Data processing: generic control systems or specific application – Generic control system – apparatus or process – Optimization or adaptive control
Reexamination Certificate
1998-08-10
2002-03-05
Wiley, David (Department: 2155)
Data processing: generic control systems or specific application
Generic control system, apparatus or process
Optimization or adaptive control
C700S042000, C700S048000, C700S073000, C710S060000
Reexamination Certificate
active
06353766
ABSTRACT:
BACKGROUND OF THE INVENTION
Field of the Invention
The invention relates to a method for generating control parameters from a response signal of a controlled system with a computer. It relates, further, to a system for adaptive setting of a PID controller with the aid of a neural network.
PID controllers are known from Otto Föllinger, “Regelungstechnik” [Control Technology], 5
th
Ed., Hüttig Verlag, Heidelberg 1995, pages 204-206. They are frequently used in process engineering. For this purpose, the controllers must be calculated in order to be able to control the assigned controlled system. In production engineering, there are, as a rule, no models of the controlled system available for calculating controllers. In order to set a PID controller, an identification is frequently carried out which can be performed, for example, offline on the basis of a measured step response. In this process, the appropriate model with the associated model parameters is sought from a given set of models of the controlled systems. A suitable PID controller can be calculated on the basis of this model.
It would be expedient if the identification step and the controller calculation could be combined to form a single step. A well defined mapping of the step response onto the desired controller parameters would result from this. In order to minimize the outlay arising in this case, it is obvious to approximate the same by means of a neural network. If such a trained neural network is available, suitable controller parameters can be found immediately by firstly recording the step response of the controlled system of a given technical installation and then making it available to the neural network as input. The outputs of the neural network are the controller parameters being sought.
In order to train the neural network, it is possible to prescribe controlled systems for which the associated controller parameters are directly known. It is possible to use, for example, an expert system for setting PID controllers or a library of appropriate controller parameters, which may be determined in some other way, as a basis for training the neural network. This device for training can be termed a teacher, for example. The task of the neural network is compared with that of the teacher and the network weightings are corrected so as to reduce the output error of the neural network. This is performed for a multiplicity of different prescribed system models until the error is sufficiently small for all the examples. Neural networks which satisfy these requirements are known, for example, from J. Hertz, A. Krogh, and R. Palmer, “Introduction to the Theory of Neural Computation”, Addison-Wesley Pub. Co., 1991, pages 115-147. A further solution is known from international publication WO 93/12476. There, the controller parameters are determined directly from the time signals with the aid of the neural network. As a rule such networks must be very large and are difficult to train, and the emitted controller parameters are not reliable.
European patent disclosure EP 0 520 233 A2 discloses a device for indicating the parameters of a transmission system, in which a simulation module is used to compare estimated output signals with measured output signals. Deviations are minimized for optimum parameters.
Instead of the step response of the controlled system, the following text refers to the response signal of the controlled system.
SUMMARY OF THE INVENTION
It is accordingly an object of the invention to provide a method for generating control parameters from a response signal of a controlled system and a system for an adaptive setting of a PID controller with the aid of a neural network which overcomes the above-mentioned disadvantages of the heretofore-known methods and systems of this general type and according to which, characteristics of a Bode diagram are generated from a noisy response signal, wherein the characteristics are largely uninfluenced by the noise. A further object is to specify a system for adaptively setting a PID controller with the aid of a neural network, in which the controller parameters emitted by the neural network ensure an acceptable controller setting even in the presence of a noisy response signal.
With the foregoing and other objects in view there is provided, in accordance with the invention, a method for generating control parameters from a response signal of a controlled system with a computer, the method which comprises:
sampling an input signal and a response signal for generating a sampled input signal and a sampled response signal;
smoothing and optionally differentiating the sampled input signal and the sampled response signal with the aid of a time-variant filter for generating a smoothed input signal and a smoothed response signal;
generating in each case frequency characteristics from the smoothed input signal and the smoothed response signal;
forming a difference between the frequency characteristics in a Bode diagram; and
determining control parameters with the aid of the difference.
With the foregoing and other objects in view there is also provided, in accordance with the invention, a method for generating control parameters from a response signal of a controlled system with a computer, the method which comprises:
sampling an input signal and a response signal for generating a sampled input signal and a sampled response signal;
deconvoluting the response signal with respect to the input signal for generating a smoothed impulse response from the sampled response signal;
forming frequency characteristics in a Bode diagram from the impulse response; and
determining control parameters with the aid of the frequency characteristics.
With the foregoing and other objects in view there is furthermore provided, in accordance with the invention, a method for generating control parameters from a response signal of a controlled system with a computer, the method which comprises:
sampling a response signal for generating a sampled response signal;
smoothing or smoothing and differentiating the sampled response signal with the aid of a time-variant filter for generating a smoothed response signal;
generating frequency characteristics in a Bode diagram from the smoothed response signal;
determining control parameters with the aid of the frequency characteristics.
In preferred embodiments of the methods according to the invention any of the steps of deconvoluting, smoothing or smoothing and differentiating of either the input signal or the response signal or the step of generating the impulse response may be performed in accordance with the formula:
{tilde over (x)}=
Vx
v
=
V
y
where
y
is a vector consisting of samples of the response signal,
V
is a matrix for at least one of deconvoluting, smoothing or smoothing and differentiating,
v
is a vector of a smoothed impulse response,
x
is a vector of the input signal, and {tilde over (x)} is a vector of the smoothed input signal.
In accordance with a further feature of the invention, the matrices
V
for smoothing, for smoothing and differentiating or for deconvoluting are obtained from an energy function having a term specifying a deviation of an approximation from a measured response signal, and having a term specifying a roughness of a reconstructed impulse response.
In accordance with a another feature of the invention, a smoothing deconvolution matrix
V
is obtained by minimizing the following energy function:
&egr;(
b
)=
k
((
y
−
XA
3
b
)
T
D
0
(
y
−
XA
3
b
)+
v
T
D
1
v
+
a
T
D
2
a
+
r
T
D
3
r
+
b
T
D
4
b
)
where k denotes a constant, e.g. k=0.5,
X
a convolution integration matrix which is calculated as a function of the input signal x(t),
A
an integration matrix,
r
=
A
b
,
a
=
A
r
,
v=A a
, and
D
0
,
D
1
D
2
, . . .
D
4
arbitrarily selectable diagonal matrices, and the solution of a minimization is
v
=
V
y
V
=
A
3
(
A
3T
X
T
D
0
XA
3
+
A
3T
D
1
A
3
+
A
2T
D
2
A
2
+
A
T
D
3
A
+
D
4
)
−1
A
3T
X
Backer Firmin
Greenberg Laurence A.
Lerner Herbert L.
Siemens Aktiengesellschaft
Stemer Werner H.
LandOfFree
Method for generating control parameters from a response... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for generating control parameters from a response..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for generating control parameters from a response... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2871630