JPH07160307A - Controller - Google Patents

Abstract

PURPOSE:To perform excellent control by automatically correcting the time constant of an internal model even when the time constant of a controlled system process is unknown. CONSTITUTION:A feedback control system consists of a command filter part 1, a 1st subtraction process part 3, a manipulated variable arithmetic part 4, an internal model storage part 6a, an internal model output arithmetic part 6b, and a 2nd subtraction process part 8. The gain of the internal model stored in the internal model storage part 6a is corrected by a step width calculation part 9, a noise process part 10, a response start area detection part 11, and a model gain calculation part 12. A process gain identification part 13 calculates the gain estimated value of the controlled system process. A model time constant correction part 14 calculates and outputs the corrected time constant to the internal model storage part 6a to correct the time constant of the internal model.

Description

Detailed Description of the Invention

[0001]

The present invention relates to an IMC (Internal Model)
The present invention relates to a controller using a control algorithm having a control structure, and particularly to a controller capable of automatically correcting the time constant of an internal model to perform good control even when the time constant of a control target process is unknown.

[0002]

2. Description of the Related Art Conventionally, there has been proposed a controller using an IMC structure control algorithm for performing control by incorporating an internal model in which a control target process is expressed by a mathematical expression.
If this IMC controller is used, a large dead time (time from when hot air comes out of the air conditioner to when the indoor temperature rises) in the process to be controlled (e.g., if this controller is an indoor air conditioner corresponds to the indoor environment) It has an excellent advantage that it can be dealt with even if it exists.

FIG. 19 is a block diagram of a control system using a conventional IMC controller. Reference numeral 33 denotes a first subtraction processing unit that subtracts a feedback amount, which will be described later, from a target value (indoor temperature setting value), and 32 denotes a filter unit that prevents a change in the output of the first subtraction processing unit 33 from being rapidly transmitted. ,
Reference numeral 34 is an operation unit for calculating an operation amount (temperature of hot air or cold air coming out of the indoor air conditioner) which is the output of the controller based on the output of the filter unit 32, and 36 is a mathematical expression approximating the control target process. And a second subtraction process for outputting a feedback amount by subtracting the reference control amount from the internal model 36 from the control amount, and outputting a reference control amount corresponding to the control amount (indoor temperature) as a control result. Part 40 is a process to be controlled.

Further, F, Gc, Gm, and Gp are the transfer function of the filter unit 32, the operating unit 34, the internal model 36, and the process 40 to be controlled, r is the target value, u is the manipulated variable, and d is the d.
Is, for example, a disturbance corresponding to the outdoor environment with respect to the indoor environment,
y is a control amount, ym is a reference control amount, and e is a feedback amount.

Next, the operation of such an IMC controller will be described. First, the first subtraction processing unit 33 subtracts the feedback amount e from the target value r, and the result is output to the filter unit 32. Next, the operation amount u is calculated by the operation unit 34 from the output of the filter unit 32, and is output to the control target process 40 and the internal model 36 of the controller. Then, the second subtraction processing unit 38 subtracts the reference control amount ym from the internal model 36 that approximates the control target process 40 from the control amount y of the control target process 40, and the result is the feedback amount e. A feedback control system that is fed back to the first subtraction processing unit 33 is configured as.

Ideally, the internal model 36 of such an IMC controller is mathematically expressed so as to be exactly the same as the control target process 40, and the operating unit 34 is also used.
Is ideally the inverse characteristic (1 / Gm) of the transfer function of the internal model 36, but the inverse of the dead time element of the internal model 36 is not possible, so normally the dead time element is ignore. Therefore, the control amount y can be obtained by the following equation from the target value r and the disturbance d with such a configuration. y = F * Gp * Gc * r / {1 + F * Gc * (Gp-Gm)} + (1-F * Gm * Gc) * d / {1 + F * Gc * (Gp-Gm)} ... (1 )

Here, the transfer function Gm of the internal model 36 is equal to the transfer function Gp of the control target process 40, and the transfer function Gc of the operating section 34 is the reciprocal of the transfer function of the internal model 36 (1 / Gm = 1 / Gp). Assuming an ideal state equal to, equation (1) becomes y = F × r + (1-F) × d (2)

Further, under ideal conditions where the target value r does not change suddenly, the filter unit 32 is unnecessary and F = 1 can be set, so that the control amount y becomes equal to the target value r (y =
r), it is possible to realize the control without any influence of the disturbance d. Focusing on the disturbance d, the controlled process 4
0 and the internal model 36 have the same dead time with respect to the manipulated variable u even if there is a large dead time, the feedback amount e output from the second subtraction processing unit 38 is equal to the disturbance d.
It is understood that the disturbance d can be suppressed.

Such an IMC controller is usually
Robust stability showing stability when the model identification error between the controlled object process 40 and the internal model 36 becomes large,
And similarly, the design is performed based on the design condition for the robust performance indicating the performance when the error becomes large. In addition, the internal model 36
When determining, the model identification error of the internal model 36 with respect to the control target process 40 is unavoidable to some extent, but control is not performed as expected when the estimation of the model identification error is incorrect. Is done by experts with control knowledge.

[0010]

Since the conventional IMC controller is constructed as described above, there is an error in the internal model identification because the time constant of the controlled process is unknown, that is, there is an error in the time constant of the internal model. If the estimate is not appropriate, the controller does not operate as expected due to phenomena such as vibration occurring in the control amount, and operators other than experts with control knowledge should use the IMC controller. There was a problem that we had to give up. In order to solve the above problems, the present invention is capable of performing good control by automatically correcting the time constant of the internal model even when the time constant of the controlled process is unknown.
The purpose is to provide a structure controller.

[0011]

According to the present invention, a target value filter unit for outputting an input target value for control with a characteristic that a transfer function has a time delay, and a feedback amount is subtracted from the output of the target value filter unit. The first subtraction processing unit and the first
Of the target value / disturbance filter unit that outputs the output of the subtraction processing unit of the transfer function with the characteristic of time delay, and the operation that calculates and outputs the manipulated variable from the output of the target value / disturbance filter unit based on the parameters of the internal model. And a manipulated variable calculation unit,
An internal model storage unit that stores the parameters of the internal model, an internal model output calculation unit that calculates the reference controlled variable from the manipulated variable based on the parameters of the internal model, and an internal model output calculation unit that outputs the controlled variable of the controlled process A second subtraction processing unit that subtracts the reference control amount that is output and outputs a feedback amount, and a control amount based on the target value, the control amount, and the reference control amount, and the difference between the initial value and the settling value of the reference control amount. A step width calculation unit that calculates a certain step width, a noise processing unit that performs processing to reduce noise in the control amount, step widths of the control amount and the reference control amount, and a control amount after noise processing that is output from the noise processing unit. A response start area detection unit that detects the start time and the end time of the response start area where the control quantity and the reference control quantity start to change based on the reference control quantity; and the start time and the end of the response start area. And a model gain calculating section for changing the gain in stored in the internal model storage section to calculate a corrected gain of the internal model from the controlled variable and the reference controlled variable change rate specified parameters to the corrected gain by time,
Based on the control gain, the reference control variable, and the process gain identification unit that calculates the gain estimation value of the control target process based on the correction gain, and the correction gain, the gain estimation value, and the time constant in the parameters stored in the internal model storage unit. A model time constant correction unit that calculates a correction time constant of the internal model and changes the time constant stored in the internal model storage unit to this correction time constant.

Further, instead of the process gain storage unit that stores the estimated gain value output from the process gain identification unit in the previous target value tracking control, and the model time constant correction unit, the gain stored in the process gain storage unit is used. The correction time constant of the internal model is calculated based on the estimated value, the estimated gain value output from the process gain identification unit in the current target value tracking control, the correction gain, and the time constant in the parameters stored in the internal model storage unit. , And a model time constant calculation unit for changing the time constant stored in the internal model storage unit to this modified time constant.

Further, a target value filter section for outputting the input target value of the control with a characteristic of a time delay of the transfer function, and a first subtraction processing section for subtracting the feedback amount from the output of the target value filter section, Based on the target value / disturbance filter unit that outputs the output of the first subtraction processing unit with the characteristic that the transfer function has a time delay, and the target value
A manipulated variable computing unit including an operating unit that computes and outputs a manipulated variable from the output of the disturbance filter unit, an internal model storage unit that stores parameters of the internal model, and a reference control from the manipulated variable based on the parameters of the internal model. An internal model output calculation unit that calculates the amount, a second subtraction processing unit that subtracts the reference control amount output from the internal model output calculation unit from the control amount of the controlled process, and outputs a feedback amount, and a specific control For the sample target value generator that alternately generates two sample target values and outputs them to the target value filter unit during the sample input operation for obtaining the time constant estimated value of the controlled object process with respect to the quantity, , A step width that is the difference between the control value and the initial value of the reference control amount and the settling value is calculated based on the sample target value, the control amount, and the reference control amount. Up width calculation unit, a noise processing unit that performs noise reduction processing of the control amount, step widths of the control amount and the reference control amount, a control amount after noise processing output from the noise processing unit, and a reference control amount. A response start area detection unit that detects a start time point and an end time point of the response start area based on which the control amount and the reference control quantity start to change, and a control amount and a reference control amount specified by the start time point and the end time point of the response start area A model gain calculation unit that calculates a modified gain of the internal model from the rate of change and changes the gain in the parameters stored in the internal model storage unit to this modified gain;
A process gain identification unit that calculates a gain estimation value of a control target process based on a reference control amount and a modified gain, and a process gain storage unit that stores the gain estimation value output from the process gain identification unit in the previous sample target value tracking control. And the gain estimation value stored in the process gain storage unit, the gain estimation value output from the process gain identification unit in the current sample target value tracking control, the correction gain, and the parameters stored in the internal model storage unit. A model time constant calculation unit that calculates the time constant estimated value of the controlled process based on the time constant and changes the time constant stored in the internal model storage unit to this time constant estimated value, and specification based on two sample target values The process time constant storage unit that stores the target value and the estimated time constant value calculated by the model time constant calculation unit During the value tracking control operation, the time constant of the internal model corresponding to the control amount is calculated based on the specific target value and the time constant estimated value output from the process time constant storage unit, and the parameters stored in the internal model storage unit are calculated. A model time constant interpolation setting unit for changing the time constant to this calculated value.

