JP2635561B2 - Adaptive control device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial application field) The present invention relates to a method of controlling a control parameter of a controller for controlling a control object whose dynamic characteristic is unknown or constantly changing, and changing a dynamic characteristic of a control system to a dynamic characteristic of a reference model. The present invention relates to a model reference type adaptive control device having a function of self-adjusting to asymptotically match.

(Prior Art) A model reference adaptive control method is known as an effective control method for a control target whose dynamic characteristics are unknown. A control device employing this method usually has a structure as shown in FIG.
A controlled object 1 and a controller 2 constitute a closed loop system, and each signal of the control system is taken into a parameter estimating device 3.
While the parameter estimating device 3 estimates parameters related to the dynamic characteristics of the controlled object 1 in real time, the controller 2 is constantly adjusted based on the parameters, whereby the control amount y (t) of the controlled object 1 is set as a standard. Model 4 output
ym (t) asymptotically. Conventionally, various types of parameter estimation algorithms in the parameter estimation device 3 have been proposed. Most of them use techniques similar to the recursive least squares method.

However, such a conventional model reference type adaptive control device exhibits excellent characteristics under certain limited conditions, but has the following problems in practical use. In other words, in the sequential least squares method, by applying an exponentially declining weight that decreases exponentially to the past to the observed control system signal, the control object whose characteristics gradually change is slowly applied. The characteristics of the control system can be kept close to the optimum state while following. However, when the control target is an actual plant, the characteristics of the plant often change suddenly with changes in working conditions and set points, and in such a case, the estimation parameters are reduced by the sequential least squares method. It may not be able to track enough. Also,
It is possible to increase the speed of following the characteristic change by adjusting the exponential decay type weight coefficient, but in this case, the parameter adjustment becomes too sensitive to the observation data. As a result, for example, when an unknown disturbance is applied to the plant, the estimation parameters may be largely disturbed and the control system may be temporarily disturbed.
Further, when the control system is in a settling state for a long time, the target positive definiteness of the estimated parameter covariance matrix in the algorithm is broken, and a so-called parameter burst phenomenon occurs. Although a method of resetting the parameter covariance matrix to follow a sudden change in the characteristics of the control target has been proposed, this method also has a problem that a transient disturbance of the control system occurs at the time of resetting.

(Problems to be Solved by the Invention) As described above, even if it is attempted to control a controlled object, such as a plant, which has a relatively large amount of disturbance and whose dynamic characteristics change every moment by using the conventional model reference adaptive controller. It has been difficult to achieve both the followability of the estimated parameters and the stability against disturbance.

Therefore, the present invention provides a method in which a parameter estimation value is not temporarily disturbed even when a disturbance is applied, a control system can be kept stable, and a parameter estimation value can be quickly followed even when a sudden change in dynamic characteristics of a controlled object occurs. It is an object of the present invention to provide a model reference type adaptive control device having an adaptive algorithm that can be used.

[Invention Configuration] (Means for Solving the Problems) In the adaptive control device according to the present invention, a storage device for storing input / output signal data to be controlled is provided.
The L × I virtual input / output data obtained by sequentially updating and storing the L input / output signal data from the past to the present obtained by the sampling and sequentially reading this data N times is obtained. The parameter is estimated using a parameter estimation algorithm.

(Operation) In the parameter estimation algorithm in the model reference adaptive control system, parameters are estimated after exponentially decay weighting of the observed signal. This is essentially the same as the exponentially weighted sequential least-squares method in a discrete-time model reference adaptive control system such as the adaptive algorithm proposed by I, D Landau. It can be said that the exponential attenuation type weighting is performed as shown.

