small value if you want to reduce the update frequency of the control rules.

[0128] Thereafter, the control rule update necessity determination unit 261 outputs a processing execution instruction S2601 to the control rule update priority update unit 262. Upon receiving the processing execution instruction S2601, the control rule update priority update unit 262 reads (S13b) and updates (S14b) the control rule update priority for each neural net number registered in the control rule evaluation value database DB5. The control rule update priorities are assigned natural numbers such as 1, 2, 3, ... in ascending order of average UEV value registered in the control rule evaluation value database DB5.

[0129] 19 shows details of the control rule evaluation value database DB5. The control rule evaluation value database DB5 registers, for example in table format, a control rule update necessity flag RUF, a control rule update priority, an average value UEV, each degree of suitability DSU, and the dates and times when these calculations were performed, in association with each neural net number.

[0130] 20 shows an overview of the control rule update processing management unit 24. The control rule update processing management unit 24 acquires the neural net No. S15 corresponding to the control rule with the highest control rule update priority from the control rule evaluation value database DB5, and issues a processing execution instruction S16 to the control rule learning unit 11. However, the processing execution instruction S16 is limited to the case where the computer load of the control processing computer 23 is low from the computer load information S17 (e.g., CPU load, memory usage rate, etc.) of the control processing computer 23 that is responsible for the execution processing of the control rule learning unit 11, and / or the case where it is confirmed by referring to the performance data Si that the controlled plant is not in a situation where the control model to be updated is used for rolling. This is to prevent a delay in the processing of the control execution device 20 caused by an excessive load on the control processing computer 23, which causes a delay in the timing of control output to the controlled plant, and to prevent the control rule from being changed during operation. After the control rule update feasibility determination unit 241 transmits a processing execution instruction S16 to the control rule learning unit 11, the control rule update feasibility determination unit 241 outputs a processing execution instruction S2401 including the neural net No. S15 to the control rule update completion processing unit 242. Upon receiving the processing execution instruction S2401, the control rule update completion processing unit 242 deletes all information associated with the neural net No. S15 registered in the control rule evaluation value database DB5.

[0131] As a result, the shape of the rolling mill, which is the plant 1 to be controlled, is not significantly disturbed. 1) Instead of separately setting a reference shape pattern and a control operation for that pattern and learning the control operation method, a combination of a shape pattern and a control operation is learned and used to carry out the control operation. 2) Since new control rules cannot be predicted in advance, and a completely unpredictable control rule may turn out to be optimal, the optimal control rule is found by randomly operating the control terminals and observing the control results. Things can be made to happen.

[0132] The control rule evaluation value database DB1 stores the neural net used by the control execution device 20, but if the stored neural net has only been initialized with random numbers, it will take time for the neural net to learn and become capable of reasonable control. Therefore, when a control unit is constructed for the controlled plant 1, learning of the control rules is performed in advance by simulation based on the control model of the controlled plant 1 that is known at that time, and the neural net that has completed learning in the simulator is stored in the database, making it possible to control the controlled plant with a certain degree of performance from the beginning of its startup.

[0133] 21 is a diagram showing an outline of hardware of a computer 500 which realizes each system including the control execution device 20, the control method learning device 21, the control rule update determination device 22, the control processing computer 23, and a system which appropriately integrates these. In the computer 500, a processor 510 such as a CPU, a memory 520 such as a RAM (Random Access Memory), a storage 530 such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive), a network I / F (Inter / Face) 540, an input / output device 550 (e.g., a keyboard, a mouse, a touch panel, a display, etc.), and a peripheral device 560 are connected via a bus.

[0134] In computer 500, each system is realized by reading out each program for realizing each system from storage 530 and executing it by cooperation between processor 510 and memory 520. Alternatively, each program for realizing each system may be obtained from an external computer by communication via network I / F 540. Alternatively, each program may be obtained by being recorded on a non-transitory recording medium and being read out by a medium reading device.

[0135] When the apparatus of the present invention is applied to an actual plant, it is necessary to determine initial values ​​for the neural network. In this regard, it is advisable to create combinations of performance data and control operations by simulation using a control model of the plant to be controlled before implementing control in the plant to be controlled, thereby shortening the learning period for the combinations of performance data and control operations in the plant to be controlled.

[0136] The present invention is not limited to the above-described embodiment, and includes various modified examples. For example, the above-described embodiment has been described in detail to clearly explain the present invention, and is not necessarily limited to those having all the configurations described. In addition, it is also possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and to add the configuration of another embodiment to the configuration of one embodiment, so long as there is no contradiction. In addition, it is possible to add, delete, replace, integrate, or distribute a part of the configuration of each embodiment. In addition, the configurations and processes shown in the embodiments can be appropriately distributed, integrated, or replaced based on processing efficiency or implementation efficiency. [Explanation of symbols]

[0137] 1: Plant to be controlled, 20: Control execution device, 21: Control method learning device, 22: Control rule update decision device, 23: Control processing computer, 500: Computer

Claims

1. A plant control system for controlling a controlled plant, comprising: a control method learning device that learns a control rule based on a combination of performance data of the controlled plant and a control operation; a control execution device that controls the controlled plant based on the control rule learned by the control method learning device; a control rule update determination device that calculates a degree of conformity of the control rule with respect to the controlled plant based on the performance data when control of the controlled plant is performed based on the control rule, and updates the control rule based on the degree of conformity, The control rule update determination device includes: a control rule conformance evaluation unit that calculates the conformance at each predetermined timing; a control rule update evaluation unit that determines whether or not the control rule needs to be updated and an update priority based on the degree of conformance at each of the predetermined timings in a recent predetermined period; a control rule update processing management unit that outputs an update instruction for the control rule based on the necessity of the update and the update priority; A plant control system comprising:

