Professional Documents
Culture Documents
Some Approximations of Fractional Order Operators
Some Approximations of Fractional Order Operators
net/publication/228392562
CITATIONS READS
209 4,127
4 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Vicente Feliu on 08 November 2014.
Abstract
In this paper, starting from the formulation of some possible models of fraction-
al-order systems, several approximations are discussed. For continuous models,
some methods for obtaining an approximated rational function using evaluation,
interpolation and curve fitting techniques are studied. For discrete models, ap-
proximations using Lubich’s formula, the trapezoidal rule, and the application of
continued fractions expansion technique to integro-differential operators formu-
lated in the Z domain are studied. The methods are compared, in both the time
and the frequency domains, using an illustrative example.
Mathematics Subject Classification: 26A33 (main), 93C15, 93C55, 93C80
Key Words and Phrases: fractional calculus, fractional-order systems, frac-
tional-order controllers, integer-order approximations
1. Introduction
At least since the sixties some researchers have been interested in obtaining ap-
proximated integer order models for fractional order systems, or finite dimensional
models for infinite dimensional systems. Most of these researchers worked in elec-
trochemistry where the problem is (roughly speaking) to build an equivalent elec-
trical circuit for processes in which diffusion is present. Even earlier, other authors
Partially supported by FEDER Research Grant IFD97-0755-C02-01
and by VEGA Reserach Grant 1/7098/20.
2 B. M. Vinagre, I. Podlubny, A. Hernández, V. Feliu
used fractional-order models for modeling physical systems, in which the presence
of viscoelasticity was important, or in the characterization of certain class of real
materials. However, applications of fractional-order models in control theory are
relatively new comparing to mentioned applications. As far as the authors of this
paper know, except in old papers of Carlson and Halijak [3, 4], only in the last
two decades the possibility of using fractional order controller has been considered
(see, for example, [1, 2, 12, 14, 17, 18, 20, 19, 21, 23, 24, 28]). In all the cases, the
important point for the purposes of this paper, is that the equivalent circuit or
the fractional-order controller in its practically realizable form is a finite dimen-
sional integer-order system resulting from the approximation of a fractional-order
or infinite-dimensional system.
In this paper the Riemann-Liouville definition of fractional integration and
differentiation is used [13, 16, 24, 27]. For the case of 0 < α < 1 and f (t) being a
causal function of t, that is, f (t) = 0 for t < 0, the fractional integral is defined
as:
Z t
1
D−α f (t) = (t − τ )α−1 f (τ ) dτ, (1)
Γ(α) 0
Of particular interest for the purposes of this paper are the expressions of the
former operators in the Laplace domain. The Laplace transform of the Riemann-
Liouville fractional integral is:
£ D −α f (t) = s −α F (s), (3)
Defining α = 1/q, m = q/2, in each iteration, starting from the initial value
H0 (s) = 1, an approximated rational function is obtained in the form:
where W (ω) is a weighting function, G(ω) is the original frequency response, and
b
G(ω) is the frequency response of the approximated rational function.
Oustaloup’s method. The method [17, 18, 20] is based on the approximation
of a function of the form:
H(s) = sµ , µ ∈ R+ (15)
6 B. M. Vinagre, I. Podlubny, A. Hernández, V. Feliu
by a rational function:
N
Y
b 1 + s/ωk
H(s) =C (16)
1 + s/ωk0
k=−N
Chareff ’s method. This method, proposed in [5], which is very close to Ous-
taloup’s method, is based on approximation of a function of the form
1
H(s) = (18)
(1 + psT )α
where the coefficients are computed for obtaining a maximum deviation from the
original magnitude response in the frequency domain of y dB. Defining
a = 10y/10(1−α), b = 10y/10α , ab = 10y/10α(1−α) (20)
the poles and zeros of the approximated rational function are obtained by applying
the following formulae:
√
p0 = pT b, pi = p0 (ab)i , zi = ap0 (ab)i (21)
The number of poles and zeros is related to the desired bandwidth and the
error criteria used by the expression:
log ωmáx
p0
N = +1 (22)
log (ab)
APPROXIMATIONS OF FRACTIONAL ORDER OPERATORS 7
3. Discrete Models
−α
Performing the PSE of the function 1 − z −1 leads to the formula given
by Lubich for the fractional integral of order α [11]:
∞
X
α k −α
∇−α
T f (nT ) =T (−1) f ((n − k)T ) (25)
k
k=0
where T is the sample period, Y (z) is the Z transform of the output sequence
y(nT ), F (z) is the Z transform of the input sequence f (nT ), and PSE{u} denotes
the expression, which results from the power series expansion of the function u.
Doing so gives:
Y (z) n ±α o
D±α (z) = = T ∓α PSE 1 − z −1 (27)
F (z)
1 − z −1
ω(z −1 ) = 2 (28)
1 + z −1
It is known that the forward difference rule is not suitable for applications to
causal problems [6, 11].
It should be mentioned that, at least for control purposes, it is not very
important to have a closed-form formula for the coefficients, because they are
usually pre-computed and stored in the memory of the microprocessor. In such
a case, the most important is to have a limited number of coefficients because of
the limited available memory of the microprocessor system.
1 − z −1
ω(z −1 ) = 2
1 + z −1
where T is the sample period, CFE{u} denotes the function resulting from apply-
ing the continued fraction expansion to the function u, Y (z) is the Z transform of
the output sequence y(nT ), F (z) is the Z transform of the input sequence f (nT ),
p and q are the orders of the approximation, and P and Q are polynomials of
degrees p and q, correspondingly, in the variable z −1 .
Other approximations
In the paper [9] some others continuous approximations have been studied, which,
in fact, are particular cases of the methods considered here. Some methods are
not considered in the present paper, because they can be used only for particular
values of fractional order (e.g., [3] for α = 0.5).
