Factory equipment control system, control method thereof, and computer readable recording medium

Abstract

The present invention provides a plant equipment control system, a control method thereof, and a computer-readable recording medium. In the plant equipment control system, control rules are effectively corrected while reducing the risk of interference with the control of the plant equipment. The system comprises: a control method learning unit that learns combinations of performance data and control operations of the target plant equipment; a control execution unit that executes control of the target plant equipment based on the combinations of performance data and control operations learned by the control method learning unit; and a state change rule learning unit that learns combinations of performance data, control operations, and state changes of the target plant equipment. Based on a determined combination of performance data, control operations, and state changes of the target plant equipment, the state change of the target plant equipment is predicted, thereby determining whether the control output is good or bad. The control rules are learned using the good or bad determination results, performance data, and supervisory data as learning data.

Description

Technical Field

[0001] The present invention relates to a factory equipment control system, a factory equipment control method, and a computer-readable recording medium. Background Art

[0002] Conventionally, in various types of plant equipment, plant equipment control based on various control theories has been performed in order to obtain appropriate control results through such control.

[0003] Taking an example of factory equipment as an example, in rolling mill control, fuzzy control and neuro-fuzzy control are applied as control theories for shape control of the fluctuating state of the control plate. Fuzzy control is suitable for shape control using coolants, while neuro-fuzzy control is suitable for shape control of the Sendzimir rolling mill. As shown in Patent Document 1, shape control using neuro-fuzzy control calculates the difference between the actual shape pattern detected by a shape detector and the target shape pattern, as well as the similarity ratio with a predetermined reference shape pattern. Based on the calculated similarity ratio, a control output for the control terminal is calculated according to a control rule represented by the control terminal operation variable for the predetermined reference shape pattern.

[0004] Hereinafter, a conventional technique for shape control of a Sendzimir mill using neuro-fuzzy control will be described.

[0005] In the shape control of Sendzimir rolling mill, neuro-fuzzy control is used. Figure 31 As shown, in the Sendzimir mill 50, a pattern recognition unit 51 performs shape pattern recognition based on the actual shape detected by a shape detector 52, calculating which of the preset reference shape patterns the actual shape most closely matches. The data on the actual shape detected by the shape detector 52 undergoes pattern recognition preprocessing in a shape detection preprocessing unit 54.

[0006] Then, the control calculation unit 53 executes control using a control rule composed of the control operation terminal operation amount for the preset shape pattern.

[0007] Here, if Figure 32 As shown, the pattern recognition unit 51 calculates which of the shapes in Modes 1 to 8 the difference (Δε) between the actual shape pattern (ε) detected by the shape detector 52 and the target shape (εref) is closest to. The control calculation unit 53 then selects one of the control methods in Modes 1 to 8 to execute based on the calculation result.

[0008] Prior art literature

[0009] Patent Literature

[0010] Patent Document 1: Japanese Patent No. 2804161

[0011] Patent Document 2: Japanese Patent Application Laid-Open No. 2018-180799 Summary of the Invention

[0012] Problems to be solved by the invention

[0013] In the prior art described in Patent Document 1, a representative shape is pre-set as a reference shape pattern, and control is performed based on a control rule that represents the relationship between the control operation amount and the reference waveform pattern. Learning of the control rule is based on the control operation amount relative to the reference waveform pattern, and the pre-set representative reference shape pattern is used directly.

[0014] Therefore, there is a problem that shape control can only reflect a specific shape mode.

[0015] The reference shape pattern is determined in advance by humans based on knowledge of the rolling mill in question, accumulated actual shapes, and experience gained through manual intervention. However, it is difficult to encompass all shapes that can be generated by the rolling mill and the rolled material. Therefore, when a shape differs from the reference shape pattern, shape control may not be executed, leaving shape deviations unchecked. Alternatively, a similar reference shape pattern may be mistakenly recognized, leading to incorrect control operations and ultimately worsening the shape.

[0016] Therefore, in conventional shape control, since control is performed by learning control rules using a preset reference shape pattern and control rules for the reference shape pattern, there is a problem that there is a limit to improving control accuracy.

[0017] To address this issue, for example, the technology described in Patent Document 2 has been proposed. Patent Document 2 describes a process that generates interference during control and gradually improves the intelligence of the neural network through learning. However, the process described in Patent Document 2, which generates control interference, disrupts the actual operation of the controlled plant equipment during operation, making it unsuitable for practical use. Furthermore, unless the controlled plant equipment is operated to a certain extent, the neural network will not be properly adjusted, and there is a high probability that it will not be able to perform appropriate control during the initial operation.

[0018] An object of the present invention is to provide a plant control system, a plant control method, and a program that can reduce the risk of disrupting control of plant equipment and efficiently correct control rules.

[0019] Means for solving problems

[0020] In order to solve the above-mentioned problems, for example, the structure described in the claims of the patent is adopted.

[0021] The present application includes a plurality of means for solving the above-mentioned problems. As one example, a plant control system is applied to a system that recognizes a pattern of combinations of performance data of controlled plant equipment and executes control on the controlled plant equipment.

[0022] In addition, the factory equipment control system includes: a control method learning unit, which learns the combination of performance data and control operations of the controlled object factory equipment; a control execution unit, which executes control of the controlled object factory equipment based on the combination of performance data and control operations learned by the control method learning unit; and a state change rule learning unit, which learns the combination of performance data, control operations and state changes of the controlled object factory equipment.

[0023] Here, the control execution unit has:

[0024] a control rule execution unit that provides a control output according to a combination of performance data of the controlled plant equipment and a determination of the control operation;

[0025] a control output quality determination execution unit that predicts a change in the state of the controlled object based on a combination of actual performance data of the controlled plant equipment, control operations, and a determination of a change in the state of the controlled object, and estimates the quality of the control output;

[0026] a new search operation amount calculation unit that calculates a new operation search operation amount based on the quality determination of the control output quality determination execution unit; and

[0027] A control output suppression unit uses the good / bad judgment of the control output good / bad judgment execution unit to prevent the control output from being output to the control target plant when it is determined that the performance data of the control target plant deteriorates when the control output is output to the control target plant.

[0028] In addition, the state change rule learning unit has:

[0029] The state change rule learning unit extracts a combination of performance data, control operations, and the amount of state change of the controlled object during the time delay from when the control effect of the control operation appears to when the performance data is obtained from the performance data of the controlled object plant equipment, generates learning data, and uses the learning data for learning.

[0030] In addition, the control method learning unit has:

[0031] A learning data generating unit obtains supervisory data using the control output quality determination of the control output quality determination executing unit and the control output; and a control rule learning unit performs learning using the actual performance data and the supervisory data as learning data.

[0032] Effects of the Invention

[0033] According to the present invention, the control rules for shape patterns and operating methods used in shape control can be optimized to minimize risks to factory equipment and automatically and efficiently adjust to environmental changes in factory equipment over time. Consequently, the present invention improves control accuracy, shortens the startup period of the control unit, and addresses changes over time.

[0034] Furthermore, according to the present invention, by preliminarily evaluating the performance of the control law, there are effects of reducing the risk to plant equipment caused by applying a new control law and improving the control performance by selecting an optimal control law.

[0035] Other problems, structures, and effects than those described above will become clear from the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a schematic configuration diagram showing a plant equipment control system according to an example of an embodiment of the present invention.

[0037] Figure 2 This is a diagram showing a specific configuration example of a control rule execution unit according to an example of an embodiment of the present invention.

[0038] Figure 3 This is a diagram showing a configuration of an example of a control output quality determination execution unit according to one embodiment of the present invention.

[0039] Figure 4 This is a diagram showing a specific configuration example of a control rule learning unit according to an example of an embodiment of the present invention.

[0040] Figure 5 This is a diagram showing a configuration example of a state change rule learning unit according to one embodiment of the present invention.

[0041] Figure 6 This is a diagram showing an example of determining whether a control result is good or bad with respect to a control method in shape control of a Sendzimir mill.

[0042] Figure 7 This is a diagram showing a configuration of an example of a quality determination error verification unit according to an embodiment of the present invention.

[0043] Figure 8This is a diagram showing a neural network configuration when used for shape control of a Sendzimir rolling mill according to an example of an embodiment of the present invention.

[0044] Figure 9 It is a diagram illustrating a shape deviation and a control method according to an example of an embodiment of the present invention.

[0045] Figure 10 This is a structural diagram showing an example of a control input data generating unit according to an example of an embodiment of the present invention.

[0046] Figure 11 This is a diagram showing a configuration example of a control output calculation unit according to an example of an embodiment of the present invention.

[0047] Figure 12 This is a diagram showing a neural network structure when used for predicting a state change of a Sendzimir rolling mill according to an example of an embodiment of the present invention.

[0048] Figure 13 This is a diagram showing a configuration of an example of a control result quality determination unit according to an example of an embodiment of the present invention.

[0049] Figure 14 This is a diagram showing an operation amount calculation method in a new search operation amount calculation unit according to an example of an embodiment of the present invention.

[0050] Figure 15 This is a diagram showing a configuration example of a control output determination unit according to an example of an embodiment of the present invention.

[0051] Figure 16 This is a diagram showing a configuration example of a control output calculation unit according to an example of an embodiment of the present invention.

[0052] Figure 17 This is a diagram showing the processing stages and processing contents in the learning data generating unit according to an example of an embodiment of the present invention.

[0053] Figure 18 This is a diagram showing an example of data stored in a learning data database according to an example of an embodiment of the present invention.

[0054] Figure 19 This is a diagram showing an example of a neural network management table according to an example of an embodiment of the present invention.

[0055] Figure 20 This is a structural diagram showing an example of a learning data database according to an embodiment of the present invention.

[0056] Figure 21This is a diagram showing an example of data stored in a learning data database according to an example of an embodiment of the present invention.

[0057] Figure 22 This is a diagram showing an example of a verification data database according to an example of an embodiment of the present invention.

[0058] Figure 23 This is a diagram showing an example of a neural network management table according to an example of an embodiment of the present invention.

[0059] Figure 24 This is a diagram showing an example of a learning data database according to an example of an embodiment of the present invention.

[0060] Figure 25 This is a diagram showing an example of a quality determination error database according to an example of an embodiment of the present invention.

[0061] Figure 26 This is a diagram showing an example of a quality evaluation value database according to an embodiment of the present invention.

[0062] Figure 27 This is a configuration diagram showing an example in which a plant equipment control system according to an example of an embodiment of the present invention includes a control rule evaluation unit.

[0063] Figure 28 This is a diagram showing an example of a control rule evaluation data database according to an example of an embodiment of the present invention.

[0064] Figure 29 This is a diagram showing an example of a control rule evaluation value database according to an example of an embodiment of the present invention.

[0065] Figure 30 This is a block diagram showing a hardware configuration example of a plant equipment control system according to an example of an embodiment of the present invention.

[0066] Figure 31 This is a structural diagram showing an example of a Sendzimir rolling mill.

[0067] Figure 32 This is a diagram showing an example of a list of control rules in shape control of a Sendzimir mill.

