WO2011027451A1 - 設定計算学習装置及び設定計算学習方法 - Google Patents
設定計算学習装置及び設定計算学習方法 Download PDFInfo
- Publication number
- WO2011027451A1 WO2011027451A1 PCT/JP2009/065427 JP2009065427W WO2011027451A1 WO 2011027451 A1 WO2011027451 A1 WO 2011027451A1 JP 2009065427 W JP2009065427 W JP 2009065427W WO 2011027451 A1 WO2011027451 A1 WO 2011027451A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- value
- calculated
- result output
- model
- calculation
- Prior art date
Links
Images
Classifications
-
- 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
-
- 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
- B21B37/16—Control of thickness, width, diameter or other transverse dimensions
- B21B37/18—Automatic gauge control
Definitions
- the present invention relates to a setting calculation learning device and a setting calculation learning method for accurately determining setting values necessary for operating machine equipment such as setup calculation in a process line.
- the model adaptive learning used in the setup calculation is a mathematical model (hereinafter referred to as a model formula) expressing a physical phenomenon in a mathematical formula, a result calculated value (hereinafter referred to as ACAL) calculated using an input variable as a result value,
- ACAL result calculated value
- the correction term of the model formula was corrected by comparing the actual value corresponding to ACAL (hereinafter referred to as ACT) obtained from the actual value measured by a measuring instrument or the like.
- This learning method is called model learning calculation here.
- a long-term learning function that divides the conditions (for example, steel grades and dimensions of steel materials in hot rolling equipment) (referred to as lots) and absorbs errors in the model formula for each lot.
- a method in which continuous (short-term) learning is performed regardless of lots and combined with a short-term learning function that absorbs errors that occur over time is often used (for example, JP-A-4-367901). .
- ACT and ACAL are calculated from actual values obtained at the time of learning, and learning is performed based on these deviations. Therefore, the physical object to be evaluated does not necessarily match the target value in the physical environment where the actual values are collected. Therefore, the physical object to be evaluated approaches the target value by using a correction term by learning. It is not always possible to properly calculate the set value of a machine. That is, in the conventional model learning calculation, the actual value obtained at the time of learning is used to correct only the model formula (model prediction accuracy) so as to meet the conditions, so it is difficult to accurately determine the set value. It was.
- the present invention has been made in view of the above problems, and an object of the present invention is to provide a setting calculation learning device and a setting calculation learning method for accurately determining a setting value of a machine facility to be controlled.
- a first feature of the setting calculation learning device is calculated using a first model formula based on an input variable actual value corresponding to a setting value set for a control target.
- the intermediate result output result calculated value calculated using the second model formula based on the calculated intermediate result output result value and the final result output actual value measured by the measurement unit to be controlled
- a model learning calculation unit that calculates a model learning correction term based on the deviation amount from the result output actual value, and corrects the second model formula based on the calculated model learning correction term, and a final result output value
- a vernier correction term is calculated by performing a smoothing process on a deviation amount between the initial target value with respect to the final result output actual value and a temporary target value based on the initial target value and the calculated vernier correction term.
- the second feature of the setting calculation learning device is calculated using the first model formula based on the input variable actual value corresponding to the setting value set for the control target.
- the intermediate result output result calculated value calculated using the second model formula based on the calculated intermediate result output result value and the final result output actual value measured by the measurement unit to be controlled
- a model learning calculation unit that calculates a model learning correction term based on the deviation amount from the result output actual value, and corrects the second model formula based on the calculated model learning correction term, and a final result output value
- a vernier correction term by smoothing a deviation amount from the final result output actual value, and a temporary target based on the initial target value for the final result output value and the calculated vernier correction term.
- the third feature of the setting calculation learning device is that it is calculated by using a model formula based on the actual input variable value corresponding to the setting value set for the control target.
- a model learning correction term is calculated based on the deviation amount between the result output actual calculation value and the final result output actual value measured by the measurement unit to be controlled, and the final result is calculated based on the calculated model learning correction term.
- a model learning calculation unit for correcting the result output value, an initial target value with respect to the final result output value, and a deviation amount between the final result output actual value are calculated by calculating a vernier correction term, and the initial target value Corrected by a vernier adaptive calculation unit that calculates a temporary target value based on a value and the calculated vernier correction term, the initial target value, the model formula, and the model learning calculation unit Based on the final result output value is to have a set value calculation unit for calculating the setting value for obtaining a temporary target value calculated by the vernier adaptive calculator.
- the first feature of the setting calculation learning method is calculated using the first model formula based on the input variable actual value corresponding to the setting value set for the control target.
- the intermediate result output result calculated value calculated using the second model formula based on the calculated intermediate result output result value and the final result output actual value measured by the measurement unit to be controlled A model learning calculation step for calculating a model learning correction term based on the deviation amount from the result output actual value, and correcting the second model formula based on the calculated model learning correction term, and a final result output value
- a vernier correction term is calculated by performing a smoothing process on a deviation amount between the initial target value with respect to the final result output actual value, and a temporary value is calculated based on the initial target value and the calculated vernier correction term.
- Based on the vernier adaptive calculation step for calculating the standard value, the initial target value, the first model formula, and the second model formula corrected by the model learning calculation step And a set value calculating step for calculating the set value for obtaining the calculated temporary target value.
- the second feature of the setting calculation learning method is calculated using the first model formula based on the input variable actual value corresponding to the setting value set for the control target.
- the intermediate result output result calculated value calculated using the second model formula based on the calculated intermediate result output result value and the final result output actual value measured by the measurement unit to be controlled
- a model learning calculation step for calculating a model learning correction term based on the deviation amount from the result output actual value, and correcting the second model formula based on the calculated model learning correction term, and a final result output value
- a vernier correction term by smoothing the deviation amount from the final result output actual value, and based on the initial target value for the final result output value and the calculated vernier correction term.
- Vernier adaptive calculation step based on the vernier adaptive calculation step for calculating the temporary target value, the initial target value, the first model formula, and the second model formula corrected by the model learning calculation step. And a set value calculating step for calculating the set value for obtaining the temporary target value calculated by the above.
