JPH0744221A - Dead time compensator for feedback controller - Google Patents

Abstract

PURPOSE:To improve controllability of a feedback controller. CONSTITUTION:A control output y(n) of a controlled system 2 is fed back to an adjusting part 8 by switching changeover switches 5 and 6 and a control input u(n), control output y(n) and differentiated value dy(n) are time sequentially stored in a memory 3. A state prediction part 7 predicats a control output y(t+L) at certain time after the lapse of dead time L based on the control output y(t) at time (t) stored in the memory 3 or the like. Weighting at the state prediction part 7 is adjusted so as to minimize error between this control output and a control output y*(t+L) at the same time stored in the memory 3. By switching the changeover switches 5 and 6, the state prediction part 7 predicts a control output y(n+L) at certain time after the lapse of the dead time L by applying the weight adjusted to the present control output y(n) or the like, and this predicted control output y(n+L) is fed back to the adjusting part 8.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device for compensating the influence of dead time generated in a feedback control device.

[0002]

2. Description of the Related Art Conventionally, feedback control has been widely used as a control method for controlling a controlled object to a desired target value. As is well known, as shown in FIG. 3, the feedback control device feeds back the control output (control amount) y (n) at the current time n output from the controlled object 2 to the adjusting unit 8, and causes the adjusting unit 8 to perform feedback. By calculating the control input (operation amount) u (n) according to the deviation e between the fed back control output y (n) and the target value v by the operation amount calculation unit 1 and adding this to the controlled object 2, Control output y
(N) is controlled so as to reach the target value v.

This feedback control device corrects the control input u (n) so that the deviation e between the target value v and the control output y (n) becomes smaller than the other. .

[0004]

However, the control input u
(N) or the output y (n) includes a large dead time, the control output y (n) includes the control input u.
This means that the information necessary for correcting (n), that is, the effect of the control input u (n) added up to the time n or the sign thereof is not included.

As described above, when the control output that does not include any information necessary for correcting the control input is fed back and the control is performed, the control becomes unstable, excessive overshoot occurs, and the settling time increases. Therefore, there is a fear that the control performance may be deteriorated.

The present invention has been made in view of the above circumstances, and predicts the control amount after the dead time elapses and feeds it back to compensate the influence of the dead time and improve the control performance. It is intended to provide a device capable of

[0007]

In order to achieve the above object, in the present invention, the control amount output from the controlled object is fed back to the adjusting unit, and the deviation between the control amount fed back to the adjusting unit and the target value. In the feedback control device for controlling the controlled object so that the controlled variable reaches the target value, by adding the manipulated variable to the controlled object, a detection means for detecting each of the controlled variable and the manipulated variable. Storage means for chronologically storing the control amount and the operation amount detected by the detection means, a predetermined weighting is given to each of the control amount at the predetermined time and the operation amount at the predetermined time, and the weighted control is given. Predicting calculation means for predictively calculating a control amount at a time after a lapse of a dead time from the predetermined time based on an amount and an operation amount; First switching means for switching and outputting to either the storage means or the prediction calculation means, and either the control amount output from the controlled object or the control quantity calculated by the prediction calculation means is switched to the adjustment section. In the learning mode, the first switching means is switched to the storage means side, and the second switching means is switched to the control target side, and is detected by the detection means. The control means is caused to perform a process of storing the control amount and the operation amount stored in the storage means in time series, and then the predetermined time is stored based on the control amount at the predetermined time stored in the storage means and the operation amount at the predetermined time. Causing the prediction calculation means to perform a prediction calculation of the control amount at the time after the dead time has elapsed,
Then, the error between the control amount at the time after the dead time from the predetermined time calculated by the prediction calculation means and the control amount at the time after the dead time from the predetermined time stored in the storage means is the minimum. In the control mode, the first switching means is switched to the prediction calculation means side, and the second switching means is changed to the prediction calculation means side. By switching, it causes the prediction calculation means to perform a calculation for predicting the control amount at the time after the dead time from the current time by the adjusted weighting, and the control amount output from the prediction calculation means to the adjustment part. I try to give feedback.

[0008]

According to this structure, in the learning mode, the first
And the second switching means is switched to the storage means side.
The switching means is switched to the controlled side, and the control amount and the manipulated variable in the conventional feedback control system in which the predictive control is not performed are stored in time series in the storage means.

Then, the predictive calculation means performs a calculation for predicting the control quantity at the time after the dead time has elapsed from the predetermined time, based on the control quantity at the predetermined time and the operation amount at the predetermined time stored in the storage means.

Next, the error between the control amount calculated by the predictive calculation means at the time after the dead time has elapsed from the predetermined time and the control amount stored at the time after the dead time from the predetermined time stored in the storage means is minimized. As described above, the weighting adjustment in the prediction calculation means is performed.