[0068] Description of Reference Numerals

[0069] 1…Controlled plant equipment, 2…Control input data generation unit, 3…Control output calculation unit, 4…Control output suppression unit, 5…Control output determination unit, 6…Control result good / bad judgment unit, 7…Learning data generation unit, 10…Control rule execution unit, 16…Control operation disturbance generation unit, 17…Control output good / bad judgment execution unit, 18…Control output operation method selection unit, 20…Control execution unit, 21…Control method learning unit, 22…Good / bad judgment rule learning unit, 23…Control rule evaluation unit, 31…State change rule learning unit, 33…New search operation amount calculation unit, 34…Good / bad judgment error verification unit, 35… 5…Control rule quality determination data collection unit, 36…Control rule evaluation data calculation unit, 37…Control rule database update unit, 50…Sendzimir rolling mill, 51…Pattern recognition unit, 52…Shape detector, 53…Control calculation unit, 54…Shape detection preprocessing unit, 101…Neural network, 102…Neural network selection unit, 110…Neural network processing unit, 111…Neural network, 112…Neural network learning control unit, 113…Neural network selection unit, 114…Input data generation unit, 115…Supervisory data generation unit, 171…Neural network, 172…Neural network 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 network processing unit, 311…Neural network, 312…Neural network learning control unit, 313…Neural network selection unit, 314…Input data generation unit, 315…Supervisory data generation unit, 341…Neural network, 342…Neural network selection unit, 343…Verification data generation unit, 344…State change good or bad conversion unit, 345…Good or bad judgment error calculation unit, 501…Rolling phenomenon model, 502…Shape correction good or bad judgment unit, 503…Shape deviation correction amount prediction data, 504…Shape deviation actual Performance data, 505…shape deviation prediction data, 602…shape change good or bad judgment unit, 703…control output operation method selection unit, 801…learning data generation unit, 802…control rule learning unit, DB1…control rule database, DB2…learning data database, DB3…output judgment database, DB4…good or bad judgment database, DB5…state change rule database, DB6…learning data database, DB7…good or bad judgment error database, DB8…verification data database, DB9…good or bad evaluation value database, DB10…control rule evaluation data database, DB11…control rule evaluation value database. DETAILED DESCRIPTION

[0070] Hereinafter, a plant equipment control system according to an example of an embodiment of the present invention (hereinafter referred to as “this example”) will be described with reference to the drawings.

[0071] First, before explaining this example, the process leading to the present invention and its outline will be explained by taking an example of a case where a plant equipment control system is applied to a shape control device of a rolling mill.

[0072] First, in order to obtain a plant equipment control system capable of efficiently correcting control rules while reducing risks to plant equipment, which is one of the objectives of the present invention, the following requirements (1), (2), (3), and (4) are necessary.

[0073] Requirement (1): In order to improve the control rules, if a control operation with good control results cannot be learned, the control operation is significantly changed, and if the control results are good, it is adopted as a new control operation method; if a control operation with good control results can be learned, the control operation is retained unchanged or slightly changed, and if the control results of the control operation are good, it is adopted as a new control operation method.

[0074] Requirement (2): Based on actual machine data, a combination of rolling performance data, control operations, and changes in shape patterns is learned. This allows a model to be constructed that can more accurately estimate the quality of control results compared to simulators using mechanical models. Regular automatic learning is then used to consistently construct a model that is appropriate for the latest factory equipment status.

[0075] Requirement (3): Using a model that estimates changes in shape based on control operations, the reliability of the control output suppression function for factory equipment, which was previously performed only in a simple mechanical model, is improved.

[0076] · Requirement (4): In the prior art, in the function of generating control rule learning data by determining whether a control result is good or not once, by using a model that estimates shape changes based on control operations, the influence of noise contained in factory equipment data can be suppressed, and even fine adjustments with small effects can be included as objects of learning data. At the same time, by preventing erroneous judgments of control effects, changes in learning data can be suppressed, and control performance can be stabilized.

[0077] To achieve these requirements (1) to (4), it is preferable to construct a neural network within the control device that can learn a combination of rolling performance data, control operations, and shape changes based on the control operations. Furthermore, the control device needs to estimate the quality of the control results resulting from the output of the control operations for the shape pattern generated by the rolling mill using the values obtained by inputting the output of the control operations into the neural network. Furthermore, the control device uses the estimated value of the quality of the control results to select a method for calculating the control operation amount used to search for a new control operation.

[0078] Using a simple model of the rolling mill, for example, verification is performed. If outputs are deemed to significantly deteriorate the shape, the control device prevents shape degradation by not outputting them to the mill's control terminals. In this case, by using an estimated value of the control result to determine whether the output is satisfactory or not, the control device can improve protection reliability and optimize the range of suppression, thereby expanding the scope of control functionality. Furthermore, even when a simple model of the rolling mill is unavailable, the control device can use an estimated value of the control result to determine whether the output is satisfactory or not, thereby expanding the scope of application of the control device.

[0079] In the early stages of application when the accuracy of estimating the quality of control results is insufficient, control operations estimated to be poor are also output to factory equipment. Therefore, the scope of learning needs to be expanded for the combination of rolling performance data, control operations, and shape changes based on control operations.

[0080] When the accuracy of estimation of the control result is sufficiently high, the control result can be estimated even without outputting an operation variable to the plant equipment. Therefore, the control device can generate learning data for the control law.

[0081] The control device uses a neural network capable of estimating shape changes based on control operations to estimate the quality of control results. This reduces the impact of noise on plant equipment data and enables quality determination of even small-effect fine-tuning data. This allows the control device to generate learning data. Furthermore, by preventing erroneous quality determinations caused by noise, the control device can improve the accuracy of the learning data.

[0082] In addition, when the accuracy of the estimated judgment of whether the control result is good or bad is reduced due to environmental changes in factory equipment caused by time-related changes, the control device can estimate the accuracy of the control result that is good or bad based on the latest factory equipment status by relearning using the actual performance data of the most recent factory equipment.

[0083] To confirm the estimated accuracy of control result quality determinations, test data is prepared separately from the data used for neural network learning and used for accuracy verification. Furthermore, the control device can verify the predicted error in quality determinations based on the quality prediction value, which uses the predicted value of shape change output by inputting the rolling performance data and control operations included in the accuracy verification test data, and the control result quality error contained in the test data.

[0084] Figure 1 The structure of the factory equipment control system of this example is shown.

[0085] Figure 1The plant equipment control system includes a control execution unit 20, a control method learning unit 21, a state change rule learning unit 22, a plurality of databases DB (DB1 to DB8), and a management table TB of each database DB.

[0086] The control execution unit 20 receives the performance data Si from the controlled plant 1 and provides the control operation variable output SO determined according to the control rule (neural network) to the controlled plant 1 to control the controlled plant 1. The controlled plant 1 is the already described Figure 31 The Sendzimir rolling mill 50 is shown.

[0087] Here, control rules refer to Figure 26 As described above, for example, the shape pattern A(ε) detected and the target shape (ε ref The control execution unit 20 selects and executes a control method of any one of the prepared multiple patterns based on the calculation result of the control rule.

[0088] The control method learning unit 21 inputs the control input data S1 generated by the control execution unit 20 and performs learning, and reflects the learned control rules in the control rules in the control execution unit 20 .

[0089] The state change rule learning unit 22 inputs the performance data Si before and after the control operation of the controlled plant 1 and performs learning, and reflects the learned state change rule on the state change rule in the control execution unit 20 .

[0090] The control execution unit 20 includes a control input data generation unit 2, a control rule execution unit 10, a control output calculation unit 3, a control output suppression unit 4, a control output determination unit 5, a control output good or bad determination execution unit 17, a new search operation amount calculation unit 33 and a control output operation method selection unit 18.

[0091] The control execution unit 20 generates input data S1 for the control rule execution unit 10 from the performance data Si of the rolling mill as the control target plant 1 using the control input data generation unit 2 .

[0092] The control rule execution unit 10 uses a neural network (control rule) that represents the relationship between the actual performance data Si of the controlled object and the control terminal operation command S2 to execute the control rule execution process, generating the control terminal operation command S2 from the input data S1. Based on the control terminal operation command S2, the control output calculation unit 3 calculates the control operation variable S3 for the control terminal. Thus, the control execution unit 20 generates the control operation variable S3 using the neural network based on the actual performance data Si of the controlled plant 1.

[0093] Furthermore, the control output quality determination execution unit 17 uses a neural network (state change rule) representing the relationship between the actual performance data Si of the controlled object, the control operation variable S3, and changes in the shape of the control operation to perform a control output quality determination execution process for generating a control output quality determination estimate value S9 based on the actual performance data Si of the controlled object and the control operation variable S3. Furthermore, the control output quality determination execution unit 17 generates a control result quality determination estimate value S11 based on the actual performance data Si of the controlled object and a selected control operation variable S8 described later.

[0094] The new search operation amount calculation unit 33 performs a new search operation amount calculation process of calculating a new search control operation amount S12 based on the control operation amount S3 and the control output quality determination estimated value S9.

[0095] The control output operation method selection unit 18 generates a selected control operation amount S8 and a control method selection flag S14 based on the control operation amount S3 and the new search control operation amount S12.

[0096] Furthermore, the control output determination unit 5 within the control execution unit 20 performs the following control output determination process: using the performance data Si from the controlled plant 1 and the selected controlled manipulated variable S8 from the control output operation method selection unit 18, it determines control manipulated variable output permission data S4 to the controlled manipulated terminal. Based on the control manipulated variable output permission data S4 and the control result quality determination estimate S11, the control output suppression unit 4 determines whether the selected controlled manipulated variable S8 can be output to the controlled manipulated terminal, and outputs the selected controlled manipulated variable S8 determined to be acceptable as the controlled manipulated variable output SO to the controlled plant 1. Consequently, the selected controlled manipulated variable S8 determined to be abnormal is not output from the control execution unit 20 to the controlled plant 1.

[0097] The control execution unit 20 configured as described above refers to the control rule database DB1 , the output determination database DB3 , the quality determination database DB4 , the state change rule database DB5 , and the quality determination error database DB7 in order to execute the processing.

[0098] The control rule database DB1 is 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 to be described later in a manner that allows access.

[0099] The control rule database DB1 stores control rules (neural networks) that are learning results of the control rule learning unit 802. The control rule execution unit 10 refers to the control rules stored in the control rule database DB1.

[0100] The learning data database DB2 stores the learning data obtained by the control rule learning unit 802 .

[0101] The output determination database DB3 is connected to the control output determination unit 5 in the control execution unit 20 so as to be accessible.

[0102] The quality determination database DB4 stores data for determining quality.

[0103] The state change rule database DB5 stores state change rules (neural networks) that are the learning results of the state change rule learning unit 31. This state change rule database DB5 is connected to any one of the control output quality determination execution unit 17 within the control execution unit 20, the state change rule learning unit 31 within the state change rule learning unit 22 (described later), and the state change rule quality determination error verification unit 34 in a manner that allows access. The control output quality determination execution unit 17 and the state change rule quality determination error verification unit 34 refer to the state change rules stored in the state change rule database DB5.

[0104] The learning data database DB6 stores the learning data learned by the state change rule learning unit 31 .

[0105] The good / bad judgment error database DB7 stores the good / bad judgment errors required for performing the good / bad judgment.

[0106] Figure 2 A specific configuration example of the control rule execution unit 10 of this example is shown.

[0107] The control rule execution unit 10 receives the input data S1 generated by the control input data generation unit 2. The control rule execution unit 10 processes the input data S1 and provides the control operation terminal operation instruction S2 to the control output calculation unit 3. The control rule execution unit 10 includes a neural network 101. The neural network 101 outputs the control operation terminal operation instruction S2 in accordance with the control output calculation unit 3. Figure 26 The shape control rule shown is the control operation end operation instruction S2.

[0108] 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 database DB1 , selects the optimal control rule as the control rule for the neural network 101 , and causes the neural network 101 to execute the optimal control rule.

[0109] In this way, the control rule execution unit 10 selects the required neural network from among the multiple neural networks divided according to operator groups and control objectives. The control rule database DB1 may also include data from the controlled plant equipment 1, including performance data (such as operator group data) Si that can be used to select the neural network and the criteria for determining whether it is good or bad.

[0110] Furthermore, since executing a neural network becomes a control law, in this specification, the terms neural network and control law are used synonymously.

[0111] Figure 3 The specific structure of the control output quality determination execution unit 17 is shown.