- a third feature of the setting calculation learning method is that the intermediate characteristic calculated using the model formula based on the input variable actual value corresponding to the setting value set for the control target
- a model learning correction term is calculated based on the deviation amount between the result output actual calculation value and the final result output actual value measured by the measurement unit to be controlled, and the final result is calculated based on the calculated model learning correction term.
- a model learning calculation step for correcting the result output value, an initial target value for the final result output value, and a deviation amount between the final result output actual value is smoothed to calculate a vernier correction term, and the initial target value is calculated.
- FIG. 1 is a configuration diagram illustrating a configuration of a setup calculation system to which the setting calculation learning device according to the first embodiment is applied.
- a setup calculation system 10 to which a setting calculation learning device 1 according to the first embodiment is applied includes a setting calculation learning device 1, mechanical equipment 3, and a result collection and collection device 4. Yes.
- the machine facility 3 is a facility having one or more devices that operate based on a set value, such as a hot rolling facility. Further, the mechanical equipment 3 is provided with various measuring instruments such as a thermometer, a pressure gauge, and a speedometer.
- the achievement collection device 4 collects measurement values measured by various measuring instruments of the mechanical equipment 3.
- FIG. 2 is a configuration diagram showing the configuration of the setting calculation learning device 1 according to the first embodiment.
- the setting calculation learning device 1 includes a setup calculation device 2 and a model adaptive learning device 5.
- the setup calculation device 2 uses a pre-registered model formula to evaluate a physical object to be evaluated, that is, a value to be evaluated among measurement values measured by a measuring instrument provided in the result collection device 4.
- the set value of the mechanical equipment 3 is obtained so as to approach.
- the set value set by the setup calculation device 2 is output to the machine equipment 3.
- the setup calculation device 2 includes a set value calculation unit 8, and the set value calculation unit 8 includes an initial target value V ori AIM , a first model formula f, and a model learning calculation unit described later.
- the second model equation g corrected by 6 On the basis of the second model equation g corrected by 6, a set value X i for obtaining a temporary target value V AIM calculated by a vernier adaptive calculation unit 7 described later is calculated.
- the model adaptive learning device 5 includes a model learning calculation unit 6 and a vernier adaptive calculation unit 7 in terms of its functions.
- the model learning calculation unit 6 calculates the intermediate result output actual value calculated using the first model formula f based on the input variable actual value X i ACT corresponding to the set value X i set for the mechanical equipment 3.
- Y ACAL and the intermediate result output corresponding to the intermediate result output actual calculation value Y ACAL calculated using the second model formula g based on the final result output actual value V ACT measured by the measuring unit of the mechanical equipment 3
- a model learning correction term Z NEW is calculated based on the deviation amount from the actual value Y ACT, and the second model equation g is corrected based on the calculated model learning correction term Z NEW .
- Vernier adaptive calculation unit 7 calculates the initial target value V ori AIM on the final result output value V, and vernier correction term Z ver NEW by the shift amount of the final result output actual value V ACT to the smoothing process, the initial A temporary target value V AIM is calculated based on the target value V ori AIM and the calculated vernier correction term Z ver NEW .
- FIG. 3 is a flowchart showing the procedure of the setup calculation process by the setup calculation system 10 to which the setting calculation learning device 1 according to the first embodiment is applied.
- the setup calculation device 2 of the setting calculation learning device 1 corresponds to the initial target value V ori AIM set as an initial value by an external input.
- An initial set value that is an input variable of the mechanical equipment 3 is calculated and set in the mechanical equipment 3 (step S103).
- the performance collecting device 4 collects the measured values measured by the various measuring instruments of the mechanical equipment 3 (step S105).
- the model learning calculation section 6 of the model adaptive learning apparatus 5 set computing learning device 1, the middle and the result output actual value Y ACT is actual value for the intermediate results output value Y OUT of the first model type f, the The intermediate result output result calculation value Y ACAL, which is the actual result calculation value for the intermediate result output value Y IN of the model formula f of 1, is calculated (step S107).
- the model learning calculation unit 6 determines the final result output actual value V ACT that is the actual value with respect to the final result output value V of the physical object to be evaluated, and the measured value W i ACT collected by the actual result collection device 4. Then, based on the other condition input b k , the intermediate result output result value Y ACT is calculated using the following (Formula 1).
- model learning calculation unit 6 performs the following (Formula 2) based on the input variable actual value X i ACT that is the actual value of the set value X i that is the input variable to be obtained as a solution and the other condition input a j. ) Is used to calculate the intermediate result output result calculation value Y ACAL .
- the model learning calculation unit 6 calculates a model learning deviation amount (step S109). Specifically, the model learning calculation unit 6 uses the following (Equation 3) based on the intermediate result output result value Y ACT calculated in step S107 and the intermediate result output result calculation value Y ACAL . A model learning deviation amount Z CUR is calculated.
- the model learning calculation unit 6 uses the following (Equation 4) to calculate a model learning correction term Z NEW used in the current setup calculation process (step S111).
- the current setup calculation process refers to the process of steps S105 to S121 being executed.
- step S123 which will be described later, it is determined that the calculation period of the setup calculation has been reached, and the step executed as the next loop.
- the process of S105 to S121 is referred to as the next setup calculation process, and the process of steps S105 to S121 executed previously is referred to as the previous setup calculation process.
- Z NEW Z OLD + ⁇ ⁇ (Z CUR ⁇ Z OLD ) (Formula 4) here, Z NEW : Model learning correction term used in the current setup calculation process Z OLD : Model learning correction term used in the previous setup calculation process ⁇ : Smoothing coefficient
- the model learning calculation part 6 reflects in the 2nd model formula g (step S113). Specifically, the model learning calculation unit 6 calculates the following (Formula 6) based on Y IN which is an intermediate result output value before correction and the model learning correction term Z NEW calculated in step S111. And Y OUT which is an intermediate result output value is calculated, and this calculated Y OUT is used as a second model equation g for calculating a final result output value V as shown in the following (Equation 5). Adapt to.