2. 2. The plant control system according to claim 1, The control rule update evaluation unit The necessity of updating is determined based on a comparison result between the degree of conformity calculated at each of the predetermined timings during the most recent predetermined period and a predetermined reference value. A plant control system comprising:

3. 2. The plant control system according to claim 1, The control rule conformity evaluation unit includes: When an output error between a control output when control of the controlled plant is performed based on the control rule and a control output included in the performance data is equal to or smaller than a threshold value, the control rule is determined to be suitable for the controlled plant and the degree of suitability is set to 1, and when the output error is greater than the threshold value, the control rule is determined to be incompatible with the controlled plant and the degree of suitability is set to 0; The control rule update evaluation unit The necessity of updating and the priority of updating are determined based on an average of the degrees of compatibility calculated at the respective predetermined times during the most recent predetermined period. A plant control system comprising:

4. 2. The plant control system according to claim 1, The control rule update processing management unit The update instruction is output on the condition that the control rule to be updated is not being used by the control execution device in the controlled plant. A plant control system comprising:

5. 5. The plant control system according to claim 4, The update instruction is output on the condition that the control processing computer on which the control execution device and the control method learning device are operating is not in a predetermined high load state. A plant control system comprising:

6. 2. The plant control system according to claim 1, The control execution device is a control rule execution unit that gives a control output in accordance with a predetermined combination of performance data of a controlled plant and a control operation; a control output determination unit that determines whether or not the control output output by the control rule execution unit is acceptable and notifies the control method learning device that the performance data and the control operation are incorrect; a control output suppression unit that, when the control output determination unit determines that the performance data of the control target plant will deteriorate if the control output is output to the control target plant, prevents the control output from being output to the control target plant, The control method learning device includes: a control result quality judging section for judging whether the performance data has improved or deteriorated compared to before the control, after a time delay until the control effect appears in the performance data, when the control execution device actually outputs a control output to a controlled plant, a learning data creating section for obtaining teacher data by using the quality of the control result judged by the control result quality judging section and the control output, and a control rule learning section for learning the performance data and the teacher data as learning data, a control rule execution unit that executes control of the plant control method by learning the control method learning device, and that obtains a combination of actual data and control operations for a plurality of control targets in accordance with a state of the plant to be controlled, and that uses the obtained combination of actual data and control operations as a predetermined combination of actual data and control operations for the plant to be controlled in the control rule execution unit.

7. 7. A plant control system according to claim 6, comprising: A plant control system that learns and controls combinations of performance data and control operations by using information related to the size of the performance data and information that standardizes the performance data to make it easier to recognize patterns, in order to change the combination of performance data and control operations depending on the size of the performance data of the plant to be controlled.

8. The plant control system according to claim 6 or 7, a control rule execution unit that holds a predetermined combination of performance data and control operations of a controlled plant as a first neural net, and a control rule learning unit that holds the combination of performance data and control operations as a second neural net, and a second neural net obtained as a result of learning in the control method learning device is used as the first neural net in the control rule execution unit.

9. A plant control system according to any one of claims 6 to 8, A plant control system, characterized in that the control execution device includes a control operation disturbance generating unit that applies a disturbance to the control output, and the control method learning device learns even when a disturbance is applied.

10. The plant control system according to any one of claims 6 to 9, a control method learning device that obtains a plurality of combinations of performance data and control operations by learning under a plurality of predetermined specifications, and the control execution device that selects one of the plurality of combinations of performance data and control operations in accordance with an operating state of a plant to be controlled, and provides the control output.

11. 9. The plant control system according to claim 8, A plant control system characterized in that a neural network that learns the combination of performance data and operation methods to be used is changed according to the magnitude of the performance data.

12. A plant control system according to any one of claims 6 to 11, A plant control system characterized by changing criteria for judging whether a control result is good or bad based on the state of the controlled plant or the experience of the operator of the controlled plant, determining the relationship between performance data and operation methods for the controlled plant, and storing the relationship in a database, thereby controlling the controlled plant using different control methods depending on the state of the controlled plant or the experience of the operator of the controlled plant, etc.

13. A plant control system according to any one of claims 6 to 12, a control model of a plant to be controlled is used to create a combination of the performance data and the control operation through a simulation before control is performed in the plant to be controlled, thereby shortening a learning period for the combination of the performance data and the control operation in the plant to be controlled.

14. A rolling mill control device to which the plant control system according to any one of claims 6 to 13 is applied, 13. A rolling mill control device, comprising: a rolling mill control device; and a control target plant, the control target plant being a rolling mill; and a control result data representing a delivery shape of the rolling mill.

15. A plant control method executed by a plant control system that controls a controlled plant, comprising: a control method learning device for the plant control system learning a control rule based on a combination of performance data and a control operation of the controlled plant; a control execution device of the plant control system controls the controlled plant based on the control rule learned by the control method learning device; A control rule update determination device of the plant control system calculates a degree of conformity of the control rule with respect to the controlled plant based on the performance data when the controlled plant is controlled based on the control rule, and updates the control rule based on the degree of conformity. Including each process, In the process of updating the control rule, a control rule conformance evaluation unit of the control rule update determination device calculates the conformance at each predetermined timing, a control rule update evaluation unit of the control rule update determination device determines whether or not the control rule needs to be updated and an update priority based on the conformance at each of the predetermined timings in a most recent predetermined period; A control rule update processing management unit of the control rule update determination device outputs an update instruction for the control rule based on the necessity of the update and the update priority. A plant control method comprising the steps of:

16. A plant control program for causing a computer to function as the plant control system or the rolling mill control device according to any one of claims 1 to 14.

Images