KR20140143225A - Setting calculation system learning device and learning method - Google Patents

Setting calculation system learning device and learning method Download PDF

Info

Publication number
KR20140143225A
KR20140143225A KR1020147031254A KR20147031254A KR20140143225A KR 20140143225 A KR20140143225 A KR 20140143225A KR 1020147031254 A KR1020147031254 A KR 1020147031254A KR 20147031254 A KR20147031254 A KR 20147031254A KR 20140143225 A KR20140143225 A KR 20140143225A
Authority
KR
South Korea
Prior art keywords
function
learning
value
downstream
calculation
Prior art date
Application number
KR1020147031254A
Other languages
Korean (ko)
Other versions
KR101622068B1 (en
Inventor
하루키 이나미
Original Assignee
도시바 미쓰비시덴키 산교시스템 가부시키가이샤
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 도시바 미쓰비시덴키 산교시스템 가부시키가이샤 filed Critical 도시바 미쓰비시덴키 산교시스템 가부시키가이샤
Publication of KR20140143225A publication Critical patent/KR20140143225A/en
Application granted granted Critical
Publication of KR101622068B1 publication Critical patent/KR101622068B1/en

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
    • B21B—ROLLING OF METAL
    • B21B37/00—Control devices or methods specially adapted for metal-rolling mills or the work produced thereby
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B21/00—Systems involving sampling of the variable controlled
    • G05B21/02—Systems involving sampling of the variable controlled electric
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning

Landscapes

  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Automation & Control Theory (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Mechanical Engineering (AREA)
  • Feedback Control In General (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

There is provided a learning apparatus of a setting calculation system capable of appropriately updating a learning term of each function constituting a mathematical expression model even when a part of actual values necessary for learning of the mathematical model can not be obtained. To this end, a learning apparatus for updating a learning term of a mathematical expression model by using a measured value in a setting calculation system that calculates a set value of a mechanical equipment using an mathematical model, comprising: a learning device for learning each of an upstream side and a downstream side An actual value judging section for judging whether or not the first actual value inputted to the downstream side function in the calculation of the learning term of the downstream side function is abnormal; A second calculation unit that calculates a second calculation result of the second calculation based on the output from the first calculation unit and the second calculation unit, And a supplementary learning calculation section for distributing the error of the calculated results to the respective learning terms of the upstream side and the downstream side functions.

Description

{SETTING CALCULATION SYSTEM LEARNING DEVICE AND LEARNING METHOD}

The present invention relates to a learning apparatus and a learning method of a setting calculation system.

Generally, for example, as a method of determining a set value for controlling a machine tool in a process line or the like of a rolling plant, a mathematical expression model in which a physical phenomenon occurring in an environment including a control object is represented and reproduced by an expression, A method of determining a set value by determining a set value for obtaining a desired result on the mathematical expression model is known.

In the determination of the set value using the mathematical expression model, increasing the precision of reproduction of the target physical phenomenon in the mathematical expression model to be used leads to determination of a better set value. Thus, in order to improve the precision of the mathematical expression model, it has been conventionally done to incorporate the term into the mathematical expression model and to modify the mathematical expression model based on the actual value.

As a learning method of such a mathematical model, a calculation value (an actual calculation value) of an actual value calculated by a mathematical expression model is compared with an actual value obtained from a measured value actually measured by a meter or the like, and the learning of the mathematical expression model The method of updating the term is often adopted.

In the conventional learning method, factors causing a difference in the actual calculated value by the mathematical expression model and actual actual values are called the time series fluctuation of the process line due to the change of the equipment or the environment, the type of the material processed in the process line, It is known that the calculation values of the mathematical expression model are corrected based on the learning coefficients calculated for each of these two kinds of fluctuations (for example, see Patent Document 1 Reference).

Patent Document 1: Japanese Patent No. 2839746

However, when the physical phenomenon (that is, the physical phenomenon occurring in the control object) to be reproduced by the mathematical expression model is complex, a mathematical expression model for reproducing the physical phenomenon by the synthesis function synthesizing a plurality of simpler functions . In such a case, it is preferable from the viewpoint of improving the reproduction precision of the mathematical expression model that a learning term is provided for each function constituting the combining function to perform learning for each individual function.

However, when measured values can not be obtained due to problems such as the environment around the measuring instrument, there is a possibility that actual values necessary for learning can not be obtained for some of the functions constituting the mathematical expression model. As the synthesis functions composing the mathematical expression model are synthesized with more functions and the more kinds of actual values required for learning are, the more likely the probability that some of the actual values will be defective will increase.