[0014]

According to the present invention, the feedback control system is constituted by the target value filter unit, the first subtraction processing unit, the manipulated variable calculation unit, the internal model storage unit, the internal model output calculation unit, and the second subtraction processing unit. Has been done. Then, the step width calculation unit calculates the step width of the control amount and the reference control amount, the response start region detection unit detects the response start region of the control amount and the reference control amount, and the model gain calculation unit calculates the internal model. The corrected gain of is calculated and output to the internal model storage unit, and the gain of the internal model is corrected. Next, the process gain identification unit calculates the estimated gain value of the control target process, and the model time constant correction unit calculates the corrected time constant of the internal model based on the corrected gain, the estimated gain value, and the time constant of the internal model. And output to the internal model storage unit, the time constant of the internal model is corrected.

Further, the estimated gain value output from the process gain identification unit in the previous target value tracking control is stored in the process gain storage unit, and the estimated gain value stored in the process gain storage unit in the model time constant calculation unit. , In the current target value tracking control, the gain of the controlled process is estimated from the gain estimation value output from the process gain identifying unit, and the correction time constant of the internal model is calculated from this result and output to the internal model storage unit. This modifies the time constant of the internal model.

Further, at the time of sample input operation, two sample target values are alternately generated from the sample target value generating section,
Based on the estimated gain value in the previous sample target value tracking control, the estimated gain value in the current sample target value tracking control, the modified gain, and the time constant of the internal model stored in the process gain storage unit in the model time constant calculation unit An estimated time constant value of the control target process is calculated and stored in the process time constant storage unit. Then, in the normal target value tracking control operation, the model time constant interpolation setting unit calculates the time constant of the internal model according to the control amount based on the time constant estimated value and outputs it to the internal model storage unit. The time constant of is modified.

[0017]

1 is a block diagram of an IMC structure controller showing an embodiment of the present invention, and FIG. 2 is a block diagram of a control system using the IMC structure controller. Figure 1
1 is a target value input section for inputting a target value r set by an operator (not shown) to this controller, and 2 is a transfer function for the target value r from the target value input section 1.
The target value filter unit 3 that outputs with the characteristic of the next delay is a first subtraction processing unit that subtracts the feedback amount e from the output of the target value filter unit 2, and 4 is a first subtraction processing unit based on a parameter from an internal model storage unit that will be described later. The operation amount calculation unit 5 for calculating the operation amount u from the output of the subtraction processing unit 3 of 1 is an operation amount calculation unit 4
This is a signal output unit for outputting the manipulated variable u output from the control target process to the control target process not shown in FIG.

Further, 6a is an internal model storage unit for storing the parameters of the internal model of the controller, and 6b is an operation as an internal model based on the parameters output from the internal model storage unit 6a and outputs a reference control amount ym. An internal model output operation unit, 7 is a control amount input unit for inputting a control amount y from a control target process to this controller, and 8 is an internal model output operation unit 6b based on the control amount y output from the control amount input unit 7. Output reference control amount ym
Is a second subtraction processing unit that subtracts and outputs the feedback amount e.

Further, 9 is a step width which is a difference between an initial value of the control amount y and the reference control amount ym to a set value at the end of change based on the target value r, the control amount y and the reference control amount ym. Step width calculation unit, 10 is a noise processing unit that performs processing for reducing noise of control amount y, 11 is a step width of control amount y and reference control amount ym, control amount after noise processing output from noise processing unit 10 , A response start area detection unit for detecting a start time point and an end time point of a response start area, which will be described later, based on the reference control amount ym, and 12 a control amount y and a reference control specified by the start time point and the end time point of the response start area. It is a model gain calculation unit that calculates a correction gain of the internal model from the rate of change of the amount ym and corrects the gain of the internal model stored in the internal model storage unit 6a.

Further, 13 is a controlled variable y, a reference controlled variable ym,
And a process gain identifying unit that calculates a gain estimated value of the process to be controlled based on the corrected gain, and 14 determines the corrected time constant of the internal model based on the corrected gain, the estimated gain value, and the time constant stored in the internal model storage unit 6a. Calculate,
This is a model time constant correction unit that changes the time constant stored in the internal model storage unit 6a to this correction time constant.

In FIG. 2, reference numeral 4a is inside the manipulated variable calculating unit 4 and the output of the first subtraction processing unit 3 has a transfer function of 1
Target value / disturbance filter unit that outputs with the characteristics of the next delay, 4b
Is also inside the target value / disturbance filter unit 4a.
An operation unit for calculating the manipulated variable u from the output of F, an internal model 6 including an internal model storage unit 6a and an internal model output operation unit 6b, F1 a transfer function of the target value filter unit 2, and F2 a target value / disturbance filter unit. 4a is a transfer function. Also,
du is a manipulated variable disturbance, and can be treated as equivalent to the controlled variable disturbance d by setting the disturbance d = Gp × du.

FIG. 2 shows the target value filter unit 2 of FIG.
The basic configuration of the controller of this IMC structure including the first subtraction processing unit 3, the manipulated variable calculation unit 4, the internal model storage unit 6a, the internal model output calculation unit 6b, and the second subtraction processing unit 8 has a control target process. 40, the disturbance d, and the manipulated variable disturbance du are rewritten as a control system.

Next, the operation of the basic configuration of such a controller will be described. The target value r is set by the operator of this controller or the like, and is input to the target value filter unit 2 via the target value input unit 1. The target value filter unit 2 outputs the target value r with the characteristic of the transfer function F1 as shown in the following equation, the time constant of which is T1. F1 = 1 / (1 + T1 × s) (3)

The time constant T1 is the dead time L of the internal model 6 which will be described later except for the preset initial value.
With the change of m, it is set as the following equation. T1 = 4 × α × Lm (4) Here, α is a proportional constant, for example, α = 0.3.

Next, the first subtraction processing unit 3 subtracts the feedback amount e output from the second subtraction processing unit 8 from the output of the target value filter unit 2. The target value / disturbance filter unit 4a in the manipulated variable calculation unit 4 includes the first subtraction processing unit 3
Of the transfer function F2 with the time constant T2. F2 = 1 / (1 + T2 × s) (5)

The time constant T2 is also set by the target value filter unit 2
Similar to the time constant T1 of (1), it is changed as shown in the following equation with the change of the dead time Lm excluding the initial value. T2 = α × Lm (6) That is, the time constant T1 is set to four times the time constant T2 as a standard setting.

Similarly, the operation unit 4 in the operation amount calculation unit 4
b is the manipulated variable u from the output of the target value / disturbance filter unit 4a.
Of the internal model storage unit 6
The following expression is obtained by the gain and time constant of the internal model 6 output from a, and is the reciprocal of the transfer function Gm of the internal model 6 excluding the elements of the dead time Lm as in the example of FIG. Gc = (1 + Tm × s) / Km (7) where Km and Tm are gains of the internal model 6, respectively.
It is a time constant.

Therefore, the transfer function of the operation amount computing section 4 as a whole is given by the following equation. F2 × Gc = (1 + Tm × s) / {Km × (1 + T2 × s)} (8) In this way, the manipulated variable u is calculated from the output of the first subtraction processing unit 3 and the signal output unit It is output to the control target process 40 via 5 and is also output to the internal model output calculation unit 6b.

Next, the controlled process 40 assumes that it has the elements of the first-order delay and the dead time, and its transfer function Gp.
Can be expressed by the approximate transfer function as follows. Gp = K * exp (-L * s) / (1 + T * s) (9) Here, K, L, and T are the gain, dead time, and time constant of the control target process 40, respectively.

The internal model 6 is a mathematical expression of the control target process 40 as described above by these parameters including the gain Km, the time constant Tm, and the dead time Lm stored in the internal model storage unit 6a. Then, the internal model output operation unit 6b calculates the reference control amount ym from the operation amount u output from the operation amount operation unit 4. The transfer function Gm is given by the following equation. Gm = Km × exp (−Lm × s) / (1 + Tm × s) (10)

Next, the second subtraction processing unit 8 subtracts the reference control amount ym from the internal model output calculation unit 6b from the control amount y from the control target process 40 input via the control amount input unit 7. Then, the feedback amount e is output. Then, this feedback amount e is input to the first subtraction processing unit 3 as described above. This is the operation as the feedback control system which is the basic configuration of the controller of this IMC structure.