In the control mode, the first switching means is switched to the predictive computing means side and the second switching means is switched to the predictive computing means side so that the control amount and the manipulated variable at the present time are switched to the predictive computing means. In addition, by the weighting adjusted in the learning mode, the calculation for predicting the control amount at the time after the dead time from the current time is performed. Then, the predicted control amount output from the predictive calculation means is fed back to the adjusting unit.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a dead time compensating device of a feedback control device according to the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of the embodiment.

The control device shown in the figure basically constitutes a feedback control system in the same manner as shown in FIG.

That is, the control output (control amount) y (n) at the current time n output from the controlled object 2 is subjected to a predetermined process described later, that is, the predicted control amount y (n + L) is adjusted. The control input (operation amount) u according to the deviation e between the control output y (n + L) fed back to the adjusting unit 8 and fed back to the adjusting unit 8 and the target value v.
The control amount y (n) is controlled to reach the target value v by computing (n) in the manipulated variable computing unit 1 and adding it to the controlled object 2.

In order to perform the predictive calculation of the predictive control amount, the embodiment has the following configuration different from the conventional one.

That is, the control output y (n) has a differentiating circuit 4
Added to. The differential circuit 4 calculates a differential value dy (n) / dt (hereinafter abbreviated as dy (n)) of the control output y (n), and this is added to the changeover switch 6. The inputs / outputs u (n) and y (n) of the controlled object 2 are directly applied to the changeover switch 6.

The change-over switch 6 changes over and outputs the input control input u (n), differential value dy (n) and control output y (n) to the memory 3 or the state predicting section 7. When the changeover switch 6 is changed over to the contact 6 a, the control input u (n) or the like inputted to the switch 6 is outputted to the memory 3. Further, when the changeover switch 6 is changed over to the contact 6b, the control input u inputted to the switch 6 is inputted.
(N) and the like are output to the state prediction unit 7.

In the memory 3, the control input u (n) and the like input from the switch 6 are stored in time series. That is, the control input u (n), the differential value dy (n), and the control output y (n) for each time n are stored at a predetermined sampling period.

In the state predictor 7, the control input u (n), the differential value dy (n), and the control output y (n) at the predetermined time n.
Given a predetermined weighting to each of the control values u (n) and the like, and based on the weighted control input u (n)
For predictive calculation. Changeover switch 6 has contact 6a
When it is switched to, the prediction calculation is performed based on the control input u (t) at the predetermined time t stored in the memory 3, and the predicted control amount y (t + L) is output. Further, when the changeover switch 6 is switched to the contact 6b, the prediction calculation is performed based on the current control input u (n) of the controlled object 2 and the like, and the predicted control amount y (n +
L) is output.

The controlled variable y (n) currently output from the controlled object 2 and the predicted controlled variable y (n + L) output from the state predicting section 7 are added to the changeover switch 5.

The change-over switch 5 switches and outputs the input control amount y (n) or predicted control amount y (n + L) to the adjusting section 8. When the changeover switch 5 is changed over to the contact 5a, the control amount y input to the switch 5 is
(N) is fed back to the adjusting unit 8. When the changeover switch 5 is changed over to the contact 5b, the predictive control amount y (n + L) input to the switch 5 is fed back to the adjusting section 8.

Processing in the learning mode While the changeover switch 6 is changed over to the contact 6a, that is, the side for outputting to the memory 3, the changeover switch 5 changes the control amount y (n) from the contact 5a, that is, the controlled object 2. When switched to the input side, the processing in the learning mode is started.

In this case, normal normal feedback control is performed in which the controlled variable y (n) output from the controlled object 2 is fed back to the adjusting section 8. Then, while such feedback control is being performed, the time-series data (u (n), dy) of the input / output state of the controlled object 2 is obtained.
(N) and y (n)) are stored in the memory 3 every predetermined sampling period.

The state predicting section 7 stores data (u (t), dy) stored in the memory 3 and showing the input / output state at time t.
By inputting (t) and y (t), the calculation as shown in FIG. 2 is performed.

As shown in FIG. 2, the state predictor 7 is composed of a multilayer neural network, and the element i on the input side is
Input / output data u (t), dy (t), y (t), respectively
Is input and yi is output. The sum xj of each yi multiplied by the predetermined synapse weight Wij is input to the element j on the output side.

That is, xj is expressed by the following equation (1).

Xj = ΣWij · yi (1) Then, the output yj of the element j on the output side is expressed by a function value when the total sum xj of the inputs is given to the monotonically increasing function f as a variable. That is, yj is represented by the following equation (2).

Yj = f (xj) (2) Here, in the back propagation, when the desired output y * j to be output by the output element j with respect to a certain input vector (pattern) is set, the following E is evaluated. There is a learning rule that minimizes the error function.

[0030] Here, yj is the predicted control amount y (t + L) at the time after the elapse of a certain dead time L. Therefore, the control amount y * at the time after the same dead time L stored in the memory 3 has elapsed
(T + L) is read out and substituted into y * j in the above equation (3). Then, the synaptic weight Wij is adjusted so as to minimize the error function E.