The problem that the learning of the mathematical model can not be completed due to the lack of some of the actual values required for the learning and the prediction accuracy of the mathematical expression model is temporarily lowered is known as the learning apparatus and learning method of the conventional setting calculation system exist. In the situation immediately after the learning is started, if the learning of the mathematical expression model can not be completed due to a deficiency in a part of the actual values required for learning, it is necessary until the mathematical expression model can predict the originally expected precision There is also a problem that the time is lengthened.

SUMMARY OF THE INVENTION The present invention has been made to solve such problems, and it is an object of the present invention to provide a learning method and a learning method for learning a mathematical model, The learning term of the setting calculation system can be appropriately updated and degradation of the accuracy of the mathematical expression model can be suppressed.

The learning apparatus of the setting calculation system according to the present invention is a learning apparatus for updating a learning term of the mathematical expression model using an actual value in a setting calculation system for calculating a set value of a mechanical equipment using an mathematical model, A learning term calculation unit for calculating a learning term of each of an upstream side function and a downstream side function constituting the learning term by using the measured value; An output from the upstream function is input to the downstream function to determine whether the first actual value is abnormal or not, A function calculating section for calculating a performance calculation value output from the function on the downstream side, The error of the measured value of the results calculated for the second measured value, and a complementary learning unit operable to distribute the calculated learning, wherein learning and wherein the downstream-side function of the upstream function.

In a learning method of a setting calculation system according to the present invention, in a setting calculation system for calculating a set value of a mechanical equipment using an equation model, a learning method of updating a learning term of the equation model by using an actual value, A first step of calculating a learning term of each of an upstream side function and a downstream side function constituting a model by using the actual values; and a second step of calculating a learning term of the downstream function in the first step, A second step of determining whether or not the first measured value, which is an actually measured value to be input, is abnormal; and a second step of determining whether or not the first measured value is abnormal, A third step of calculating a performance calculation value output from the downstream-side function, and a third step of calculating an actual-value calculation value corresponding to the output from the downstream- The error of the results calculated for the second measured value, and a step of claim 4 for dispensing the anti-learning and learning, wherein the downstream-side function of the upstream function.

In the learning apparatus and the learning method of the setting calculation system according to the present invention, even when a part of the actual values required for learning the mathematical expression model can not be obtained, learning using the obtained actual values is performed, Term can be appropriately updated, and it is possible to suppress deterioration in precision of the mathematical expression model.

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a diagram for explaining the entire configuration of a setting calculation system according to Embodiment 1 of the present invention; Fig.
Fig. 2 is a flowchart for explaining the operation of the learning apparatus of the setting calculation system according to the first embodiment of the present invention; Fig.
3 is a diagram for explaining a detailed configuration centered on a learning apparatus included in a setting calculation system according to Embodiment 1 of the present invention.

The present invention will be described with reference to the accompanying drawings. Throughout the drawings, the same reference numerals denote the same or corresponding parts, and redundant descriptions thereof are appropriately simplified or omitted.

Embodiment Mode 1.

1 to 3 relate to a first embodiment of the present invention. FIG. 1 is a diagram for explaining the entire configuration of a setting calculation system, FIG. 2 is a flowchart for explaining an operation of a learning apparatus of the setting calculation system, 3 is a diagram for explaining a detailed configuration around a learning apparatus provided in the setting calculation system.

1 shows the entire configuration of a setting calculation system 2 for calculating a set value of a machine tool 1 constituting a process line or the like of a rolling plant, for example. The setting calculation system 2 includes a setting calculation device 3, a measured value collection device 4, and a learning device 5.

In the setting calculation device 3, a mathematical expression model in which a physical phenomenon caused by the operation of the mechanical equipment 1 is modeled by an equation is registered in advance. The setting calculation device 3 simulates the operation result of the machine tool 1 by using this mathematical expression model so that the set value of the machine tool 1 is set such that the result of operation of the machine tool 1 is closer to the target value . The setting value calculated by the setting calculation device 3 is output to the machine tool 1. [ Then, the machine tool 1 operates according to the set value calculated in the setting calculation system 2.

The measured value collecting device 4 calculates actual values of predetermined types of physical quantities necessary for learning the mathematical expression model in the learning device 5 among the physical quantities in the environment in which the mechanical equipment 1 or the mechanical equipment 1 operates To collect. In an environment in which the mechanical equipment 1 or the mechanical equipment 1 is installed, a measuring instrument for measuring the physical quantity is provided in advance. The measured value collecting device 4 collects actual values of the physical quantities measured by a meter or the like.

The learning device 5 performs learning of the mathematical model used in the setting calculation device 3 based on the measured values collected by the measured value collection device 4. The learning apparatus 5 includes an equation model learning calculation section 6, a measured value determination section 7, and a supplementary learning calculation section 8.

