Professional Documents
Culture Documents
2
Systems and Control, Marrakech, Morocco, October
23-25, 2019
Redouane CHAIBIa , Ismail ER RACHIDa , El Houssaine TISSIRa , AbdelAziz HMAMEDa , Fernando TADEOb
Abstract— The static output feedback (SOF) control for systems, as solving SOS-based conditions of polynomial sys-
continuous-time polynomial Takagi-Sugeno (T-S) systems is tems is less conservative than using LMI-based conditions for
solved in this paper. Using a Polynomial Lyapunov Function, T-S systems. It is worth to note that the static output feedback
sufficient conditions are derived, in the form of sum-of-squares
(SOS). Thanks to some special derivations, these conditions control for polynomial systems is not widely studied as its
do not include neither transformation matrices nor equality linear counterpart. We can cite [10], where the robust H∞
constraints, which simplifies the numerical solution. Moreover, control of uncertain polynomial systems was studied. The
the proposed conditions can provide controllers that are less stabilization and H∞ dynamic output feedback control for
conservative than those derived from existing results. This is discrete-time polynomial systems were discussed in [11] and
illustrated with some examples at the end of the paper, that
show the applicability of the proposed design approach. [12] respectively. Based on polynomial Lyapunov functions,
polynomial SOF fuzzy controllers has been recently derived
I. I NTRODUCTION in terms of SOS in [13]. Moreover, a static output feedback
H∞ controller for polynomial systems with parametric un-
Takagi-Sugeno (T-S) models [1] are receiving a great
certainties is studied in [14]. In these addressed works, the
deal of attention thanks to their properties as universal
conditions are bi-linear in the decision variables, so iterative
approximators for nonlinear systems, making possible to
algorithms based on SOS decomposition that depend on
extend linear system tools to nonlinear systems. Indeed, a T-
the initial values, has been developed to numerically solve
S system represents exactly a nonlinear system (in a certain
the stabilization problem. However, how to select the initial
region) as a set of linear models interpolated by membership
values is still an open problem.
functions. Numerous techniques for T-S systems have been
Motivated by these earlier developments, we propose in
discussed in the literature in both the continuous-time and
this paper an SOF control of continuous-time polynomial
discrete-time: [2], where the stability and stabilization of T-
systems in the form of SOS. The SOF controller guarantees
S fuzzy time-varying delay systems is investigated with pa-
that the closed-loop system is asymptotically stable. The
rameter uncertainties; in [3] sufficient conditions for discrete-
main advantage of the proposed method is to overcome the
time T-S fuzzy systems, subject to actuator saturation were
problem of bi-linear matrix inequalities, and to reduce the
presented, with the associated stabilization conditions solved
conservatism with respect to the existing methods. Moreover,
using LMIs. Static output feedback (SOF) control strategies
the theoretical proof, shows that the proposed design condi-
are important in practice as they are simpler to implement,
tions include the existing ones published in [5], [6] as special
and more reliable than state-feedback controllers (See, for
cases. The rest of this paper is arranged as follows: Section
instance, [4] and the references therein). It must be pointed
II presents a system description and some preliminaries.
out that the stabilization problem with SOF controllers has
The main results are given in Section III, while Section
not been sufficiently investigated for T-S systems as it is
IV presents some illustrative examples to demonstrate the
quite hard. In this context we can cite [5]–[8] and [9].
validity and the advantages of the proposed methodologies.
In particular, a numerical procedure based on the cone
Some conclusions are drawn in Section V.
complementary algorithm was given for the design of SOF
stabilizing controllers of T-S systems in [7]. The design of II. P ROBLEM FORMULATIONS
discrete-time SOF controllers in the presence of bounded Consider a polynomial T-S system, in which the i-th rule
delays was solved in [8]. is represented as follows:
Generally speaking, polynomial systems are a generaliza- Plant Rule i: IF ς1 (t) is Mi1 AND . . . AND ςs (t) is Mis
tion of the T-S systems that makes possible to deal with THEN
a wider class of nonlinear systems. Sum of squares (SOS)
approaches are currently being actively investigated for these ẋ(t) = Ai (x(t))x̃(x(t)) + Bi (x(t))u(t)
(1)
y(t) = Ci (x(t))x̃(x(t))
a LESSI, Department of Physics Faculty of Sciences Dhar El Mehraz,
B.P. 1796 Fes-Atlas Morocco c.redouane.chaibi@gmail.com, where i = 1, 2, . . . , r, with r the number of rules, ς (t) =
ismail.errachid@gmail.com, elhoussaine.tissir@usmba.ac.ma, [ς1 (t)ς2 (t) . . . ς p (t)] are known premise variables, Mi j are
hmamed − abdelaziz@yahoo. f r,
b Instituo de Procesos Sostenibles, Universidad de Valladolid, 47005 fuzzy sets, u(t) ∈ Rm1 is the control input, y(t) ∈ Rm2
Valladolid, Spain. Fernando.Tadeo@uva.es is the measurement output, Ai (x(t)), Bi (x(t)), and Ci (x(t))
+ Bi (x(t))K j (x(t))Cl (x(t))]x̃(x(t)) −ν2T Ωii j (x) + Ωi ji (x) + Ω jii (x) + εi j (x)I ν2 is SOS
(13)
where 1 ≤ i = j ≤ r
hi = hi (ς (t)), h j = h j (ς (t)), hk = hk (ς (t)).
