JP2009193192A - Method and device for model prediction control - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method and a device for model prediction control, capable of compensating external disturbance of which application can be predicted, without need of predicting an operational amount corresponding to the external disturbance. <P>SOLUTION: A model predicted error between an actually measured value of a control amount PV when the external disturbance is applied and a control amount Y using an internal model is learned in advance and stored in a memory 5. Thereafter, the learned model predicted error is read out from the memory 5, and an estimation amount C of a change amount of the model predicted error from the present time to after a future prediction horizon is calculated by an estimation section 6. In an MPC controller 3, the model prediction control corresponding to the external disturbance to compensate the external disturbance is performed using the estimation amount C. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a model predictive control method and a model predictive control apparatus that predict a control amount using a model to be controlled and perform disturbance compensation.

  The disturbance includes a disturbance that occurs as a part of the operation sequence (hereinafter also referred to as “standard disturbance”), and the disturbance is a disturbance that occurs as a part of the operation sequence and can be predicted to be applied.

  Examples of such fixed disturbances include, for example, a packaging machine that lowers the cylinder temperature due to the introduction of a new material at the start of operation in an injection molding machine or seals the open end of a plastic sheet by thermocompression with a heater block. In this case, the temperature of the heater block is lowered due to heat being taken away by the plastic sheet at the start of operation, and the chamber temperature is lowered due to the introduction of the work into the chamber of the heating device. Further, some standard disturbances require a plurality of disturbances until the heat transfer method, which is a cause of the disturbance, settles down to a certain level.

  In feedback control, the correction of the manipulated variable is started after the influence of disturbance appears in the controlled variable. Therefore, the longer the dead time is, the greater the delay in response, and the greater the disturbance of the controlled variable due to the disturbance. .

A technology that predicts the manipulated variable equivalent to the disturbance and adds the predicted disturbance manipulated variable to the manipulated variable to cancel the disturbance in order to reduce the disturbance of the controlled variable due to such a regular disturbance. Has also been proposed (see, for example, Patent Document 1).
JP 2000-227801 A

  Although the change in the control amount due to the disturbance can be easily measured, in Patent Document 1, since the disturbance is predicted as the operation amount, the operation amount corresponding to the disturbance is predicted based on the change in the control amount due to the disturbance. Computation processing, that is, computation processing for converting the change in the control amount due to the disturbance into the manipulated variable is necessary, and there is a problem that many computations are required to accurately compensate the disturbance.

  The present invention has been made in view of the above points, and it is not necessary to predict an operation amount corresponding to a disturbance, and a model prediction control method and model prediction capable of compensating for a disturbance capable of predicting an application. An object is to provide a control device.

  (1) A model predictive control method according to the present invention is a model predictive control method that predicts a controlled variable using a model to be controlled and compensates for a disturbance that can be predicted to be applied, when the disturbance is applied. A learning step for learning in advance a model prediction error between an actually measured value of the controlled variable and a predicted value of the controlled variable using the model, and control is performed so as to compensate the applied disturbance using the learned model predicted error. A disturbance response control step.

  The disturbance that can be predicted to be applied is a fixed disturbance that occurs as part of the operation sequence. The disturbance may be a disturbance that changes little by little, for example, one that requires a plurality of disturbances until the heat transfer method that causes the disturbance settles down. In such a disturbance, since the change cannot be dealt with by only one learning step, a plurality of learning steps are executed until the change converges, and the model prediction error sequentially updated by the plurality of learning steps is reduced. It is sufficient to use the disturbance response control step. In addition, while executing the learning step to cope with the changing disturbance, the disturbance corresponding control step is executed using the model prediction error immediately before learning, that is, the learning step and the disturbance corresponding control step are simultaneously executed. Also good.

  In the learning step, it is preferable to learn the model prediction error over at least a period in which a disturbance is applied.

  When multiple types of disturbances are applied in the operation sequence, model prediction errors corresponding to each type of disturbance are learned in advance, and each type of disturbance is compensated using model prediction errors. You just have to do it.