[0112] The control output quality determination execution unit 17 receives inputs of the actual performance data Si input from the controlled plant 1 and the control manipulated variable S3 generated by the control output calculation unit 3. Based on these input data, the control output quality determination execution unit 17 generates a control output quality determination estimated value S9 and supplies it to the new search manipulated variable calculation unit 33.

[0113] Furthermore, the control output quality determination execution unit 17 receives inputs of the actual performance data Si input from the controlled plant 1 and the selected control manipulated variable S8 generated by the control output manipulation method selection unit 18. Based on these input data, the control output quality determination execution unit 17 generates a control result quality determination estimate S11 and provides it to the control output suppression unit 4 and the learning data generation unit 801.

[0114] The control output quality determination execution unit 17 includes a neural network 171 and a neural network selection unit 172 .

[0115] The neural network 171 estimates a predicted value S20 of the shape change when the control operation amount S3 (control mode) is output for the performance data Si based on past control performance.

[0116] The neural network selection unit 172 selects the optimal state change rule as the state change rule in the neural network 171 by referring to the state change rules stored in the state change rule database DB5 .

[0117] In this way, the control output quality determination execution unit 17 selects a necessary neural network from a plurality of neural networks separated according to differences in the properties of the material to be controlled.

[0118] The state change rule database DB5 may include, as data from the controlled plant 1, actual performance data Si (such as steel type and plate width data) on the properties of the material to be controlled. Furthermore, since executing a neural network generates state change rules, the terms "neural network" and "state change rules" are used synonymously in this specification.

[0119] The control result quality determination unit 6 uses the performance data Si from the controlled plant 1, the shape change S20 estimated by the neural network, the quality determination data S5 stored in the quality determination database DB4, and the quality determination error data S21 stored in the quality determination error database DB7 to perform a control result quality determination process to determine whether the performance data Si is changing in an improving or deteriorating direction. The control result quality determination unit 6 then outputs control result quality data S9 or S11 indicating the determination result.

[0120] Figure 6 This is a diagram showing a specific example of determining whether a control result is good or bad in a control method for shape control in a Sendzimir mill. Figure 6 express Figure 26 The control result of each shape control rule is shown as good or bad.

[0121] return Figure 1 As described above, the control method learning unit 21 performs learning of the neural network 101 used in the control execution unit 20.

[0122] The control method learning unit 21 includes a learning data generating unit 801 and a control rule learning unit 802 .

[0123] The learning data generation unit 801 within the control method learning unit 21 performs the following learning data generation processing: It generates new supervisory data S7a for neural network learning using the control terminal operation command S2 generated by the control execution unit 20, the selected control operation variable S8, the control method selection flag S14, the control result good / bad judgment estimate S11 generated by the control output good / bad judgment execution unit 17, and the control operation variable output permission data S4 generated by the control output judgment unit 5. The learning data S7a generated by the learning data generation unit 801 is provided to the control rule learning unit 802.

[0124] The supervisory data S7 a corresponds to the control operation terminal operation instruction S2 output by the control rule execution unit 10 .

[0125] Figure 4 A specific configuration example of the control rule learning unit 802 is shown.

[0126] The control rule learning unit 802 includes an input data generating unit 114 , a supervisory data generating unit 115 , a neural network processing unit 110 , and a neural network selecting unit 113 .

[0127] Control rule learning unit 802 receives external inputs including input data S1 from control input data generator 2 and new supervisory data S7a from learning data generator 801. Control rule learning unit 802 also refers to data stored in control rule database DB1 and learning data database DB2.

[0128] In the control rule learning unit 802 , the input data S1 is input to the neural network processing unit 110 via the input data generating unit 114 .

[0129] Furthermore, in the control rule learning unit 802, the new supervisory data S7a from the learning data generation unit 801 is provided to the neural network processing unit 110 by the supervisory data generation unit 115 as the combined supervisory data S7c including the past supervisory data S7b stored in the learning data database DB2. These supervisory data S7a and S7b are appropriately stored in the learning data database DB2 for use.

[0130] Similarly, input data S8a from control input data generator 2 is provided to neural network processing unit 110 by input data generator 114 as total input data S8c including past input data S8b stored in learning data database DB2. These input data S8a and S8b are also appropriately stored in learning data database DB2 for use.

[0131] The neural network processing unit 110 is composed of a neural network 111 and a neural network learning control unit 112 .

[0132] The neural network 111 takes in the input data S8c from the input data generator 114, the supervisory data S7c from the supervisory data generator 115, and the control rule (neural network) selected by the neural network selector 113, and stores the finally determined neural network in the control rule database DB1.

[0133] The neural network learning control unit 112 controls the input data generation unit 114 , the supervisory data generation unit 115 , and the neural network selection unit 113 at appropriate timings to obtain input to the neural network 111 and store the processing results in the control rule database DB1 .

[0134] Here, Figure 2 The neural network 101 in the control rule execution unit 10 and Figure 4The neural networks 111 in the control method learning unit 21 are all neural networks of the same concept, but differ as described below.

[0135] That is, the neural network 101 in the control rule execution unit 10 is a neural network with predetermined contents, and is a neural network that obtains the control operation terminal operation instruction S2 as a corresponding output when input data S1 is provided.

[0136] On the other hand, the neural network 111 in the control method learning unit 21 is a neural network that finds the input-output relationship through learning when the input data S1 and the input data S8c and the supervision data S7c about the control operation terminal operation instruction S2 are set as learning data.

[0137] The basic processing in the control method learning unit 21 is considered as follows.

[0138] First, when the control manipulated variable output permission data S4 indicates "yes" and the control result good / bad judgment estimate S11 indicates "good" (the performance data Si changes toward a positive direction), the control execution unit 20 outputs the control manipulated variable output SO to the controlled plant 1. Here, the learning data generation unit 801 determines that the selected control manipulated variable S8 output by the control output operation method selection unit 18 is correct and generates learning data so that the output of the neural network corresponds to the selected control manipulated variable S8.

[0139] On the other hand, when the content of the control operation quantity output possibility data S4 is "No", or the content of the control result good or bad judgment estimated value S11 obtained by outputting the control operation quantity output SO to the controlled object factory equipment 1 is "No" (the actual performance data Si changes in the direction of deterioration), the learning data generation unit 801 determines that the selected control operation quantity S8 output by the control output operation method selection unit 18 is wrong.

[0140] In this case, learning data generation unit 801 checks whether control operation variable S3 is selected in control output operation method selection unit 18 based on control method selection flag S14. If control operation variable S3 is selected during this check, learning data generation unit 801 determines that control operation terminal operation instruction S2 output by control rule execution unit 10 is incorrect and generates learning data without outputting the neural network output. In this case, the neural network output is configured such that two outputs, one in the positive direction and one in the negative direction, are issued to the same control terminal as control outputs, and learning data is generated without outputting the output-side control operation terminal operation instruction S2.

[0141] In addition, as a result of data processing performed by the neural network learning control unit 112, Figure 4The control rule learning unit 802 shown performs processing as follows.

[0142] First, the control rule learning unit 802 performs learning of the neural network 101 used in the control rule execution unit 10 using learning data that is a combination of data S8c obtained from the input data S1 to the control execution unit 20 and supervisory data S7c generated by the supervisory data generation unit 115 .

[0143] In fact, the control rule learning unit 802 has a neural network 111 that is the same as the neural network 101 of the control rule execution unit 10. It performs application tests under various conditions to learn the response at that time and obtains control rules that confirm that better results are produced as a result of learning.

[0144] Since learning requires multiple learning data sets, multiple past learning data sets are retrieved from the learning data database DB2, which accumulates previously generated learning data, and then learned and processed. The current learning data is then stored in the learning data database DB2. Furthermore, the learned neural network is stored in the control rule database DB1 for use in the control rule execution unit 10.

[0145] The learning of the neural network can be performed by using the past learning data each time new learning data is generated, or by using the past learning data after the learning data accumulates to a certain level (for example, 100).

[0146] With such a configuration, the control output operation method selection unit 18 selects a new search operation amount, thereby outputting the new search operation amount to the target plant equipment, and generating learning data based on the control result, thereby enabling learning of a new control method.

[0147] return Figure 1 As explained above, the state change rule learning unit 22 executes the neural network 171 ( Figure 3 ) learning. In the controlled plant equipment 1, when the equipment position changes, it takes time for the actual control effect to appear as a change in the performance data Si. Therefore, learning is performed using data obtained by delaying the time. In addition, Figure 1 etc., recorded as "Z -1 The processing unit DL of " indicates that there is an appropriate time delay when transmitting each data.

[0148] The state change rule learning unit 22 includes a state change rule learning section 31 , a good / bad judgment error verification section 34 , and a good / bad judgment database DB4 .

[0149] Figure 5 The specific structure of the state change rule learning unit 31 is shown.

[0150] The state change rule learning unit 31 includes an input data generating unit 314 , a supervisory data generating unit 315 , a neural network processing unit 310 , a neural network selecting unit 313 , and a learning data generating unit 316 .

[0151] The state change rule learning unit 31 obtains the rolling performance data Si and the time-delayed rolling performance data Si-1 of the control target plant 1 as input from the outside.

[0152] Furthermore, the state change rule learning unit 31 is connected to the state change rule database DB5 , the learning data database DB6 , and the verification data database DB8 in an accessible manner.

[0153] The learning data generator 316 extracts rolling state variables and control operation variables from the time-delayed rolling performance data Si-1, and outputs these as input data S12a to the input data generator 314. Furthermore, the learning data generator 316 extracts shape deviations from the rolling performance data Si and from the time-delayed rolling performance data Si-1, calculates shape variation based on the difference, and outputs this to the supervisory data generator 315 as supervisory data S13a.

[0154] The supervised data S13a is provided to the neural network processing unit 310 in the supervised data generating unit 315 as the combined supervised data S13c including the past supervised data S13b stored in the learning data database DB6. These supervised data S13a and S13b are appropriately stored in the learning data database DB6 and used.

[0155] Similarly, input data S12a is provided to neural network processing unit 310 by input data generator 314 as total input data S12c including past input data S12b stored in learning data database DB6. These input data S12a and S12b are appropriately stored in learning data database DB6 and used.

[0156] At this time, the learning data generation unit 316 stores the supervisory data S13a and input data S12a generated at a constant ratio not in the learning data database DB6 but in the verification data database DB8. The verification data database DB8 also stores the pre-change shape deviation extracted from the time-delayed rolling performance data Si-1 and the neural network number set in the neural network 311.

[0157] The neural network processing unit 310 is composed of a neural network 311 and a neural network learning control unit 312 .

[0158] The neural network 311 takes in the input data S12 c from the input data generation unit 314 , the supervisory data S13 c from the supervisory data generation unit 315 , and the control rule (neural network) selected by the neural network selection unit 313 .

[0159] The neural network learning control unit 312 controls the input data generation unit 314, the supervisory data generation unit 315, and the neural network selection unit 313 at appropriate timings to obtain input to the neural network 311. Furthermore, the neural network learning control unit 312 stores the processing results in the state change rule database DB5 via the neural network selection unit 313.

[0160] Here, Figure 3 The neural network 171 of the control execution unit 20 and the neural network 341 of the good / bad judgment accuracy verification unit 34 described later are Figure 5 The neural networks 311 in the state change rule learning unit 22 shown are all neural networks of the same concept, but differ in the following aspects.

[0161] Neural network 171 in control execution unit 20 and neural network 341 in quality determination accuracy verification unit 34 are neural networks with predetermined contents. Specifically, neural network 171 and neural network 341 are neural networks that, when given rolling state variables Si and selected control variables S8 or S3, or verification input data S24, calculate predicted shape changes S20 and S25 as corresponding outputs. These neural networks are used in so-called unidirectional processing.

[0162] In contrast, the neural network 311 in the state change rule learning unit 22 is a neural network that obtains the input-output relationship through learning when the data S12c and the supervisory data S13c obtained by extracting the control operation amount from the time-delayed rolling performance data Si are set as learning data.