In such a control system, the step width calculating section 9, the noise processing section 10, the response start area detecting section 11,
The model gain calculation unit 12 corrects the gain Km of the internal model 6 in the first half of the control response to the input target value r, and then the process gain identification unit 13 corrects the gain of the control target process 40 when the control response is settled. The estimated value is calculated, and the model time constant correction unit 14 calculates the time constant T of the internal model 6.
Correct m.

Next, such an operation will be described. Figure 3
FIG. 3A is a diagram showing target value followability of the control amount y for explaining the operation of the step width calculation unit 9, and FIG. 3B is a diagram showing target value followability of the reference control amount ym. y0, y
1 is the initial value, set value, ym0, ym of the controlled variable y, respectively.
1 is the initial value and set value of the reference controlled variable ym, ST is the controlled variable y
Is a step width, which is the difference between the settling value y1 and the initial value y0, and STm is the settling value ym for the reference control amount ym.
The step width is the difference between 1 and the initial value ym0.

In FIG. 3A, at time 0 (initial state), the target value r, which is a step input, is input, and the operation amount u is output from the controller to the controlled object process 40. As a result, the control amount y is initialized to y0. Shows that the settling value y1 is finally changed to the target value r, and the settling state is entered. Further, since the manipulated variable u is output to the internal model output operation unit 6b, the reference control amount ym output from the internal model output operation unit 6b is similarly set.
The state of settling to m1 is shown in FIG.

Then, the step width calculation unit 9 calculates the step width ST of such a control amount y according to the following equation. ST = | r−r0 | (11) where r0 is the initial value of the target value r. Originally, from the definition of the step width ST, ST = | y1-y0 | = | r
-Y0 |, and if no noise is added to the controlled variable y, r0 = y0 and ST = | r-y0 | = | r-r0 |
Becomes However, noise may be added to the control amount y as described later, and at this time, r0 ≠ y0. Therefore, the equation (11) is used.

Therefore, if no noise is added to the control amount y, ST = | r−y0 | may be used. Similarly, the step width STm of the reference control amount ym is calculated by the following equation. STm = | ym1-ym0 | ... (12)

Here, the control amount y of the controlled process 40 and the reference control amount ym output from the internal model 6 can be obtained from the target value r and the disturbance d by the following equation. y = F1 * F2 * Gp * Gc * r / {1 + F2 * Gc * (Gp-Gm)} + (1-F2 * Gm * Gc) * d / {1 + F2 * Gc * (Gp-Gm)} ... (13) ym = F1 * F2 * Gm * Gc * r / {1 + F2 * Gc * (Gm-Gp)} + (-F2 * Gm * Gc) * d / {1 + F2 * Gc * (Gm-Gp)}.・ ・ (14)

If the current value (initial value here) of the gain Km of the internal model 6 is Km0, then equation (13)
From (14), the initial value ym0 of the reference control amount ym is given by the following equation. ym0 = (y0−d) × Km0 / K (15) From the equation (15), if the disturbance d = 0, the control amount initial value y0
And the reference control amount initial value ym0 match the ratio between the gain K (here estimated value) of the controlled object process 40 and the current gain value Km0 of the internal model 6 as in the following equation. K / Km0 = y0 / ym0 (y0 ≠ 0, ym0 ≠ 0) (16)

Therefore, the gain K of the controlled object process 40
Can be estimated as K = Km0 × y0 / ym0 (17) Similarly, the ratio between the control amount settling value y1 and the reference control amount settling value ym1 matches the ratio between the gain K and the current gain value Km0. K / Km0 = y1 / ym1 (18) Further, the control amount settling value y1 matches the target value r as described above. y1 = r (19)

Therefore, the equation (12) is expressed by the equations (16)-
From (19), it can be transformed into the following equation. STm = | ym1-ym0 | = | y1 * Km0 / K-y0 * Km0 / K | = | y1-y0 | * Km0 / K = | (r * ym0 / r0) -ym0 | ... (20) Thus The step width calculation unit 9 calculates the step width ST of the control amount y and the step width STm of the reference control amount ym from the formula (1
1) and (20), and outputs them to the response start area detection unit 11.

Next, the response start area detector 11 determines the control amount y and the reference control amount y based on the step widths ST and STm.
Although a response start region of m described later is detected, noise may be added to the control amount y output from the control amount input unit 7. In such a case, the detection of the response start region becomes inaccurate. The gain Km of the internal model 6 will not be corrected properly. Such noise of the control amount y includes, for example, measurement noise by a sensor (not shown) that measures the control amount y and outputs it to the control amount input unit 7.

Therefore, the noise processing section 10 which is a low-pass filter reduces the noise by damping the control amount y as in the following equation. yd (i) = {y (i) + Td × yd (i-1)} / (1 + Td) (21) where yd (i) is the control after the damping process at the sampling time i of this control system. The quantities y and y (i) are the control quantities y at the sampling time i. Further, Td is a damping time constant, for example, Td = 5 seconds when the sampling cycle (control cycle) is 1 second.

Then, the control amount y after this damping process
d (i) is input to the response start area detection unit 11. Next, the response start area detection unit 11 outputs the operation amount u from the operation amount calculation unit 4 in response to the input of the response start area of the control amount y and the reference control amount ym, that is, the target value r, and thus the control amount y. , Reference control variables ym are initial values y0 and ym, respectively.
A region that starts changing from 0 is detected.

FIG. 4A is a diagram showing the response start area of the control amount y for explaining the operation of the response start area detector 11.
FIG. 4B is also a diagram showing the response start region of the reference control amount ym, where i1 and i2 are the start and end points of the response start region of the control amount y, and im1 and im2 are the responses of the reference control amount ym. The start time and the end time of the start area. Further, each number of 1, 2, ... Ny, Ny + 1 ... N, n + 1 is the sampling time i, and the start time of the response start area is 1. FIGS. 4A and 4B correspond to enlarged views of the change start portion in FIGS. 3A and 3B, respectively.

First, the response start area detection unit 11 outputs the step width S of the control amount y output from the step width calculation unit 9.
T, based on the control amount yd (i) after the damping process output from the noise processing unit 10, the start time point i1 of the response start region of the control amount y is detected as in the following equation. | Yd (i) −yd (0) |> β × ST (22) where β is a proportional constant, for example β = 0.05.

That is, the start time point i1 of the response start area of the controlled variable y is the current controlled variable y (see FIG. 4A) as shown in FIG.
In (a), the control amount yd obtained by performing noise processing on y (i))
The difference between (i) and its initial value yd (0) is the threshold value β × S.
It is the first sampling point that exceeds T.

Then, the start time point im1 of the response start area of the reference control amount ym is similarly detected by the following equation. | Ym−ym0 |> β × STm (23) That is, the start time point i of the response start region of the reference control amount ym
m1 is the first sampling time when the difference between the current reference control amount ym and its initial value ym0 exceeds the threshold value β × STm as shown in FIG.

Next, the end time point i2 of the response start area of the controlled variable y is from the start time points i1 to n of the response start area of the controlled variable y.
It is either the time point after sampling (for example, n = 9) or the first sampling time point that satisfies the following equation, whichever is detected first. | Yd (i) −yd (0) |> δ × ST (24) Here, δ is a proportional constant, for example, δ = 0.20. Then, when the sampling time point by the equation (24) is set to the end time point i2, the number of samplings from the start time point i1 to the end time point i2 is set to Ny.

That is, the end time point i2 of the response start region of the controlled variable y is the time point after 9 samplings in this embodiment (n + 1 time point in FIG. 4A) or the controlled variable yd (i) after the damping process. Of the initial sampling time (Ny + 1 time point in FIG. 4A) when the difference between the initial value yd (0) and the initial value yd (0) exceeds 20% of the step width ST, and thus the Ny + 1 time point in FIG. 4A. Is the end time point i2 of the response start area.

Then, the end time point im2 of the response start area of the reference control amount ym is n sampling points after the start time point im1 of the response start area of the reference control amount ym, or the same as the sampling number Ny obtained above. From the start time to the time after Ny sampling, whichever is earlier. Therefore, the start time i of the response start region of the controlled variable y
The number of samplings from 1 to the end time i2 and the start time im1 to the end time im of the response start region of the reference control amount ym
The sampling numbers up to 2 match.

As described above, the detection of the end time is the earlier of the two points. The reason is that the detection by the fixed sampling number n alone causes the control amount y to approach the settling state earlier and the gain Km described later. This is because the correction of may become too late. Response start area detector 1
1 outputs the start time points i1, im1 and the end time points i2, im2 of the response start area thus detected to the model gain calculation unit 12.

Next, the model gain calculation unit 12 stores the control amount y after the initial value y0, and the control amount y specified by the start time point i1 and the end time point i2 of the response start area, that is, the response start area. Control amount y (i1)
y (i2) is analyzed by the method of least squares, and the control amount change rate (slope) Ay is calculated. Similarly, the reference control amount ym after the initial value ym0 is stored, and the reference control amounts ym (im1) to ym (im2) in the response start area are analyzed by the least square method to determine the reference control amount change rate Aym. calculate.

Then, the model gain calculating section 12 calculates the correction gain Km1 of the internal model 6 as in the following equation. Km1 = ρ × Km0 × Ay / Aym (25) where ρ is a safety factor, for example, ρ = 2.0.