On the other hand, it is necessary to mention the diffusive realizations pro-
posed in [15] and [8]. While the starting point in these works is quite different
10 B. M. Vinagre, I. Podlubny, A. Hernández, V. Feliu
from the starting point of the methods considered in the present paper, the re-
sulting approximations can be viewed, in the Laplace domain, as rational approx-
imations of the fractional-order operators. Furthermore, these approximations
exhibit a common feature, which we observe in all good rational approximations:
they have poles and zeros interlaced on the negative real axis of the s plane, and
the distance between successive poles and zeros decreases as the approximation
is improved by increasing the degree of the numerator and denominator polyno-
mials. Probably, this fact was noted for the first time in [10], where the following
idea appeared: a dense interlacing of simple poles and zeros along a line in the s
plane is, in some way, equivalent to a branch cut; and sα , 0 < α < 1, viewed as
an operator, has a branch cut along the negative real axis for arguments of s on
(−π, π) but is otherwise free of poles and zeros.
1 1/2
Carlson’s method. Starting from: H(s) = s , H0 (s) = 1, after two itera-
tions we obtain:
1/2
Matsuda‘s method. With G(s) = 1s , finitial = 1, ff inal = 100,
fk ={ 1, 1.7783, 3.1623, 5.6234, 10, 17.783, 31.623, 56.234, 100 }, we obtain:
ωh = 102 , ωb = 10−2 ,
−20 −20
−40 −40
−2 −1 0 1 2 −2 −1 0 1 2
10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
Phase (deg); Magnitude (dB)
−20 −20
−20 −20
−40 −40
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
20 20
H (s) H (s)
0 5 0 6
−20 −20
−10
−20 −20
−30
−40 −40
−50
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
Phase (deg); Magnitude (dB)
20 20
0 H7(s) H8(s)
0
−20
−40 −20
0
−20
−40 −20
−60
−80 −40
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
7 7
6 H (s) 6 H (s)
1 2
5 5
4 4
3 3
2 2
1 1
0 0
0 10 20 30 0 10 20 30
Time (sec.) Time (sec.)
7 7
6 H (s) 6 H (s)
3 4
5 5
4 4
3 3
2 2
1 1
0 0
0 10 20 30 0 10 20 30
Time (sec.) Time (sec.)
7 7
6 H (s) 6 H (s)
5 6
5 5
4 4
3 3
2 2
1 1
0 0
0 10 20 30 0 10 20 30
Time (sec.) Time (sec.)
7 7
6 H7(s) 6 H8(s)
5 5
4 4
3 3
2 2
1 1
0 0
0 10 20 30 0 10 20 30
Time (sec.) Time (sec.)
20 7
10 6
−20 4
20 3
0
2
−20
−40 1
−60 0
−1 0 1 2 0 5 10 15 20 25 30
10 10 10 10
Frequency (rad/sec) Time (sec)
10 7
H10(z) 6
Phase(deg); Magnitude (dB)
0
H (z)
5 10
−10
−20 4
0 3
−20 2
−40 1
−60 0
−2 −1 0 1 2 0 5 10 15 20 25 30
10 10 10 10 10
Frequency (rad/sec) Time (sec.)
20 7
10 6 H (z)
Phase(deg); Magnitude (dB)
H (z) 11
0 11
5
−10
−20 4
0 3
−20 2
−40 1
−60 0
−3 −2 −1 0 1 2 0 5 10 15 20 25 30
10 10 10 10 10 10
Frequency (rad/sec) Time (sec.)
Figure 5: Bode plots and step responses of H9 (z), H10 (z), H11 (z).
5. Controller Realization
In general, there are two possibilities for realizing a controller: a hardware real-
ization based on the use of a physical device, or a software (or digital) realiza-
tion based on a program, which will run on a computer or microprocessor. In
electronics, hardware realizations imply the use of electronic devices or circuits,
implementing the required function as an admittance or impedance function.
ways can be used: producing a microelectronic specific device that, for construc-
tion, has the required admittance or impedance (see, for example, [7]), or realizing
an approximate rational function by using a finite lumped-element network, in a
ladder, tree, cascaded, or lattice topology. Since this paper deals with rational
approximations of the fractional-order operators, only the last way is discussed
below.
In order to be realizable by a finite lumped-element network, the rational
approximation must be a positive real rational function. We see that the contin-
uous approximations given in Section 2, fulfill this requirement for being rational
functions with poles and zeros interlaced on the negative real axis of the s plane.
In view of this, they can be realized as the driving point impedance or admit-
tance of RL or RC networks, that is, as passive RL or RC filters. Furthermore,
these realizations can be complemented by using active electronic devices, such
as operational amplifiers. That is, a more flexible way for realizing the required
functions can be obtained by considering the possibility of using active filters. In
laboratory conditions, we have successfully tested an active filter, which realizes
the fractional integrator of order 0.5 by using a lattice RC network, obtained from
a former version of the Carlson’s method [3].
6. Conclusion
discrete models, our results seem to indicate that there are two suitable techniques
for obtaining discrete approximations of the fractional operators: the discretiza-
tion of an appropriate continuous approximation, or the use of the CFE of the
generating function defined by the trapezoidal rule.
We would like to mention that there is a demand and motivation for further
work towards practical implementation of fractional-order controllers.
References
[20] A. O u s t a l o u p, F. L e v r o n, B. M a t h i e u, and F. M. N a n o t.
Frequency-band complex noninteger differenciator: Characterization and sys-
thesis. IEEE Trans. on Circuits & Systems I: Fundamental Theory and Ap-
plications, 47, No. 1 (2000), 25–39.