As described above, the predicted control amount y (t + L) calculated by the state predicting unit 7 and the same time t stored in the memory 3 are stored.
By feeding back the error with the controlled variable y * (t + L) of the controlled object 2 at + L to the state prediction unit 7,
The synapse weight Wij of the state prediction unit 7 is adjusted.

When the adjustment of the synapse weight Wij is completed in this way, the changeover switches 5 and 6 are switched to the contact points 5b and 6b, respectively, and the processing shifts to the predictive control mode.

Processing in predictive control mode In this predictive control mode, output y of controlled object 2
Instead of directly feeding back (n) to the adjusting unit 8, the predicted control amount y (n +) calculated by the state predicting unit 7 is calculated.
L) is fed back to the adjusting unit 8.

In other words, the state predicting unit 7 has the current input / output state (u (n), dy (n), y (n)) of the controlled object 2 in it.
Is input, and the synapse weights Wij adjusted in the learning mode are given to them, and the control amount y (n + at the time n + L after the elapse of the dead time L from the current time n is given.
L) is predicted and calculated, and the predicted control amount y (n + L) is fed back to the adjusting unit 8. In this case, the error between the actual control amount y (n + L) after the dead time elapses and the predicted control amount y (t + L) calculated by the state prediction unit 7 is set so that the evaluation function E becomes minimum in the learning mode. Since the synaptic weight Wij is adjusted to be the minimum, the control amount y (n + L) can be accurately predicted, and the influence of the dead time can be reduced.

As described above, according to the embodiment, the dynamic characteristics of the dead time system are automatically acquired in the learning mode, and at the time of control, the control amount y (n + L) after the dead time L has elapsed is predicted and fed back. Therefore, even a control system with dead time can be treated as having no dead time, the instability of feedback control is improved compared to the conventional method, and the amount of overshoot is reduced,
The control performance can be improved without shortening the settling time.

In the embodiment, the predictive calculation is performed based on the differential value dy of the control output in addition to the control input u and the control output y. The point is that the input / output state of the controlled object 2 is wasteful. It suffices that the control output after the elapse of time can be predicted, and in some cases, the prediction calculation can be performed without using the differential value dy as a parameter.

[0037]

As described above, according to the present invention, the control amount after the dead time has elapsed is predicted and is fed back, so that the influence of the dead time is compensated, and the control instability and excessive control are increased. It is possible to dramatically improve the control performance without causing such overshoot and increasing the settling time.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of a dead time compensation device of a feedback control device according to the present invention.

FIG. 2 is a diagram used to explain the contents of calculation in a state prediction unit shown in FIG.

FIG. 3 is a block diagram showing a general configuration of a feedback control system.

[Explanation of symbols]

 2 Control object 3 Memory 5 Changeover switch 6 Changeover switch 7 State prediction unit 8 Adjustment unit

Claims (2)

[Claims] 1. A control amount output from a control target is fed back to an adjusting unit, and an operation amount corresponding to a deviation between a control amount fed back to the adjusting unit and a target value is added to the control target. In a feedback control device that controls the controlled object so that the control amount reaches the target value, a detection unit that detects each of the control amount and the operation amount, and a control amount and an operation amount detected by the detection unit Storage means for storing in series, giving a predetermined weight to each of the control amount at a predetermined time and the operation amount at the predetermined time, and after a dead time from the predetermined time based on the weighted control amount and operation amount Predictive calculation means for predictively calculating the control amount at the time, and the detection result of the detection means is switched and output to either the storage means or the predictive calculation means. And a second switching means for switching and outputting to the adjustment section any one of the control amount output from the controlled object and the control amount calculated by the predictive calculation means. Occasionally, the first switching means is switched to the storage means side and the second switching means is switched to the control target side, and the control amount and the operation amount detected by the detection means are stored in the storage means. A process of storing the control amounts at a predetermined time and a predictive calculation of the control amount at a time after the dead time from the predetermined time are performed based on the control amount at the predetermined time and the operation amount at the predetermined time stored in the storage means. Is executed by the predictive calculation means, and then from the control amount at the time after the dead time has elapsed from the predetermined time calculated by the predictive calculation means and the predetermined time stored in the storage means. The weighting adjustment in the prediction calculation means is performed so that the error with the control amount at the time after the dead time elapses is minimized, and in the control mode, the first switching means is switched to the prediction calculation means side. At the same time, the second switching unit is switched to the prediction calculation unit side, and the prediction calculation unit is made to calculate the control amount at the time after the dead time has elapsed from the current time by the adjusted weighting. A dead time compensating device in a feedback control device configured to feed back the control amount output from the prediction calculation means to the adjusting unit. 2. A differential value of the control amount other than the control amount and the operation amount is detected by the detection means, and the differential value of the control amount is stored in the storage means in time series, and the prediction calculation means is provided. 2. The dead time compensating device in the feedback control device according to claim 1, wherein the differential value of the control amount is given a predetermined weighting to perform the prediction calculation.