The corrected gain Km1 is output to the internal model storage unit 6a, so that the gain initial value Km0 stored in the internal model storage unit 6a is updated to the corrected gain Km1. Such gain Km (K
The correction of m0) is performed once when the response start region ends for both the control amount y and the reference control amount ym. Therefore, the correction of the gain Km by the step width calculation unit 9, the noise processing unit 10, the response start area detection unit 11, and the model gain calculation unit 12 is performed in the first half of the response to the input of the target value r.

The reason why the gain Km of the internal model 6 is corrected as described above is as follows. The target value r, which is a step input, is input to the operation unit 4b through a transfer function filter unit having a delay of about 1st to 2nd order (in the embodiment, the target value filter unit 2 and the target value / disturbance filter unit 4a cause a 2nd order delay). In this case, the first half, which is the response to the input high frequency included in the target value r, is the most unstable state in the overall response.

Therefore, in the IMC controller, it is important for the stability that the error between the internal model 6 and the controlled process 40 is small in the first half of the response including the response start region. Therefore, the gain Km of the internal model 6 is corrected so that the rate of change of the control amount y of the controlled process 40 and the rate of change of the reference control amount ym of the internal model 6 in the first half of the response are close to each other, as shown in Expression (25). . If the gain Km is corrected in the first half of the response according to the detection result of the response start area, the effect is sufficiently obtained.

Although the change rate of the control amount y and the reference control amount ym may not match from the latter half of the response to the final settling due to the correction of the gain Km, in the latter half of the response approaching the settling state. Since the input of the target value r to the operation unit 4b has a frequency sufficiently lower than that in the first half of the response, the control stability is not so affected.

When the control amount change rate Ay is calculated, the control amount y (i) before the damping process is used instead of the control amount yd (i) after the damping process. This is the control rate change rate A
This is because a safe operation is obtained when y is calculated larger, and the influence of noise can be eliminated by using the least squares method. Further, a safer operation is obtained by using the safety coefficient ρ for the calculation of the correction gain Km1.

By correcting the gain Km of the internal model 6 as described above, a stable response can be obtained even when the gain identification of the controlled object process 40 includes an error, and the controlled variable y is gradually settled. Approaching, and eventually settling.

In the settling state, the relation of the equation (18) is established, and the current value Km0 of the gain Km of the internal model 6 is the corrected gain Km1 at the settling time. Therefore, when the estimated gain value of the controlled process 40 is Kp, the process gain identification unit 13 calculates the estimated gain value Kp by the following equation. Kp = K = Km1 × y1 / ym1 (26)

Next, the model time constant correction unit 14 corrects the time constant Tm of the internal model 6 as follows. In the gain correction by the model gain calculation unit 12, the gain Km of the internal model 6 is corrected so that the change rate of the control amount y of the controlled process 40 and the change rate of the reference control amount ym of the internal model 6 in the response start region are close to each other. ing. This is 1
This is an equivalent adjustment of the rate of change in the next-delay response, and approximately satisfies the relationship shown in the following equation.

(D / dt) t = t1 [K × {1-exp (-t / T)}] = (d / dt) t = t1 [(Km1 / ρ) × {1-exp (-t / Tm0)}] (27) Here, (d / dt) is the time differential operator, t1 is the average time of the response start region (start time point i in FIGS. 4A and 4B).
1, im1 and the end time points i2, im2), and Tm0 is the current value of the time constant Tm of the internal model 6 stored in the internal model storage unit 6a.

Then, the equation (27) becomes the following equation. K / {T × exp (−t1 / T)} = (Km1 / ρ) / {Tm0 × exp (−t1 / Tm0) (28) At this time, in the equation (28), the average time t1 is an error. Is an uncertain parameter including, but exp (-t1 / T) and exp (-t1 / Tm0) at the average time t1.
Are both close to 1.

At the approximate level, if K> Km1 / ρ, T>Tm0; if K = Km1 / ρ, T = Tm0,
If K <Km1 / ρ, the relationship of T <Tm0 is qualitatively established. Therefore, the following simplified relational expression can be set on condition that the above qualitative relationship is established. K / T = (Km1 / ρ) / Tm0 (29)

In the controller of the present embodiment, the process gain identification unit 13 obtains the estimated gain value Kp of the control target process 40 at the time of settling.
If the gain K of 0 is equal to the response start region at the time of settling, the time constant estimated value Tp of the controlled process 40 can be calculated by the following equation. Tp = T = ρ × K × Tm0 / Km1 = ρ × Kp × Tm0 / Km1 (30) Therefore, the model time constant correction unit 14 uses the equation (30) to estimate the time constant Tp of the control target process 40. To calculate.

The time constant estimated value Tp is output as the modified time constant Tm1 from the model time constant modifying unit 14 to the internal model storage unit 6a, so that the time constant Tm0 currently stored in the internal model storage unit 6a. Is updated to this correction time constant Tm1. In this way, one operation after the target value r is input is completed.

That is, the controller of the present embodiment has a step width calculating section 9 and a noise processing section 1 as the overall operation.
0, the response start area detection unit 11, and the model gain calculation unit 12 determine the gain K of the internal model 6 in the first half of the control response.
Then, the process gain identification unit 13 calculates the estimated gain value Kp of the control target process 40 at the time of settling, and the model time constant correction unit 14 corrects the time constant Tm of the internal model 6.

By such an operation, the time constant Tm of the internal model 6 can be corrected even when the time constant T of the controlled process 40 is unknown, but since the correction by the equation (30) is based on the simple estimation, It is not accurately estimated, but estimated in the direction of approaching the time constant T. Therefore, although the time constant identification accuracy is insufficient with one operation, sufficient identification accuracy can be obtained by repeating the above operation.

FIG. 5 is a diagram showing the target value followability when the controller of this embodiment is used for controlling the height of the liquid level in the tank. In FIG. 5, at 0 seconds, the target value r (broken line) is input as a step input of 4 cm of liquid level, and then 3
It is a simulation result in which the target value r was input as heights 1, 4, and 1 cm at 00, 600, and 900 seconds, respectively, and the control amount y (solid line), which is the height of the liquid surface of the control result, was obtained.

Here, the gain K of the process 40 to be controlled, which is the liquid in the tank, is 8, the time constant T is 20 seconds, the dead time L is 10 seconds, and the gain Km of the internal model 6 of the controller of this embodiment is 4. , The time constant Tm is 10 seconds, and the dead time Lm is 10 seconds. That is, the gain Km and the time constant Tm
There is a model identification error in. Further, the time constant T1 of the target value filter unit 2 of the controller is 24 seconds, the time constant T2 of the target value / disturbance filter unit 4a is 6 seconds, and the control cycle Δt is 1 second.

As shown in FIG. 5, it can be seen that the controller of this embodiment can obtain a stable response even if there is a model identification error. The correction time constant Tm1 obtained by the model time constant correction unit 14 is Tm1 = 16.6 seconds (identification error 20.5) at the first time when the target value r = 4 cm.
%), Tm1 = 19 seconds (5.4%) for the second time with the target value r = 1 cm, and 19.7 seconds (1.6% for the same time) with the third time.
%) At the fourth time, Tm1 = 19.8 seconds (1%), and sufficient identification accuracy can be obtained by performing several times.

In the embodiment of FIG. 1, the time constant Tm of the internal model 6 is corrected as described above, but the gain K changes in conjunction with the nonlinearity which changes in association with the control amount y and in time. For the time-varying control target process 40, the correction time constant Tm1 does not approach the time constant T of the control target process 40, and the identification error increases.

This is because the calculation of the time constant estimated value Tp of the controlled object process 40 by the model time constant correction unit 14 is based on the assumption that the gain K of the controlled object process 40 is equal to the response start region at the settling time. This is because if these differ greatly, the theoretical premise will not hold. Therefore, in such a case, another countermeasure is required.

FIG. 6 is a block diagram of a controller having an IMC structure showing another embodiment of the present invention. The same parts as those in FIG. 1 are designated by the same reference numerals. Reference numeral 14a denotes a model time constant calculation unit, which has a gain estimated value calculated by the process gain identification unit 13 in the previous target value tracking control, a gain estimated value similarly calculated in the current target value tracking control, a corrected gain, and A modified time constant Tm1 of the internal model 6 based on the time constant Tm stored in the internal model storage unit 6a.
Is calculated, and the time constant Tm is changed to this corrected time constant Tm1. A process gain storage unit 15 stores the gain estimated value calculated by the process gain identification unit 13 in the previous target value tracking control and outputs the estimated value to the model time constant calculation unit 14a.

Also in this embodiment, the target value input unit 1, the target value filter unit 2, the first subtraction processing unit 3, the manipulated variable calculation unit 4, the signal output unit 5, the internal model storage unit 6a, the internal model output calculation. The operation of the basic configuration of the controller including the unit 6b, the control amount input unit 7, and the second subtraction processing unit 8 is exactly the same as the example of FIG. 1, and the step width calculation unit 9, the noise processing unit 10, and the response start The area detection unit 11 and the model gain calculation unit 12 correct the gain Km of the internal model 6 in the first half of the control response, and the process gain identification unit 13 calculates the estimated gain value Kp of the control target process 40 when the control response is settled. The operation to do is exactly the same.

