1

Sampled-data in Space Control of Scalar Semilinear

Parabolic and Hyperbolic Systems
Igor Furtat, Pavel Gushchin

Abstract—The paper describes a novel method of sampled-data spatial sampling is used to obtain the implementable control
in space (spatial variable) control of scalar semilinear systems law.
of parabolic and hyperbolic type with unknown parameters The observability of distributed systems with sampled-data
arXiv:2009.13874v1 [eess.SY] 29 Sep 2020

and distributed disturbances. A finite set of sampled-data in

the spatial variable measurements is available. The control law space is studied in [16]. The sampled-data in space control
depends on the function which depends on the spatial variable of infinite-dimensional systems with known parameters is
and on a finite set of state measurements. A special choice considered in [17], [18]. Differently from [17], [18] in [19]–
of this function can affect on some properties of the closed- [21] a sampled-data in space control of parabolic systems
loop system. In particular, the paper describes the examples of with unknown distributed parameters is considered. Also in
this function that provides reduced control costs in comparison
with some other control methods. The exponential stability of [19]–[21] the analysis of exponential stability of the closed-
the closed-loop system and robustness with respect to unknown loop system is proposed in terms linear matrix inequalities
parameters and disturbances is proposed in terms of linear (LMIs). However, the results of [17]–[21] are obtained without
matrix inequalities (LMIs). The simulations confirm theoretical disturbances.
results and show the efficiency of the proposed control law
In the present paper, as in [17]–[21], we propose the
compared with some existing ones.
sampled-data in space control law. Differently from [17]–[21],
Index Terms—sampled-data control, semilinear partial differ- the main contribution of our paper is as follows:
ential equation, Lyapunov functional, exponential stability, linear
matrix inequality. (i) we consider scalar semiliniar parabolic and hyperbolic
systems under distributed disturbances and unknown
distributed parameters;
I. I NTRODUCTION (ii) the proposed control law allows one to form various
configurations of the control signal in a spatial variable
The paper considers systems presented by semilinear dif- in order to obtain various properties in the closed-loop
ferential equations of parabolic and hyperbolic type with system. In particular, stabilization with less control costs
distributed disturbances. Such systems describe convection- is considered;
diffusion processes, a rotating column of a compressor with an (iii) the exponential stability of the closed-loop systems
air injection drive, heat distribution in a rod, string oscillation, under disturbances and unknown parameters using the
etc. LMI approach is proposed.
The finite-dimensional control using Fourier transform and
The paper is organized as follows. Problem formulation is
Galerkin method are considered in [1]–[4]. In [5] a control
presented in Section II. Section III describes the control law
law based on moving sensors and actuators along the spatial
design. Sections IV, V consider the analysis of exponential
variable is proposed for linear parabolic systems. Also for such
stability of the closed-loop systems in terms of LMIs. The
systems the adaptive control based on the backstepping method
well-posedness problem is considered in Section VI. Section
is proposed in [6]–[9]. However, the backstepping control law
VII illustrates an efficiency of the proposed method and its
is complicated in calculation and implementation.
advantages compared with the existing methods. Section VIII
Differently from [1]–[9], in the present paper we propose
collects some conclusions.
a method for design a sampled-data in space control law. For
Notation. Throughout the paper Rn denotes the n dimen-
finite-dimensional systems a similar approach has been studied
sional Euclidean space with the norm | · |; Rn×m is the set of
over the past few decades as a discretization (quantization)
all n × m real matrices; P > 0 and P ∈ Rn×n means that P
of measured signal, see, e.g. [10]–[15]. Unlike continuous
is symmetric and positive definite; the symmetric elements
control law such discrete one does not take into account the
of the symmetric matrix will be denoted by ∗. Functions,
behavior of the plant between samples, but in some cases it
continuously differentiable in all arguments, are referred to as
allows solving a number of technical problems, e.g. control
of class C 1 . Subscripts indicate partial derivatives zξ = ∂z ∂ξ
via digital communication channels, control with restriction 2

on information communication channels, etc. In our paper the and zξξ = ∂∂ξz2 . L2 (0, l) is the Hilbert space of square
R l z(ξ), ξ ∈ [0, l] with the corresponding
integrable functions
I. Furtat was with the Institute for Problems in Mechanical Engineering norm kzk2L2 = 0 z 2 (s)ds. H1 (0, l) is the Sobolev space of
of the Russian Academy of Sciences, V.O., Bolshoj pr., 61, St. Petersburg, absolutely continuousR scalar functions z : [0, l] → R with
199178 Russia e-mail: [email protected]. l
P. Gushchin is with the Gubkin University, 65 Leninsky Prospekt, Moscow, the norm kzk2H1 = 0 zs2 (s)ds and zξ ∈ L2 (0, l). H2 (0, l)
119991, Russia. is the Sobolev space of scalar functions z : [0, l] ∈ R with
 2

