You are on page 1of 9

Automatica 104 (2019) 8–16

Contents lists available at ScienceDirect

Automatica
journal homepage: www.elsevier.com/locate/automatica

Brief paper

Output feedback anti-disturbance control of input-delayed systems


with time-varying uncertainties✩

Shoulin Hao a , Tao Liu a , , Bin Zhou b
a
Institute of Advanced Control Technology, Dalian University of Technology, Dalian, 116024, China
b
Center for Control Theory and Guidance Technology, Harbin Institute of Technology, Harbin, 150001, China

article info a b s t r a c t

Article history: For input-delayed systems with time-varying uncertainties in industrial applications, a robust predictor-
Received 1 December 2017 based anti-disturbance control design is proposed in this paper based on only output measurement,
Received in revised form 18 November 2018 which can be applied to improve system performance in the presence of constant, asymptotically
Accepted 27 January 2019
stable, step, ramp, or harmonic type disturbances, typically encountered in engineering practice. A
Available online xxxx
novel predictor, named state & disturbance observer–predictor (SDOP), is firstly introduced to estimate
Keywords: the ‘delay-free’ system state and disturbance dynamics simultaneously, with no need to store the
Input delay control history as required in the existing predictor-based control methods. To allow for long input
Time-varying uncertainties delay, another design of sequential SDOPs (SSDOPs) is proposed such that each SSDOP only needs to
State & disturbance observer–predictor estimate the future system state and disturbance in terms of a specified step size for implementation.
(SDOP)
Moreover, a recursive sub-optimal H∞ design of SSDOPs is given such that the computation burden is
Sequential SDOPs
independent of the number of SSDOPs and thus could be significantly reduced. Consequently, two
anti-disturbance control schemes are developed based on an SDOP or SSDOPs to improve system
performance. For the nominal system, the input delay is allowed to be arbitrarily long by increasing
the number of SSDOPs. For the presence of time-varying plant uncertainties, asymptotic disturbance
rejection performance can be achieved. A sufficient condition for robust stability of the closed-loop
system is established in terms of matrix inequality, which can be effectively solved to determine
the controller parameters. An illustrative example from the literature is used to demonstrate the
effectiveness and merit of the proposed method. ,
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction scheme (Smith, 1959) developed in frequency domain. Due to


the fact that the standard SP control structure could not be used
Time delay widely exists in industrial process operations due to control open-loop unstable processes because of the internal
to mass transportation, energy exchange, and signal process- instability (Normey-Rico & Camacho, 2007), a few modified dead-
ing etc. Liu and Gao (2012). The presence of time delay, if not time compensator based control schemes had been developed for
considered in the controller design, may cause performance de- open-loop unstable systems with time delay, such as the filtered
terioration even instability of the control system, in particular for
Smith predictor (Normey-Rico & Camacho, 2007) and the general-
the presence of input delay. Advanced control design for indus-
ized predictor (Albertos & Garcia, 2009). However, robust stability
trial systems with input delay has therefore been continuously
was difficult to be explicitly analyzed in these frequency domain
explored in the past decades (Du, Lam, & Shu, 2010; Fridman,
2014; Kojima, Uchida, Shimemura, & Ishijima, 1994; Krstic, 2010; control methods with respect to time-varying plant uncertain-
Richard, 2003). The pioneer work of control design for time- ties. Concerning the time domain approaches, predictor based
delay systems could be dated back to the Smith predictor (SP) feedback control methods were developed to address the input
delay (Artstein, 1982; Krstic, 2010), focusing on infinite dimen-
✩ The work was supported in part by the NSF China grants 61633006, sional control design to result in a delay-free closed-loop system
61473054, and 61773140, and the Fundamental Research Funds for the Central such that its stability could be relatively independent of the time
Universities of China. The material in this paper was not presented at any delay. By assuming the open-loop system with input delay is not
conference. This paper was recommended for publication in revised form by exponentially unstable, a truncated predictor feedback control
Associate Editor Akira Kojima under the direction of Editor Ian R. Petersen.
∗ Corresponding author. method was proposed in the recent paper (Zhou, Lin, & Duan,
E-mail addresses: haoshoulin@gmail.com (S. Hao), liurouter@ieee.org 2012) by ignoring the distributed terms involved with conven-
(T. Liu), binzhou@hit.edu.cn (B. Zhou). tional predictor-based feedback control schemes. To cope with