The equation model learning calculation unit 6 calculates an earnings calculation value using an equation model such as that used in the setting calculation apparatus 3. [ Here, the earnings calculation value is an output from the mathematical expression model when the measured value collected by the measured value collection device 4 is input. The equation model learning calculation section 6 compares the calculated value of the performance calculated from the mathematical expression model with the actual value directly obtained from the measured value collected by the measured value collection device 4, Calculate the learning term.

The measured value determination section 7 determines whether the measured values collected by the measured value collection device 4 are normal or anomalous. Here, the measured value is an abnormal state, which includes a state of missing (missing) in which the measured value itself does not exist, in addition to a state deviating from a range where the measured value is normal (this range is predetermined for each measurement target).

When the actual value is judged to be abnormal by the actual value judging section 7, the learning device 5, when there is an actual value calculated corresponding to the actual value judged to be abnormal, is used instead of the actually measured value, Proceed with the calculation of the term.

Here, the "performance calculation value corresponding to the measured value" will be described again. First, as described previously, when the physical phenomenon to be reproduced by the mathematical expression model is complex, a mathematical expression model for reproducing the physical phenomenon may be constructed by combining a plurality of simpler functions. In such a case, the output from a certain function (hereinafter referred to as " upstream function ") becomes the input of another function (hereinafter referred to as " downstream function "

When the upstream side function and the downstream side function having such a relationship exist, the updating of the learning term in the downstream side function is performed when the measured value collected by the measured value collection device 4 is input to the downstream side function, Using the performance calculation value output from the side function. However, when the measured value to be input to the downstream-side function is abnormal, it is impossible to input the measured value into the downstream-side function, so that it becomes impossible to learn the downstream-side function.

Thus, by inputting the performance calculation value output from the upstream-side function into the downstream-side function instead of the measured value, the learning in the downstream-side function proceeds. That is, the above-mentioned " performance calculation value corresponding to the measured value " is the calculated value of the upstream side function used as the input of the downstream-side function in the mathematical expression model.

In this way, when the learning in the downstream function is advanced by inputting the performance calculation value output from the upstream-side function to the downstream-side function instead of the measured value, if it is intrinsic (that is, if the normal measured value to be input to the downstream- , The error to be absorbed in the learning term of the upstream side function is also absorbed in the learning term on the downstream side. Then, when a correct one is obtained as an actual value to be used for inputting the downstream function, when the learning in each of the upstream function and the downstream function is resumed, the individual learning of the upstream function and the downstream function And the learning accuracy in the term is deteriorated.

Therefore, in the learning apparatus 5 of the present invention, the error between the output of the downstream-side function calculated by using the performance calculation value output from the upstream-side function instead of the measured value and the actual value corresponding to this output, And a supplementary learning calculation section 8 for distributing the learning term to learning terms of the upstream side function and the downstream side function as the distribution ratio obtained by a predetermined procedure.

The complementary learning calculator 8 calculates the error between the output of the downstream function and the actual value corresponding to this output when the actual value to be input to the downstream function is determined to be abnormal by the measured value determining section 7 . Then, this error is distributed to both learning terms of the upstream side function and the downstream side function as a distribution ratio obtained by a predetermined procedure.

The procedure for obtaining the distribution ratio of the error at this time will be described. First, as the learning of the mathematical model progresses, the error between the output of the mathematical model and the actual value becomes smaller. Therefore, in a state in which the learning of the mathematical expression model proceeds sufficiently, the change when the learning term is updated is small. And, in the state where the learning term is stable as described above, the relative size of the learning term of each function can be evaluated with the size of each function as a scale. Here, the size of the function is the size of the output from the function whose reference to the function is input.

Thus, the supplemental learning calculation unit 8 obtains the magnitude of the output from the function based on the input to the function, with respect to each of the upstream function and the downstream function. The ratio of the size of the upstream-side function thus obtained to the size of the downstream-side function is defined as the above-mentioned distribution ratio. Here, as a value to be input to the function when the size of each function is obtained, actually measured values collected by the measured value collecting device 4 are used. However, when the measured value to be input is abnormal and can not be used, another value such as the calculated value of the performance may be substituted.

As a distribution ratio thus obtained, the supplementary learning calculation section 8 distributes the error to both the learning term of the upstream side function and the downstream side function. Then, the equation model learning calculation section 6 updates each learning term of the upstream side function and the downstream side function on the basis of the error distributed by the supplementary learning calculation section 8.