[0163] Figure 7 The specific structure of the quality determination error verification unit 34 is shown.

[0164] The good / bad judgment error verification unit 34 includes a verification data generation unit 343, a neural network 341, a neural network selection unit 342, a state change good / bad conversion unit 344, a good / bad evaluation value database DB9, and a good / bad judgment error calculation unit 345.

[0165] The verification data generation unit 343 sequentially reads out the verification data S22 corresponding to the state change rule (neural network No.) for which error verification is to be performed from the verification data database DB8, outputs the verification input data S24 to the neural network 341, and outputs the good or bad conversion verification data S23 to the state change good or bad conversion unit 344.

[0166] The neural network 341 receives the verification input data S24 from the verification data generation unit 343 and outputs the predicted shape change amount S25 predicted based on past control performance to the state change goodness / badness conversion unit 344 .

[0167] The neural network selection unit 342 refers to the state change rules stored in the state change rule database DB5 and selects a state change rule for error verification from a plurality of neural networks divided according to differences in the properties of the material to be controlled.

[0168] The state change goodness or badness prediction value conversion unit 344 receives the goodness or badness conversion verification data S23 from the verification data generation unit 343, and receives the predicted shape change amount S25 from the neural network 341, and calculates the verification data goodness or badness evaluation value and the predicted goodness or badness evaluation value based on them, and saves the goodness or badness evaluation value S26 in the goodness or badness evaluation value database DB9.

[0169] The good or bad judgment error calculation unit 345 reads the good or bad evaluation value data S27 from the good or bad evaluation value database DB9 in units of neural network No., calculates the good or bad judgment error of the verification data good or bad evaluation value and the predicted good or bad evaluation value, and writes it together with the verification result flag as the good or bad judgment error data S28 to the good or bad judgment error database DB7.

[0170] Next, a specific example of a plant equipment control method will be described with respect to shape control in a Sendzimir rolling mill. Shape control will be described using the following specifications A and B.

[0171] Specification A is a specification regarding priority, and contains information on priority in the board width direction.

[0172] For example, in shape control, it is often difficult to control the mechanical properties to the target value across the entire area in the plate width direction. Therefore, the operator assigns priorities in the plate width direction based on past experience and performs operations. Therefore, the following two priority specifications A1 and A2 are set in the plate width direction. Among them, the priority specification A1 is a specification that "gives priority to the plate ends." In addition, the priority specification A2 is a specification that "gives priority to the center." Which specification to use is determined based on the operator's experience and the rolling conditions such as speed conditions and acceleration and deceleration.

[0173] Control is performed according to the two priority levels of the specifications A1 and A2. That is, when the plant equipment control system performs control, either the specification A1 or A2 regarding the priority level is considered.

[0174] Specification B corresponds to pre-determined conditions. For example, the relationship between shape modes and control methods varies depending on the conditions. Therefore, for example, specification B1 may be defined as the plate width, while specification B2 may be defined as the steel type. As these specifications change, the degree of influence on the shape of the shape manipulation terminal also changes.

[0175] In this example, the control target plant 1 is a Sendzimir mill, and the performance data represents actual shapes. The Sendzimir mill is a rolling mill equipped with clustered rolls for cold rolling hard materials such as stainless steel. To achieve strong pressure contact on hard materials, the Sendzimir mill uses small-diameter work rolls. Therefore, control to achieve flat steel plates is difficult with the Sendzimir mill. To address this, the Sendzimir mill employs a clustered roll structure and various shape control units.

[0176] The Sendzimir mill typically features six upper and lower split rolls and two AS-U rolls, in addition to the upper and lower first intermediate rolls having a single taper for displacement. In the example described below, the actual shape data Si uses detection data from a shape detector, and the input data S1 uses the shape deviation, the difference from the target shape. Furthermore, the control operation variable S3 is the roll displacement amount for AS-Us #1 to #n and the upper and lower first intermediate rolls.

[0177] Figure 8 The following shows the structure of a neural network used for shape control of a Sendzimir mill. Here, the neural network 101 is shown and used in the control rule execution unit 10. Furthermore, the neural network 111 is shown and used in the control rule learning unit 802. Neural networks 101 and 111 have the same structure.

[0178] In the example of shape control of the Sendzimir mill, the performance data Si from the controlled plant 1 is the performance data of the Sendzimir mill including the data of the shape detector (here, the data of the shape deviation, which is the difference between the output performance shape and the target shape). The control input data generator 2 receives the normalized shape deviation 201 and the shape deviation stage 202 as input data S1. Thus, the input layer of the neural network 101, 111 is composed of the normalized shape deviation 201 and the shape deviation stage 202. In addition, Figure 8 In the example, the shape deviation stage 202 is used as input to the neural network input layer, but the neural network can also be switched according to the stage.

[0179] The output layers of neural networks 101 and 111 correspond to the AS-U and first intermediate roll, which are the shape control operating terminals of the Sendzimir mill, and are composed of an AS-U operating level 301 and a first intermediate operating level 302. Regarding the respective operating levels, each AS-U has an AS-U opening direction (the direction in which the roll gap (the gap between the upper and lower working rolls of the rolling mill) opens) and an AS-U closing direction (the direction in which the roll gap closes).

[0180] Regarding the first intermediate rolls, the upper and lower first intermediate rolls have a first intermediate roll opening direction (the direction in which the first intermediate rolls move from the center of the rolling mill toward the outside) and a first intermediate roll closing direction (the direction in which the first intermediate rolls move toward the center of the rolling mill).

[0181] For example, if the shape detector has 20 regions and the shape deviation stage 202 has three stages (large, medium, and small), the input layer has 23 inputs. Furthermore, if the AS-U has seven saddles and the upper and lower first intermediate rollers are displaceable in the sheet width direction, the output layer has 18 inputs, with 14 AS-U operation levels 301 and 4 intermediate operation levels. The number of intermediate layers and the number of neurons in each layer should be appropriately determined.

[0182] Furthermore, the shape control operation terminals of the Sendzimir mill serving as the output layer constitute a neural network output in such a manner that two outputs, a + direction and a - direction, are output to each control operation terminal.

[0183] Figure 9 Represents the shape deviation and control method in this example.

[0184] Figure 9 (a) shows the control method when the shape deviation is large. Figure 9 (b) shows the control method when the shape deviation is small. Figure 9 The height direction (vertical axis direction) of (a) and (b) represents the size of the shape deviation, the horizontal axis direction represents the plate width direction, the two sides of the plate width represent the plate ends, and the center represents the plate center.

[0185] like Figure 9 As shown in (a), when the shape deviation is large, priority is given to correcting the overall shape rather than local shape deviation in the plate width direction.

[0186] On the other hand, Figure 9 As shown in (b), when the shape deviation is small, priority is given to reducing the local shape deviation.

[0187] Thus, since the control method needs to be changed according to the size of the shape deviation, Figure 8As shown, a shape deviation stage 202 is provided and provided to the neural networks 101 and 111 to determine the magnitude of the shape deviation. The shape deviation is independent of its magnitude; for example, a normalized shape deviation between 0 and 1 can be used. This is merely an example; the shape deviation can be directly input to the input layer of the neural network without normalization, or the neural network itself can be modified based on the magnitude of the shape deviation. For example, two neural networks can be prepared, one for use when the shape deviation is large and the other for use when the shape deviation is small.

[0188] In this example of factory equipment control, the Figure 8 Neural networks 101 and 111 with such structures learn how to operate the shape pattern and use the learned neural network to perform shape control. Even neural networks with the same structure can have different characteristics depending on the learning conditions and can produce different control outputs for the same shape pattern.

[0189] Therefore, by using multiple neural networks according to the actual shape and other conditions, it is possible to optimally control various conditions. This corresponds to specification B. Figure 2 The structure of represents a specific example of the case where such specifications are implemented.

[0190] That is, in Figure 2 In the example configuration, different neural networks 101 are prepared for use in the control rule execution unit 10 based on rolling performance, the name of the rolling mill operator, the type of steel being rolled, the plate width, and the like, and these neural networks are pre-registered in the control rule database DB1. A neural network selection unit 102 selects a neural network that meets the conditions at that time and sets it as the neural network 101 of the control rule execution unit 10.

[0191] Furthermore, as the condition at this time point in the neural network selection unit 102, the data of the plate width can be taken from the actual performance data Si in the control target plant 1, and the neural network can be selected based on this data. Figure 8 The input layer and output layer shown in FIG, the number of intermediate layers and the number of units in each layer can also be different.

[0192] Figure 10 The structure of the control input data generating unit 2 for generating data S1 (normalized shape deviation 201 , shape deviation stage 202 ) to be input to the input layer of the neural network 101 , 111 is shown.

[0193] The control input data generator 2 receives as input the shape detector data from a shape detector that detects the shape of a plate during rolling in the Sendzimir mill, the controlled plant 1, as performance data Si. The control input data generator 2 then uses a shape deviation PP value calculator 210 to calculate a shape deviation PP value (Peak to Peak value) SPP, which is the difference between the maximum and minimum values of the detection results for each shape detector region.

[0194] The shape deviation stage calculation unit 211 classifies shape deviation into three stages: large, medium, and small, based on the shape deviation PP value SPP. Shape refers to the distribution of elongation in the width direction of the rolled material, using the I-unit, which represents elongation in 10-5 units. For example, the classification is performed as shown in the following mathematical formula.

[0195] Here, the classification is as follows: if [Formula 1] holds, the shape deviation stage is (large = 1, medium = 0, small = 0), if [Formula 2] holds, the shape deviation stage is (large = 0, medium = 1, small = 0), and if [Formula 3] holds, the shape deviation stage is (large = 0, medium = 0, small = 1). The shape deviation of each region is normalized using SPM, where SPM = SPP.

[0196] [Formula 1]

[0197] S PP ≥50 I-UNIT

[0198] [Formula 2]

[0199] 50I-UNIT>S PP ≥10I-UNIT

[0200] [Formula 3]

[0201] 10I-UNIT>S PP

[0202] As described above, the control input data generator 2 generates the normalized shape deviation 201 and the shape deviation level 202 as input data to the neural network 101. The normalized shape deviation 201 and the shape deviation level 202 are input data S1 to the control rule execution unit 10.

[0203] Figure 11 The configuration of the control output calculation unit 3 is shown.

[0204] The control output calculation unit 3 generates an operation instruction, or control operation amount S3, for each shape control operation terminal based on the control operation terminal operation instruction S2, which is the output from the neural network 101 within the control rule execution unit 10. In the case of shape control of the Sendzimir mill, the control operation terminal operation instruction S2 corresponds to the AS-U operation level 301 and the first intermediate operation level 302.

[0205] exist Figure 11 In FIG. 1 , one data example is shown for each of the AS-U operation levels 301 and the first intermediate operation level 302 , which exist in a plurality of numbers. Each data example is composed of a pair of data of an opening direction level and a closing direction level.

[0206] In the control output calculation unit 3, the input AS-U operation level 301 has outputs of each AS-U in the open direction and the closed direction, so the difference between them is calculated by the subtractor 303. Then, the output of the subtractor 303 is multiplied by the conversion gain G by the multiplier 304. ASU Multiplying them, the operation command to each AS-U is generated and output. Since the control output to each AS-U is the AS-U position change (unit is length), the conversion gain G ASU It is the conversion gain from degree to position change.

[0207] Similarly, the first intermediate operation level 302 inputted has outputs outside and inside the first intermediate level, and thus the difference between them is calculated by the subtractor 305. Then, the output of the subtractor 305 is multiplied by the conversion gain G by the multiplier 306. 1ST Multiplying by , the operation command for the shift of each first intermediate roller is generated and output. Since the control output to each first intermediate roller is the change amount of the first intermediate roller shift position (unit is length), the conversion gain G 1ST It is the conversion gain from degree to position change.