Rl
absolutely continuous zξ , the norm kzk2H2 = 0
2
zss (s)ds and Here ∆ is known value. Assume that N sensors are placed
zξξ ∈ L2 (0, l). inside these sub-intervals, i.e. only the signals z(x̄j , t), x̄j ∈
(xj , xj−1 ), j = 0, ..., N − 1 are available for measurement.
II. P ROBLEM FORMULATION Our objective is to design the control law (sampled-data in
A. Models space) which ensures the exponential stability of the closed-
loop system for (1) or (4).
1) Consider a semilinear scalar equation of parabolic type
in the form
∂
III. C ONTROL LAW DESIGN
zt (x, t) = ∂x [a1 (x)zx (x, t)]
+ a2 (x)zx (x, t)
Introduce the control law in the form
+ φ(z, x, t)z(x, t) + u(x, t) + f (x, t), (1)
x ∈ [0, l], l > 0, u(x, t) = −KF j (x, t), x ∈ [xj , xj+1 ),
(6)
j = 0, ..., N − 1,
with Dirichlet boundary conditions
where K > 0, the function F j (x, t) satisfies the following
z(0, t) = z(l, t) = 0 (2)
conditions:
or with mixed boundary conditions (i) F j (x, t) is of class C 1 ;
(ii) F j (x̄j , t) = z(x̄j , t), where x̄j ∈ (xj , xj−1 );
zx (0, t) = γz(0, t), z(l, t) = 0, γ ≥ 0. (3)
(iii) Fxj (x, t) is bounded for any x ∈ [0, l] and t ≥ 0.
Here t ≥ 0, z : [0, l] × [0, ∞) → R is the state, u(x, t) is the The condition (i) will be required to solve the boundary-
control. The functions a1 (x), a2 (x), φ(z(x, t), x, t) and f (x, t) value problem in Section VI. The conditions (ii) and (iii)
are unknown and of class C 1 . Also these functions satisfy the will be required to proof the closed-loop system stability in
following conditions: a1 (x) ≥ a1 > 0, a2 ≤ a2 (x) ≤ a2 , Sections IV and V.
φ ≤ φ(z, x, t) ≤ φ, |f (x, t)| ≤ f¯ with known bounds a1 , a2 , Consider a few examples of the function F j (x, t).
a2 , φ, φ, and f¯. The value of γ in (3) may be unknown. Example 1. Let F j (x, t) = ϕj (x, t)z(x̄j , t), where j =
Remark 1: System (1) describes convection-diffusion pro- 0, ..., N −1, ϕj (x, t) is of class C 1 , ϕjx (x, t)z(x̄j , t) is bounded
cesses under u(x, t) = 0. In [2] system (1) describes a rotating for any x ∈ [xj , xj+1 ), t ≥ 0 and ϕj (x̄j , t) = 1.
stand of a compressor with an air injection drive u(x, t), where Example 2. In [19]–[21] u(x, t) 6= 0 over the interval
z(x, t) is the axial flow through the compressor. Also the [xj , xj+1 ) when z(x̄j , t) 6= 0. Now we consider a case when
boundary-value problem (1), (2) with u(x, t) = 0 describes u(x, t) 6= 0 only on a part of the interval [xj , xj+1 ) for
the heat distribution in a uniform one-dimensional rod with a z(x̄j , t) 6= 0. Let in example 1 the function ϕj (x, t) is given
in the form
fixed temperature at the ends, where a1 and φ are coefficients  
α(x−x̄ )

of thermal conductivity and heat transfer with the environment 
 0.5 + 0.5 cos 1+z2 (x̄ j,t) if

  i 
π(1+z 2 (x̄j ,t)) π(1+z 2 (x̄j ,t))

respectively, a2 = 0, z(x, t) is the temperature at time t in the 
x ∈ x̄j −
 ; x̄ j + ;

 α α
point x. 

ϕj (x, t) = 0 if  (7)
2) Additionally, consider a semilinear scalar differential  π(1+z 2 (x̄j ,t))

x ∈ xj ; x̄j −


equation of hyperbolic type in the form 

 α
 

π(1+z 2 (x̄j ,t))

∂ ∪ x̄j +

; xj+1 .
ztt (x, t) = ∂x [a1 (x)zx (x, t)]+ a2 (x)zx (x, t) 
α
+ φ(z, x, t)z(x, t)
(4) Here α > 0 is a sufficiently largen number that can be selected
− b(z, x, t)zt (x, t) + u(x, t) + f (x, t), x̄ −x xj+1 −x̄j
o
x ∈ [0, l], from the condition α > max π(1+zj 2 (x̄jj ,t)) , π(1+z 2 (x̄ ,t))
j
.
j j 1
It is obvious, that ϕ (x̄j , t) = 1, ϕ (x, t) is of class C ,
with Dirichlet boundary conditions (2) or with mixed boundary
ϕjx (x, t)z(x̄j , t) is bounded for any x ∈ [xj , xj+1 ), t ≥ 0
conditions (3). In (4) the function b(z, x, t) is of class C 1 which
and z(x̄j , t) ∈ R.
satisfies the condition 0 < b ≤ b(z, x, t) ≤ b with known
bounds b and b. Other functions in (4) take the same values
as in (1).
Remark 2: The boundary-value problem (4), (2) with
u(x, t) = 0 describes vibrations of a uniform string with fixed
ends and energy dissipation, where a1 , b and φ are coefficients
of elasticity, dissipation and stiffness respectively, a2 = 0,
z(x, t) and zt (x, t) are deflection and speed of the string at
time t in the point x.