The subsequent operation of the controller of this embodiment will be described. The process gain storage unit 15
The estimated gain value Kp of the control target process 40 calculated by the process gain identification unit 13 with the previous target value input
(Hereinafter, referred to as Kp0) is stored. On the other hand, the gain estimated value Kp obtained from the process gain identification unit 13 at the current target value input is set to Kp1.

Then, the model time constant calculating section 14a is provided with the gain estimated value Kp stored in the process gain storing section 15.
Based on 0 and the estimated gain value Kp1 obtained from the process gain identification unit 13 at the current control response, the correction time constant Tm1 of the internal model 6 is calculated by the following equation. Tm1 = Tp = ρ × {(1-Q) × Kp0 + Q × Kp1} × Tm0 / Km1 (31) Here, Q is the process gain estimation load, and 0 ≦ Q ≦
1, for example, Q = 0.1.

The equation (31) is based on the equation (30), but it is assumed that the gain K of the controlled process 40 changes substantially linearly before and after the control in which the target value r is input. , The gain estimation value Kp of the controlled process 40 obtained before and after the control is used to estimate the gain K in the response start region. Since the response start region is in the first half of the control response and is close to that before control, the process gain estimated load Q in equation (31) has a value close to zero.

The corrected time constant Tm1 is output from the model time constant calculation unit 14a to the internal model storage unit 6a, so that the time constant Tm0 currently stored in the internal model storage unit 6a is changed to the corrected time constant Tm1. Will be updated. Thus, similarly to the example of FIG. 1, also in the controller of the present embodiment, the step width calculation unit 9 and the noise processing unit 1
0, the response start area detection unit 11, and the model gain calculation unit 12 correct the gain Km of the internal model 6, the process gain identification unit 13 calculates the estimated gain value Kp of the control target process 40, and the process gain storage unit 15 By repeating the correction of the time constant Tm of the internal model 6 by the model time constant calculation unit 14a, sufficient identification accuracy and stable response can be obtained even for the non-linear controlled object process 40.

FIG. 7 is a diagram showing the target value followability when the controller of this embodiment is used for controlling the height of the liquid level in the tank, as in the example of FIG. Here, it is assumed that the gain K of the controlled process 40 is K = 0.8 × y + 6.4, that is, it changes in association with the control amount y, and other parameters are the same as in the example of FIG.

As shown in FIG. 7, according to the controller of this embodiment, a stable response can be obtained even for the non-linear controlled object process 40 in which the gain K changes in association with the controlled variable y. . Further, the model time constant calculation unit 14a
The correction time constant Tm1 obtained by
Tm1 = 21.7 seconds (identification error −7.9)
%), Tm1 = 19.% in the second time with the target value r = 1 cm.
Sufficient identification accuracy can be obtained by performing the operation several times, such as 2 seconds (4.1% in the same time) and 19.2 seconds in the third time (4.1% in the same time).

In the example of FIGS. 1 and 6, the time constant Tm of the internal model 6 is corrected as described above. However, the time constant T changes in conjunction with the controlled variable y and the nonlinearity and the time change. For the time-varying controlled process 40 that changes, the correction time constant Tm1 does not approach the time constant T of the controlled process 40, and the identification error increases. This is the model time constant correction unit 1
4. Modified time constant Tm1 by the model time constant calculation unit 14a
Is based on the assumption that the time constant T of the control target process 40 is equal to the response start region after settling,
This is because if these differ greatly, the theoretical premise will not hold. Therefore, in such a case, another measure is required.

FIG. 8 is a block diagram of a controller having an IMC structure showing another embodiment of the present invention. The same parts as those in FIG. 1 are designated by the same reference numerals. Reference numeral 14b denotes a model time constant calculation unit, which is used for a sample input operation described later, and which is a gain estimated value stored in the process gain storage unit 15 in the previous sample target value tracking control, and a process gain identification unit in the current sample target value tracking control. The time constant estimated value of the controlled process 40 is calculated based on the gain estimated value output from 13 and the corrected gain Km1 and the time constant Tm of the internal model 6, and the time constant Tm stored in the internal model storage unit 6a is calculated as Change to the estimated value of the time constant.

Further, 16 is a sample target value generator for alternately generating two sample target values during the sample input operation for obtaining the time constant estimated value of the controlled object process 40 for a specific control amount, and 17 is two A process time constant storage unit that stores a specific target value based on a sample target value and a time constant estimated value calculated by the model time constant calculation unit 14b, 1
Reference numeral 8 indicates a time constant Tm of the internal model 6 according to the control amount y based on the specific target value and the time constant estimated value output from the process time constant storage unit 17 during the normal target value follow-up control operation. A model time constant interpolation setting unit that changes the time constant Tm stored in the storage unit 6a to this calculated value.

In this embodiment as well, the operation of the basic configuration of the controller is exactly the same as the example of FIG. 1, but it is an offline operation for obtaining the estimated value of the time constant T of the controlled process 40 for a specific control amount y. During sample input operation,
The sample target value generator 16 repeatedly outputs the sample target value as follows.

9 and 10 are diagrams showing the target value followability of the control amount y and the reference control amount ym for explaining the operation of the sample target value generation unit 16, and r11 to rj2 are sample target value generation units 16 respectively. Sample target value output from, y11 to y
j2 is a control amount set value set by the output of the sample target values r11 to rj2.

The sample target value generator 16 enters a sample input operation in response to a command from the outside, for example, and repeatedly generates two specific sample target values r11 and r12 as shown in FIG. Output to 1.

At this time, the operation of the step width calculation unit 9, noise processing unit 10, response start area detection unit 11, model gain calculation unit 12, and process gain identification unit 13 for the sample target values r11 and r12 is the same as the example of FIG. This is exactly the same, and the gain Km of the internal model 6 is corrected in the first half of the control response every time the target value r11 or r12 is input, and the gain estimated value Kp of the controlled process 40 is calculated when the settling state is reached.

Next, the model time constant calculating section 14b is operated as shown in FIG.
In the same manner as in the above example, the process gain identification unit 13 receives the previous target value input (when r11 is currently input, it is r12).
Based on the gain estimation value Kp0 calculated by the above and stored in the process gain storage unit 15 and the gain estimation value Kp1 obtained from the process gain identification unit 13 at the current target value input, the following equation corresponding to the equation (31) is obtained. The time constant estimated value Tp of the controlled process 40 is calculated by. Tp = R × ρ × {(1-Q) × Kp0 + Q × Kp1} × Tm0 / Km1 + (1-R) × Tm0 (32)

Here, R is the process time constant estimated load, and 0 ≦ R ≦ 1, and only for the first time of the repetitive control, R = 1,
In the subsequent repeat control, R = 0.5. The first repeat control means the time when the value of r11 is first changed to r12. That is, in the sample input operation, when the sample target value changes from r11 to r12 is the first control, the gain Km of the internal model 6 is corrected and the estimated gain value Kp of the controlled process 40 is calculated as described above. Then, subsequently, the model time constant calculating unit 14b calculates the time constant estimated value Tp.

The time constant estimated value Tp is output as the modified time constant Tm1 from the model time constant calculating unit 14b to the internal model storage unit 6a, so that the time constant Tm0 currently stored in the internal model storage unit 6a. Is updated to this correction time constant Tm1.

The estimated time constant value Tp is also output to the process time constant storage unit 17, and sample target values r11 and r
It is stored in the process time constant storage unit 17 in a form corresponding to the specific target value r1 which is the median value of 12 (the time constant estimated value Tp at this time is defined as Tp (r1)). Thus, the first iteration control is completed, the sample target value generator 16 outputs the sample target value r11, and the second iteration control is performed in the same manner as above.

The equation (32) is for estimating the average value of the time constant T on the assumption that the time constant T of the controlled object process 40 changes substantially linearly between the two target values. The process time constant estimated load R is R =
The reason for setting 1 is that the current value Tm0 of the time constant of the internal model 6 stored in the internal model storage unit 6a at this time is uncertain. Therefore, in the first iteration control,
The first term on the right side of Expression (32) [R × ρ × {(1-Q) × Kp
Only 0 + Q × Kp1} × Tm0 / Km1] is used.

Then, after the second repeat control,
Since the current value Tm0 of the time constant of the internal model 6 has been corrected to Tm1 by the first iteration control and is almost certain, R = 0.5 is set, and the second term on the right side of the equation (32) ( 1-R) × Tm0 is also used together with the first term.

Further, the process gain estimated load Q is also used similarly to the equation (31) because the gain K and the control amount y are controlled in the control target process 40 in which the time constant T changes according to the control amount y. It is preferable that the gain K be treated as if it changes.

After the above-described iterative control is performed several times between the sample target values r11 and r12, as the next iterative control, the sample target value generator 16 outputs the sample target values r21 and r22 as shown in FIG. The output is repeated and the same repetitive control is performed. The specific target value r2 which is the median of the sample target values r21 and r22 and the time constant estimated value Tp (r2
) Is stored in the process time constant storage unit 17.

Such repetitive control is repeated up to the sample target values rj1 and rj2, and the specific target values r1 to rj and the time constant estimated values Tp (r1) to T corresponding to each of them.
p (rj) is stored in the process time constant storage unit 17.
This completes the sample input operation.

The sample input operation as described above is performed with respect to the control target process 40 in which the time constant T changes according to the control amount y, with respect to each of the specific target values ri corresponding to a plurality of control amounts yi discontinuously. This means that the constant estimated value Tp (ri) has been obtained.