  According to the model predictive control method of the present invention, by obtaining a model prediction error that changes in accordance with a disturbance in advance, it is possible to compensate for a regular disturbance using the model prediction error thereafter. In addition, the model prediction error can be calculated as the difference between the measured value of the controlled variable when the disturbance is applied and the predicted value using the model, so that the controlled variable when the disturbance is applied as in the conventional example. Arithmetic processing for converting the change into an operation amount corresponding to disturbance is not required.

  (2) In one embodiment of the model predictive control method of the present invention, the learning step stores the model predictive error, and the disturbance response control step uses the stored model predictive error to Control may be performed so that the amount of change in the model prediction error between the model prediction error and the future model prediction error is estimated, and the disturbance is compensated using the estimated amount of change in the model prediction error.

  The model prediction error may be stored as it is, or may be stored in an approximate form such as a polygonal line function or a polynomial in order to reduce the storage capacity.

  According to this embodiment, the amount of change in the model prediction error from the current time to after the future prediction horizon is estimated using the model prediction error stored in advance, and disturbance compensation is performed using the amount of change in the model prediction error. be able to.

  (3) In the disturbance response control step of the embodiment of (2), the estimated amount of change in the model prediction error, the predicted value of the control amount predicted using the model, the future target value and the current time The operation amount may be obtained based on the deviation from the control amount.

  The future target value may be a target value on the reference trajectory.

  According to this embodiment, in model predictive control in which a predicted value of a future control amount is obtained and an operation amount is determined so that the predicted value is as close as possible to a future target value in a prediction interval, disturbance using a model prediction error is performed. Compensation can be performed.

  (4) In a preferred embodiment of the present invention, the learning step and the disturbance response control step may be started in response to a timing signal prior to the application of the disturbance.

  The timing signal prior to the application of the disturbance may be used to control not only the start of the learning step and the disturbance response control step, but also the end, for example, each step is started when the timing signal is turned on, You may make it complete | finish each step when it turns OFF.

  The disturbance that occurs as a part of the operation sequence can predict the applied timing, and according to this embodiment, learning of the model prediction error in the learning step is performed using the timing signal prior to the application of the disturbance. Alternatively, disturbance compensation in the disturbance handling control step can be started.

  (5) A model predictive control apparatus according to the present invention is a model predictive control apparatus that predicts a control amount using a model to be controlled and compensates for a disturbance that can be predicted to be applied, when the disturbance is applied. Disturbance response control means for controlling so as to compensate for an applied disturbance using a model prediction error between an actually measured value of the controlled variable and a predicted value of the controlled variable using the model is provided.

  A model prediction error between the measured value of the controlled variable when a disturbance is applied and the predicted value of the controlled variable using the model is obtained in advance, stored in the model predictive control device or other higher-level device, and stored model prediction It is preferable to perform disturbance compensation using an error.

  According to the model prediction control apparatus of the present invention, by obtaining in advance a model prediction error that changes in accordance with a disturbance, it is possible to compensate for a fixed disturbance using the model prediction error thereafter. In addition, the model prediction error can be calculated as the difference between the measured value of the controlled variable when the disturbance is applied and the predicted value using the model, so that the controlled variable when the disturbance is applied as in the conventional example. Arithmetic processing for converting the change into an operation amount corresponding to disturbance is not required.

  (6) In one embodiment of the present invention, the calculation means for calculating the model prediction error between the measured value of the controlled variable when a disturbance is applied and the predicted value of the controlled variable using the model, Storage means for storing the model prediction error, and the disturbance response control means may be controlled to compensate the applied disturbance using the model prediction error of the storage means.

  According to this embodiment, the model prediction error that changes in accordance with the disturbance is stored in the storage unit in advance, and thereafter, the stored model prediction error can be used to compensate for the disturbance.

  (7) In the embodiment of the above (6), the disturbance response control means estimates the amount of change in the model prediction error between the current model prediction error and the future model prediction error using the model prediction error. You may make it provide an estimation part.

  According to this embodiment, the amount of change in the model prediction error from the current time to the future after the prediction horizon is estimated using the model prediction error stored in advance, and disturbance compensation is performed using the amount of change in the model prediction error. It can be carried out.