B. Objective
Divide the segment [0, l] into N sampling intervals (not
necessarily equal length) and denote
Fig. 1. The plots of ϕj (x, t) from example 2 (for z(x̄j , t) = const > 0) on
0 = x0 < x1 < ... < xN = l, ∆ ≥ xj+1 − xj , the interval x ∈ [xj , xj+1 ) at various values of α > 0 ((α = α1 ) < (α =
(5) α2 )).
j = 0, ..., N − 1.
 3

Example 3. In (7) transition between the values of ϕj where

depends on z(x̄j , t). Now let us consider new function that
Ψ11 a2 1 0
 
eliminates this dependency: 2
−1
 ∗ Ψ22 0 − 4∆
π 2 K R̄

α α Ψ=  ∗
,
∗ −β1
 −
(1+z 2 (x̄j ,t))(β 2 −(x−x̄j )2 )
+
βj (1+z 2 (x̄j ,t)) 0 

 e j
4∆2 −1 (11)
∗ ∗ ∗ −β2 + π2 K R̄


ϕj (x, t) = if x ∈ (x̄j − βj ; x̄j + βj ) ; (8)

 0 if Ψ11 = −2K + 2φ + 2δ + K R̄,
2
−1
Ψ22 = −2a1 + 4∆

π 2 K R̄ .

x ∈ [xj ; x̄j − βj ] ∪ [x̄j + βj ; xj+1 ) .
Then the following inequality holds
Here α > 0, βj ≤ min{xj+1 − x̄j , x̄j − xj }. Differently
γ
from (7), transition between the function values in (8) does kz(·, t)k2L2 ≤ e−2δt kz(·, 0)k2L2 + 2δ , (12)
not depend on z(x̄j , t), but the transition depends only on
where
βj . We have ϕj (z(x̄j , t), x̄j , t) = 1, ϕj (z(x̄j , t), x, t) is of
N −1 Z xj+1
class C 1 and ϕjx (z(x̄j , t), x, t)z(x̄j , t) is bounded for any x ∈ l
Z X
[xj , xj+1 ), t ≥ 0 and z(x̄j , t) ∈ R. The graphs of (8) are γ = β1 sup f 2 (x, t)dx + β2 sup (Fxj (x, t))2 dx.
t≥0 0 t≥0 j=0 xj
given in Fig. 2.
Before proving this theorem, consider two auxiliary lemmas.
Lemma 1: (Extended Wirtinger inequality). Let z ∈ H1 (0, l)
be a scalar function, 0 = χ0 < χ1 < ...χn−1 < χn = l and
∆ ≥ χi − χi+1 , i = 0, ..., n − 1. If z(χi ) = 0, i = 1, ..., n − 1,
then Rl 2 2 Rl

0
z (ξ)dξ ≤ 4∆ π2 z 2 (ξ)dξ.
0 ξ
(13)
Proof 1: Rewrite the left hand side of (13) in the form
Rl 2 Rχ
0
z (ξ)dξ = 0 1 z 2 (ξ)dξ
Pn−1 R χ Rl
+ i=1 χii+1 z 2 (ξ)dξ + χn z 2 (ξ)dξ.
Using Wirtinger’s inequality, we obtain
Fig. 2. The plots of ϕj (z(x̄j , t), x, t) from Example 3 (for z(x̄j , t) = R χ1 2 Rl Pn−1 R χ
const > 0) on the interval x ∈ [xj , xj+1 ) at various values of α > 0 0
z (ξ)dξ + χn z 2 (ξ)dξ + i=1 χii+1 z 2 (ξ)dξ ≤
2
 R χ1 2 Rl 2 
((α = α1 ) < (α = α2 )).
≤ 4∆ π 2 0
zξ (ξ)dξ + χn
zξ (ξ)dξ
2 Pn−1 R χi−1 2 Rl