Next, in the actual target value tracking control operation, the step width calculation unit 9, the noise processing unit 10, the response start region detection unit 11, the model gain calculation unit 12, the process gain identification unit 13, and the process gain storage unit 15 are performed. The model time constant calculator 14b and the sample target value generator 16 do not operate.

The controlled variable y starts to change when the manipulated variable u is output in response to the input of the actual target value r, but the model time constant interpolation setting unit 18 is stored in the process time constant storage unit 17. Based on the estimated time constant value Tp (ri), the corrected time constant Tm1 (y) of the internal model 6 corresponding to the control amount y output from the control amount input unit 7 is linearly interpolated and calculated according to the following equation. .

Tm1 (y) = Tp (ri) * (y-ri + 1) / (ri-ri + 1) + Tp (ri + 1) * (y-ri) / (ri + 1-ri) (ri .Ltoreq.y.ltoreq.ri + 1) (33) When y <r1 is calculated by the following equation. Tm1 (y) = Tp (r1) (34) Similarly, when rj≤y, it is calculated by the following equation. Tm1 (y) = Tp (rj) (35)

That is, the model time constant interpolation setting unit 18
When the controlled variable y is between the specific target values ri and ri + 1, the modified time constant Tm1 (y) corresponding to the controlled variable y is set to the estimated time constant Tp.
(Ri) and Tp (ri + 1) are linearly interpolated and calculated.
The corrected time constant Tm1 (y) is output from the model time constant interpolation setting unit 18 to the internal model storage unit 6a, so that the time constant Tm0 currently stored in the internal model storage unit 6a is changed to the time constant Tm1 ( y) is updated.

The calculation of the time constant Tm1 (y) and the update of the time constant Tm0 are performed every control cycle Δt (for example, Δt = 1 second). Therefore, the corrected time constant Tm of the internal model 6 is adjusted according to the actual control amount y based on the estimated time constant value Tp (ri) obtained during the sample input operation.
By calculating 1 (y), the control target process 40
Good control can be performed even when the time constant T of changes in conjunction with the control amount y. In this embodiment, the linear interpolation as described above is used to obtain the correction time constant Tm1 (y), but other interpolation methods may be used.

FIG. 11 is a diagram showing the target value followability when the controller of this embodiment is operated for sample input.
4 and 0.1 are input as sample target values. Here, the time constant T of the controlled process 40 is T = 2
Xy + 16, that is, it changes in association with the control amount y, and other parameters are the same as in the example of FIG.

While the average value of the time constant T of the controlled process 40 is 20 seconds, the estimated time constant value Tp obtained by the model time constant calculation unit 14b is Tp = 14.9 seconds (identification error 34.4%),
With the target value r = 0.1, Tp = 17.3 seconds (15.8% in the second time) and similarly Tp = 17.3 seconds (1% in the third time).
5.8%), Tp = 19 seconds at the 4th time (5.4%) and sufficient accuracy can be obtained by repeating the control several times. Therefore, the actual control using the time constant estimated value Tp is good. It is possible to obtain various characteristics.

FIG. 12 is an IMC showing another embodiment of the present invention.
FIG. 2 is a block diagram of a controller having a structure, and the same parts as those in FIG. 1 are denoted by the same reference numerals. Reference numeral 19 denotes a time constant T1 of the target value filter unit 2 and a target value not shown in FIG.
An initial filter safety setting unit that initializes the time constant T2 of the disturbance filter unit 4a, and 20 increases the dead time of the internal model and stores it in the internal model storage unit 6a while the control amount y does not change as a control response. This is a sequential dead time updating unit that repeats updating the dead time Lm.

Reference numeral 21 denotes a certainty factor calculation unit for calculating the certainty factor, which is the dead time estimation accuracy, based on the control amount y and the change rates Ay and Aym of the reference control amount ym. Reference numeral 22 denotes a dead time detector, which starts the response start area i1 and im1 and outputs the end time i2 and i of the response start area output from the response start area detector 11.
The control target process 4 based on the control amount y specified by m2, the change rates Ay and Aym of the reference control amount ym, and the certainty factor.
An estimated value of the dead time L of 0 is calculated, and the dead time Lm of the internal model 6 is changed to this estimated value.

In the present embodiment, the operation of the basic configuration of the controller is exactly the same as in the example of FIGS. 1 and 6, and the step width calculation unit 9, noise processing unit 10, response start area detection unit 1
1, model gain calculation unit 12, process gain identification unit 1
3, the operation of the process gain storage unit 15 and the model time constant calculation unit 14a is exactly the same as the example of FIG.

In the controller of this embodiment, the maximum gain (upper limit value) Kmu that is assumed as the gain K of the controlled object process 40 is set as the gain Km as the initial setting of the parameters of the internal model 6 stored in the internal model storage unit 6a. Similarly, the time constant Tm is set to the minimum time constant (lower limit value) Tmd assumed as the time constant T, and the dead time Lm is set to the minimum possible dead time 0 as the dead time L.

The time constant T1 of the target value filter unit 2, FIG.
Then, the time constant T of the target value / disturbance filter unit 4a (not shown)
2 is set according to the change of the dead time Lm of the internal model 6 as described in the example of FIG. 1, but the initial filter safety setting unit 19 sets the minimum time constant Tmd by the following equation. The time constants T1 and T2 corresponding to are output to the target value filter unit 2 and the target value / disturbance filter unit 4a to perform the initial setting in consideration of safety. T1 = 4 × α × Sf × Tmd (36) T2 = α × Sf × Tmd (37) Here, Sf is a safety estimation coefficient, for example, Sf = 50.
Is.

In such a controller, the dead time updating unit 20 sequentially corrects the dead time Lm of the internal model 6 stored in the internal model storage unit 6a as follows. FIG. 13 is a diagram showing a change start portion of the control amount y for explaining the operation of the sequential dead time updating unit 20, and FIG.
This corresponds to an enlarged view of the change start portion in (a). Lm1
Is the dead time Lm sequentially changed by the dead time update unit 20, and Δt is the control cycle (sampling cycle) of this control system.
And the ∘ mark on the controlled variable y indicates the sampling time.

The sequential dead time updating unit 20 receives the control amount yd after the damping process input from the noise processing unit 10.
Based on (i), it is determined by the following equation whether or not the control amount y has changed as a control response. | Yd (i) −yd (0) | <ε × ST = ε × | r−r0 | (38) Here, ε is a proportional constant, for example, ε = 0.05.

That is, as shown in FIG. 13, when ε × ST is used as a threshold value and the difference between the current controlled variable yd (i) and the initial value yd (0) is smaller than this threshold value, the controlled variable y is changed. It is determined that there is not, and in the above cases, it is determined that a change has appeared. In this way, 5% of the step width ST is used to judge the change in the control amount y.
The threshold value is provided to prevent erroneous detection of a change in the control amount y due to disturbance or the like.

When it is determined that the control amount y has not changed, the control period Δt is added to the dead time Lm stored in the internal model storage unit 6a. Since the initial value of the dead time Lm is initially set to 0, the dead time Lm is Δt.
Becomes The added dead time Lm is sequentially output from the dead time update unit 20 to the internal model storage unit 6a, so that the current dead time Lm stored in the internal model storage unit 6a is updated.

Then, the change of the dead time Lm as described above is repeated while the change in the control amount y is not detected by the determination by the equation (38), and is stopped when the change is detected. As a result, the dead time Lm finally obtained is shown in FIG.
Is Lm1. In actual control, the dead time Lm is output to the internal model storage unit 6a every time the control cycle Δt is added, and the stored dead time Lm is changed and then output to the internal model output calculation unit 6b. Reference controlled variable y
The m is calculated based on the changed dead time Lm.

Therefore, the control amount y of the controlled process 40
The reference control amount ym is held at the initial value ym0 as the dead time Lm is repeatedly lengthened until a change appears in the control amount y, and a change in the control amount y due to the control response appears in the dead time L.
The change appears in the reference control amount ym when the change of m is completed. Since the dead time is the time from the input of the target value r to the change in the control amount y, as described above, the dead time Lm of the internal model 6 is successively increased while the control amount y does not change as a control response. By updating, the dead time Lm of the internal model 6 can be corrected even when the dead time L of the control target process 40 is unknown, and deterioration of the control characteristics can be avoided.

Next, since there is a detection limit based on the threshold value ST in the sequential dead time updating unit 20, the certainty factor calculating unit 2
1. The dead time detection unit 22 corrects the dead time Lm of the internal model 6 with higher accuracy. First, the certainty factor calculation unit 21 calculates the certainty factor, which is the dead time estimation accuracy, as follows.

The certainty factor calculation unit 21 stores the control amount y after the initial value y0 output from the control amount input unit 7 and the reference control amount ym after the initial value ym0 output from the internal model output calculation unit 6b. The response similar to that of the model gain calculation unit 12 in the example of FIG. 1 is obtained by the least squares method based on the start time points i1, im1 and the end time points i2, im2 of the response start area output from the response start area detection unit 11. The control amount change rate Ay of the start region and the reference control amount change rate Aym are calculated.