Next, the setting calculation device 3 and the learning device 5 will be described in more detail with reference to Fig. Here, for convenience of explanation, it is assumed that the mathematical expression model is composed of an upstream function and a downstream function, and as shown in Fig. 2, the upstream function is also expressed by the model equations (1a) and 1b, and the downstream-side function is also made up of the model equation 2a and the model equation 2b.

The setting calculation device 3 calculates the set value of the machine tool 1 so that the output of the mathematical expression model showing the result of operation of the machine tool 1 is closer to the target value. The computation of the output of the mathematical expression model will be described following the order. First, the function of the model equation (1a) is denoted by f, the input physical quantity is denoted by V 0 , the set value of the mechanical equipment (1) is denoted by X 1 i (i = 1, 2, (Intermediate result) output (Y 1 ) from the model equation (1a) is expressed by the following expression (1), where a 1 j (j = 1, 2, ...)

[Equation 1]

Figure pct00001

During this process, Y 1 Z is obtained as a result of the correction by the learning term expressed by the following equation (2) with respect to the result output (Y 1 ). Then, by using the obtained Y 1 Z as an input to the model equation (1b) and calculating the following equation (3), an output V 1 from the upstream-side function is obtained. Where G is a function of the model equation (1b), W 1 l (l = 1, 2, ...), and G 1 is a function of the learning term. It is the other input variables, b 1 k (k = 1 , 2, ...) is the other of the input condition.

[Equation 2]

Figure pct00002

[Equation 3]

Figure pct00003

In this manner, V 1 output from the upstream-side function is input to the downstream-side function, and calculation of the equation model proceeds. The function of the model equation 2a is represented by f, the set value of the mechanical equipment 1 is represented by X 2 i (i = 1, 2,...), And the other conditions input by a 2 j (j = 1, 2, ...), the intermediate result output (Y 2 ) from the model equation (2a) is expressed by the following equation (4).

[Equation 4]

Figure pct00004

Then, Y 2 Z obtained by performing the correction by the learning term expressed by the following expression (5) with respect to the result output (Y 2 ) is input to the model equation (2b) ), The output V 2 from the downstream side function is obtained. This V 2 is the final result output from the mathematical model. Where G is the function of the model equation (2b), W 2 l (l = 1, 2, ...), and G 2 is the coefficient of the learning term. It is the other input variables, b 2 k (k = 1 , 2, ...) is the other of the input condition.

[Equation 5]

Figure pct00005

[Equation 6]

Figure pct00006

The setting calculation device 3 determines the set value of the machine tool 1 by obtaining X 1 i and X 2 i equal to the target value V AIM obtained by the final result output V 2 thus obtained. That is, X 1 i and X 2 i satisfying the equations (1) to (7) are obtained after the following Equation (7).

[Equation 7]

Figure pct00007

The equation model learning calculation section 6 provided in the learning apparatus 5 is constituted by a performance calculation value calculation section 6a and a learning term calculation section 6b. The performance calculation value calculation unit 6a calculates the performance calculation value using the mathematical expression model used in the setting calculation apparatus 3. The learning term calculation section 6b compares the actual value calculated by the actual value calculation section 6a with the actual value obtained by the actual value collected by the actual value collection device 4 so that the difference between these values becomes smaller Calculate the learning term of the equation model.

With regard to the calculation of the learning term in the mathematical expression model learning calculation section 6, the calculation of the performance calculation value in the performance calculation value calculation section 6a will first be described. Since the calculation of this earnings calculation value is basically the same in the upstream function and the downstream function, subscripts " 1 " and " 2 ", respectively indicating the upstream function and the downstream function are omitted. Let the function of the model equation (a) be f, the measured value of the input physical quantity be V ACT , the actual value of the set value of the mechanical equipment 1 be X ACT i (i = 1, 2, j (j = 1, 2, ...) when said model equation (a) result (performance calculated) (Y ACAL) from the middle of is represented by the following (equation 8) formula.

[Equation 8]

Figure pct00008

Y ACAL Z is obtained as a result of performing the correction by the learning term expressed by the following expression (9) on the result output (Y ACAL ). Then, using the obtained Y ACAL Z as the input of the model equation (b), the following equation (10) is calculated to obtain the performance calculation value (V ACAL ). In the equation (9), H is an error correction function, Z is a coefficient of a learning term, g is a function of the model equation (b), W l (l = 1, 2, , And b k (k = 1, 2, ...) are other condition inputs.

[Equation 9]

Figure pct00009

[Equation 10]

Figure pct00010

Next, the calculation of the learning term in the learning term calculation section 6b will be described. The calculation of this learning term is also the same in the upstream and downstream functions, as in the case of the performance calculation. In the example described here, the learning term is enforced for the intermediate result output from the model equation (a). Therefore, comparison between the calculated value of the performance for learning and the actual value is made at the output of the result from the model equation (a).