- the vernier adaptive calculation unit 7 of the model adaptive learning device 5 of the setting calculation learning device 1 is evaluated with the final result output actual value VACT that is the actual value with respect to the final result output value V of the physical object to be evaluated.
- V ori AIM is the initial value of the target value for the final result output values V of the physical object, using the following the (equation 7), calculates a vernier shift amount Z ver CUR (step S115 ).
- the vernier adaptive calculation unit 7 performs a smoothing process on the vernier deviation amount Z ver CUR calculated based on the target value V ori AIM and the final result output result value V ACT using the following (Formula 8).
- a vernier correction term Z ver NEW is calculated (step S117).
- the vernier adaptive calculation unit 7 reflects the target value (step S119). Specifically, the vernier adaptive calculation unit 7 is evaluated using the following (Equation 9) based on the initial target value V ori AIM and the vernier correction term Z ver NEW calculated in step S117. The corrected temporary target value V AIM for the final output value V of the physical object is calculated.
- the setting value calculation unit 8 of the setup calculation device 2 of the setting calculation learning device 1 calculates a setting value (step S121). Specifically, the set value calculation unit 8 uses the following (Formula 10) based on the first model formula f and the second model formula g in which the model learning correction term Z NEW is reflected in Step S113. Using (Formula 13), a set value X i for obtaining the temporary target value V AIM calculated in step S119 is calculated.
- step S123 the model adaptive learning device 5 of the setting calculation learning device 1 determines that the calculation cycle of the setup calculation has been reached.
- the process proceeds to step S125 (step S123).
- step S123 When it is determined in step S123 that the setup calculation calculation period has been reached (NO), when the setup calculation process is requested to stop (step S125), the setting calculation learning device 1 ends the setup calculation process. To do.
- the model learning calculation unit 6 of the model adaptive learning device 5 corrects the second model formula
- the vernier adaptive calculation unit 7 calculates a temporary target value
- the set value calculation unit 8 of the setup calculation device 2 calculates the temporary target value calculated based on the first model formula and the corrected second model formula. Therefore, the set value for the machine equipment 3 to be controlled can be determined with high accuracy.
- the final result output value V is set as the plate thickness of the rolled sheet, and finish rolling is performed. It is possible to accurately calculate the set value of the roll gap of the finish rolling mill that performs.
- the vernier adaptive calculation unit 7 smoothes the deviation amount between the final result output value V and the final result output result value V ACT to thereby obtain the vernier correction term Z ver NEW .
- the setting calculation learning device 1 to be calculated will be described as an example.
- the setting calculation learning device 1 according to the first embodiment of the present invention shown in FIG. Since it is the same as the configuration of the applied setup calculation system 10, the description thereof is omitted.
- FIG. 4 is a configuration diagram showing a configuration of the setting calculation learning device 1 according to the second embodiment.
- the setting calculation learning device 1 includes a setup calculation device 2 and a model adaptive learning device 5.
- the configuration of the setup calculation device 2 is the same as the configuration of the setup calculation device 2 provided in the setting calculation learning device 1 according to the first embodiment of the present invention shown in FIG. Omitted.
- the model adaptive learning device 5 includes a model learning calculation unit 6 and a vernier adaptive calculation unit 7 in terms of its functions.
- the model learning calculation unit 6 calculates the intermediate result output actual value calculated using the first model formula f based on the input variable actual value X i ACT corresponding to the set value X i set for the mechanical equipment 3.
- Y ACAL and the intermediate result output corresponding to the intermediate result output actual calculation value Y ACAL calculated using the second model formula g based on the final result output actual value V ACT measured by the measuring unit of the mechanical equipment 3
- a model learning correction term Z NEW is calculated based on the deviation amount from the actual value Y ACT, and the second model equation g is corrected based on the calculated model learning correction term Z NEW .
- the vernier adaptive calculation unit 7 calculates a vernier correction term Z ver NEW by performing a smoothing process on the deviation amount between the final result output value V and the final result output result value V ACT, and an initial target for the final result output value V A temporary target value V AIM is calculated based on the value V ori AIM and the calculated vernier correction term Z ver NEW .
- the setup calculation processing by the setup calculation system 10 to which the setting calculation learning device 1 according to the second embodiment is applied is the setup calculation system to which the setting calculation learning device 1 according to the first embodiment shown in FIG. 3 is applied.
- the setup calculation processes of FIG. 10 the processes of steps S115 to S119 are different, and these processes will be described below.
- step S113 of the flowchart shown in FIG. 3 when the model learning calculation unit 6 of the setting calculation learning device 1 reflects Y OUT that is an intermediate result output value in the second model equation g, the setting calculation learning device 1
- the vernier adaptive calculation unit 7 of the model adaptive learning device 5 of FIG. 5A is the final result output actual value VACT that is the actual value for the final result output value V of the physical object to be evaluated, and the final result output value V of the physical object to be evaluated.
- the vernier shift amount Z ver CUR is calculated using the following (Formula 15) (step S115).
- the vernier adaptive calculation unit 7 uses the following (Equation 16) to smooth the vernier deviation amount Z ver CUR calculated based on the final result output value V and the final result output result value V ACT.
- the vernier correction term Z ver NEW is calculated (step S117).
- the vernier adaptive calculation unit 7 reflects the target value (step S119). Specifically, the vernier adaptive calculation unit 7 is evaluated using the following (Formula 17) based on the initial target value V ori AIM and the vernier correction term Z ver NEW calculated in step S117. The corrected temporary target value V AIM for the final output value V of the physical object is calculated.
- the model learning calculation unit 6 of the model adaptive learning device 5 corrects the second model formula
- the vernier adaptive calculation unit 7 calculates a temporary target value based on the final result output value
- the set value calculation unit 8 of the set-up calculation device 2 based on the first model formula and the corrected second model formula. Since the set value for obtaining the calculated temporary target value is calculated, the set value of the machine equipment 3 to be controlled can be accurately determined.