Then, the confidence factor CF is calculated by the following equation,
It is output to the dead time detection unit 22. CF = exp {-(1-Ay / Aym) ² / AC} (39) Here, AC is a positive number set value, for example, AC = 10.
It is 0. The confidence factor CF is Ay / A as shown in FIG.
This is a bell-shaped function in which CF = 1 when ym = 1, and shows that the dead time estimation accuracy is highest when Ay / Aym = 1.

The reference control amount ym changes as a control response from the time when the dead time Lm of the internal model 6 is changed to the dead time Lm1 calculated by the dead time updating unit 20 as described above. Therefore, the above ratio Ay / Aym
Calculating the confidence factor CF based on
The result of the control response by m1 will be evaluated.

Next, the dead time detecting section 22 estimates the dead time Lm2 of the controlled process 40 as follows.
To calculate. FIG. 15A is a diagram showing a response start region of the control amount y for explaining the operation of the dead time detection unit 22, and FIG. 15B is a diagram showing a response start region of the reference control amount ym.

In FIG. 15A, Lm2 is the estimated value of the dead time L of the controlled process 40 to be calculated, i0 is the time when the controlled variable y starts to change, and A is the straight line from this point i0 on the controlled variable y. The time from the intersection of the extension line of H2 to H1 and the initial value y0, B is the time from this intersection to the start time point i1 of the response start region. In FIG. 15 (b),
im0 is the time when the reference control amount ym starts to change, C is the time from the intersection of the extension line of the straight lines H4 to H3 on the reference control amount ym and the initial value ym0 to the start time im1 of the response start region, and D is the point im0. From the start time im1.

Since the dead time detector 22 is supplied with the control amount y and the reference control amount ym that are not subjected to noise processing, noise is added to these values. Figure 15
(A) and (b) show the influence of this noise in the response start region as at the sampling time of the circle, and for other sections, only the straight lines that approximate these are shown.

Here, the dead time Lm2 can be estimated as the elapsed time from time 0 to the change start time i0 of the controlled variable y. Then, the proportional constant ε of the change start determination threshold value (equation (38)) by the sequential dead time update unit 20 and the proportional constant of the detection threshold value (equation (22)) at the start time point i1 by the response start area detection unit 11 are calculated. Since β is ε = β = 0.05, the dead time Lm1 sequentially obtained by the dead time update unit 20 is the time from time 0 to the start time point i1 as shown in FIG.

Further, since the reference control amount ym starts to change after the dead time Lm1 is determined as described above, Lm1 is the time from time 0 to change start time im0 as shown in FIG. 15 (b).

Therefore, the estimated value Lm2 of the dead time is as shown in FIG.
As shown in FIG. 5 (a), it can be calculated by the following equation from the dead time Lm1, the time A, and the time B which are sequentially obtained by the dead time update unit 20 and are currently stored in the internal model storage unit 6a. Lm2 = Lm1-BA (40)

The dead time detector 22 stores the control amount y after the initial value y0, and the start time i of the response start area
1, the control amount y specified by the end time point i2, that is, the control amounts y (i1) to y (i2) in the response start region
Is analyzed by the method of least squares to obtain a linear function expression as follows. y (i) = Ay × i + By (41)

That is, the expression (41) is an expression representing the straight lines H2 to H1 in FIG. 15A, Ay is the rate of change of the control amount which is the inclination of this straight line, and By is the constant term which is also the intercept. In addition, the dead time detection unit 22 similarly sets the initial value ym0.
The subsequent reference control amount ym is stored, and the reference control amounts ym (im1) to ym (im2) in the response start area are analyzed by the least square method to obtain the following equation. ym (i) = Aym × i + Bym (42)

This is an equation representing the straight lines H4 to H3 in FIG. 15B by the reference control amount change rate Aym and the constant term Bym. In this way, the control amount change rate Ay, the reference control amount change rate Aym, and the constant terms By and Bym can be calculated.

Next, the time B is given by the following equation from these obtained values. B = i1- (r0-By) / Ay (43) Further, the time A can be obtained by the following equation. A = (Aym / Ay) ^1/2 × (D−C) (44)

Then, the times C and D are given by the following equations. C = im1- (ym0-Bym) / Aym ... (45) D = im1-Lm1 ... (46)

Therefore, the expression (40) is the following expression from the certainty factor CF calculated by the certainty factor calculation unit 21 and the equations (43) to (46). Lm2 = CF * [Lm1-i1 + (r0-By) / Ay- (Aym / Ay) ^1/2 * {-Lm1 + (ym0-Bym) / Aym}] ... (47)

When the dead time Lm2 is calculated as a negative value, the following equation is used. Lm2 = CF × [Lm1-2 × {i1- (r0-By) / Ay}] (48) Even if Lm2 is a negative value according to the equation (48), Lm2
Let 2 = 0. In this way, the calculation is performed by multiplying the certainty factor CF by setting the dead time Lm of the internal model 6 to be smaller than the dead time L of the control target process 40 in the IMC controller. Because it is safe.

Then, the dead time detection unit 22 outputs this dead time Lm2 to the internal model storage unit 6a after the control is set, and changes the dead time of the internal model 6 to Lm2.
Thus, by using the confidence factor CF, it is possible to prevent an inappropriate dead time correction from being performed and to obtain a safe response.

Also, the calculation of the dead times Lm1 and Lm2 by the normal dead time updating section 20 and the dead time detecting section 22 is carried out.
The correction is performed only once, but the detection result may not be sufficient even with the dead time Lm2 based on the certainty factor CF as described above. Therefore, the dead time detection unit 22 performs reliability evaluation using the following equation and again It is determined whether or not to calculate the dead time Lm1 and Lm2. Ay / Aym> Alm1 (49) Ay / Aym <Alm2 (50) Here, Alm1 is the upper limit of the change rate ratio of the response start region, A
lm2 is the lower limit of the change rate ratio of the response start region, for example, Alm1 = 2 and Alm2 = 0.1.

If the equations (49) and (50) are not satisfied, that is, if the change rate ratio Ay / Aym of the response start area is within the upper and lower limits, it is assumed that the dead time detection result is sufficiently reliable. In response, the dead time update unit 20
Calculation of the dead time Lm1 due to
The calculation of 2 is also not executed. If Expression (49) or (50) is satisfied, it is considered unreliable, and the dead time update unit 20 is made to sequentially calculate the dead time Lm1 even when the next target value r is input, and then the dead time Lm2.
The calculation of is executed.

As described above, the dead time Lm of the internal model 6
Is corrected, the dead time L of the control target process 40 is unknown, and even if noise is added to the control amount y, the dead time can be corrected with high accuracy, and an inappropriate dead time correction is performed. You can prevent being exposed.

Therefore, when the operation of the controller of this embodiment is governed, the target value input unit 1, the target value filter unit 2, the first subtraction processing unit 3, the manipulated variable calculation unit 4, the signal output unit 5, and the internal model storage. Unit 6a, internal model output calculation unit 6
b, the control amount input unit 7, and the second subtraction processing unit 8 operate as a basic configuration as a controller, and the step width calculation unit 9, the noise processing unit 10, the response start area detection unit 11, and the model gain calculation unit. 12 corrects the gain Km of the internal model 6 in the first half of the control response to the input target value r, and then the process gain identifying unit 13 calculates the estimated gain value Kp of the controlled process 40 when the control response is settled. Then, the process gain storage unit 15 and the model time constant calculation unit 14a correct the time constant Tm of the internal model 6. On the other hand, the sequential dead time update unit 20 corrects the dead time Lm of the internal model 6 in the first half of the control response, and the certainty factor calculation unit 21,
Further, the dead time detection unit 22 corrects the dead time Lm with higher accuracy and evaluates the reliability.

By doing so, the gain Km and the time constant Tm of the internal model 6 are increased even if the model identification error is large due to unknown characteristics of the controlled process 40 and noise is added to the controlled variable y. Since the dead time Lm can be corrected, highly accurate and reliable control can be performed, and a general-purpose and easy-to-use controller with a wide control target can be realized.

FIG. 16 is a diagram showing the target value followability when the controller of this embodiment is used for controlling the height of the liquid level in the tank as in the example of FIG. 5, and FIG. 17 is the control amount of this controller. The figure which shows the control amount y output from the input part 7, FIG.
Similarly, FIG. 8 is a diagram showing the target value followability of the conventional IMC controller. The control amount y shown in FIG. 16 is the same as the example of FIGS. 5 and 7, but the control amount y shown in FIG. 17 is measured by a sensor (not shown) in the control amount y shown in FIG. It is input, and it is shown that the measurement noise by the sensor is added.

The conventional IMC controller is a controller having only the basic configuration of the example of FIG.
No measurement noise was added to the C controller. Here, the gain K of the controlled process 40 is K = 0.8 ×
y + 6.4, that is, it changes in association with the control amount y. Further, the time constant T of the controlled process 40 is set to 20.
Seconds, the dead time L is set to 20 seconds, the gain Km of the internal model of this embodiment and the conventional controller is set to 40 seconds, and the time constant T is set.
Let m be 5 seconds and dead time Lm be 0 seconds.

Further, the time constant T1 of the target value filter unit 2 of this embodiment and the conventional IMC controller is T1 = 4 × α ×
Sf × Tm = 120 seconds, the time constant T2 of the target value / disturbance filter unit 4a is T2 = α × Sf × Tm = 30 seconds, and the control cycle Δt is 1 second. The measurement noise is random noise having a maximum amplitude of 0.2 (5% of the step width ST) and an average value of 0.