Therefore, first, the actual value (Y ACT ) corresponding to the intermediate result output from the model equation (a) is calculated from the following equation (11) based on the measured value (V ACT ). This (Equation 11) where, g - 1 is the inverse of the model formula (b), W ACT l ( l = 1, 2, ...) is the other variable input of the measured value, b k (k = 1, 2, ...) Is the other condition input.

[Equation 11]

Figure pct00011

The error (Z CUR ) is calculated from the following equation (12) from Y ACAL obtained from the actual values (Y ACT ) and the equation (8) obtained in this manner. In this equation (12), h is an error calculation function.

[Equation 12]

For example, the difference between Y ACT and Y ACAL may be taken as a specific form of the error calculation function h (i.e., Z CUR = Y ACT -Y ACAL ), Y The ratio of ACT to Y ACAL may be taken (i.e., Z CUR = Y ACT / Y ACAL ). The concrete form of the error correcting function H of the equation (5) and (9) also changes according to the concrete form of the error calculating function h. That is, when the error calculation function h takes the difference of the input variables, the error correction function H takes the sum of the input variables, and the error calculation function h takes the ratio of the input variables , The error correcting function H takes the product of the input variable.

Then, a new learning term (Z NEW ) is calculated by smoothing the error by the following equation (13) and reflecting it to the learning term. In the equation (13), Z NEW is the learning term used in the setting calculation in the next setting calculation apparatus 3, Z OLD is the learning term used in the setting calculation in the previous setting calculation apparatus 3, Smoothing factor.

[Equation 13]

Figure pct00013

This is the calculation method of the learning term in the case where the measured value necessary for the calculation of the learning term can be normally obtained. On the other hand, as described above, when the measured value inputted to the downstream-side function is abnormal in the calculation of the term of the downstream-side function, the calculation result is input to the downstream-side function by inputting the calculated value of the output from the upstream- To obtain the performance calculation value necessary for learning in Calculation of the calculated value of the performance required for learning to the downstream function when the measured value input to the downstream function is abnormal in the calculation of the learning term of the downstream function will be described below.

In this case, the input to the model equation (2a) of the downstream-side function is V ACAL calculated using the equation (10) in the upstream-side function. Therefore, the function of a downstream function model equation (2a) of f, the set value of the measured value of the mechanical equipment (1) X 2ACT i (i = 1, 2, ...), the other of the condition input a 2 j (j = 1 , 2, ...), the intermediate result output (Y 2 ACAL ) from the model equation (2a) is expressed by the following expression (14). In the equation (14), V 1 ACAL is the calculated value of the performance calculated using the equation (10) on the upstream side function.

[Equation 14]

Figure pct00014

The learning term is calculated from the actual value (Y 2ACT ) obtained by substituting the actual value (V ACT ) = V 2ACT in the thus calculated Y 2 ACAL and the equation (11). Since the Y 2 ACAL includes the errors of both the upstream and downstream functions as described above, the complementary learning calculator 8 calculates the error (Y ACT -Y 2 ACAL ) as the upstream function And the learning function of the downstream side function.

The calculation method of the distribution ratio of the error in the supplementary learning calculation unit 8 is changed by the type of the error calculation function h in the equation (12). More specifically, when the error calculation function h takes the difference of the input variables, the distribution of the error in the supplementary learning calculation section 8 can be calculated based on the following equations (15) and (16) It is done by a proportional distribution.

[Equation 15]

Figure pct00015

[Equation 16]

Figure pct00016

In the case where the error calculation function h takes the ratio of the input variable, the distribution of the error in the complementary learning calculation section 8 is proportional to the following Expression (17) and Expression (18) Distribution.

[Equation 17]

[Equation 18]

Figure pct00018

In the equations (15) to (18), Z 1CUR is the error distributed to the learning term of the upstream function, Z 2CUR is the error distributed to the learning term of the upstream function, f is the model equation (a) .

After the error is proportionally distributed to each learning term of the upstream side function and the downstream side function, an error distributed to each of the learning terms of the upstream side function and the downstream side function by the equation (13) It is smoothed and then reflected in the learning term. The updated new learning term is used for calculating the set value in the setting calculation apparatus 3 at the next setting timing.