[0208] Based on the above, the control output calculation unit 3 can calculate the control operation variable S3. The control operation variable S3 is composed of the #1 to #n AS-U position change amounts (n is based on the number of saddles of the AS-U roller), the upper first intermediate shift position change amount, and the lower first intermediate shift position change amount.

[0209] Figure 12 The following shows the structure of a neural network used in the control output quality determination execution unit 17, the state change rule learning unit 31, and the quality determination error verification unit 34 for predicting state changes in the Sendzimir rolling mill. The neural network used in the control output quality determination execution unit 17 is neural network 171, in the state change rule learning unit 31 is neural network 311, and in the quality determination error verification unit 34 is neural network 341, but the structures are the same in all cases.

[0210] Based on the actual control performance data Si of the controlled plant 1, the control operation variables are extracted from rolling state variables such as the control device position and the control operation variable S3, or the selected control operation variable S8 described later, or data obtained by time-delaying the rolling performance data Si, and used as input to the input layer. The control operation variable or the selected control operation variable S8 extracted from the control operation variable S3 or the rolling performance data Si is composed of the operation variables of each control operation device. The rolling state variables can use state variables that have a significant impact on the predicted state changes after the control operation, such as the rolling speed and the position of each control device.

[0211] The output layer also outputs the shape change amount S20 or shape change supervisory data S13c predicted when outputting the control operation to the controlled plant equipment 1. The number of layers in the intermediate layer and the number of neurons in each layer are set appropriately.

[0212] In this example of factory equipment control, the Figure 12 Neural networks 171, 311, and 341 with such structures learn shape changes in response to changes in the control device's position, and use the learned neural networks to predict shape changes. Even neural networks with the same structure can have different characteristics depending on the performance data used for learning, allowing them to generate different shape changes in response to the same change in the control device's position.

[0213] Therefore, by using a plurality of neural networks according to other conditions of the rolling performance data, it is possible to perform optimal shape change prediction for a variety of conditions. This corresponds to specification B. Figure 3 The structure of represents a specific example of the case where such specifications are implemented.

[0214] That is, in Figure 3 In the configuration example, different neural networks 171 are prepared for use in the control output quality determination execution unit 17 depending on the type of steel, plate width, and other factors of the rolled material, and these neural networks are pre-registered in the state change rule database DB5. A neural network selection unit 172 selects a neural network that meets the conditions at that point in time and sets it as the neural network 171 of the control output quality determination execution unit 17.

[0215] Furthermore, as the condition at this time point in the neural network selection unit 172, the data of the plate width can be taken from the actual performance data Si in the control target plant 1, and the neural network can be selected based on this data. Figure 12 For input layers and output layers as shown, the number of intermediate layers and the number of units in each layer can also be different.

[0216] Figure 134 shows the operation of the control result quality determination unit 6. The shape change quality determination unit 602 uses the quality determination evaluation function Jc shown in [Formula 4].

[0217] [Formula 4]

[0218] ε pred (i) = ε ∫b (i)+ε chg (i)

[0219]

[0220] In [Equation 4], εchg(i) is the predicted shape change S21 output by the neural network 171, εfb(i) is the shape deviation performance data included in the performance data Si, εpred(i) is the predicted shape deviation after the control operation, and wC(i) is the weight coefficient in the plate width direction used for quality judgment. The weight coefficient wC(i) for quality judgment is set from the quality judgment database DB4 based on the specifications A1 and A2 related to the control priority. The quality of the control result is determined based on the quality judgment evaluation function Jc.

[0221] The upper threshold LCU and the threshold plus / minus LCL are pre-set based on the threshold condition (LCU ≥ 0 ≥ LCL). At this time, if the result of comparison with the quality evaluation function Jc is Jc > LCU, the quality evaluation estimated value S9 (S11) = 0 (shape deteriorates). If Jc < LCL, the quality evaluation estimated value S9 (S11) = 1 (shape improves). If LCU ≥ 0 ≥ LCL, the quality evaluation estimated value S9 (S11) = -1 (excluding the quality evaluation target).

[0222] To determine the upper threshold LCU and the threshold plus / minus LCL, the good / bad judgment error data S28 corresponding to the neural network number used in the control rule execution unit is read from the good / bad judgment error database. The good / bad judgment standard error εn and the verification result flag fn for neural network n are referenced. The verification result flag fn indicates whether verification has been performed with a sufficient number of data. If verification has not reached a sufficient number of data, the reliability of the good / bad judgment value is low, so it is best not to use it. The upper threshold LCU and the threshold plus / minus LCL are set to sufficiently large values, and in all cases, the good / bad judgment estimated value S9 (S11) is determined to be -1 (except for those subject to good / bad judgment). If verification has reached a sufficient number of data, the upper threshold LCU and the threshold plus / minus LCL are set based on the good / bad judgment standard error, allowing for accurate threshold setting.

[0223] IF flagn=0,THEN LCU=-LCL=th big

[0224] IF flagn=1,THEN LCU=-LCL=εn

[0225] In this threshold setting, th big The absolute values of the upper threshold value LCU and the plus / minus threshold value LCL are set as the standard error, but they can be changed to 2 times or 0.5 times the standard error, etc., depending on the situation, to adjust the reliability of the control output.

[0226] Thus, the weight coefficient wC(i) in the plate width direction changes depending on the control priority specifications A1 and A2, and thus differs from the quality evaluation function Jc. Therefore, it is considered that the judgment result of the quality estimation value S9 (S11) also differs. Therefore, the quality judgment rule learning unit 22 performs judgment on the quality estimation value S9 (S11) for the two control priority specifications A1 and A2.

[0227] Figure 14 The operation amount calculation method in the new search operation amount calculation unit 33 is shown.

[0228] The new search operation amount calculation unit 33 calculates a new search control operation amount S12 according to the following policy using the control output quality determination estimated value S9 output by the control output quality determination execution unit 17 .

[0229] That is, when the control output quality determination estimation value S9 = 1, the quality determination of the control operation is estimated to be good, and therefore the new search operation amount calculation unit 33 performs a fine adjustment as the new search operation amount.

[0230] When the control output quality determination estimated value S9 = 0, the quality determination of the control operation is estimated to be poor, so the new search operation amount calculation unit 33 searches for a new appropriate operation method by significantly changing the control operation.

[0231] When the control output quality determination estimated value S9 = -1, the control output is not subject to quality determination, so the control change operation is not performed.

[0232] Based on the above policy, the formula for calculating the new search operation amount Crand is set as follows.

[0233] IF(S9=1)THEN Crand=Cref*(1+β*th1)

[0234] IF(S9=0)THEN Crand=Cref+γ*th2*G

[0235] IF(S9=-1) THEN Crand=Cref

[0236] Here, β and γ represent random values generated between -1 and 1. th1 represents the degree of fine adjustment. For example, when the range of ±10% of the original instruction is set as fine adjustment, th1 is set to 0.1.

[0237] th2 is a setting that significantly changes the operation method. For example, if th2 is set to 0.1, a 10% offset is added to the original instruction, which may cause the operation polarity to change and output instructions to devices that were not originally operated.

[0238] The values of β and γ are different for each operating device, and the operation amount of each device is changed independently. G represents the maximum operating position control command for each control operating device. By multiplying it with the command % above, the % value is converted into the operating position control command.

[0239] The control output operation method selection unit 18 selects either the control operation variable S3 or the new search control operation variable S12 and outputs it as the selected control operation variable S8. The probability Prand of using the new search control operation variable S12 can be set by the user from 0 to 1 to probabilistically determine whether the control operation variable S3 or the new search control operation variable S12 is selected. The value δ, which randomly takes a value between 0 and 1, is determined by the following equation.

[0240] IF(δ>Prand)THEN C″ref=Cref, α=1

[0241] ELSE C″ref=Crand、α=0

[0242] Here, C″ref represents the selected control operation amount S8 output by the control output operation method selection unit 18 to the subsequent operation unit. δ uses a common value in the calculation of the operation amounts of all devices, and the operation amount on the same side is used in all devices. α is the control method selection flag S14, which takes 1 when the control operation amount S3 is selected and takes 0 when the new search control operation amount S12 is selected. The control method selection flag S14 is output to the subsequent operation unit together with the selected control operation amount S8. As a setting method of Prad, it is set to 0 through random operation in the control in the actual machine when no risk is wanted to be brought to the factory equipment. In order to improve the control rules, a ratio other than 0 is set when the new search operation amount is wanted to be output.

[0243] Figure 15 The configuration of the control output determination unit 5 is shown.

[0244] The control output determination unit 5 is composed of a rolling phenomenon model 501 and a shape correction adequacy determination unit 502. The control output determination unit 5 receives performance data Si from the controlled plant 1, control operation variables S3 from the control output calculation unit 3, and information from the output determination database DB3, and provides control operation variable output permission data S4 to the control operation terminal.

[0245] The control output determination unit 5 with such a configuration inputs the selected control manipulated variable S8 calculated by the control output operation method selection unit 18 to a known model of the control target plant 1, thereby predicting the change in shape when output to the rolling mill, which is the control target plant 1. The known model of the control target plant 1 is here the rolling phenomenon model 501. In this prediction, if shape deterioration is predicted, the control output determination unit 5 suppresses the control manipulated variable output SO to prevent significant shape deterioration.

[0246] More specifically, the control output determination unit 5 inputs the selected control operation amount S8 into the rolling phenomenon model 501 , predicts the shape change caused by the selected control operation amount S8 , and calculates shape deviation correction amount prediction data 503 .

[0247] Meanwhile, the control output determination unit 5 adds the shape deviation correction amount prediction data 503 to the shape detector data Si from the controlled plant 1 to obtain shape deviation prediction data 505, and then evaluates the shape deviation prediction data 505. This allows the control output determination unit 5 to predict how the shape will change when the control manipulated variable S3 is output to the controlled plant 1. The shape detector data Si here refers to the current point in time's actual shape deviation performance data 504.

[0248] In the control output determination unit 5, the shape correction good or bad determination unit 502 determines whether the shape is changing towards improvement or deterioration based on the current shape deviation performance data 504 and the shape deviation prediction data 505, and obtains the control operation amount output possibility data S4.

[0249] Specifically, the shape correction quality determination unit 502 determines the quality of the shape correction as follows. First, as shown in the specifications A and B regarding the priority of shape control, a weight coefficient w(i) in the width direction is pre-set in the output determination database DB3 for each of specifications A1 and A2 to take into account the control priority in the width direction. Using this weight coefficient, the quality of the shape change is determined using, for example, an evaluation function J such as the following [Formula 4]. In [Formula 4], w(i) is the weight coefficient, εfb(i) is the actual shape deviation data 504, εest(i) is the predicted shape deviation data 505, i is the shape detector area, and rand is the random number term.

[0250] [Formula 5]

[0251]

[0252] When using the evaluation function J in [Equation 5], the evaluation function J becomes positive when the shape improves and negative when it deteriorates. Furthermore, rand is a random number term that randomly changes the evaluation result of the evaluation function J. This allows the evaluation function J to become positive even when the shape deteriorates, making it possible to learn the relationship between the shape pattern and the control method even when the rolling phenomenon model 501 is incorrect.

[0253] Here, when the model of the control target plant 1 is unreliable as at the beginning of a trial run, the random term rand increases its maximum value. When the control method is learned to some extent and stable control is to be performed, the random term rand is appropriately changed so as to become zero.

[0254] The shape correction good or bad judging unit 502 calculates the evaluation function J. When J≥0, the control operation amount outputs the permission data S4=1 (yes). When J<0, the control operation amount outputs the permission data S4=0 (no). The control operation amount outputs the permission data S4.