Remark 3: In [19]–[21] the control law u(x, t) = +∆ π 2 i=1 χi zξ

2
(ξ)dξ ≤ 4∆ π2 z 2 (ξ)dξ.
0 ξ
−Kz(x̄j , t), x̄j = 0.5(xj + xj+1 ), x ∈ [xj , xj+1 ), j = Lemma 2: Let the function V : [t0 , ∞) → [0, ∞) be
0, ..., N − 1 is proposed. This control law is a special case differentiable on [t0 , ∞), t0 ≥ 0. Consider the following
of (6) when F j (x, t) = z(x̄j , t) (or ϕj (x, t) = 1 in example differential inequality
1). If the function F j (x, t) is chosen as in examples 2 and 3,
then the area under the curve ϕj (x, t) = 1 can be significantly V̇ (t) ≤ −δV (t) + f (t), (14)
larger than the area under the proposed curves (see Figs. 1,
where δ > 0 and sup |f (t)| = β. Then the following
2). Thus, the proposed control law can stabilize system (1) at t≥t0
a lower cost of the control signal. inequality holds
V (t) ≤ e−δ(t−t0 ) V (t0 ) + βδ , t ≥ t0 . (15)
IV. M AIN RESULT FOR A PARABOLIC TYPE SYSTEM (1) Proof 2: Denote by

Substituting (6) into (1), represent the closed-loop system y(t) = e−δ(t−t0 ) V (t0 ) + βδ , t ≥ t0 .
as follows
It is easy to verify that the function y(t) is a solution of the
zt (x, t) = ∂ differential equation
∂x [a1 (x)zx (x, t)]
+ a2 (x)zx (x, t)
+ f (x, t) − (K − φ(z, x, t))z(x, t) ẏ(t) = −δy(t) + β, t ≥ t0 . (16)
(9)
+ K[z(x, t) − F j (x, t)],
x ∈ [xj , xj+1 ), j = 0, ..., N − 1. Using the comparison principle, we will show that V (t) ≤ y(t)
for any t ≥ t0 . Let ε1 > ε2 > ... > εn > ... is a sequence of
Theorem 1: Consider the closed-loop system (9) under the positive numbers such that lim εn = 0+ . Then the function
n→∞
boundary conditions (2) or (3). Given R̄ > 0, ∆ > 0 and
β εn
δ > 0 there exists K > 0 such that the following two linear ẏn (t) = −δy(t) + δ + δ
(17)
matrix inequalities hold is a solution of the differential equation
Ψ(a2 = a2 ) ≤ 0 and Ψ(a2 = a2 ) ≤ 0, (10) ẏ(t) = −δy(t) + β + εn . (18)
 4

Suppose that there exists t∗ > t0 such that holds. Using Lemma 2, the solution of the differential inequal-
ity (26) can be defined as
t∗ = inf{t > t0 : V (t) ≥ yn (t)}. (19)
γ
V (t) ≤ V (0)e−2δt + 2δ . (27)
∗ ∗ ∗
Then V (t ) ≥ yn (t ) and V (t) < yn (t) at t0 ≤ t ≤ t .
The condition V̇ (t∗ ) < ẏn (t∗ ) holds from (14) and (18). On Expression (12) follows from (27). Theorem 1 is proved.
the other hand, the condition V̇ (t∗ ) ≥ ẏn (t∗ ) follows from
V (t) < yn (t) at t < t∗ and V (t∗ ) = yn (t∗ ). We have a V. M AIN RESULT FOR A SYSTEM OF HYPERBOLIC TYPE (4)
contradiction. Therefore, V (t) < yn (t) for any t ≥ t0 and Substituting (6) into (4), rewrite the closed-loop system in
n = 1, 2, ... Consequently, V (t) ≤ lim yn (t) = y(t) for all the form
n→∞
t ≥ t0 . Lemma 2 is proved. ∂
ztt (x, t) = ∂x [a1 (x)zx (x, t)] + a2 (x)zx (x, t)
Proof 3: To analyze the stability of the closed-loop system
− b(z, x, t)zt (x, t) + f (x, t)
(9), consider the following Lyapunov functional
− (K − φ(z, x, t))z(x, t) (28)
+ K[z(x, t) − F j (x, t)],
Rl
V (t) = 0 z 2 (x, t)dx. (20)
x ∈ [xj , xj+1 ), j = 0, ..., N − 1.
Differentiating V (t) in time along the trajectories of (9), we
have Theorem 2: Consider the closed-loop system (28) under the
Rl boundary conditions (2) or (3). Given p ∈ (−0.5; 0.5), R̄ > 0,
∂
V̇ (t) + 2δV (t) = 2 0 [z(x, t) ∂x [a1 (x)zx (x, t)] ∆ > 0, and δ > 0 there exists K > 0 such that the following
Rl
+ a2 (x)zx (x, t)z(x, t) + 2δ 0 z 2 (x, t)dx LMIs hold
(21)
− (K − φ(z, x, t))z 2 (x, t) + z(x, t)f (x, t)]dx