The flow of operation in the learning apparatus 5 constructed as described above will be described again with reference to Fig. First, in step S1, the measured value collecting device 4 collects measured values. Next, the process proceeds to step S2, and the measured value determining section 7 confirms whether or not V 2 AC is abnormal among the measured values collected by the measured value collecting device 4. If V 2ACT is not abnormal, the process proceeds to step S 3. In this step S3, the measured value determining section 7 confirms whether or not V1ACT of the measured values collected by the measured value collecting device 4 is abnormal. If V 1ACT is not abnormal, the process proceeds to step S 4.

In this step S4, the performance calculation value calculation unit 6a calculates the performance calculation value ( Y1ACAL ) of the upstream side function using the formula (8). In the succeeding step S5, the learning term calculation section 6b calculates the actual value ( Y1ACT ) of the upstream-side function using the equation (11). Next, the process proceeds to step S6, and the learning term calculation section 6b calculates the error ( Z1CUR ) of the upstream side function using the equation (12).

After step S6, the process proceeds to step S7. In this step S7, the performance calculation value calculating section 6a substitutes the measured value V 1ACT into V ACT of the formula (8) to calculate the performance calculation value (Y 2ACAL ) of the downstream function. In the succeeding step S8, the learning term calculation section 6b calculates the actual value (Y 2ACT ) of the downstream side function using the formula (11). Next, the process proceeds to step S9, and the learning term calculation section 6b calculates the error (Z 2CUR ) of the downstream side function using the equation (12).

After step S9, the process proceeds to step S10. In this step S10, the learning term calculation section 6b substitutes Z 1CUR and Z 2CUR into the Z CUR of the expression (13) to calculate the learning term (Z 1NEW ) of the upstream function and the learning term (Z 2 NEW ). Then, the process advances to step S11 to update the learning term of the mathematical expression model of the setting calculation apparatus 3 with the calculated learning terms ( Z1NEW and Z2NEW ), and the series of learning term updating processing ends.

On the other hand, if V1ACT is abnormal in step S3, the process proceeds to step S20. In this step S20, the performance calculation value calculating unit 6a calculates the performance calculation value ( Y1ACAL ) of the upstream side function using the formula (8). In step S21, the performance calculation value calculation unit 6a calculates the performance calculation value V1ACAL of the upstream side function using the formula (9) and the formula (10).

After step S21, the process proceeds to step S22. In this step S22, the performance calculation value calculating unit 6a calculates the performance calculation value (Y 2 ACAL ) of the downstream side function using V 1 ACAL by the formula (14). In the succeeding step S23, the learning term calculation section 6b calculates the actual value (Y 2ACT ) of the downstream side function using the formula (11).

Then, the process proceeds to step S24, where the supplementary learning calculation unit 8 calculates Z 1CUR and Z 2CUR (Equation 15) by using the equation 15 and the equation 16 or the equation 17 and the equation 18, . After step S24, the process proceeds to step S10 described above.

If it is determined in step S2 that V2ACT is abnormal, the flow advances to step S30. In this step S30, the measured value determining section 7 confirms whether or not V1ACT of the measured values collected by the measured value collecting device 4 is abnormal. When V 1ACT is also abnormal, the learning term is not updated in both the upstream side function and the downstream side function.

On the other hand, if V1ACT is not abnormal in step S30, the process proceeds to step S31. In this step S31, the performance calculation value calculating unit 6a calculates the performance calculation value ( Y1ACAL ) of the upstream side function using the formula (Eq. 8). In the succeeding step S32, the learning term calculation section 6b calculates the actual value ( Y1ACT ) of the upstream side function using the formula (11). Next, the process proceeds to step S33, and the learning term calculation section 6b calculates the error ( Z1CUR ) of the upstream side function using the equation (12).

After step S33, the process proceeds to step S34. In this step S34, the learning term calculation unit 6b calculates the learning term ( Z1NEW ) of the upstream side function using the equation (13). Then, the process advances to step S35 to update the learning term of the mathematical expression model of the setting calculation apparatus 3 with the calculated learning term (Z 1 NEW ), and the series of learning term updating processing ends.

The learning apparatus of the setting calculation system configured as described above includes a learning term calculation unit for calculating learning terms of each of an upstream side function and a downstream side function constituting the mathematical expression model using measured values, An output from the upstream function is determined as a downstream side function when the first measured value is determined to be abnormal; Side function and a second actual value corresponding to an output from the downstream-side function, and a second calculation unit that calculates an error of an actual calculation value with respect to a second actual value corresponding to an output from the downstream- And a supplementary learning calculation section that distributes the learning term to the learning term of the function.

Therefore, even when the actual value input to the downstream function can not be obtained for learning, the performance computation value output from the upstream function is input to the downstream function instead of the measured value to advance learning in the downstream function, The error is distributed to the learning term of the upstream side function and appropriate learning is realized in both the upstream side function and the downstream side function.