  (8) In the embodiment of the above (7), the disturbance response control means obtains a predicted value of the control amount using the model, obtains a deviation between the target value and the control amount at the current time, and determines the control amount. A prediction control unit that obtains an operation amount may be provided based on the predicted value, the deviation, and the estimated value of the change amount of the model prediction error estimated by the estimation unit.

  According to this embodiment, in model predictive control in which a predicted value of a future control amount is obtained and an operation amount is determined so that the predicted value is as close as possible to a future target value in a prediction interval, disturbance using a model prediction error is performed. Compensation can be performed.

  (9) In any one of the above embodiments (6) to (8), the calculation of the model prediction error by the calculation unit and the control for compensating the disturbance by the disturbance response control unit are performed prior to the application of the disturbance. You may make it start in response to a signal.

  According to this embodiment, the calculation of the model prediction error can be started using the timing signal prior to the application of the disturbance, and the model prediction error can be stored. Thereafter, the disturbance response control is performed using the timing signal. Disturbance compensation using the model prediction error by the means can be started.

  (10) In one embodiment of the present invention, the controlled variable is a temperature and controls the temperature of the controlled object.

  According to this embodiment, it can be suitably implemented as a temperature control device.

  According to the present invention, by obtaining a model prediction error that changes in accordance with a disturbance in advance, it is possible to compensate for a regular disturbance using the model prediction error thereafter. In addition, the model prediction error can be calculated as the difference between the measured value of the controlled variable when the disturbance is applied and the predicted value using the model, so that the controlled variable when the disturbance is applied as in the conventional example. Arithmetic processing for converting the change into an operation amount corresponding to disturbance is not required.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(Embodiment 1)
FIG. 1 is a block diagram showing a schematic configuration of a model predictive control apparatus according to an embodiment of the present invention.

  The model predictive control apparatus 1 of this embodiment has a model of the controlled object 2 as an internal model, and uses this internal model to control based on the set value SP and the controlled variable PV from the controlled object 2. The MPC controller 3 is provided as a predictive control unit that outputs the manipulated variable MV to the target 2, and normal model predictive control is performed by the MPC controller 3 when disturbance compensation is not performed.

  The model predictive control apparatus 1 is configured as follows so that model predictive control corresponding to a disturbance that compensates for a fixed disturbance that occurs as a part of an operation sequence can be performed.

  That is, a calculation unit 4 that calculates a model prediction error Emp between an actually measured value of the control amount PV that is actually measured when a fixed disturbance is applied in advance and a predicted value of the control amount Y that is predicted by using an internal model; A memory 5 for storing the model prediction error, and learning and storing the model prediction error when a fixed disturbance is applied in the learning mode in advance. This learning mode can be set by a user operation, for example.

  The disturbance is a regular disturbance, and the timing to which it is applied can be predicted. Therefore, the learning mode starts at the timing when the disturbance notification signal is turned on prior to the timing when the disturbance is applied, and includes a period during which the disturbance is applied. The model prediction error is calculated over the period L and stored in the memory 5.

  When the disturbance is, for example, a decrease in cylinder temperature due to the introduction of new material at the start of operation in the injection molding machine, for example, a timing signal for starting operation can be used as the disturbance notification signal.

  Next, in the normal operation mode, when the disturbance notification signal is turned ON, a disturbance predictive model predictive control that compensates the disturbance over a certain period K (K + d + H ≦ L) to be described later including a period in which the disturbance is applied. To do. In the model prediction control corresponding to the disturbance, the estimation unit 6 estimates the change amount C of the model prediction error using the model prediction error read from the memory 5 as described later, and the MPC controller 3 determines the model prediction error. Using the change amount C, the manipulated variable MV is calculated so as to compensate for the disturbance as will be described later, and disturbance predictive model prediction control is performed. The MPC controller 3 and the estimation unit 6 constitute disturbance response control means 7 that performs model prediction control corresponding to disturbance.

  The MPC controller 3, the calculation unit 4, the memory 5, the estimation unit 6, and the like are configured by a computer, for example.