In the apparatus of the present invention, observation data having a length L is connected N times, so that the observation data has weights as shown in FIG. 2 (b), that is, as shown in FIG. 2 (c). This is equivalent to performing weighting. by this,
Since an estimated value obtained by averaging sections of data length L is obtained, L
If is taken to some extent, it is possible to prevent the estimation parameters from being greatly disturbed by the influence of temporary disturbance. Also, by increasing the number of times of reading N, the weight for the section of the data length L can be increased, and the weight for the observation data before t ₀ −L can be sharply reduced with respect to the current time t _0. When the dynamic characteristic of the target changes suddenly, it becomes possible to quickly follow the estimated parameter without being affected by the past data. Therefore, it is possible to achieve both the stability of the estimated parameter with respect to the disturbance and the followability with respect to the characteristic fluctuation of the controlled object.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

In FIG. 1, reference numeral 1 denotes a control object, and reference numeral 2 denotes a reference model representing preferable characteristic specifications of the control system. The control target 1 is driven by an operation amount u (t), and generates a control amount y (t).
In this embodiment, a zero-order holder 3 for holding a sample value u (k) of an operation amount at a constant value during a sampling period τ [sec] and a sampler for sampling a control amount y (t) every τ [sec] 4, the control target 1 is apparently a discrete time system. Therefore, various signals r (k), y (k), u (k) and the like used in the following description mean a discrete time sequence with a sampling period τ [sec].

The control system outputs the control target value r (k) and the output of the reference model 2.
Output error ε between ym (k) and control amount y (k) of controlled object 1
(K), control amount y (k), etc., and operation amount u (k)
5, a filter 6 for performing a certain type of filtering on the manipulated variable u (k), a 0th-order folder 3, and a control target 1
And a closed loop system composed of a sampler 4.

The operation amount u (k) and the control amount y (k) are stored in the data memory 8 once via the data flow switching unit 7 or directly applied to the filter 9 and then sent to the parameter estimating unit 10. . The parameter estimating unit 10 estimates parameters relating to the control target 1 or directly control parameters. These parameters are checked by the parameter check unit 11 to determine whether they are appropriate values. Then, the parameters of the controller 5 are adjusted based on the parameters.

On the other hand, reference numeral 12 in the figure denotes a control performance evaluation unit, and the control performance evaluation unit 12 outputs an error ε between the reference model 2 and the control target 1.
It has a function of monitoring (k) and monitoring and evaluating control performance. In the figure, reference numeral 13 denotes a signal excitability monitoring unit, which monitors and evaluates the excitability of the signal u (k), y (k) used for parameter estimation, that is, how many frequency components it has. Has a function.

In the figure, reference numeral 14 denotes a control system control unit. The control unit 14 monitors the control performance monitoring unit 12 and the signal excitability monitoring unit 13, adjusts a check criterion in the parameter check unit 11 based on the evaluation result, and identifies an identification signal described later. It has a function of controlling the activation of the generator 15, adjusting the data fetch timing in the data flow switching unit 7, adjusting the data length L, the number of readings N, and the like. When an activation signal is given from the control system control unit 14, the identification signal generation unit 15 generates a persistent exciting signal such as an M-sequence signal, and outputs this signal via a changeover switch 16 to a control target r (k) or an operation signal. selectively added to u (k).

In the figure, reference numeral 17 denotes an input / output device, which is used for helping an operator to read out various information related to the control system collected in the control system control unit 14 or changing a value. .

Next, an adaptive algorithm in the model reference adaptive control device configured as described above will be described.

Filter 6, Holder 3, Control target 1 and Sampler 4
Can be considered as a discrete-time system such as

G (Z^-1) = ｛B (Z^-1) / A (Z^-1)｝ ・ Z^-d = ｛B₀Z^-d+ B₁Z^-d-1+ …… + b_mZ^-dm｝ / ｛1 + a₁Z^-1 + …… + a_nZ^-n規 Normative model 2 representing desirable dynamic characteristics is Gm (Z^-1) = ｛B (Z^-1) / Am (Z^-1)｝ ・ Z^-d = ｛₀Z^-d+₁Z^-d-1+ ... + Z^-D-｝ / ｛1 +₁Z^-1+ …… + Z⁻…… (1) Where Z^-1Is the time τ [sec]
It is a time transition operator.

The adaptive control operation is executed by the following algorithm.

First, the control amount y (k) and the operation amount u (k) are
Through.