[0255] As already explained, the control output quality determination execution unit 17 receives inputs such as rolling performance data such as the position of the controlled equipment and the selected control operation variable S8 from the control performance data Si of the controlled plant 1, and outputs a control result quality determination estimate value S11. The control result quality determination estimate value S11 is 1 if the control result is estimated to be improving, 0 if it is estimated to be deteriorating, and -1 if the control result is not subject to quality determination.

[0256] The control output suppressing unit 4 determines whether to output the control manipulated variable output SO to the controlled plant 1 based on the control manipulated variable output permission data S4, which is the determination result of the control output determining unit 5, and the control result goodness / badness determination estimated value S11. The control manipulated variable output permission data S4 refers to the position change outputs #1 to #nAS-U, the upper first intermediate shift position change output, and the lower first intermediate shift position change output, and is determined under the following conditions.

[0257] IF (control method selection flag = 1) THEN

[0258] IF (Control operation variable output permission data S4 = 0 OR Control result good or bad judgment estimated value S11 = 0) THEN

[0259] #1~#nAS-U position change output=0

[0260] Upper first intermediate shift position change output = 0

[0261] Lower first intermediate shift position change output = 0

[0262] ELSE

[0263] #1~#nAS-U position change output = #1~#nAS-U position change

[0264] Upper first intermediate shift position change output = Upper first intermediate shift position change

[0265] Lower first intermediate shift position change output = Lower first intermediate shift position change

[0266] ENDIF

[0267] ELSE

[0268] IF((control operation variable output availability data S4=0 OR control result good / bad judgment estimated value S11=0) AND(PTRIAL<η)) THEN

[0269] #1~#nAS-U position change output=0

[0270] Upper first intermediate shift position change output = 0

[0271] Lower first intermediate shift position change output = 0

[0272] ELSE

[0273] #1~#nAS-U position change output = #1~#nAS-U position change

[0274] Upper first intermediate shift position change output = Upper first intermediate shift position change

[0275] Lower first intermediate shift position change output = Lower first intermediate shift position change

[0276] ENDIF

[0277] ENDIF

[0278] Furthermore, η is a variable that takes a random value between 0 and 1, and PTRIAL represents the probability of disabling output suppression and outputting a new search operation to the plant. When the control method selection flag S14 is 0, including verifying the effectiveness of the control method in an unknown region, output suppression for the plant is ignored with a certain probability, and output is performed to the plant.

[0279] While the above example illustrates the use of both the control operation variable output permission data S4, which is the determination result of the control output determination unit 5, and the control result success / failure determination estimate S11, a situation is also conceivable in which insufficient information on the device is available due to the controlled object, and thus the control operation variable output permission data S4 cannot be constructed by the control output determination unit 5 using a simulation of the controlled object. In this case, the following processing is performed using only the control result success / failure determination estimate S11.

[0280] IF (control method selection flag = 1) THEN

[0281] IF (control result good or bad judgment estimated value S11 = 0) THEN

[0282] #1~#nAS-U position change output=0

[0283] Upper first intermediate shift position change output = 0

[0284] Lower first intermediate shift position change output = 0

[0285] ELSE

[0286] #1~#nAS-U position change output = #1~#nAS-U position change

[0287] Upper first intermediate shift position change output = Upper first intermediate shift position change

[0288] Lower first intermediate shift position change output = Lower first intermediate shift position change

[0289] ENDIF

[0290] ELSE

[0291] IF((control result good or bad judgment estimated value S11=0) AND(PTRIAL<η)) THEN

[0292] #1~#nAS-U position change output=0

[0293] Upper first intermediate shift position change output = 0

[0294] Lower first intermediate shift position change output = 0

[0295] ELSE

[0296] #1~#nAS-U position change output = #1~#nAS-U position change

[0297] Upper first intermediate shift position change output = Upper first intermediate shift position change

[0298] Lower first intermediate shift position change output = Lower first intermediate shift position change

[0299] ENDIF

[0300] ENDIF

[0301] The control execution unit 20 performs the aforementioned calculations based on performance data Si from the controlled plant 1 (rolling mill), outputting the control operation variable output SO to the controlled plant 1 (rolling mill), thereby executing shape control. Furthermore, the control method learning unit 21 utilizes the data used by the control execution unit 20.

[0302] Next, the operation performed by the learning data generating unit 801 will be described.

[0303] like Figure 1 As shown, the learning data generation unit 801 generates supervision data S7a for the neural network 111 used in the control rule learning unit 802 based on the control result good or bad judgment estimate value S11 from the control output good or bad judgment rule execution unit 17, the control operation end operation instruction S2, the selected control operation amount S8, the control method selection flag S14, and the judgment result of the control output inhibition unit (control operation amount output possibility data S4).

[0304] In this case, the supervisory data S7a becomes Figure 8 The output from the output layer of the neural network 111 is shown as the AS-U operation level 301 and the intermediate operation level 302. The learning data generation unit 7 uses the control operation terminal operation instruction S2 (AS-U operation level 301, intermediate operation level 301) as the output of the neural network 101, and the #1 to #n AS-U position change amount outputs, the upper first intermediate shift position change amount output, and the lower first intermediate shift position change amount output as the selected control operation amount S8 to generate supervision data S7a for the neural network 111 used in the control rule learning unit 802.

[0305] When explaining the operation of the learning data generating unit 801, Figure 16 express Figure 11 The relationship between the data and symbols of each part in the control output calculation unit 3 is shown. Here, for the control operation terminal operation instruction S2 as the output of the neural network 101, the AS-U operation level 301 is representatively shown, the data on the positive side of the operation level is denoted as OPref, the data on the negative side of the operation level is denoted as OMref, the conversion gain is denoted as G, and the control operation amount output S3 is denoted as Cref.

[0306] The difference between the positive and negative operation degree data OPref and OMref is obtained by subtractor 701 and multiplied by conversion gain G by multiplier 702 to obtain the control operation variable output Cref. The control operation variable output Cref is provided to the control output operation method selection unit 18 to obtain the selected operation command value C″ref.

[0307] Here, for convenience, the output from the output layer of the neural network 101 of the control law execution unit 10 is set to the positive side of the operation degree and the negative side of the operation degree.

[0308] Figure 17 The processing stages and processing contents in the learning data generating unit 7 are shown.

[0309] In the first processing stage 71 , the operation command value C″ref refers to the selected control operation amount S8 which is the output value of the control output operation method selection unit 18 .

[0310] In the next processing stage 72, the operation instruction value C″ref is corrected and set to C′ref based on the estimated value S11 for determining whether the control result is good or not, the control operation variable output permission data S4, and the control method selection flag 14. Specifically, when (the estimated value S11 for determining whether the control result is good or not = 0 or the control operation variable output permission data = 0) and the control method selection flag S14 = 1, the following [Formula 6] is used, and when the estimated value S10 for determining whether the control result is good or not = 1 and the control operation variable output permission data = 1, the following [Formula 7] is used to set the correction value C′ref of the operation instruction value C″ref. In addition, when (the estimated value S11 for determining whether the control result is good or not = 0 or the control operation variable output permission data = 0) and the control method selection flag S14 = 0, a new search method is selected, and when it is determined that the control effect is low, new supervision data generation is not implemented.

[0311] [Formula 6]

[0312] IF C″ref>0THEN C’ref=C″ref-Δcref

[0313] IF C″ref<0THEN C’ref=C″ref+Δcref

[0314] [Formula 7]

[0315] IF C″ref>0THEN C’ref=C″ref+Δcref

[0316] IF C″ref<0THEN C’ref=C″ref-Δcref

[0317] In the processing stage 73 , the operation degree correction amount ΔOref is calculated using [Formula 8] based on the corrected operation command value C′ref.

[0318] [Formula 8]

[0319]

[0320] In the processing stage 74, the supervisory data OP′ref and OM′ref for the neural network 111 are obtained by [Formula 9].

[0321] [Formula 9]

[0322]

[0323] Thus, in the learning data generating unit 7, as shown in FIG. Figure 16 As shown, based on the control result goodness judgment estimate S11 of the control output goodness judgment rule execution unit 17, the control operation quantity output possibility data S4 of the control output suppression unit 4, and the control method selection flag S14, the operation instruction value correction value C′ref is calculated for the operation instruction value C″ref actually output to the control object factory equipment 1.

[0324] Specifically, when the control result goodness estimation value S11 = 1 and the control operation amount output permission data S4 = 1, if it is determined that the operation is good, the operation command value is increased in the same direction by ΔCref.

[0325] Conversely, if the control result quality estimation value S11 = 0 or the control manipulation variable output permission data S4 = 0, if the operation is judged to be unfavorable, the control method selection flag is 0, and the manipulation variable based on the output of the control rule execution unit 10 is selected, new supervisory data is generated by reducing the manipulation command value in the opposite direction by ΔCref. Since the conversion gain G is a predetermined value and is known, the correction amount ΔOref can be calculated if the values of the positive and negative manipulation degrees are known. Here, ΔCref is previously determined to an appropriate value through simulation, etc., and then set. Through the above steps, the supervisory data OP′ref and OM′ref used in the control rule learning unit 802 can be calculated using [Equation 9].

[0326] In addition, Figure 16Although it is explained with a simple example, it is actually performed for all of the AS-U operation levels 301 for #1 to #nAS-U and the first intermediate operation levels 302 for the upper first intermediate roller shift and the lower first intermediate roller shift, as supervision data (AS-U operation level supervision data, one intermediate operation level supervision data) of the neural network 111 used in the control rule learning unit 802.

[0327] Figure 18 An example of data stored in the learning data database DB2 is shown.

[0328] Learning the neural network 111 requires multiple combinations of input data S8a and supervisory data S7a. A set of learning data, composed of supervisory data S7a generated by the learning data generator 7 and input data S1 (S8a) input to the control rule execution unit 10 by the control execution unit 20, is accumulated in the learning data database DB2. Supervisory data S7a herein includes AS-U operation level supervisory data and the first intermediate operation level. Furthermore, input data S1 (S8a) includes the normalized shape deviation 201 and the shape deviation stage.

[0329] also, Figure 1 The plant equipment control system uses various databases DB1, DB2, DB3, and DB4, but each database DB1, DB2, DB3, and DB4 is managed and operated in association with each other through a neural network management table TB.

[0330] Figure 19 It shows the structure of the neural network management table TB.

[0331] The neural network management table TB divides the specifications based on (B1) plate width, (B2) steel type, and the specifications A1 and A2 for control priority. As for (B1) plate width, four categories are used: 3-foot width, meter width, 4-foot width, and 5-foot width, for example. As for steel type, approximately 10 categories are used: steel type (1) to steel type (10). Furthermore, for the specification A for control priority, two categories are used: A1 and A2. In this case, there are 80 categories, and 80 neural networks are used according to the rolling conditions.

[0332] The neural network learning control unit 112 follows Figure 19 The neural network management table TB shown in FIG. Figure 18 The learning data shown as a combination of input data and supervision data is stored in association with the corresponding neural network No. Figure 20 In the learning data database DB2 shown.

[0333] Each time shape control is executed on the controlled plant 1, the control execution unit 20 generates two sets of learning data. This is because two types of supervisory data are generated to determine the quality of the control results using two evaluation criteria, namely, specification A1 and specification A2, regarding control priority, for the same input data and control output. Once supervisory data reaches a certain level (e.g., 200 sets), or when new supervisory data is accumulated in the learning data database DB2, the neural network learning control unit 112 instructs the neural network 111 to begin learning.

[0334] In the control rule database DB1, according to Figure 19 The management table TB shown stores a plurality of neural networks. The neural network learning control unit 112 specifies the neural network number to be learned, and the neural network selection unit 113 extracts the neural network from the control rule database DB1 and sets it as the neural network 111.