PN −1 R xj+1 Ψ̄ φ = {φ, φ}, a2 = {a2 , a2 }, b = {b, b} ≤ 0, (29)
+ 2K j=0 xj z(x, t)[z(x, t) − F j (x, t)].
where
Taking into account the boundary conditions (2) or (3) and 
Ψ̄11 0.5pa2 Ψ̄13 0.5p 0

integrating by parts the first term in (21), we get 2
 ∗
 Ψ̄22 a2 0 − 4∆
π2
K R̄−1  
 ∗
Ψ̄ =  ∗
Rl ∂ Ψ̄33 1 0 ,
2 0 z(x, t) ∂x [a1 (x)zx (x, t)]dx  ∗ ∗ ∗ −β1 0


l Rl 2
= 2a1 (x)z(x, t)zx (x, t) − 2 0 a1 (x)zx2 (x, t)dx (22) ∗ ∗ ∗ ∗ −β2 + 4∆π2
K R̄ −1
(30)
Rl 2 0
≤ −2a1 0 zx (x, t)dx. Ψ̄11 = −0.5p(K − φ) + 0.25K R̄p2 + 2δ,
Ψ̄13 = 1 − 0.5pb − K + φ + K R̄p,
Using Young’s inequality for the penultimate term in (21), 2

we obtain Ψ̄22 = −pa1 + 4∆π2

K R̄−1 ,
PN −1 R x Ψ̄33 = 0.5p − 2b + 0.5K R̄.
2K j=0 xjj+1 z(x, t)[z(x, t) − F j (x, t)]dx
Rl Then inequality (12) holds by taking into account parameters
≤ K R̄ 0 z 2 (x, t)dx (23) from (29).
PN −1 R x
+K R̄−1 j=0 xjj+1 [z(x, t) − F j (x, t)]2 dx. Proof 4: Introduce the following Lyapunov functional
Rl
Applying Lemma 1 to (23) and taking into account V (t) = 0 [azx2 (x, t) + z 2 (x, t)
(31)
z(x̄j , t) = F j (x̄j , t), one gets + pz(x, t)zt (x, t) + zt2 (x, t)]dx.

PN −1 R xj+1 The inequality z 2 + pzzt + zt2 ≥ 0 holds for p ∈ (−0.5; 0.5).

−1 j 2 Therefore, V (t) ≥ 0. Differentiating V (t) in time along the
K R̄ j=0 xj [z(x, t) − F (x, t)] dx ≤
4∆2
≤ π2 K R̄ −1
PN −1 R xj+1 2
[zx (x, t) (24) trajectories of (28), we have
j=0 xj
Rl
−2zx (x, t)Fxj (x, t) + (Fxj (x, t))2 ]dx. V̇ (t) + 2δV (t) = 2 0 [a1 (x)zx (x, t)zxt (x, t)
+ z(x, t)zt (x, t) + 0.5pzt2 (x, t)+
Denote by ηj = col{z(x, t), zx (x, t), f (x, t), Fxj (x, t)}. ∂
[0.5pz(x, t) + zt (x, t)][ ∂x [a1 (x)zx (x, t)]+
Applying (22)–(24) to (21), we have
a2 (x)zx (z, x, t) − b(z, x, t)zt (x, t) (32)
Rl + f (x, t) − (K − φ)z(x, t)]dx
V̇ (t) + 2δV (t) − β1 0 f 2 (x, t)dx PN −1 R x
PN −1 R x + 2K j=0 xjj+1 [0.5pz(x, t) + zt (x, t)]
− β2 j=0 xjj+1 Fxj (x, t)dx (25) Rl
PN −1 R x × [z(x, t) − F j (x, t)] + 2δ 0 z 2 (x, t)dx.
≤ j=0 xjj+1 η T Ψηdx,
Taking into account the boundary conditions (2) or (3),
where Ψ is given by (11). The expression (11) is affine with consider the integration by parts
respect to a2 . According to [21], if LMIs (10) holds in the Rl ∂
vertices a2 = a2 and a2 = a2 , then LMI Ψ ≤ 0 holds for any 2 0 zt (x, t) ∂x [a1 (x)zx (x, t)]dx
l
a2 ∈ [a2 , a2 ]. Therefore, the inequality = 2a1 (x)zx (x, t)zt (x, t)
Rl Rl 0 (33)
V̇ (t) + 2δV (t) − β1 0 f 2 (x, t)dx− −2 0 a1 (x)zx (x, t)zxt (x, t)dx
PN −1 R xj+1 j (26) Rl
β2 j=0 xj Fx (x, t)dx ≤ 0 = −2a1 0 zx (x, t)zxt (x, t)dx.
 5