  As shown in FIG. 2, the model predictive control by the MPC controller 3 measures a control amount PV (n) at the current time n, and gradually approaches the set value SP with the control amount PV (n) at the current time as a starting point. The reference trajectory shown by is calculated. Here, for simplicity, the control horizon Hu is set to 1.

  Next, using the internal model, the manipulated variable MV (n) at the current time n is determined so that the predicted value PV (n + H) of the controlled variable PV after the predicted horizon H matches the reference trajectory.

  The obtained operation amount MV (n) is actually added to the control object 2 and the value is held until the next sampling time n + 1.

  If the control amount PV (n + 1) is measured at the time n + 1, the operation amount is determined so that the future predicted value and the reference trajectory coincide with each other after the predicted horizon H, again considering the time n + 1 as the current time. The manipulated variable is added to the control object 2 until the sampling time. Thereafter, this procedure is repeated.

  Hereinafter, prediction control in the MPC controller 3 of this embodiment will be described in detail.

The MPC controller 3 performs predictive control using the internal model as described above. In this embodiment, the internal model is represented as an Nth-order ARX model expressed by the following equation discretized by the sampling time. Yes.
Y (n) = a1 * Y (n-1) + a2 * Y (n-2) +... + AN * Y (n-N)
+ B1 * U (n−1) + b2 * U (n−2) +... + BM * U (n−M)
here,
Y (n): internal model output value at time n U (n): manipulated variable at time n a1 to aN, b1 to bM: coefficients of internal model N, M: internal model order.

  Although the dead time can be included in the model, in this embodiment, the dead time is separated from the internal model and handled separately as described later.

  The determination of the ARX model is performed, for example, by measuring time series data of input / output with respect to the controlled object 2, that is, time series data of the manipulated variable MV and the controlled variable PV in advance and using a least square method or the like.

  In this embodiment, as a reference trajectory indicated by a broken line in FIG. 2, a trajectory that makes the deviation at the current time n exponentially approach 0 with a time constant Tr is used.

That is, the target value R (n + H) on the reference trajectory after the predicted horizon H can be obtained by the following equation.
R (n + H) = SP (n + H) −λ ^H * {SP (n) −PV (n)}
λ = exp (−Tc / Tr)
here,
PV (n): Control amount SP (n) at time n, SP (n + H): Target value R (n + H) at time n, n + H: Target value Tc on the reference trajectory ahead of the predicted horizon H: Sampling time.

Accordingly, the increment from the control amount PV (n) at the current time n, that is, the increment of the control amount PV necessary to make the control amount PV coincide with the target value R (n + H) on the reference trajectory after the predicted horizon H. (Deviation) ΔP (n + H) is
ΔP (n + H) = SP (n + H) −λ ^H * {SP (n) −PV (n)} − PV (n)
= (1-λ ^H ) {SP (n) −PV (n)} + SP (n + H) −SP (n)
It becomes.

  Next, calculation of the operation amount MV will be described.

  In the case of a linear control object, the behavior of the model output can be obtained by the following two additions.

(1) Free response The model output Yf (n + H) after the predicted horizon H when the current state is the initial value and 0 as the future manipulated variable MV is obtained by repeated calculation from the above ARX model formula. .
Yf (n + 1) = a1 * Y (n) + a2 * Y (n-1) +... + AN * Y (n-N + 1)
Yf (n + 2) = a1 * Yf (n + 1) + a2 * Y (n) +... + AN * Y (n−N + 2)
...
Yf (n + H) = a1 * Yf (n + H-1) + a2 * Yf (n + H-2) +... + AN * Y (n-N + H)
(2) Step response The initial state is set to 0, and the model output S (H) at time H in the step response of MV = 1 (100%) is obtained.
S (1) = b1
S (2) = a1 * S (1) + (b1 + b2)
...
S (H) = a1 * S (H-1) + a2 * S (H-2) + ... + aN * S (H-N) + (b1 + b2 + ... + bM)
If not MV = 1 (100%) but generally MV (n), the step response output at time H is MV (n) * S (H).

here,
MV (n): manipulated variable Yf (n + H) at time n: free response output of model after predicted horizon H S (H): step response output of model at time H

From the previous section, when the manipulated variable MV (n) is continued after time n, the model output (predicted value of the controlled variable) at time n + H is as follows.
Y (n + H) = Yf (n + H) + MV (n) * S (H)
The increment from Y (n), ie the expected increase in model output ΔM (n + H) after the predicted horizon H, is
ΔM (n + H) = Yf (n + H) + MV (n) * S (H) −Y (n)
It becomes.