(K) = F ₁ (Z ⁻¹ ) y (k) (2) (k) = F ₁ (Z ⁻¹ ) u (k) (3) However, the filter 9 is composed of F ₁ (Z ⁻¹⁾ _{) = {1 + g 1 Z} -1 + g 2 Z -2 + ... + g ng Z -ng} / {1 + f 1 Z -1 + f 2 Z -2 + ... + f nt Z -nt} ... (4) in becomes a digital filter For example, F ₁ (Z ⁻¹ ) = 1−Z ⁻¹ (5) is used to eliminate the influence of bias disturbance applied to the control target 1.

Next, the following equation is used as the estimation model. From observation data (k), (k) (k = 0,1,2, ...)
The parameter estimation unit 10 successively estimates the estimation parameter θ (k).

However, D (Z ^-1 ) = 1 + d ₁ Z ^-1 + d ₂ Z ^-2 + ... + d _nd Z ^-nd (7) is that the root x of D (x) = 0 is centered on the origin of the complex plane It is assumed that the selection is made to be outside the unit circle. At this time, the parameter estimation algorithm is as follows.

At time k = 0, (Matrix of only the diagonal is α, let α≫1) θ ₍₀₎ = [1,0, ..., 0] ^T ... (9) At time k> 0, step 1 data sequence (k), (k ), The observation vector, φ _(kd) = [ _(kd) , _(kd-1) ,… _(k-2d-m-1) , _(kd) , _(kd-1) ,…
_(kn-1-d) ]… (10) Further, D (Z ^-1 ) (k) = _(k) + d _{1 (k-1)} + d _{2 (k-2)} +... + D _{(nd) (k-nd)} .

Step 2 Ask for.

Here, 0 <λ ₁ (k) 1, 0 <λ ₂ (k) 2 μ (k) −δδ> 0 (15) Next, the controller 5 uses the estimation algorithms (8) to (1)
Estimated parameter vector θ sequentially obtained by 5)
(K), the past value u (i) of the operation input (i = k−1, k−
2,...) And the control amount y (i) (i = k, k−1, k−2,...) From the operation input u (k), uF (k), u at time k.
(T) is calculated by the following calculation.

Step 1 ym _{(k + d)} = {Bm (Z ^-1 ) / Am (Z ^-1 )} · r (k)... (16), the d-step future value of the output ym (k) of the reference model 2 is Ask.

Step 2 estimated value at time k, θ (k) = [ b r0, b r1 ... b rm + d-1, s 0, s 1, ... s n-1] T ... (1
7) From the above, u (k) is obtained by the following equation.

u (k) = 1 / b r0 [ym (k + d) + dym (k + d-1) + ... + d nd ym (k + d-nd) -s 0 y (k) -s 1 y (k- ₁₎ −… −s _(n-1) y _{(k−n + 1)} −b _r1 u _(k-1) −b _r2 u _(k-2) −… −b _{rm + d-1} u _{(km- d + 1)} ]…
(18) uF (k) = F ₂ (Z ⁻¹ ) u (k) (19) u (t) = H (s) u (k) (20) (H (s) is a zero-order hold element However, when | b _r0 | <ε, b _r0 = sgn (B _r0 ) · ε _br .

In general, filter 6 of the operation ^{_{end, F 2 (Z -1) =}} [1+ 1 Z -1 + 2 Z -2 + ...... + ng Z -ng]
_{^{/ [1+ 1 Z -1 + 2}} Z -2 + ...... + t Z -t] ... (2
1) For example, F ₂ (Z ⁻¹ ) = 1 / (1−Z ⁻¹ ) (22) is used in order to eliminate a steady-state error with respect to a bias disturbance applied to the control target 1. In order to prevent the input to the operation terminal from becoming too large, the amplitude limiter, Is also available.

Next, the operation of the algorithm that is a feature of the present invention in the control system control unit 14, the data flow switching unit 7, and the data memory 8 will be described.

The control system control unit 14 sends to the data flow switching unit 7 a trigger signal for setting a timing according to the state of the control system, a data length L, and the number N of repeated readings. The data flow switching unit 7 starts storing the input / output signals u (k) and y (k) of the control target 1 in the data memory 8 at the same time as receiving the trigger signal at time k. When the L pieces of sampling data are stored in the data memory 8, the L pieces of data y (i), u (i) (i = k, k + 1,...)
.. K + L-1) are read out and sent to the parameter estimating unit 10 through the filter 9. This reading is performed N times. During this time, new input / output data u (i), y (i) (i = k + L, k
+ L + 1,...) Are stored in another storage area of the data memory 8. When the sampling time τ is sufficiently longer than the operation time of the above algorithm, the data is read out N times, and the operations (2), (3) and (10), (11), (12) of the parameter estimation algorithm described above. ),
(13), (14), and (15) end instantly.