Using Young’s inequality for the penultimate term in (32), B. The closed-loop system (28)
we obtain Now we consider the closed-loop system (28) with the
PN −1 R x boundary conditions (2). Rewrite the boundary-value problem
2K j=0 xjj+1 [0.5pz(x, t) + zt (x, t)]
× [z(x, t) − F j (x, t)]dx ≤ (28), (2) as an abstract inhomogeneous Cauchy problem in the
Rl (34) Hilbert space H = L2 (0, l) in the form
≤ K R̄ 0 [0.5pz(x, t) + zt (x, t)]2 dx
PN −1 x
+ K R̄−1 j=0 xjj+1 [z(x, t) − F j (x, t)]2 dx.
R ˙ = Āξ(t) + F̄ (t, ξ(t)), ξ0 = ξ(0) ∈ Ā,
ξ(t) (38)

Considering z(x̄j , t) = F j (x̄j , t) and applying where

 ξ = col{z, zt }, the operator Ā =
0 1
Lemma 1 to (34), we have (24). Denote by ∂ ∂ ∂ has the dense domain
η̄j = col{z(x, t), zx (x, t), zt (x, t), f (x, t), Fxj (x, t)}. ∂x [a1 (x) ∂x ] + a2 (x) ∂x 0
D(Ā) = {ξ ∈ H2 (0, l) : ξ(0) = ξ(l) = 0}, F̄ (t, ξ(t)) =
Applying (22), (33), (34) and (24) to (32), rewrite result as
[0 1]T [f (t) + u(t) + φ(t, ξ1 (t))ξ1 (t) − b(t, ξ1 (t))ξ2 (t)], the
follows
Rl function u(t) is given by (6). According to Theorem 1, [23]
V̇ (t) + 2δV (t) − β1 0 f 2 (x, t)dx and [24], the infinitesimal operator Ā generates a strictly
PN −1 R x
− β2 j=0 xjj+1 Fxj (x, t)dx (35) continuous exponentially stable semigroup (C0 -semigroup)
PN −1 R xj+1 T T (t). Therefore, further considerations for (38) are similar to
≤ j=0 xj η̄ Ψ̄η̄dx,
those for (36) in Subsection VI-A.
where Ψ̄ is given by (30). The expression (30) is affine with
respect to the parameters φ, a2 and b. According with [21], if VII. E XAMPLES
LMIs (10) hold in the vertices φ = {φ, φ}, a2 = {a2 , a2 }, A. The simulations of systems (1) and (4)
b = {b, b}, then LMI Ψ̄ ≤ 0 holds for any φ = [φ, φ], Let l = 1. To simulate systems (1) and (4) we divide
a2 = [a2 , a2 ], b = [b, b]. Therefore, inequality (26) holds. the segment [0, 1] into 160 sub-intervals of the same length
Considering Lemma 2, the solution of differential inequality with a sampling step in the spatial variable D = 1/160.
(35) can be defined as (27). Then, inequality (12) follows from The first and the second derivatives in the spatial vari-
(27) but taking into account parameters from (29). Theorem 2 able of z(x, t) are calculated in the points 0 = x0 <
is proved. x1 < ...xl < ... < x160 according with the following
expressions zx (xk , t) = z(xk+1 ,t)−z(x
D
k ,t)
and zxx (xk , t) =
z(xk+1 ,t)−2z(xk ,t)+z(xk−1 ,t)
VI. W ELL - POSEDNESS OF THE CLOSED - LOOP SYSTEM D2 .
In this section, we show that there exist solutions of the For design the control law (6) we divide the segment [0, 1]
closed-loop systems (9) and (28) satisfying Dirichlet boundary into N = 2 and N = 10 equal sub-intervals (see (5)).
conditions (2). The well-posedness under the mixed conditions Consider systems (1) and (4) under the Dirichlet bound-
(3) can be proved similarly. ary conditions (2) and a1 (x) ∈ [0.5, 2], a2 (x) ∈ [−5, 5],
φ(z, x, t) ∈ [−5, 5], b(z, x, t) = [−5, −1], |f (x, t)| ≤ 20
for any x and t. The initial conditions are given as follows
A. The closed-loop system (9) z(x, 0) = sin(πx) zx (x, 0) = 0.
Consider the closed-loop system (9) with the boundary The matrix inequalities (10) and (29) are feasiable for K ≥
conditions (2). The boundary-value problem (9), (2) can be 100.
represented as an abstract inhomogeneous Cauchy problem in
the Hilbert space H = L2 (0, l) as follows B. The results of simulations of system (1)
ż(t) = Az(t) + F (t, z(t)), z0 = z(0) ∈ D(A). (36) In (1) choose a1 (x) = 1 + sin(x), a2 (x) = −2 − sin(1.3x),
φ(z, x, t) = 5+cos(3z), and f (x, t) = 0.2[sin(30t)+sin(2t)].
∂ ∂ ∂
Here the operator A = ∂x [a1 (x) ∂x ] + a2 (x) ∂x has the Consider two partitions for N = 2 and N = 10 (see (5)). Let
dense domain D(A) = {z ∈ H2 (0, l) : z(0) = z(l) = 0}, x̄j = 0.5(xj + xj+1 ). Choose F j (x, t) in the control law (6)
F (t, z(t)) = f (t) + u(t) + φ(t, z(t))z(t), the function u(t) is from example 3, where α = 103 , as well as βj = 1/8 for
given by (6). According to Theorem 1, [23] and [24], infinites- N = 2 and βj = 1/80 for N = 10. In Figs. 3–5 the solutions
imal operator A generates a strictly continuous exponentially of (1) and the spatio-temporal graphs of u(x, t) are illustrated
stable semigroup (C0 -semigroup) T (t). Then, the boundary- for:
value problem (36) can be represented as a boundary-value 1) the control law u(x, t) = −Kz(x̄j , t) from [19]–[21]
problem on the semi-infinite interval [0, ∞) and its solutions for K = 100;
can be found as solutions of the following integral equation 2) the proposed control law (6) for K = −100;
Rt 3) the proposed control law (6) for K = −500.
z(t) = T (t)z(0) + 0 T (t − s)F (s, z(s))ds. (37)
Figs. 3, 4 show, that the solutions of z(x, t) almost the same
Since the function F (t, z(t)) is continuously differentiable for the proposed control law and the one from [19]. However,
w.r.t. t, then, according to Theorem 3.1.3 [24], there is a unique the proposed control law provides exponential stability under
solution of (36) and this solution satisfies the integral equation distributed disturbances. If the coefficient K is increased by
(37). 5 times, then the magnitude of the control signal is also
 6