That is, even when a part of the actual values necessary for the learning of the mathematical expression model can not be obtained, it is possible to appropriately update the learning term of each function constituting the mathematical expression model by performing learning using the obtained actual values, It is possible to suppress deterioration in precision. For this reason, it can contribute to improvement in the calculation accuracy of the set value of the mechanical equipment.

Further, the learning apparatus of the setting calculation system configured as described above realizes, for example, a function of completing each part of the equation model learning calculation unit, the performance calculation value calculation unit, the learning term calculation unit, the measured value determination unit and the supplementary learning calculation unit It is also possible to execute the information processing for executing the above-described processing on hardware resources having a central processing unit, a storage device and the like.

[Industrial Availability]

INDUSTRIAL APPLICABILITY The present invention can be used in a learning apparatus and a learning method for updating a learning term of a mathematical expression model using an actual value in a setting calculation system for calculating a set value of a mechanical equipment using an mathematical model.

1: Hardware
2: Setting calculation system
3: Setting calculation device
4: Measured value collecting device
5: Learning device
6: Formula model learning calculation unit
6a: Calculation section
6b:
7:
8: Complementary learning calculation unit

Claims (5)

A learning apparatus for updating a learning term of the mathematical expression model using an actual value in a setting calculation system for calculating a set value of a mechanical equipment using an mathematical model,
A learning term calculation unit for calculating a learning term of each of an upstream side function and a downstream side function constituting the above described mathematical expression model using said measured value;
An actual-value determining unit that determines whether or not the first measured value, which is an actually measured value, input to the downstream-side function in the calculation of the learning term of the downstream-side function in the learning term calculating unit is anomalous;
A performance calculation value calculation unit that inputs an output from the upstream function to the downstream function and calculates a performance calculation value output from the downstream function when the first measured value is determined to be abnormal;
And a supplementary learning calculation section for distributing the error of the performance calculation value to the second measured value which is the measured value corresponding to the output from the downstream side function to the learning term of the upstream side function and the learning term of the downstream side function Wherein the setting calculation system comprises: The method according to claim 1,
The complementary learning unit may calculate a ratio of a magnitude of the output from the upstream function based on the input to the upstream function and a magnitude of the output from the downstream function based on the input to the downstream function, And distributes the error to the learning term of the upstream function and the learning term of the downstream function. 3. The method of claim 2,
The supplementary learning unit,
The error is obtained by the difference between the second measured value and the calculated value of the performance,
And a learning term of the upstream function and a learning term of the downstream function are set such that the absolute value of the difference between the input to the upstream function and the output from the upstream function, Based on the absolute value of the difference between the input and the output from the downstream-side function. 3. The method of claim 2,
The supplementary learning unit,
Wherein said error is obtained by a ratio between said second measured value and said calculated calculated value,
And a learning term of the downstream function and an absolute value of a ratio between an input to the upstream function and an output from the upstream function and an absolute value of the ratio of the absolute value of the ratio of the input to the upstream function to the output of the downstream function, Based on the absolute value of the ratio between the input and the output from the downstream-side function. A learning method for updating a learning term of the mathematical expression model using an actual value in a setting calculation system for calculating a set value of a mechanical equipment using an mathematical model,
A first step of calculating a learning term of each of an upstream side function and a downstream side function constituting the mathematical expression model using the measured value,
A second step of determining whether or not the first measured value, which is an actually measured value inputted to the downstream function, is abnormal in the calculation of the term of the function of the downstream function in the first step,
A third step of inputting the output from the upstream function to the downstream function and calculating a calculated value of the output from the downstream function when the first measured value is determined to be abnormal;
And a fourth step of distributing the error of the actual calculated value to the actual measured value corresponding to the output from the downstream function to the learning term of the upstream function and the learning term of the downstream function Wherein the set-up calculation system comprises:
KR1020147031254A 2012-05-16 2012-05-16 Setting calculation system learning device and learning method KR101622068B1 (en)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2012/062535 WO2013171862A1 (en) 2012-05-16 2012-05-16 Setting calculation system learning device and learning method

Publications (2)

Publication Number Publication Date
KR20140143225A true KR20140143225A (en) 2014-12-15
KR101622068B1 KR101622068B1 (en) 2016-05-17

Family

ID=49583306

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1020147031254A KR101622068B1 (en) 2012-05-16 2012-05-16 Setting calculation system learning device and learning method

Country Status (4)

Country Link
JP (1) JP5846303B2 (en)
KR (1) KR101622068B1 (en)
CN (1) CN104303114B (en)
WO (1) WO2013171862A1 (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10124381B2 (en) * 2014-02-17 2018-11-13 Toshiba Mitsubishi-Electric Industrial Systems Corporation Rolling process learning control device