When disturbance compensation is not performed, the above-described control amount PV for causing the increment ΔM (n + H) of the model output expected after the predicted horizon H to match the control amount PV with the target value on the reference trajectory after the predicted horizon H is obtained. The operation amount MV may be obtained so as to be equal to the increment ΔP (n + H).
That is, an operation amount MV that satisfies ΔM (n + H) = ΔP (n + H) may be obtained.

  In this embodiment, since the compensation for the fixed disturbance is performed, a case where the disturbance compensation is performed will be described.

An estimated value CH (n) of the change amount of the model prediction error between time points n and n + H is obtained as described later using the model prediction error learned in the learning mode described above, and the change amount of the model prediction error is estimated. Disturbance compensation is performed using the value CH (n). That is,
From ΔM (n + H) + CH (n) = ΔP (n + H),
Yf (n + H) + MV (n) * S (H) -Y (n) + CH (n) = (1-λ H) {SP (n) -PV (n)} + SP (n + H) -SP (n)
Solving for MV (n)
MV (n) = [(1-λ ^H ) {SP (n) -PV (n)} + SP (n + H) -SP (n) -Yf (n + H) + Y (n) -CH (n)] / S ( H)
In this embodiment, as described above, since the dead time is not included in the internal model, it is necessary to correct the calculation formula for the MV in the previous section to take into account the dead time d.

Therefore, in this embodiment, for actual process data, data at time n + d is used instead of time n.
MV (n) = [(1-λ ^H ) {SP (n + d) −PV (n + d)} + SP (n + H + d) −SP (n + d) −Yf (n + H) + Y (n) −CH (n)] / S ( H)
Here, a predicted value of PV (n + d) is required, and is usually obtained by the following calculation formula as a reasonable approximation.
PV (n + d) = PV (n) + Y (n) -Y (nd)
Again, disturbance compensation can be performed by obtaining the estimated value Cd (n) of the change amount of the model prediction error between time points n and n + d as described later using the model prediction error learned in the learning mode. ing.
PV (n + d) = PV (n) + Y (n) −Y (n−d) + Cd (n) from MV (n) = [(1−λ ^H ) {SP (n + d) − (PV (n) + Y ( n) -Y (nd) + Cd (n))} + SP (n + H + d) -SP (n + d) -Yf (n + H) + Y (n) -CH (n)] / S (H)
here,
Cd (n): Estimated value of model prediction error change from time n to n + d CH (n): Estimated value of model prediction error change from time n + d to H.

  In this embodiment, the estimated values of the two model prediction error variations are calculated as follows.

  That is, in the learning mode described above, the model prediction error obtained by the following equation is stored as Emp (i) (0 ≦ i <L) for the period L from the timing of the disturbance notification signal ON.

Emp (i) = Y (nd) -PV (n)
At the time of fixed disturbance response control, estimated values CH (n) and Cd (n) of the amount of change in model prediction error are calculated from Emp (i) (0 ≦ i <K) for the period K from the timing of the disturbance notification signal ON. Obtained and used for prediction calculation. However, K + d + H ≦ L
CH (n) = Emp (i + d + H) −Emp (i + d),
Cd (n) = Emp (i + d) −Emp (i)
FIG. 3 is a diagram illustrating estimated values CH (n) and Cd (n) of changes in model prediction error at the current time i. The current time i in FIG. 3 corresponds to the elapsed time after the disturbance notification signal is turned on.