increased approximately 5 times. In this case the rate of

exponential convergence and quality of disturbance rejection in
the steady state are higher than the ones from [19] at K = 100.
Now we analyze the control costs. Figs. 6, 7
illustrate
PN −1 Rthe integral difference in the form
t
I = j=0 0 (|uF &B (x̄j , s)| − |uproposed (x̄j , s)|) ds, where
uF &B (x̄j , t) is the control law from [19], uproposed (x̄j , t) is
the proposed one. The advantages of the proposed algorithm
is clearly seen, i.e. the control costs of the proposed control
law are less than ones from [19]. Moreover, the proposed
control law can be approximated by finite actions, while
the control law from [19]–[21] requires implementation
throughout the whole spatial variable. The simulations for
the function F j (x, t) from example 2 with α = 100 are
comparable with the results obtained for the function F j (x, t)
from example 3.

Fig. 4. The spatio-temporal graphs of z(x, t) and u(x, t) for the proposed
control law for N = 2 (a,b), N = 10 (c,d) and K = 100.

Fig. 5. The spatio-temporal graphs of z(x, t) u(x, t) for the proposed control
law for N = 10 and K = 500.

measurements of the output signal. Also the control law

depends on a function depending on the spatial coordinate and
Fig. 3. The spatio-temporal graphs of z(x, t) and u(x, t) from [19]–[21] for
the current measurement. This function allows one to achieve
N = 2 (a,b) and N = 10 (c,d) for K = 100. different properties, for example, to provide reduced control
costs. The exponential stability of the closed-loop systems and
robustness with respect to unknown parameters and external
C. The results of simulations of system (4) disturbances are considered. The simulations have confirmed
the theoretical results and have showed the efficiency of the
In (4) choose a1 (x) = 1 + sin(x), a2 (x) = −2 − sin(1.3x),
proposed algorithm compared with ones from [19]–[21].
b(z, x, t) = −2 − sin(1.1z), φ(z, x, t) = 5 + cos(3z) and
f (x, t) = 0.2[sin(30t) + sin(2t)]. In Figs. 8, 9 the solutions
of (4) and spatio-temporal graphs of u(x, t) are given for ACKNOWLEDGMENT
the proposed control law (6) with the function F j (x, t) from The results were proposed with the support of a grant from
example 3 for α = 103 , as well as βj = 1/8 for N = 2 the Russian Science Foundation (No. 18-79-10104) in IPME
and βj = 1/80 for N = 10 and K = −100, K = −500. RAS.
The proposed control law provides exponential stability under
disturbances. The simulations for the function F j (x, t) from R EFERENCES
example 2 with α = 100 are comparable with the results
[1] U.O. Candogan, H. Ozbay, and H.M. Ozaktas, Controller Implementation
obtained for the function F j (x, t) from example 3. for a Class of Spatially-varying Distributed Parameter Systems, IFAC
Proceedings Volumes (Proc. of the 17th IFAC world congress). 2008. V.
VIII. C ONCLUSIONS 41. No. 2. P. 7755-7760.
[2] G. Hagen and I. Mezic, Spillover stabilization in finite-dimensional
A sampled-data in space control law is proposed for scalar control and observer design for dissipative evolution equations, SIAM
semilinear differential equations of parabolic and hyperbolic Journal on Control and Optimization. 2003. V. 42. No. 2. P. 746-768.
[3] E. Smagina and M. Sheintuch, Using Lyapunovs direct method for wave
types with interval-indefinite parameters and external bounded suppression in reactive systems, Systems & Control Letters. 2006. V. 55.
disturbances. The control law is only used a finite set of No. 7. P. 566-572.
 7

Fig. 6. The plots of integral difference for N = 2 (a) and N = 10 (b)

between the control law [19], [20] and the proposed control law for K = 100.