Family Cites Families (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH07107682B2 (en) * 1989-02-15 1995-11-15 株式会社東芝 Target system modeling method and target system modeling apparatus
JP2839746B2 (en) * 1991-06-17 1998-12-16 株式会社神戸製鋼所 Learning control method in process line
US5566275A (en) * 1991-08-14 1996-10-15 Kabushiki Kaisha Toshiba Control method and apparatus using two neural networks
JPH06342422A (en) * 1993-05-31 1994-12-13 Kawasaki Steel Corp Signal processing method
JPH07200005A (en) * 1993-12-28 1995-08-04 Mitsubishi Electric Corp Learning control method
JPH09265302A (en) * 1996-03-28 1997-10-07 Kawasaki Steel Corp Process model identification method
US5764509A (en) * 1996-06-19 1998-06-09 The University Of Chicago Industrial process surveillance system
JP2000242323A (en) * 1999-02-24 2000-09-08 Hitachi Ltd Plant operation guidance system
JP2002091505A (en) * 2000-09-14 2002-03-29 Toshiba Corp Model identifying device
JP4119684B2 (en) * 2002-05-27 2008-07-16 東芝三菱電機産業システム株式会社 Learning control device for rolling mill setting calculation device
JP4208509B2 (en) * 2002-07-18 2009-01-14 東芝三菱電機産業システム株式会社 Model learning device for rolling process
US7720641B2 (en) * 2006-04-21 2010-05-18 Exxonmobil Research And Engineering Company Application of abnormal event detection technology to delayed coking unit
JP5069608B2 (en) * 2008-05-26 2012-11-07 株式会社日立製作所 Sheet width control device and control method for hot rolling mill
CN101966535B (en) * 2009-07-28 2012-11-14 宝山钢铁股份有限公司 Cold rolling strip shape forward control setting method based on incoming material plate profile
JP2012043235A (en) * 2010-08-20 2012-03-01 Hitachi Ltd Plant model construction support system

Also Published As

Publication number Publication date
CN104303114B (en) 2017-06-16
CN104303114A (en) 2015-01-21
KR101622068B1 (en) 2016-05-17
WO2013171862A1 (en) 2013-11-21
JP5846303B2 (en) 2016-01-20
JPWO2013171862A1 (en) 2016-01-07

Similar Documents

Publication Publication Date Title
JP6125506B2 (en) Method and system for detection and correction of problematic advanced process control parameters
JP6193891B2 (en) Apparatus and method for performing incoherent closed-loop step tests using adjustable trade-off factors
KR101832653B1 (en) Rolling process learning control device
JP4972623B2 (en) Distribution system state estimation apparatus, method and program
Freyberger Asymptotic theory for differentiated products demand models with many markets
TWI736696B (en) Imaging overlay metrology method, computer program product, metrology module, and imaging overlay metrology system
JP2010264499A (en) Method of optimizing parameter of control model
KR101622068B1 (en) Setting calculation system learning device and learning method
JP2009072807A (en) Method and device for controlling planar shape of rolled stock and method of manufacturing thick steel plate
CN114264865A (en) Online self-calibration method for current collection device
JP2007245204A (en) Learning method for rolling-load model and device therefor
CN109807183A (en) Method for control steel plate thickness and device
WO2022012416A1 (en) Method and system for eccentric load error correction
US8041511B2 (en) Method for optimizing calibration maps for an algorithm of estimation of a control quantity of an internal combustion engine
CN108599976B (en) Network quality evaluation model construction method and device and computer equipment
CN113532614A (en) Method, processor and weighing system for predicting sensor data
KR102570644B1 (en) Method for compensating a cutting error of vessel main panel
JP2008077376A (en) Simulation apparatus, method and program
US20220269250A1 (en) Device and Method for Automatic Calculation of Measurement Confidence in Flexible Modular Plants and Machines
JP3983988B2 (en) Pass schedule calculation device in reverse mill
TWI533097B (en) Parameter tuning method and computer program product
JP4379698B2 (en) Method for manufacturing sheet product and sheet product manufacturing apparatus
JP6816629B2 (en) Manufacturing process learning controller
CN117092975A (en) Production control method and system based on digital twin
JP2000126809A (en) Set up device of rolling mill

Legal Events

Date Code Title Description
A201 Request for examination
E902 Notification of reason for refusal
E701 Decision to grant or registration of patent right
GRNT Written decision to grant
FPAY Annual fee payment

Payment date: 20190311

Year of fee payment: 4