  The estimated value CH (n) of the change amount of the model prediction error is calculated by subtracting the model prediction error Emp (i + d) at the i + d time point from the model prediction error Emp (i + d + H) at the i + d + H time point as shown in the above formula. Then, the estimated value Cd (n) of the model prediction error change amount is obtained by subtracting the model prediction error Emp (i) at the current i time from the model prediction error Emp (i + d) at the i + d time as shown in the above equation. Is calculated by

After calculating the manipulated variable MV (n) at time n, the next model output Y (n + 1) is obtained for the next calculation.
Y (n + 1) = a1 * Y (n) + a2 * Y (n-1) +... + AN * Y (n-N + 1)
+ B1 * MV (n) + b2 * MV (n−1) +... + BM * MV (n−M + 1)
However, MV (n) uses the value after the manipulated variable limit process.

  As described above, in model predictive control corresponding to disturbance, disturbance compensation is performed using estimated values CH (n) and Cd (n) of changes in model prediction error, and normal model predictive control without disturbance compensation is performed. In this period, the manipulated variable is calculated with the estimated values CH (n) and Cd (n) of the model prediction error change amount set to zero.

  As another embodiment, in a normal model prediction control period in which disturbance compensation is not performed, the estimated values CH (n) and Cd (n) of the model prediction error change amount may be estimated by a known method.

  FIG. 4 is a flowchart showing an outline of the processing procedure of the model predictive control method of this embodiment.

  First, the control object 2 is identified and an internal model is determined (step n1). Next, it is determined whether or not the disturbance notification signal is ON (step n2). When the disturbance notification signal is not ON, the manipulated variable MV is calculated as normal model predictive control without disturbance compensation (step n3). Then, the value of the internal model is updated, and the process returns to step n2 (step n4).

  In step n2, when the disturbance notification signal is ON, it is determined whether or not the learning mode is set (step n5), and when the learning mode is set, the model prediction error Emp is stored and the process proceeds to step n3 (step n6). ).

  In this embodiment, steps n2, n5, n6, n3, and n4 are repeated over the period in which the disturbance notification signal is ON, and the model prediction error is over the period in which the disturbance notification signal is ON. Emp is stored.

  If it is determined in step n5 that the learning mode is not set, it is determined that the control is disturbance response control, and an estimated value of a future model prediction error change amount is calculated from the time series data of the model prediction error stored in the learning mode (step n7). .

  Next, using the estimated value of the change amount of the model prediction error, the operation amount MV is calculated according to the calculation formula of the operation amount MV with disturbance compensation as described above (step n8), and the value of the internal model is updated. The process returns to step n2 (step n9).

  In this embodiment, steps n2, n5, n7, n8, and n9 are repeated over the period when the disturbance notification signal is ON, and the disturbance response signal is output over the period when the disturbance notification signal is ON. Model predictive control is performed.

  In this embodiment, when the disturbance notification signal is turned off, the model prediction control corresponding to the learning mode or disturbance is terminated. However, as another embodiment, for example, a specified time has elapsed, an end signal, or model prediction is used. The model prediction control corresponding to the learning mode or the disturbance may be terminated by automatically determining that the change amount of the error or the model prediction error has converged to a certain value or less.

  FIG. 5 is a diagram showing a simulation result showing the effect of performance improvement.

  FIG. 4A shows a change in the control amount PV, and FIG. 4B shows a change in the manipulated variable MV. In the figure, the first period T1 is a conventional model prediction control that does not perform disturbance compensation for a fixed disturbance, and the period T2 is a model prediction control according to the present embodiment that performs disturbance compensation for a fixed disturbance. Each is shown. Disturbance response control is performed from time 190 to 200 on the horizontal axis of the figure.

In this simulation, the control target is 1.54e ^−s /(1+19.6s) (1 + ^s ), the internal model is 1.54e ^−0.98s / (1 ⁺ 19.5s) (1 ⁺ 0.98s), and the horizontal axis of the figure This shows a case where a disturbance (square pulse-shaped disturbance) is applied to the control target input for a certain period of time from time 130 and 190.

  As shown in FIG. 5A, in the conventional example without disturbance response control, the start of the correction operation when a disturbance is applied is delayed, and the disturbance of the control amount PV becomes large. In this embodiment to be performed, it can be seen that the disturbance of the control amount PV can be greatly reduced in order to start the correction operation quickly in response to the disturbance notification signal.