- the model learning calculation unit has been described by taking the setting calculation learning device 1 that corrects the second model formula based on the calculated model learning correction term as an example. Not exclusively.
- the setting calculation learning device 1 in which the model learning calculation unit corrects the final result output value based on the calculated model learning correction term will be described as an example.
- the setting calculation learning device 1 according to the first embodiment of the present invention shown in FIG. Since it is the same as the configuration of the applied setup calculation system 10, the description thereof is omitted.
- FIG. 5 is a configuration diagram showing the configuration of the setting calculation learning device 1 according to the third embodiment.
- the setting calculation learning device 1 includes a setup calculation device 2 and a model adaptive learning device 5.
- the setup calculation device 2 uses a pre-registered model formula so that the physical object to be evaluated, that is, the measured value measured by the measuring instrument provided in the result collecting device 4 is close to the target value.
- the set value of 3 is obtained.
- the set value set by the setup calculation device 2 is output to the machine equipment 3.
- the setup calculation device 2 includes a set value calculation unit 8, and the set value calculation unit 8 is corrected by an initial target value V ori AIM , a model formula, and a model learning calculation unit 6 described later. Based on the final result output value V, a set value X i for obtaining the temporary target value V AIM calculated by the vernier adaptive calculation unit 7 is calculated.
- the model adaptive learning device 5 includes a model learning calculation unit 6 and a vernier adaptive calculation unit 7 in terms of its functions.
- the model learning calculation unit 6 calculates the intermediate result output result calculation value Y ACAL calculated using the model formula f based on the input variable result value X i ACT corresponding to the set value X i set for the mechanical equipment 3.
- the model learning correction term Z NEW is calculated based on the deviation amount from the final result output result value V ACT measured by the measuring unit of the mechanical equipment 3, and the final result is calculated based on the calculated model learning correction term Z NEW.
- the result output value V is corrected.
- Vernier adaptive calculation unit 7 calculates the initial target value V ori AIM on the final result output value V, and vernier correction term Z ver NEW by the shift amount of the final result output actual value V ACT to the smoothing process, the initial A temporary target value V AIM is calculated based on the target value V ori AIM and the calculated vernier correction term Z ver NEW .
- FIG. 6 is a flowchart showing the procedure of the setup calculation process by the setup calculation system 10 to which the setting calculation learning device 1 according to the third embodiment is applied.
- a hot rolling facility is employed as the mechanical facility 3
- a finish rolling mill that performs finish rolling when the final sheet thickness value V, which is a final result output value, is the sheet thickness of the rolled sheet.
- a setting calculation learning apparatus 1 that accurately calculates the set value of the roll gap will be described as an example.
- the setup calculation device 2 of the setting calculation learning device 1 corresponds to the initial target value V ori AIM set as an initial value by an external input.
- An initial set value that is an input variable of the mechanical equipment 3 is calculated and set in the mechanical equipment 3 (step S203).
- the result collection device 4 collects the measured values measured by the various measuring instruments of the mechanical equipment 3 (step S205).
- the model learning calculation unit 6 of the model adaptive learning device 5 of the setting calculation learning device 1 performs an intermediate result output actual calculation value Y ACAL that is an actual calculation value for the intermediate output plate thickness value Y of the gauge meter plate thickness model formula f. Is calculated (step S207). Specifically, the model learning calculation unit 6 receives an input variable actual value X 1 ACT that is an actual value of the roll gap setting value X 1 that is an input variable to be obtained as a solution, and an actual value of the rolling load. Based on the input variable result value X 2 ACT and the other condition input a j , the intermediate result output result calculation value Y ACAL is calculated using the following (Formula 19).
- the model learning calculation unit 6 calculates a model learning deviation amount Z CUR (step S209). Specifically, the model learning calculation unit 6 is based on the intermediate result output calculation value Y ACAL calculated in step S207 and the plate thickness actual value V ACT supplied from the result collection device 4 as follows ( The model learning deviation amount Z CUR is calculated using Equation 20).
- the model learning calculation unit 6 calculates the model learning correction term Z NEW used in the current setup calculation process using the following (Formula 21) (step S211).
- Z NEW Z OLD + ⁇ ⁇ (Z CUR ⁇ Z OLD ) (Formula 21) here, Z NEW : Model learning correction term used in the current setup calculation process Z OLD : Model learning correction term used in the previous setup calculation process ⁇ : Smoothing coefficient
- the model learning calculation unit 6 reflects the model learning correction term Z NEW calculated in step S211 on the midway output sheet thickness value Y using the following (Equation 22), and sets the final sheet thickness value V to Correction is performed (step S213).
- the vernier adaptive calculation unit 7 of the model adaptive learning device 5 of the setting calculation learning device 1 performs the plate thickness actual value V ACT that is the actual value for the final plate thickness value V and the plate thickness target value for the final plate thickness value V. based of the initial thickness and the target value V ori AIM is the initial value, using the following the (formula 23), calculates a vernier shift amount Z ver CUR (step S215).
- the vernier adaptive calculation unit 7 uses the following (Equation 24), smoothing the vernier shift amount Z ver CUR, which is calculated based on the thickness target value V ori AIM and the plate thickness actual value V ACT Thus, the vernier correction term Z ver NEW is calculated (step S217).
- the vernier adaptive calculation unit 7 reflects the plate thickness target value (step S219). Specifically, the vernier adaptive calculation unit 7 uses the following (Equation 25) based on the initial plate thickness target value V ori AIM and the vernier correction term Z ver NEW calculated in step S217, The corrected temporary plate thickness target value V AIM for the plate thickness value V is calculated.
- the setting value calculation unit 8 of the setup calculation device 2 of the setting calculation learning device 1 calculates a roll gap setting value (step S221). Specifically, the set value calculation unit 8 uses the following (Equation 10) to (Equation 13) based on the gauge meter plate thickness model formula f and the final plate thickness value V corrected by the model learning calculation unit 6. ) is used to calculate the roll gap set value X 1 for obtaining a provisional thickness target value V AIM calculated in step S219.