−ν2T Ωi jl (x) + Ωil j (x) + Ω jil (x) + Ω jli (x) + Ωli j (x)
We now provide some definitions and lemmas that would
+ Ω jil (x) + Ω jli (x) + Ωli j (x) + Ωl ji (x) P(x̂)T (x)Āi jl + ĀTijl T (x)T P(x̂) + Ṗ(x̂) < 0 (24)
r r r
Let us consider the Lyapunov function given by [15]
=∑∑ ∑ hi h j hl Ωi jl (x) < 0
i=1 j=1 l=1
V (x(t)) = x̃(t)T P(x̂)x̃(t) (25)
which is verified if
⎡ ⎤ the derivative of (25) with respect to time satisfies
Ω11
i jl ∗ ∗
r r r
Ωi jl (x) = ⎣ Ω21 Ω22 ∗ ⎦<0 V̇ (x(t)) = ∑ ∑
i jl
Ω31
i jl
Ω32 Ω33
(16)
∑ hi h j hl {x̃(t)T [ĀTijl T (x)T P(x̂) (26)
i jl i jl i jl i=1 j=1 l=1
0.4
ẋ(t) = A3 x(t) + B3 u(t)
T HEN 0.2
y(t) = C3 x(t)
Plant Rule 4: IF x1 (t) is M12 and x3 (t) is M22 0
ẋ(t) = A4 x(t) + B4 u(t) -0.2
T HEN
y(t) = C4 x(t)
0 2 4 6 8 10 12 14 16 18 20
Time (sec)
1.5 1
x 2 (t)
1 0.5
0
0.5
-0.5
0
-1
-0.5 -1.5
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (sec) Time (sec)
Fig. 2. Evolution of the system state x2 (t) Fig. 5. Evolution of the closed loop states in Example 1
0.5
Theorem 1
0.4
Theorem 1 [5] it is possible to compare our results with previous results in
0.3
the literature. For instance, the LMI methods in [6], [5] and
[21] are infeasible when fixing a = 4 and b = 4; on the other
0.2
hand, the proposed Theorem 1 gives a feasible solution even
0.1
when the Lyapunov matrix P and the SOF gain matrices are
x 3 (t)
0
not polynomials (i.e., for degree 2d = 0). The corresponding
-0.1
SOF gains are:
-0.2
K1 = 0.002667 −2.818 , K2 = 0.004089 −2.805
-0.3
-0.4 K3 = 0.009141 −5.494 , K4 = 0.02927 −6.183
-0.5
0 2 4 6 8 10 12 14 16 18 20 Fig. 5 shows the trajectories for the closed-loop nonlinear
Time (sec)
system (27), with the SOS approach for Theorem 1 when
Fig. 3. Evolution of the system state x3 (t)
2d = 0 and x0 = [1 3 − 0.5 − 1]T . It is confirmed that
the closed-loop system (5) is asymptotically stable with the
derived controller.
0.5
Theorem 1
Theorem 1 [5]
Example 2: This example presents the design of SOF
controller for a system in [13]. Consider the two-rules T-
0 S system of the form (2), with the parameters from [13]:
⎡ ⎤ ⎡ ⎤
1 1 sin b
b −0.1 1 + x12
⎢ 1 −2 0 0 ⎥ ⎢ ⎥
A1 (x) = ⎢ ⎥ , B1 (x) = ⎢ 0 ⎥ ,
x 4 (t)
-0.5
⎣ 1 x2 −0.3 0 ⎦ ⎣ 0 ⎦
1
0 0 sin b
−1 0
⎡ b ⎤
-1
1 1 1 −0.1
⎢ 1 −2 0 0 ⎥
A2 (x) = ⎢
⎣ 1 x2 −0.3
⎥ , B2 (x) = B1 (x)
-1.5 1 0 ⎦
0 2 4 6 8 10 12 14 16 18 20
0 0 1 −1
Time (sec)
Evolution of the system state x4 (t)
0 1 0 1 + x12
Fig. 4. C1 (x) = C2 (x) = ,
1 0 0 0