As is clear from the comparison between FIGS. 16 and 18, the conventional IMC controller vibrates in the controlled variable y even though no measurement noise is added. It can be seen that a stable response can be obtained even when the controlled object process 40 is non-linear, the model identification includes an error, and the controlled variable y includes measurement noise.

[0144]

According to the present invention, the gain of the internal model is automatically corrected during the target value tracking control operation, and then the time constant of the internal model is automatically corrected even when the model identification of the controlled process includes an error. Therefore, it is possible to perform control with high accuracy and reliability, and it is possible to cope with changes in the characteristics of the process to be controlled. Further, in such a case, deterioration of control characteristics can be avoided even if noise is added to the control amount. Further, since it is possible to prevent the occurrence of troubles due to the identification error of the process to be controlled, it is possible to reduce the work load on the operator who does not have specialized control knowledge.

Further, a process gain storage unit and a model time constant calculation unit are provided to estimate the gain of the control target process from the gain estimation value of the previous target value tracking control and the current gain estimation value of the target value tracking control. Since the time constant of the internal model is automatically corrected from the result, it is possible to perform highly accurate and reliable control even for a non-linear controlled object process whose gain changes in conjunction with the control amount.

Further, a sample target value generation unit, a process time constant storage unit, and a model time constant interpolation setting unit are provided to obtain a time constant estimated value of a control target process for a specific control amount during a sample input operation which is an offline operation. The time constant of the internal model is stored in the process time constant storage section, and in the normal target value tracking control operation, the time constant of the internal model is calculated from this estimated value of the time constant to correct the internal model time constant. It is possible to perform highly accurate and reliable control even for a non-linear controlled object process that changes in accordance with the control amount.

[Brief description of drawings]

FIG. 1 is a block diagram of a controller having an IMC structure showing an embodiment of the present invention.

FIG. 2 is a block diagram of a control system using the controller having the IMC structure of FIG.

FIG. 3 is a diagram showing target value followability of a control amount and a reference control amount.

FIG. 4 is a diagram showing a response start region of a control amount and a reference control amount.

FIG. 5 is a diagram showing target value followability when the controller of FIG. 1 is used to control the height of the liquid level in the tank.

FIG. 6 is a block diagram of an IMC structure controller according to another embodiment of the present invention.

FIG. 7 is a diagram showing target value followability when the controller of FIG. 6 is used to control the height of the liquid level in the tank.

FIG. 8 is a block diagram of an IMC structure controller according to another embodiment of the present invention.

FIG. 9 is a diagram showing target value followability of a control amount and a reference control amount.

FIG. 10 is a diagram showing target value followability of a control amount and a reference control amount.

FIG. 11 is a diagram showing target value followability when the controller of FIG. 8 is operated for sample input.

FIG. 12 is a block diagram of an IMC structure controller according to another embodiment of the present invention.

FIG. 13 is a diagram showing a change start portion of a control amount y.

FIG. 14 is a diagram showing a certainty factor calculated by a certainty factor calculation unit of the controller of FIG. 12;

FIG. 15 is a diagram showing a response start region of a control amount and a reference control amount.

FIG. 16 is a diagram showing target value followability when the controller of FIG. 12 is used to control the height of the liquid level in the tank.

17 is a diagram showing a control amount output from a control amount input unit of the controller of FIG.

FIG. 18 is a diagram showing a target value followability of a conventional IMC controller.

FIG. 19 is a block diagram of a control system using a conventional IMC controller.

[Explanation of symbols]

 2 Target value filter unit 3 First subtraction processing unit 4 Manipulation amount calculation unit 6a Internal model storage unit 6b Internal model output calculation unit 8 Second subtraction processing unit 9 Step width calculation unit 10 Noise processing unit 11 Response start area detection unit 12 model gain calculation unit 13 process gain identification unit 14 model time constant correction unit 14a, 14b model time constant calculation unit 15 process gain storage unit 16 sample target value generation unit 17 process time constant storage unit 18 model time constant interpolation setting unit

Claims (3)

[Claims] 1. A reference control amount corresponding to a control amount of a control target process, which is a control result, is calculated by calculating an operation amount to be output to a control target process from a control target value, and using an internal model expressing the control target process by a mathematical expression. Control is performed by calculating and feeding back the difference between the control amount and the reference control amount I
In a controller having an MC structure, a target value filter unit that outputs a target value of input control with a characteristic of a time delay of a transfer function, and a first subtraction processing unit that subtracts a feedback amount from the output of the target value filter unit. , A manipulated variable is calculated from the output of the target value / disturbance filter unit based on the parameters of the target value / disturbance filter unit in which the transfer function outputs the output of the first subtraction processing unit with the characteristic of time delay. An operation amount calculation unit including an operation unit that outputs the internal model, an internal model storage unit that stores parameters of the internal model, and an internal model output that calculates a reference control amount from the operation amount based on the parameters of the internal model. The calculation unit and the control amount of the controlled process are subtracted from the reference control amount output from the internal model output calculation unit to output the feedback amount. A second subtraction processing unit; a step width calculation unit that calculates a step width that is a difference between an initial value of the control amount and the reference control amount and a settling value based on the target value, the control amount, and the reference control amount; A noise processing unit that performs processing for reducing noise in the control amount, a step width of the control amount and the reference control amount, a control amount after noise processing output from the noise processing unit, and a control amount and a reference based on the reference control amount. A response start area detection unit that detects a start time point and an end time point of the response start area where the control amount starts to change, and the change rate of the control amount and the reference control amount specified by the start time point and the end time point of the response start area A model gain calculation unit that calculates a correction gain of the internal model and changes the gain in the parameters stored in the internal model storage unit to the correction gain, and the control amount, the reference control amount, and the correction gain. A process gain identification unit that calculates a gain estimation value of the control target process based on the correction gain, the gain estimation value, and a correction time constant of the internal model based on the time constant in the parameters stored in the internal model storage unit. A model time constant correction unit for changing the time constant stored in the internal model storage unit to the correction time constant. 2. The controller according to claim 1, wherein the process gain storage unit that stores the estimated gain value output from the process gain identification unit in the previous target value tracking control, and the model time constant correction unit are used instead of the process gain storage unit. Based on the estimated gain value stored in the process gain storage unit, the estimated gain value output from the process gain identification unit in the current target value tracking control, the modified gain, and the time constant in the parameters stored in the internal model storage unit. A controller comprising: a model time constant calculation unit that calculates a correction time constant of the internal model and changes the time constant stored in the internal model storage unit to the correction time constant. 3. A reference control amount corresponding to the control amount of the control target process, which is a control result, is calculated by calculating an operation amount to be output to the control target process from a control target value, and using an internal model expressing the control target process by a mathematical expression. Control is performed by calculating and feeding back the difference between the control amount and the reference control amount I
In a controller having an MC structure, a target value filter unit that outputs a target value of input control with a characteristic of a time delay of a transfer function, and a first subtraction processing unit that subtracts a feedback amount from the output of the target value filter unit. , A manipulated variable is calculated from the output of the target value / disturbance filter unit based on the parameters of the target value / disturbance filter unit in which the transfer function outputs the output of the first subtraction processing unit with the characteristic of time delay. An operation amount calculation unit including an operation unit that outputs the internal model, an internal model storage unit that stores parameters of the internal model, and an internal model output that calculates a reference control amount from the operation amount based on the parameters of the internal model. The calculation unit and the control amount of the controlled process are subtracted from the reference control amount output from the internal model output calculation unit to output the feedback amount. A sample for alternately generating two sample target values and outputting them to the target value filter unit during the sample input operation for obtaining the time constant estimated value of the controlled process for the specific control amount with the second subtraction processing unit. With respect to each of the target value generation unit and the sample target value, the step width which is the difference between the control value and the initial value of the reference control amount and the settling value is calculated based on the sample target value, the control amount, and the reference control amount. A step width calculation unit, a noise processing unit that performs processing for reducing noise of the control amount, a step width of the control amount and a reference control amount, a control amount after noise processing output from the noise processing unit, and a reference control amount A response start area detection unit that detects a start time point and an end time point of the response start area where the control amount and the reference control amount start to change based on A model gain calculation unit that calculates a correction gain of the internal model from the change rate of the control amount and the reference control amount that are more specified, and changes the gain in the parameters stored in the internal model storage unit to the correction gain, Process gain identification unit that calculates the estimated gain value of the controlled process based on the controlled variable, reference control amount, and modified gain, and the process that stores the estimated gain value output from the process gain identification unit in the previous sample target value tracking control A gain storage unit and a gain estimation value stored in this process gain storage unit,
In the current sample target value tracking control, the time constant estimated value of the control target process is calculated based on the gain estimated value output from the process gain identification unit, the corrected gain, and the time constant in the parameter stored in the internal model storage unit. , A model time constant calculation unit that changes the time constant stored in the internal model storage unit to this time constant estimated value, and a specific target value based on the two sample target values and a time calculated by the model time constant calculation unit. The process time constant storage unit that stores the constant estimated value and the internal model time corresponding to the control amount based on the specific target value and the time constant estimated value output from the process time constant storage unit during normal target value tracking control operation. A model time constant interpolation setting unit that calculates a constant and changes the time constant in the parameters stored in the internal model storage unit to the calculated value. Controller that.