- V AIM V (Formula 26)
- V p (Y, Z NEW )
- Y f (X 1 , X 2 ,..., A 1 , a 2 ,...)
- Equation 28 Next, if the model adaptive learning device 5 of the setting calculation learning device 1 determines that the calculation cycle of the setup calculation has been reached, the process proceeds to step S205, and if it is determined that the calculation cycle of the setup calculation has not been reached, the processing is performed. The process proceeds to step S225 (step S223).
- step S223 If it is determined in step S223 that the setup calculation cycle has been reached (NO), when the setup calculation process is requested to stop (step S225), the setting calculation learning device 1 ends the setup calculation process. To do.
- the model learning calculation unit 6 outputs the final result based on the calculated model learning correction term Z NEW.
- the value V is corrected
- the vernier adaptive calculation unit 7 calculates the temporary plate thickness target value V AIM
- the set value calculation unit 8 of the setup calculation device 2 uses the gauge meter plate thickness model formula f and the model learning calculation unit 6.
- the roll gap set value X 1 for obtaining the temporary plate thickness target value V AIM calculated by the vernier adaptive calculation unit 7 is calculated, so that the apparatus load is reduced.
- the set value of the machine equipment 3 to be controlled can be determined with high accuracy.
- a hot rolling facility is employed as the mechanical facility 3, and the final thickness value V, which is the final result output value, is set as the thickness of the rolled sheet.
- the setting calculation learning device 1 that accurately calculates the setting value of the roll gap of the finish rolling mill that performs finish rolling is described as an example.
- the mechanical equipment 3 is not limited to the hot rolling equipment, and is set. Any facility that has one or more devices that operate based on the set values may be used.
- the present invention can be applied to a control device that accurately determines a set value necessary for operating mechanical equipment such as a hot rolling device that rolls metal hot.
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Mechanical Engineering (AREA)
- Feedback Control In General (AREA)
- Control Of Metal Rolling (AREA)
Abstract
Description
図1は、第1の実施形態に係る設定計算学習装置が適用されたセットアップ計算システムの構成を示した構成図である。
図3は、第1の実施形態に係る設定計算学習装置1が適用されたセットアップ計算システム10によるセットアップ計算処理の処理手順を示したフローチャートである。
ここで、
YACT:第1のモデル式fの途中結果出力値YOUTに対する実績値(途中結果出力実績値)
g-1:第2のモデル式gの逆関数
VACT:評価される物理対象の最終結果出力値Vに対する実績値(最終結果出力実績値)
Wi ACT (i=1,2,3,・・・):その他の変数入力の実績値(実績収集装置4により収集された計測値)
bk (k=1,2,3,・・・):その他の条件入力
とする。
ここで、
YACAL=:第1のモデル式fの途中結果出力値YINに対する実績計算値(途中結果出力実績計算値)
f:学習の偏移量を評価される(補正項が施される)物理量の第1のモデル式
Xi ACT(i=1,2,3,・・・):解として求めるべき入力変数である設定値Xiの実績値(入力変数実績値)
aj (j=1,2,3,・・・):その他の条件入力
とする。
ここで、
ZCUR:モデル学習偏移量
h:減算又は除算(i.e. ZCUR=YACT-YACAL 又は ZCUR=YACT/YACAL)
とする。
ここで、
ZNEW:今回のセットアップ計算処理で使用するモデル学習補正項
ZOLD:前回のセットアップ計算処理で使用したモデル学習補正項
β:平滑化係数
とする。
YOUT=p(YIN,ZNEW) (数式6)
ここで、
YOUT:補正されたYIN
p:(数式3)が減算であれば加算、(数式3)が除算であれば乗算(i.e. YOUT=YIN+ZNEW又は YOUT=YIN×ZNEW)
とする。
ここで、
Zver CUR:バーニア偏移量
q:減算又は除算(i.e. Zver CURR=VACT-Vori AIM 又は Zver CUR=VACT/Vori AIM)
VACT:評価される物理対象の最終結果出力値Vに対する実績値(最終結果出力実績値)
Vori AIM:評価される物理対象の最終結果出力値Vに対する初期目標値
とする。
ここで、
Zver NEW:今回のセットアップ計算で使用するバーニア補正項
α:平滑化係数