[0335] The neural network learning control unit 112 instructs the input data generation unit 114 and the supervisory data generation unit 115 to retrieve input data and supervisory data corresponding to the corresponding neural network from the learning data database DB2, and uses these data to perform learning of the neural network 111. Various methods have been proposed for learning neural networks, and any method may be used.

[0336] When the learning of the neural network 111 is completed, the neural network learning control unit 112 writes the neural network 111 as a learning result back to the position of the corresponding neural network No. in the control rule database DB1, thereby completing the learning.

[0337] Learning can be targeted at Figure 19 All neural networks defined as shown may be executed simultaneously at a fixed time interval (e.g., every day), or only the neural network No. having accumulated a certain amount (e.g., 100 sets) of new learning data may be learned at that time point.

[0338] Next, the operation of the state change rule learning unit 22 will be described.

[0339] The state change rule learning unit 22 uses the time delay data of the rolling performance data Si of the control target plant 1. Here, the time delay Z -1 It refers to e-TS, which means delaying the preset time T.

[0340] Because the controlled plant 1 has a time response, there is a time delay between a change in the controlled device position and a change in the actual performance data. Therefore, learning uses the shape change calculated by subtracting the actual performance data before the controlled device position change from the actual performance data at the time when the delay time T has passed since the controlled device position change.

[0341] In shape control, it takes several seconds for the shape meter to detect a shape change after issuing an operating command to the AS-U or the first intermediate roll. Therefore, it is preferable to set T to approximately 2 to 3 seconds. Furthermore, the delay time also varies depending on the type of shape detector and the rolling speed. Therefore, it is preferable to set T to the optimal time until a change in the control operation end results in a shape change.

[0342] After the control operation, the shape change amount calculated by subtracting the shape deviation extracted from the performance data before the control device position change from the shape deviation extracted from the performance data at the time point after the delay time T is used as supervision data S13a for the neural network 311.

[0343] Figure 21 An example of data stored in the learning data database DB6 is shown.

[0344] Multiple combinations of input data S12a and supervisory data S13a are required to learn the neural network 311. Therefore, a set of learning data, consisting of supervisory data S13a (shape change data) and input data S12a obtained by extracting rolling state variables and control operation variables from the time-delayed rolling performance data Si, is accumulated in the learning data database DB6.

[0345] Here, rolling state variables that significantly influence the tendency of shape change due to control operations are selected and extracted from the rolling performance data Si. For example, rolling speed, control device position, and tension before and after the rolling mill have a significant impact on the tendency of shape change due to control operations, and are therefore preferably selected. However, if the number of rolling state variables is excessively increased, the learned relationships become complex, and the amount of learning data required increases. As a result, the time required for neural network learning increases, or the structure of the neural network becomes complex, increasing the computational load and causing delays in control operations. Therefore, it is possible to prioritize variables with significant influence based on usage conditions.

[0346] In this case, the learning data is not stored in the learning data database DB6 at a fixed rate but is stored in the verification data database DB7 and can be used for the verification of the good / bad judgment rule in the good / bad judgment rule accuracy verification unit 34 . Figure 22 1 shows an example of data stored in the verification data database DB 8. In addition to the same combination of input data and supervisory data as the learning data, the pre-change shape deviation and neural network number extracted from the time-delayed rolling performance data Si are stored as additional data.

[0347] also, Figure 1 The factory equipment control system uses various databases DB5, DB6, but in Figure 23 2 shows the structure of the neural network management table TB2 for managing and operating the databases DB5 and DB6 in association with each other.

[0348] Specifically, if Figure 23 As shown, management table TB2 divides specifications based on (B1) plate width and (B2) steel type. For example, four plate widths (B1) are used: 3-foot width, meter width, 4-foot width, and 5-foot width. For steel type, approximately 10 steel types (1) to (10) are used. In this case, 40 neural networks are used in 10 different divisions, depending on the rolling conditions.

[0349] The neural network learning control unit 312 follows Figure 23 The neural network management table TB2 will Figure 21 The learning data shown as a combination of input data and supervision data is stored in association with the corresponding neural network No. Figure 24 In the learning data database DB6 shown.

[0350] In the controlled plant 1, learning data is generated each time the shape control equipment operates for a certain period of time. When supervisory data reaches a certain level (e.g., 200 sets) or when new data is accumulated in the learning data database DB6, the neural network learning control unit 312 instructs the neural network 311 to start learning.

[0351] State change rule database DB5 Figure 21 The management table TB2 shown in FIG. is used to store multiple neural networks. Then, the neural network learning control unit 312 specifies the neural network number to be learned, and the neural network selection unit 313 retrieves the corresponding neural network from the state change rule database DB5 and sets it as the neural network 311. The neural network learning control unit 312 retrieves the input data and supervisory data corresponding to the corresponding neural network from the learning data database DB6, and instructs the input data generation unit 314 and supervisory data generation unit 315 to use this data to execute the learning of the neural network 311. Various methods have been proposed for learning neural networks, and any method may be used.

[0352] When the learning of the neural network 311 is completed, the neural network learning control unit 312 writes the neural network 311 as the learning result back to the position of the neural network No. in the control rule database DB6, thereby completing the learning.

[0353] Learning for Figure 23Alternatively, only the neural network No. for which a certain amount of new learning data (e.g., 100 sets) has been accumulated may be trained at that time.

[0354] Furthermore, by including rolled material information such as steel type and plate width in the input data for the state change rule, it is possible to learn within a single neural network including differences in shape change tendencies due to specification B. In this case, there is no need to switch the state change rule based on rolling conditions when executing the state change rule.

[0355] Similar to the control output quality determination execution unit 17, the quality determination error verification unit 34 includes a neural network 341 that performs calculations in only one direction. The verification data generation unit 343 reads the rolling state variables and control operation variables extracted from the time-delayed rolling performance data Si from the verification data database DB8 as verification input data S24 and outputs them to the neural network 341. The neural network 341 then outputs the predicted shape change S25. Simultaneously, the verification data generation unit 343 reads the shape change and the pre-change shape deviation from the verification data database DB8 as quality conversion verification data S23 and outputs them to the state change quality conversion unit 344.

[0356] The state change good or bad conversion unit 344 receives the shape change and the shape deviation before the change as the good or bad conversion verification data S23 from the verification data generation unit 343, and receives the predicted shape change S25 based on the input data of the verification data from the neural network 341. The state change good or bad conversion unit 344 calculates the shape deviation spda(i) after the control device position change based on the supervisory data and the predicted shape deviation after the control device position change based on the output of the neural network using the following formula as shown in [Formula 10] below.

[0357] Here, spd is the shape deviation before the change included in the verification data S23 for quality conversion, Δsp is the shape change included in the verification data S23 for quality conversion,

[0358] represents the predicted shape change S25, and i represents the shape detector number in the plate width direction.

[0359] [Formula 10]

[0360] spd a (i)=spd(i)+Δsp(i)

[0361]

[0362] Then, the state change goodness / badness conversion unit 344 calculates the goodness / badness evaluation value eva and the goodness / badness evaluation value evp based on the output of the neural network using the following [Formula 11]. Here, wc represents the weight in the plate width direction stored in the goodness / badness judgment database DB4. Here, for each specification A, the goodness / badness evaluation value eva based on the supervisory data and the goodness / badness evaluation value evp based on the output of the neural network are calculated and stored together with the information of specification A in Figure 25 The quality evaluation value database DB9 is as shown.

[0363] [Formula 11]

[0364]

[0365]

[0366] The good or bad judgment error calculation unit 345 calculates the good or bad judgment standard error ε and the verification result flag flag for each neural network No. and specification A based on the good or bad evaluation value eva based on the supervision data and the good or bad evaluation value evp based on the output of the neural network calculated according to each verification data stored in the good or bad evaluation value DB9, using the following formula. Here, n represents the number of verification data. th is the judgment threshold for whether the number of verification data is sufficient. When the number of verification data is less than the threshold, flag = 0 (inadequate verification) is set. The control output judgment execution unit 17 of the control execution unit 20 can know that the verification of the state change rule is not yet completed when using the good or bad judgment error database DB7. The good or bad judgment standard error ε and the verification result flag flag of each neural network No. and each specification A obtained in this way are saved to Figure 26 The good / bad judgment error database DB7 as shown is stored therein.

[0367] [Formula 12]

[0368]

[0369] IF n≥th THEN flag=1

[0370] IF n<th THEN flag=0

[0371] As described above, to improve the control rules for the controlled plant 1, if a control operation with good control results cannot be learned, the control operation is significantly modified. Alternatively, if the control results are good, a new control operation method is adopted. Furthermore, if a control operation with good control results can be learned, the control operation remains unchanged, with minor modifications being made. Furthermore, if the control results for these control operations are good, adopting a new control operation method is effective.

[0372] Furthermore, by learning the combination of position and shape changes of control equipment based on actual machine data, the quality of control results can be estimated with high precision, matching the factory equipment status, compared to simulators using mechanical models. Regular automatic learning can also consistently build a model that is appropriate for the latest factory equipment status.

[0373] Furthermore, by determining whether the estimated control result is good or bad, it is possible to improve the reliability of the control output suppression function for plant equipment, which has been performed only in a simple machine model in the conventional technology.

[0374] Furthermore, while control rule learning data is typically generated based on a single control result determination, this example uses an estimated control result determination to mitigate the effects of noise contained in plant equipment data, allowing even minor adjustments with minimal effects to be included in the learning data. Furthermore, this example prevents erroneous control effect determinations, suppressing fluctuations in the learning data and achieving stable control performance.

[0375] Furthermore, the control rule database DB1 stores the neural network used by the control execution unit 20. While the stored neural network performs only initial processing using random numbers, it takes time for the neural network to learn until it can perform the desired control. Therefore, when the control unit is constructed for the controlled plant 1, control rule learning is performed in advance through simulation based on the control model of the controlled plant 1 determined at that time. Furthermore, by storing the learned neural network in the database, a certain level of performance control can be achieved from the start-up of the controlled plant.

[0376] Alternatively, the state change rules are learned by the state change rule learning unit 22 based on the actual performance data of the operation data in the actual machine. Thus, even without controlling the actual machine, it is possible to make an inference about whether the output is good or bad based on the control rules, learn the control rules based on the inference, and perform a certain degree of performance control before applying it to the controlled object factory equipment.

[0377] Figure 27The plant equipment control system of this example is configured to include a control rule evaluation unit 23 that performs control rule evaluation processing.

[0378] The control rule evaluation unit 23 includes a control rule suitability determination 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 DB10 , and a control rule evaluation value database DB11 .

[0379] The control rule evaluation unit 23 performs a control output calculation on the control input data S2 when the control rule whose performance is to be evaluated is set in the control rule execution unit 20. In the control output good or bad judgment execution unit 17, the control rule is evaluated using the estimated control output good or bad judgment estimated value S9.

[0380] The control rule evaluation unit 23 compares the results of the control rule evaluation with the evaluation of the control rule currently applied to the control. If the newly evaluated control rule has a higher evaluation, the neural network number in the database management table TB is updated to apply the new control rule to the control. The following describes the processing details of the control rule evaluation unit 23.

[0381] The control rule quality determination data collection unit 35 receives the control output quality determination estimated value S9 from the control output quality determination rule execution unit 17. If the control output quality determination estimated value S9 is not -1 (not subject to determination), the control rule quality determination data collection unit 35 stores the evaluation target neural network number used in the control execution unit 20, the control rule number selection conditions (specifications A and B), the number of determinations, and the control output quality determination estimated value S9 as control rule quality determination data S16 in the control rule evaluation data database DB10. If the control rule being evaluated is not registered in the database management table TB for current control, the evaluation target neural network number is assigned a consecutive number starting from the last neural network number registered in the database management table TB. Figure 28 1 shows an example of data stored in the control rule evaluation data database DB 10. In this example, the last number of the neural network No. registered in the database management table TB is set to 100, and the new control rules are set to 101 and so on.