Fig. 7. The plots of integral difference for N = 10 between the control law Fig. 8. The spatio-temporal graphs of z(x, t) and u(x, t) for the proposed
[19]–[21] and the proposed control law for K = 100. control law for N = 2 (a, b), N = 10 (c, d) and K = 100.

[4] J. Zhong, C. Zeng, Y. Yuan, Y. Zhang, and Y. Zhang, Numerical solution

of the unsteady diffusion-convection-reaction equation based on improved
spectral Galerkin method AIP Advances. 2018. No. 8, pp. 045314.
[5] M. A. Demetriou, Guidance of mobile actuator-plus-sensor networks for
improved control and estimation of distributed parameter systems, IEEE
Transactions on Automatic Control. 2010. V. 55. P. 1570-1584.
[6] A. Smyshlyaev and M. Krstic, On control design for PDEs with spacede-
pendent diffusivity or time-dependent reactivity, Automatica. 2005. V. 41.
P. 1601-1608.
[7] M. Krstic and A. Smyshlyaev, Adaptive boundary control for unstable
parabolic PDEs-part I: Lyapunov design, IEEE Transactions on Automatic
Control. 2008. V. 53. P. 1575-1591.
[8] M. Izadi, J. Abdollahi, and . S.Dubljevic, PDE backstepping control of
one-dimensional heat equation with time-varying domain, Automatica.
2015. V. 54. P. 41-48.
[9] S. Chen, R. Vazquez, and M. Krstic, Folding Backstepping Approach to
Parabolic PDE Bilateral Boundary Control, IFAC-PapersOnLine. 2019.
V. 52. No. 2. P. 76-81.
[10] D. F. Delchamps, Extracting State Information from a Quantized Output
Record, System Control Letters. 1989. V. 13. P. 365-372.
[11] R. W. Brockett and D. Liberzon, Quantized feedback stabilization of
linear systems, IEEE Transactions on Automatic Control. 2000. V. 45. P.
1279-1289.
[12] J. Baillieul, Feedback Coding for Information-Based Control: Operating
Near the Data Rate Limit, Proc. of the 41st IEEE Conf. Decision Control,
Las Vegas, Nevada, USA, 2002. P. 3229-3236.
[13] B.-C. Zheng and G.-H. Yang, Quantized output feedback stabilization Fig. 9. The spatio-temporal graphs of z(x, t) and u(x, t) for the proposed
of uncertain systems with input nonlinearities via sliding mode control, control law for N = 2 (a, b), N = 10 (c, d) and K = 500.
International Journal of Robust and Nonlinear Control. 2012. V. 24. No.
2. P. 228-246.
[14] I. B. Furtat, A. L. Fradkov, and D. Liberzon, Compensation of distur- [19] E. Fridman and A. Blighovsky, Robust sampled-data control of a class
bances for MIMO systems with quantized output, Automatica. 2015. Vol. of semilinear parabolic systems, Automatica. 2012. V. 48. P. 826-836.
60. P. 239-244. [20] L. Kun, E. Fridman, and Y. Xia, Networked Control Under Communi-
[15] C. Li and J. Lian, Stabilization of switched nonlinear systems with cation Constraints, Springer, 2020.
dynamic output quantization and disturbances, International Journal of [21] E. Fridman, Introduction to Time-Delay Systems. Analysis and Control,
Robust and Nonlinear Control 2020. V. 30. P. 1679-1695. Birkhauser, 2014.
[16] A. Y. Khapalov, Continuous observability for parabolic system under [22] G. H. Hardy, J. E. Littlewood, and G. Polya, Inequalities, Cambridge:
observations of discrete type, IEEE Transactions on Automatic Control. Cambridge: University Press, 1988.
1993. V. 38. No. 9. P. 1388-1391. [23] D. Henry, Geometric theory of semilinear parabolic equations, New
[17] M. B. Cheng, V. Radisavljevic, C. C. Chang, C. F. Lin, and W. C. Su, York: Springer-Verlag, 1993.
A sampled data singularly perturbed boundary control for a diffusion [24] R. Curtain and H. Zwart, An introduction to infinite-dimensional linear
conduction system with noncollocated observation, IEEE Transactions of systems theory, New York: Springer-Verlag, 1995.
Automatic Control. 2009. V. 54. No. 6. P. 1305-1310.
[18] H. Logemann, R. Rebarber, and S. Townley, Generalized sampled-data
stabilization of well-posed linear infinite-dimensional systems, SIAM
Journal on Control and Optimization. 2005. V. 44. No. 4. P. 1345-1369.