次に、バーニア適応計算部7は、目標値へ反映する(ステップS119)。具体的には、バーニア適応計算部7は、初期目標値Vori AIMと、ステップS117において算出されたバーニア補正項Zver NEWとに基づいて、下記の(数式9)を用いて、評価される物理対象の最終結果出力値Vに対する補正後の仮目標値VAIMを算出する。
ここで、
VAIM:評価される物理対象の最終結果出力値Vに対する補正後の仮目標値
r:(数式7)が減算であれば加算、(数式7)が除算であれば乗算(i.e. VAIM=Vori AIM+Zver NEW又は VAIM=Vori AIM×Zver NEW)
とする。
V=g(YOUT,W1,W2,・・・,b1,b2,・・・) (数式11)
YOUT=p(YIN,ZNEW) (数式12)
YIN=f(X1,X2,・・・,a1,a2,・・・) (数式13)
次に、設定計算学習装置1のモデル適応学習装置5は、セットアップ計算の計算周期に達したと判定すると、処理をステップS123へ移行し、セットアップ計算の計算周期に達していないと判定すると処理をステップS125へ移行する(ステップS123)。
本発明に係る第1の実施形態では、バーニア適応計算部7が、最終結果出力値Vに対する初期目標値Vori AIMと、最終結果出力実績値VACTとの偏移量をスムージング処理することによりバーニア補正項Zver NEWを算出する設定計算学習装置1を例に挙げて説明したが、これに限らない。
第2の実施形態に係る設定計算学習装置1が適用されたセットアップ計算システム10によるセットアップ計算処理は、図3に示した第1の実施形態に係る設定計算学習装置1が適用されたセットアップ計算システム10によるセットアップ計算処理のうち、ステップS115~S119の処理が異なるので、これらの処理について以下に説明する。
ここで、
Zver CUR:バーニア偏移量
q:減算又は除算(i.e. Zver CURR=VACT-Vori AIM 又は Zver CUR=VACT/Vori AIM)
VACT:評価される物理対象の最終結果出力値Vに対する実績値(最終結果出力実績値)
V:評価される物理対象の最終結果出力値
とする。
ここで、
Zver NEW:今回のセットアップ計算で使用するバーニア補正項
α:平滑化係数
次に、バーニア適応計算部7は、目標値へ反映する(ステップS119)。具体的には、バーニア適応計算部7は、初期目標値Vori AIMと、ステップS117において算出されたバーニア補正項Zver NEWとに基づいて、下記の(数式17)を用いて、評価される物理対象の最終結果出力値Vに対する補正後の仮目標値VAIMを算出する。
ここで、
VAIM:評価される物理対象の最終結果出力値Vに対する補正後の仮目標値
r:(数式15)が減算であれば加算、(数式15)が除算であれば乗算(i.e. VAIM=Vori AIM+Zver NEW又は VAIM=Vori AIM×Zver NEW)
とする。
本発明に係る第1の実施形態では、モデル学習計算部が、算出したモデル学習補正項に基づいて第2のモデル式を補正する設定計算学習装置1を例に挙げて説明したが、これに限らない。
図6は、第3の実施形態に係る設定計算学習装置1が適用されたセットアップ計算システム10によるセットアップ計算処理の処理手順を示したフローチャートである。ここでは、一例として、機械設備3として熱間圧延設備が採用され、最終結果出力値である最終板厚値Vを圧延された圧延板の板厚としたときの、仕上げ圧延を行う仕上げ圧延ミルのロールギャップの設定値を精度良く算出する設定計算学習装置1を例に挙げて説明する。
ここで、
YACAL=:ゲージメータ板厚モデル式fの途中出力板厚値Yに対する実績計算値(途中結果出力実績計算値)
f:ゲージメータ板厚モデル式
X1 ACT:ロールギャップの実績値(入力変数実績値)
X1 ACT:圧延荷重の実績値(入力変数実績値)
aj (j=1,2,3,・・・):その他の条件入力
とする。
ここで、
ZCUR:モデル学習偏移量
h:減算(i.e. ZCUR=VACT-YACAL)
とする。
ここで、
ZNEW:今回のセットアップ計算処理で使用するモデル学習補正項
ZOLD:前回のセットアップ計算処理で使用したモデル学習補正項
β:平滑化係数
とする。
ここで、
V:補正されたY、即ち最終板厚値
p:加算(i.e. V=Y+ZNEW)
とする。
ここで、
Zver CUR:バーニア偏移量
q:減算(i.e. Zver CURR=VACT-Vori AIM)
VACT:最終板厚値Vに対する実績値(最終結果出力実績値)
Vori AIM:最終板厚値Vに対する初期板厚目標値
とする。
ここで、
Zver NEW:今回のセットアップ計算で使用するバーニア補正項
α:平滑化係数
次に、バーニア適応計算部7は、板厚目標値へ反映する(ステップS219)。具体的には、バーニア適応計算部7は、初期板厚目標値Vori AIMと、ステップS217において算出されたバーニア補正項Zver NEWとに基づいて、下記の(数式25)を用いて、最終板厚値Vに対する補正後の仮板厚目標値VAIMを算出する。
ここで、
VAIM:最終板厚値Vに対する補正後の仮板厚目標値
r:加算(i.e. VAIM=Vori AIM+Zver NEW)
とする。
V=p(Y,ZNEW) (数式27)
Y=f(X1,X2,・・・,a1,a2,・・・) (数式28)
次に、設定計算学習装置1のモデル適応学習装置5は、セットアップ計算の計算周期に達したと判定すると、処理をステップS205へ移行し、セットアップ計算の計算周期に達していないと判定すると処理をステップS225へ移行する(ステップS223)。
Claims (6)
- 制御対象に対して設定する設定値に対応する入力変数実績値に基づいて第1のモデル式を用いて算出された途中結果出力実績計算値と、前記制御対象の計測部により計測された最終結果出力実績値に基づいて第2のモデル式を用いて算出された前記途中結果出力実績計算値に対応する途中結果出力実績値との偏移量に基づいてモデル学習補正項を算出し、この算出されたモデル学習補正項に基づいて前記第2のモデル式を補正するモデル学習計算部と、
最終結果出力値に対する初期目標値と、前記最終結果出力実績値との偏移量をスムージング処理することによりバーニア補正項を算出し、前記初期目標値と、前記算出されたバーニア補正項とに基づいて仮目標値を算出するバーニア適応計算部と、
前記初期目標値と、前記第1のモデル式と、前記モデル学習計算部により補正された第2のモデル式とに基づいて、前記バーニア適応計算部により算出された仮目標値を得るための前記設定値を算出する設定値計算部と、
を備えたことを特徴とする設定計算学習装置。 - 制御対象に対して設定する設定値に対応する入力変数実績値に基づいて第1のモデル式を用いて算出された途中結果出力実績計算値と、前記制御対象の計測部により計測された最終結果出力実績値に基づいて第2のモデル式を用いて算出された前記途中結果出力実績計算値に対応する途中結果出力実績値との偏移量に基づいてモデル学習補正項を算出し、この算出されたモデル学習補正項に基づいて前記第2のモデル式を補正するモデル学習計算部と、
最終結果出力値と、前記最終結果出力実績値との偏移量をスムージング処理することによりバーニア補正項を算出し、前記最終結果出力値に対する初期目標値と、前記算出されたバーニア補正項とに基づいて仮目標値を算出するバーニア適応計算部と、
前記初期目標値と、前記第1のモデル式と、前記モデル学習計算部により補正された第2のモデル式とに基づいて、前記バーニア適応計算部により算出された仮目標値を得るための前記設定値を算出する設定値計算部と、
を備えたことを特徴とする設定計算学習装置。 - 制御対象に対して設定する設定値に対応する入力変数実績値に基づいてモデル式を用いて算出された途中結果出力実績計算値と、前記制御対象の計測部により計測された最終結果出力実績値との偏移量に基づいてモデル学習補正項を算出し、この算出されたモデル学習補正項に基づいて最終結果出力値を補正するモデル学習計算部と、