  In the above-described embodiment, the model prediction error is learned in advance by one learning mode, and disturbance response control is performed using the learned model prediction error after the second time. In the case of a disturbance that gradually changes, for example, a disturbance that requires multiple times before the heat transfer that is the cause of the disturbance settles down, learning is performed to improve the accuracy of learning. The mode may be executed a plurality of times, and the disturbance response control may be performed using each model prediction error learned in each learning mode. Furthermore, disturbance response control may be performed using the latest learned model prediction error while learning the model prediction error in the learning mode.

  The present invention is applicable to model predictive control in which a control amount is predicted using a model to be controlled and disturbance compensation is performed. For example, an internal model is used like Smith dead time compensation control. It can also be applied to PID control in which PV after dead time is predicted and PID operation is performed on the predicted PV.

  The present invention is useful for predictive control using a model.

It is a block diagram of the model prediction control apparatus which concerns on one embodiment of this invention. It is a figure for demonstrating the algorithm of MPC. It is a figure which shows the estimated value CH (n) and Cd (n) of the variation | change_quantity of a model prediction error. It is a flowchart with which operation | movement description is provided. It is a figure which shows the simulation result which shows the effect of performance improvement.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Model prediction control apparatus 2 Control object 3 MPC controller 4 Calculation part 5 Memory 6 Estimation part 7 Disturbance response control means

Claims (10)

A model predictive control method that predicts a controlled variable using a model to be controlled and compensates for a disturbance that can predict application,
A learning step for learning in advance a model prediction error between an actual measurement value of a controlled variable when a disturbance is applied and a predicted value of a controlled variable using the model;
A model predictive control method comprising: a disturbance response control step for controlling to compensate for an applied disturbance using the learned model prediction error. In the learning step, the model prediction error is stored,
In the disturbance countermeasure control step, using the stored model prediction error, a change amount of the model prediction error between the current model prediction error and a future model prediction error is estimated, and the estimated change amount of the model prediction error is calculated. The model predictive control method according to claim 1, wherein control is performed so as to compensate for disturbance.   In the disturbance response control step, an operation is performed based on a change amount of the estimated model prediction error, a predicted value of the control amount predicted using the model, and a deviation between a future target value and a control amount at the current time. The model predictive control method according to claim 2, wherein the quantity is obtained.   The model prediction control method according to claim 1, wherein the learning step and the disturbance response control step are started in response to a timing signal prior to the application of the disturbance. A model predictive control device that predicts a controlled variable using a model to be controlled and compensates for a disturbance that can predict application,
Disturbance response control means for controlling to compensate for an applied disturbance using a model prediction error between an actual value of a controlled variable when a disturbance is applied and a predicted value of a controlled variable using the model A model predictive control device characterized by the above. A calculation means for calculating the model prediction error between the measured value of the controlled variable when a disturbance is applied and the predicted value of the controlled variable using the model;
Storage means for storing the calculated model prediction error,
The model prediction control apparatus according to claim 5, wherein the disturbance response control unit performs control so as to compensate for an applied disturbance using the model prediction error of the storage unit.   The model prediction according to claim 6, wherein the disturbance response control unit includes an estimation unit that estimates a change amount of a model prediction error between a current model prediction error and a future model prediction error using the model prediction error. Control device.   The disturbance response control means obtains a predicted value of the control amount using the model, obtains a deviation between the target value and the control amount at the current time, and calculates the predicted value of the control amount, the deviation, and the estimation unit. The model prediction control apparatus according to claim 7, further comprising: a prediction control unit that obtains an operation amount based on the estimated value of the change amount of the model prediction error estimated in step (8).   The calculation of the model prediction error by the calculation unit and the control for compensating for the disturbance by the disturbance response control unit are started in response to a timing signal prior to the application of the disturbance. Model predictive control device.   The model predictive control apparatus according to any one of claims 5 to 9, wherein the control amount is a temperature, and the temperature of the control target is controlled.

Images