FIG. 3 shows these timing relationships. In addition to the operation modes shown in FIG. 3, the following operation modes of the estimation algorithm can be considered.

(1) Normally, the operation of the conventional model reference adaptive control system is executed, and a trigger signal is generated only when it is detected that disturbance has been applied or the dynamic characteristic of the control target has changed,
Perform the above operation.

(2) The above operation is performed intermittently, and the parameter estimation algorithm is stopped at other times.

(3) Normally, the parameters of the controller 5 are fixed to a PID controller or the like, and the above-described operation is executed only when it is detected that the characteristics of the control target have changed.

These can all be realized by controlling the trigger signal of the control system control unit 14.

The above is the parameter estimation algorithm in the adaptive control device of the present embodiment.

The present invention is not limited to the embodiment described above. That is, in the above-described embodiment, one-input, one-output control target is targeted. However, the adaptive algorithm can be easily extended to other input / output systems to cope with it. Although the present invention is mainly based on a model reference type adaptive control system, the method in the parameter estimation algorithm is applied to other adaptive control systems such as a self-tuning regulator and a PID auto-tuning controller in exactly the same way. Can be used.

[Effect of the Invention] According to the model reference adaptive control device according to the present invention,
A storage unit for temporarily storing input / output data to be controlled is used, and in this storage unit, the L input / output signals from the past to the present obtained by sampling are stored while being sequentially updated, and are stored N times. L obtained by reading
Since the parameters are estimated using a method similar to the parameter estimation algorithm used in the conventional adaptive control system using × N data strings, the estimated value is disturbed by disturbance applied to the control target, It is possible to prevent the system from becoming temporarily unstable, and at the same time, it is possible to quickly follow the change in the dynamic characteristic of the controlled object with the estimated parameter.

[Brief description of the drawings]

FIG. 1 is a block diagram of a model reference type adaptive control device according to an embodiment of the present invention, and FIG. 2 explains the weighting of observed signals in a parameter estimation algorithm in the device of the present invention in comparison with a conventional algorithm. Figure for
FIG. 3 is a diagram for explaining the flow of observation signal data and the operation of the parameter estimation algorithm in the parameter estimation algorithm of the device of the present invention, and FIG. 4 is a block diagram of a conventional model reference adaptive control device. 1 ... Control target, 2 ... Reference model, 5 ... Control unit, 7 ... Data flow switching unit, 8 ... Data memory, 10 ... Parameter estimation unit, 11 ... Parameter check unit, 12 ... Control Performance evaluation unit, 13 ... Signal excitation monitoring unit, 14 ... Control system control unit.

Claims (2)

(57) [Claims] A controller for calculating an operation amount signal as an input of the control object from a control amount signal and a control target value signal as an output of the control object; a reference model representing preferable characteristic specifications of the control system; Parameter estimating means for estimating a parameter relating to the dynamic characteristic of the controlled object based on the input / output signal of the controlled object, and adjusting the parameter of the controller in accordance with the parameter obtained by the means; And a means for storing the input / output signal of the control object, wherein the storage device stores the input / output signal of the control object, and the storage device obtains the L from the past to the present obtained by sampling in the adaptive control device. L × N data strings obtained by storing the input / output signal data while sequentially updating them and reading the data continuously N times are converted into one input / output signal data. Regarded, adaptive control apparatus characterized by that was through the filter is introduced into the parameter estimation section so as to estimate the parameters. 2. An input / output to be stored in the storage device according to a magnitude of a disturbance applied to the control target, a temporal change rate of a dynamic characteristic of the control target, and a degree of continuous excitation of the input / output signal. Number L of signal data and number N of repeated readings
2. The adaptive control apparatus according to claim 1, further comprising means for automatically adjusting the value of the adaptive control.