前記最終結果出力値に対する初期目標値と、前記最終結果出力実績値との偏移量をスムージング処理することによりバーニア補正項を算出し、前記初期目標値と、前記算出されたバーニア補正項とに基づいて仮目標値を算出するバーニア適応計算部と、
前記初期目標値と、前記モデル式と、前記モデル学習計算部により補正された最終結果出力値とに基づいて、前記バーニア適応計算部により算出された仮目標値を得るための前記設定値を算出する設定値計算部と、
を備えたことを特徴とする設定計算学習装置。 - 制御対象に対して設定する設定値に対応する入力変数実績値に基づいて第1のモデル式を用いて算出された途中結果出力実績計算値と、前記制御対象の計測部により計測された最終結果出力実績値に基づいて第2のモデル式を用いて算出された前記途中結果出力実績計算値に対応する途中結果出力実績値との偏移量に基づいてモデル学習補正項を算出し、この算出されたモデル学習補正項に基づいて前記第2のモデル式を補正するモデル学習計算ステップと、
最終結果出力値に対する初期目標値と、前記最終結果出力実績値との偏移量をスムージング処理することによりバーニア補正項を算出し、前記初期目標値と、前記算出されたバーニア補正項とに基づいて仮目標値を算出するバーニア適応計算ステップと、
前記初期目標値と、前記第1のモデル式と、前記モデル学習計算ステップにより補正された第2のモデル式とに基づいて、前記バーニア適応計算ステップにより算出された仮目標値を得るための前記設定値を算出する設定値計算ステップと、
を有することを特徴とする設定計算学習方法。 - 制御対象に対して設定する設定値に対応する入力変数実績値に基づいて第1のモデル式を用いて算出された途中結果出力実績計算値と、前記制御対象の計測部により計測された最終結果出力実績値に基づいて第2のモデル式を用いて算出された前記途中結果出力実績計算値に対応する途中結果出力実績値との偏移量に基づいてモデル学習補正項を算出し、この算出されたモデル学習補正項に基づいて前記第2のモデル式を補正するモデル学習計算ステップと、
最終結果出力値と、前記最終結果出力実績値との偏移量をスムージング処理することによりバーニア補正項を算出し、前記最終結果出力値に対する初期目標値と、前記算出されたバーニア補正項とに基づいて仮目標値を算出するバーニア適応計算ステップと、
前記初期目標値と、前記第1のモデル式と、前記モデル学習計算ステップにより補正された第2のモデル式とに基づいて、前記バーニア適応計算ステップにより算出された仮目標値を得るための前記設定値を算出する設定値計算ステップと、
を有することを特徴とする設定計算学習方法。 - 制御対象に対して設定する設定値に対応する入力変数実績値に基づいてモデル式を用いて算出された途中結果出力実績計算値と、前記制御対象の計測部により計測された最終結果出力実績値との偏移量に基づいてモデル学習補正項を算出し、この算出されたモデル学習補正項に基づいて最終結果出力値を補正するモデル学習計算ステップと、
前記最終結果出力値に対する初期目標値と、前記最終結果出力実績値との偏移量をスムージング処理することによりバーニア補正項を算出し、前記初期目標値と、前記算出されたバーニア補正項とに基づいて仮目標値を算出するバーニア適応計算ステップと、
前記初期目標値と、前記モデル式と、前記モデル学習計算ステップにより補正された最終結果出力値とに基づいて、前記バーニア適応計算ステップにより算出された仮目標値を得るための前記設定値を算出する設定値計算ステップと、
を有することを特徴とする設定計算学習方法。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN200980161316.6A CN102483612B (zh) | 2009-09-03 | 2009-09-03 | 设定计算学习装置和设定计算学习方法 |
KR1020127005905A KR101318047B1 (ko) | 2009-09-03 | 2009-09-03 | 설정 계산 학습 장치 및 설정 계산 학습 방법 |
PCT/JP2009/065427 WO2011027451A1 (ja) | 2009-09-03 | 2009-09-03 | 設定計算学習装置及び設定計算学習方法 |
JP2011529741A JP5309219B2 (ja) | 2009-09-03 | 2009-09-03 | 設定計算学習装置及び設定計算学習方法 |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2009/065427 WO2011027451A1 (ja) | 2009-09-03 | 2009-09-03 | 設定計算学習装置及び設定計算学習方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2011027451A1 true WO2011027451A1 (ja) | 2011-03-10 |
Family
ID=43649013
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2009/065427 WO2011027451A1 (ja) | 2009-09-03 | 2009-09-03 | 設定計算学習装置及び設定計算学習方法 |
Country Status (4)
Country | Link |
---|---|
JP (1) | JP5309219B2 (ja) |
KR (1) | KR101318047B1 (ja) |
CN (1) | CN102483612B (ja) |
WO (1) | WO2011027451A1 (ja) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6486731B2 (ja) * | 2015-03-13 | 2019-03-20 | 株式会社東芝 | 機器特性モデル学習装置、機器特性モデル学習方法、及びプログラム |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001025807A (ja) * | 1999-07-15 | 2001-01-30 | Toshiba Corp | リバース式圧延機の学習制御装置 |
JP2002301509A (ja) * | 2001-04-05 | 2002-10-15 | Toshiba Corp | 圧延機の制御装置 |
JP2003340508A (ja) * | 2002-05-27 | 2003-12-02 | Toshiba Ge Automation Systems Corp | 圧延機設定計算装置用学習制御装置 |
Family Cites Families (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0661566B2 (ja) * | 1987-11-18 | 1994-08-17 | 株式会社日立製作所 | 圧延機のセットアップ装置 |
JP2839746B2 (ja) * | 1991-06-17 | 1998-12-16 | 株式会社神戸製鋼所 | プロセスラインにおける学習制御方法 |
JP3067879B2 (ja) * | 1992-01-31 | 2000-07-24 | 新日本製鐵株式会社 | ストリップ圧延における形状制御方法 |
CN1105296A (zh) * | 1993-10-05 | 1995-07-19 | 日立有限公司 | 辊轧机控制方法及装置 |
JPH08243614A (ja) * | 1995-03-14 | 1996-09-24 | Kawasaki Steel Corp | 形状・板厚精度に優れるリバース式圧延方法 |