[0382] Control rule suitability determination data S16 is generated each time a control output calculation using a control rule is performed in the control execution unit 20. The generated control rule suitability determination data S16 is stored in the control rule evaluation data database DB10. Because a large amount of data is stored for each control rule, the control rule evaluation data database DB10 determines an upper limit for the data stored for each control rule. When the upper limit is exceeded, old data is deleted and new data is stored.

[0383] The control rule evaluation data calculation unit 36 retrieves the control rule quality determination data S17 accumulated for each control rule and each specification condition (A, B) from the control rule evaluation data database DB10, and calculates the average value of the control output quality determination estimated value S9. The calculated average value is the ratio of the number of good operations to the total number of control rule outputs, and this value is used as an indicator for evaluating the performance of the control rule.

[0384] The control rule evaluation data calculation unit 36 stores the control rule evaluation data S18 calculated by the above-described method in the control rule evaluation value database DB11 . Figure 29 This figure shows an example of data stored in the control rule evaluation value database DB11. Control rule evaluation data S18 is stored when control rules B1, B2, and A are applied. Even for the same control rule, re-evaluation at different times allows calculation of evaluation values corresponding to the latest operating status of the factory equipment. In this case, the re-evaluation overwrites the previously calculated evaluation values and updates the database.

[0385] The database management table TB registers the neural network numbers (control rules) used according to the conditions. In contrast, the control rule evaluation value database DB9 manages the evaluation values of multiple control rules for the same condition. The control rule database update unit 37 references the control rule evaluation value database DB11 and compares the control rule evaluation value of the neural network number (control rule) registered in the database management table TB with the control rule evaluation values of other control rules applied to the same condition. The control rule with the highest evaluation value is determined as the control rule to be applied for future control and is updated as the neural network number (control rule) in the database management table TB.

[0386] Figure 27 Other parts of the plant equipment control system shown are Figure 1 The plant equipment control system shown is similarly constructed. Figure 27 In the case of the factory equipment control system shown, Figure 1The plant control system shown here can be used as an offline system to perform evaluations using actual rolling performance data Si on the back side of the plant control. Alternatively, the control target plant 1 can be evaluated using past performance data, with the control rule evaluation unit 23 performing evaluations based on this past performance. In this case, the control execution unit 20 does not need to actually execute the control target plant 1. Specifically, the control output suppression unit 4 does not need to supply the control output amount S0 to the control target plant 1.

[0387] According to the Figure 27 The plant equipment control system shown sets the control rule to be evaluated in the control rule execution unit 10 and provides past performance data as Si. In this way, even if the control output is not actually performed on the controlled plant equipment 1, the control rule evaluation value database DB11 and the database management table TB can be updated.

[0388] <Modification>

[0389] The present invention is not limited to the above-described embodiments, but includes various modifications. For example, the above-described embodiments are examples described in detail to facilitate understanding of the present invention, and are not necessarily limited to having all the described configurations.

[0390] For example, Figure 1 、 Figure 27 The plant equipment control system shown is configured to include a processing unit that performs data generation, learning, control, and other processing. Figure 1 、 Figure 27 The control execution unit 20, control method learning unit 21, quality determination rule learning unit 22, and control rule evaluation unit 23 shown may also be constituted by programs (software) that enable a processor to realize their respective functions, and the computer executes the programs. Figure 30 An example of the configuration of a computer in this case is shown.

[0391] That is, Figure 30 As shown, the computer constituting each unit 20 to 23 includes a CPU (Central Processing Unit) a, a ROM (Read Only Memory) b, and a RAM (Random Access Memory) c connected to a bus, a nonvolatile storage device d, and a network interface e.

[0392] The CPUa is a processing unit that reads the program code for the software that executes the processing in each unit 20-23 from the ROMb and executes it. Variables and parameters generated during the processing are temporarily written to the RAMc. The nonvolatile storage device d, a high-capacity information storage unit such as an HDD (Hard Disk Drive) or SSD (Solid State Drive), stores programs executed by each unit 20-23 and data from various databases.

[0393] Furthermore, the units 20 to 23 may be constituted by separate computers, or the programs may be installed on a small number of computers such as one and executed simultaneously.

[0394] The network interface e uses, for example, a NIC (Network Interface Card) and performs data transmission and reception with other units and the controlled plant equipment 1 .

[0395] Information such as programs that realize each processing function in this case can be stored in a nonvolatile storage device d such as HDD or SSD, or in a recording medium such as a memory, an IC card, an SD card, or an optical disk.

[0396] Furthermore, part or all of the functions performed by the units 20 to 23 may be realized by hardware such as an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit).

[0397] In addition, Figure 1 、 Figure 27 In the block diagrams shown, control lines and information lines are only those necessary for explanation, and not all control lines and information lines are necessarily shown in the product. In practice, it can be assumed that almost all components are connected to each other.

[0398] In addition, in the above embodiment, the Sendzimir mill is used as an example of the control target plant 1, but the present invention can be applied to the control of various other plants. The control rules when applied to the Sendzimir mill are also an example, and the present invention is not limited to the above embodiment.

Claims

1. A plant control system for identifying a pattern of combinations of performance data of a controlled plant and executing control, wherein: The factory equipment control system has: a control method learning unit that learns a combination of performance data and control operations of the control target plant; a control execution unit that executes control of the control target plant based on the combination of performance data and control operations learned by the control method learning unit; and a state change rule learning unit that learns a combination of performance data, control operations, and state changes of the control target plant. The control execution unit comprises: a control rule execution unit that provides a control output based on a combination of performance data of the control target plant and a determination of a control operation; a control output quality determination execution unit configured to predict a change in the state of the controlled object based on a combination of actual performance data of the controlled plant equipment, control operations, and a determination of a change in the state of the controlled object, and to estimate the quality of the control output; a new search operation amount calculation unit for calculating a new search operation amount based on the quality determination of the control output quality determination execution unit; as well as a control output suppressing unit that, using the quality determination of the control output quality determination executing unit, prohibits output of the control output to the control target plant if it is determined that performance data of the control target plant deteriorates when outputting the control output to the control target plant; The state change rule learning unit has: a state change rule learning unit that extracts, from the performance data of the controlled plant equipment, a combination of the actual performance data, the control operation, and the state change of the controlled plant equipment during a time delay until the control effect of the control operation appears in the performance data, generates learning data, and performs learning using the learning data; The control method learning unit comprises: a learning data generating unit that obtains supervisory data using the control output quality determination performed by the control output quality determination executing unit and the control output; and A control rule learning unit performs learning using the actual performance data and the supervisory data as learning data.

2. The factory equipment control system according to claim 1, characterized in that: The control method learning unit performs learning to obtain separate combinations of performance data and control operations for a plurality of control targets according to the state of the control target plant equipment. The obtained combination of the performance data and the control operation is used as the combination of the performance data and the control operation determined by the control rule execution unit for the control target plant.

3. The factory equipment control system according to claim 1, characterized in that: The control output quality determination execution unit stores a combination of performance data of the control target plant, control operations, and a determination of a state change of the control target in a first neural network. The state change rule learning unit stores a combination of performance data, control operations, and state changes of the controlled object as a second neural network. The second neural network obtained as a result of learning in the state change rule learning unit is used as the first neural network in the control output quality determination execution unit.

4. The factory equipment control system according to claim 1, characterized in that: The state change rule learning unit has: a good / bad judgment error verification unit that predicts a change in the state of the controlled object based on a combination of good / bad judgments of control outputs based on past performance data, performance data of the controlled plant equipment, control operations, and a determination of a change in the state of the controlled object, and calculates a good / bad judgment error by comparing the good / bad judgments of the control outputs. The quality determination error generated by the quality determination error verification unit is used to change a criterion for quality determination of a control result in the control output quality determination execution unit.

5. The factory equipment control system according to claim 1, characterized in that: The factory equipment control system further comprises: a control output determination unit that determines whether the control output is possible based on a simulation using a physical model, The control output inhibition unit uses both the goodness / failure judgment of the control output goodness / failure judgment execution unit and the control output feasibility judgment of the control output judgment unit, or the goodness / failure judgment of the control output goodness / failure judgment execution unit, to determine that the performance data of the control object factory equipment deteriorates when the control output is output to the control object factory equipment, and thus prevents the control output from being output to the control object factory equipment.

6. The plant equipment control system according to any one of claims 1 to 5, characterized in that: The plant equipment control system further comprises a control rule evaluation unit, The control rule evaluation unit has: a control rule quality determination data collection unit that accumulates quality determination data of the quality determination execution unit of the control execution unit in a database; and a control rule evaluation data calculation unit that calculates control rule evaluation data based on the good / bad judgment data accumulated in the database; The control rules used in the control execution unit can be evaluated without being output to the controlled plant.

7. A method for controlling plant equipment, comprising: identifying a pattern of combinations of performance data of a control target plant equipment, and controlling the control target plant equipment by a computer, wherein: The processing executed by the computer includes: a control method learning process for learning a combination of performance data and control operations of the control target plant; a control execution process for executing control of the control target plant based on the combination of performance data and control operations learned through the control method learning process; and a state change rule learning process for learning a combination of performance data, control operations, state changes of the control target plant, and whether the control result is good or not. The control execution process includes: A control rule execution process for providing a control output based on a combination of performance data of the control target plant and a determination of a control operation; a control output quality determination execution process for predicting a change in the state of the controlled object based on a combination of actual performance data of the controlled plant equipment, control operations, and a determination of a change in the state of the controlled object, and inferring the quality of the control output; A new search operation amount calculation process of calculating a new search operation amount based on the good / bad judgment in the control output good / bad judgment execution process; and a control output suppression process that uses the control output quality determination in the control output quality determination execution process to prevent the control output from being output to the control target plant if it is determined that the performance data of the control target plant has deteriorated when the control output is output to the control target plant; The state change rule learning process includes: A state change rule learning process extracts, from the performance data of the controlled plant equipment, a combination of performance data, control operations, and state changes of the controlled equipment during a time delay until the control effect of the control operation appears in the performance data, generates learning data, and performs learning using the learning data. The control method learning process includes: a learning data generating process of obtaining supervisory data using the control output quality determination and the control output in the control output quality determination execution process; and The control rule learning process performs learning using the actual data and the supervisory data as learning data.

8. A computer-readable recording medium storing a program for identifying a pattern of combinations of performance data of a controlled plant and causing the computer to execute plant control, wherein: The program causes the computer to execute: a control method learning step of learning a combination of performance data and control operations of the control target plant; a control execution step of executing control of the control target plant based on the combination of performance data and control operations learned in the control method learning step; and a state change rule learning step of learning a combination of performance data, control operations, state changes of the control target plant, and whether the control result is good or not. The control execution step includes: a control rule execution step of providing a control output based on a combination of performance data of the control target plant equipment and a determination of the control operation; a control output quality determination execution step of predicting a change in the state of the controlled object based on a determined combination of actual performance data of the controlled plant equipment, control operations, and a state change of the controlled object, and inferring the quality of the control output; A new search operation amount calculation step of calculating a new search operation amount based on the good / bad judgment in the control output good / bad judgment execution step; and a control output suppressing step of using the control output quality determination in the control output quality determination executing step to prevent the control output from being output to the control target plant if it is determined that the performance data of the control target plant deteriorates when the control output is output to the control target plant; The state change rule learning step includes: a state change rule learning step of extracting, from the performance data of the controlled plant equipment, a combination of performance data, control operations, and state changes of the controlled equipment during a time delay until the control effect of the control operation appears in the performance data, generating learning data, and performing learning using the learning data; The control method learning step includes: a learning data generating step of obtaining supervisory data using the control output goodness / badness determination in the control output goodness / badness determination executing step and the control output; and The control rule learning step performs learning using the actual performance data and the supervisory data as learning data.

Images