JP3300208B2 (ja) * | 1995-09-06 | 2002-07-08 | 株式会社神戸製鋼所 | プロセスラインにおける学習制御方法 |
JP3253013B2 (ja) * | 1997-10-06 | 2002-02-04 | 川崎製鉄株式会社 | 熱間圧延における板クラウン・形状制御方法 |
JPH11197727A (ja) * | 1998-01-13 | 1999-07-27 | Toshiba Corp | ワークロールクラウン量演算装置および方法 |
JP3942356B2 (ja) * | 2000-11-21 | 2007-07-11 | 東芝三菱電機産業システム株式会社 | リバース式圧延機における板厚制御装置 |
JP2003245706A (ja) * | 2002-02-26 | 2003-09-02 | Kobe Steel Ltd | 複数パス圧延における圧延中の再パススケジューリング方法 |
JP2004255436A (ja) * | 2003-02-27 | 2004-09-16 | Kobe Steel Ltd | タンデム式熱間圧延機の板クラウン制御方法 |
KR101161700B1 (ko) * | 2004-07-09 | 2012-07-03 | 가부시끼가이샤 히다치 세이사꾸쇼 | 열간 압연밀의 크라운제어장치 및 그 제어방법 |
JP4507193B2 (ja) * | 2005-03-31 | 2010-07-21 | 住友金属工業株式会社 | マンドレルミルの圧延制御方法 |
CN100474313C (zh) * | 2005-10-31 | 2009-04-01 | 宝山钢铁股份有限公司 | 一种粗轧带钢的宽度优化设定方法 |
CN100467153C (zh) * | 2006-11-17 | 2009-03-11 | 燕山大学 | 可进行轧辊热膨胀动态补偿的冷轧带钢厚度控制方法 |
-
2009
- 2009-09-03 WO PCT/JP2009/065427 patent/WO2011027451A1/ja active Application Filing
- 2009-09-03 KR KR1020127005905A patent/KR101318047B1/ko active IP Right Grant
- 2009-09-03 CN CN200980161316.6A patent/CN102483612B/zh active Active
- 2009-09-03 JP JP2011529741A patent/JP5309219B2/ja active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001025807A (ja) * | 1999-07-15 | 2001-01-30 | Toshiba Corp | リバース式圧延機の学習制御装置 |
JP2002301509A (ja) * | 2001-04-05 | 2002-10-15 | Toshiba Corp | 圧延機の制御装置 |
JP2003340508A (ja) * | 2002-05-27 | 2003-12-02 | Toshiba Ge Automation Systems Corp | 圧延機設定計算装置用学習制御装置 |
Also Published As
Publication number | Publication date |
---|---|
KR20120059531A (ko) | 2012-06-08 |
JPWO2011027451A1 (ja) | 2013-01-31 |
CN102483612B (zh) | 2014-01-01 |
JP5309219B2 (ja) | 2013-10-09 |
KR101318047B1 (ko) | 2013-10-14 |
CN102483612A (zh) | 2012-05-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP2009113101A (ja) | 圧延荷重の学習制御方法および装置、ならびに鋼板の製造方法 | |
CN113649420B (zh) | 一种平整机轧制力获取方法及装置 | |
CN103831304B (zh) | 一种热连轧中间坯目标宽度计算方法及系统 | |
JP6229799B2 (ja) | 平坦度制御装置 | |
JP2017224091A (ja) | 圧延ラインの数学モデル算出装置および圧延材の温度制御装置 | |
JP4983589B2 (ja) | 冷間連続圧延設備の制御装置 | |
JP5309219B2 (ja) | 設定計算学習装置及び設定計算学習方法 | |
KR101877341B1 (ko) | 교정 롤 시스템의 교정 롤 정렬 방법 | |
KR20040043911A (ko) | 신경 회로망을 이용한 압연 하중 예측 장치 및 방법 | |
JP2013226596A (ja) | 圧延荷重の学習制御装置および学習制御方法、並びにこれを用いた金属板製造方法 | |
JP6385847B2 (ja) | 圧延機の板厚制御方法 | |
JP2012217994A (ja) | 鋼板の圧延制御方法、装置及びプログラム | |
CN105522002A (zh) | 一种冷轧厚度自动控制方法 | |
JP4224375B2 (ja) | 板厚・クラウン制御方法 | |
JP2015147249A (ja) | 圧延機の制御方法、圧延機の制御装置、及び圧延材の製造方法 | |
JP2000126809A (ja) | 圧延機のセットアップ装置 | |
JP2006021243A (ja) | 熱間圧延ミルのクラウン制御装置およびその制御方法 | |
KR100929015B1 (ko) | 압연재 소성계수 보정에 의한 예측압연하중 보정방법 | |
JP5565214B2 (ja) | 圧延機の板厚制御方法 | |
JP5617307B2 (ja) | 鋼板の圧延方法及びパススケジュール算出方法 | |
JP5846303B2 (ja) | 設定計算システムの学習装置及び学習方法 | |
JP5821527B2 (ja) | ロール偏芯除去方法及びロール偏芯除去制御装置 | |
JP2005014058A (ja) | 圧延機の板厚制御方法及び板厚制御装置 | |
JPH11104720A (ja) | 圧延機のセットアップ装置 | |
TW202238298A (zh) | 精軋機出口溫度估算方法 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 200980161316.6 Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 09848982 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2011529741 Country of ref document: JP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 20127005905 Country of ref document: KR Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2855/CHENP/2012 Country of ref document: IN |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 09848982 Country of ref document: EP Kind code of ref document: A1 |