https://doi.org/10.1016/j.automatica.2019.02.047
0005-1098/© 2019 Elsevier Ltd. All rights reserved.
S. Hao, T. Liu and B. Zhou / Automatica 104 (2019) 8–16 9

time-varying input delay, prediction-based stabilization meth- (ii) To address the stability issue in the presence of time-varying
ods were recently proposed in Gonzalez (2013), Gonzalez, Sala, plant uncertainties, which was left open in the existing
and Albertos (2012). To compensate the input delay for linear references for anti-disturbance predictor feedback control
systems with both input and state delays, a nested predictor design (e.g. Hao et al., 2017; Lechappe et al., 2015, 2016;
feedback was established in Zhou (2014). By comparison, a ro- Sanz et al., 2016, 2017), a sufficient condition of robust sta-
bust controller design (Sanz, Garcia, Albertos, & Zhong, 2017) bility is established for the proposed predictor-based anti-
was developed in the continuous-time domain for input-delayed disturbance control scheme;
systems with nonlinear uncertainties, by constructing an un- (iii) To cope with long input delay, another design of sequential
certainty estimator together with a state predictor. Recently, a SDOPs (SSDOPs) is proposed such that each SSDOP only
sequential sub-predictors approach was proposed for linear sys- needs to estimate the future system state and disturbance in
tems with a long input delay (Najafi, Hosseinniab, Sheiholeslama, terms of a specified step size for implementation. Moreover,
& Karimadini, 2013), which was further extended to nonlinear a recursive sub-optimal H∞ design of SSDOPs is given to
delay systems (Mazenc & Malisoff, 2017b), time-varying delay reduce the computation burden owing to its independence
systems (Mazenc & Malisoff, 2017a), systems with both input and of the number of SSDOPs, therefore addressing the high-
state delays (Liu & Zhou, 2017; Zhou, Liu, & Mazenc, 2017) and dimension issue involved with sequential sub-predictors
systems with input, state and output delays (Cacase & Germani, studied in the existing literature (e.g. Najafi et al., 2013;
2017). Najafi, Sheiholeslama, Wang, & Hosseinniab, 2014).
Note that in industrial applications, various disturbances seep-
For clarity, this paper is organized as follows. In Section 2,
ing into system operation could have a significant influence on
the problem statement and preliminaries are presented. The pro-
the system performance. Disturbance rejection has been a major
posed anti-disturbance control scheme with SDOP is given in
concern for control system design and performance optimization Section 3, followed by another control scheme with SSDOPs to
in process industries (Liu & Gao, 2012). Robust control meth- accommodate for long input delay in Section 4. Accordingly, ro-
ods such as H∞ control approaches had been recognized to be bust stability of the closed-loop system is addressed in Section 5.
effective for disturbance attenuation of systems with external An illustrative example is shown in Section 6 to demonstrate
disturbances bounded by the H2 norm. However, different types the effectiveness of the proposed method. Some conclusions are
of disturbances (e.g. harmonics or unknown constant load dis- drawn in Section 7.
turbances (Guo & Chen, 2005)) may be involved with industrial
applications. To eliminate or reduce the influence from such dis- Notation: Throughout this paper, the following notations are
turbance having continuous magnitude or energy, a natural idea used: Rn and Z+ denote the n-dimensional Euclidean space and
is to measure these disturbances and counteract them by com- the set of positive integers, respectively. For any symmetric ma-
pensation via the corresponding controller in the control system. trix P ∈ Rn×n , P > 0 means P is a symmetric positive definite ma-
Unfortunately, disturbances could seldom be measured in prac- trix, where the symmetric elements are indicated by “∗”. Denote
tice. Disturbance estimation has therefore received much atten- by l2 [0, +∞) the∑ space of square summable infinite sequence

tion in the recent years, such as the extended state observer (Hao, defined by {δ (k)| k=0 δ T (k)δ (k) < +∞}. For p, q ∈ Z+ with
Liu, & Zhou, 2017), and disturbance observers (Guo & Chen, p ≤ q, denote I[p, q] ≜ {p, p + 1, . . . , q}.
2005; Yao & Guo, 2013). A new prediction-based scheme was
proposed in Lechappe, Moulay, Plestan, Glumineau, and Chriette 2. Problem statement and preliminaries
(2015) for control of continuous linear time-invariant (LTI) sys-
tems with a constant and known input delay subject to unknown An input-delayed system with time-varying uncertainties sub-
disturbances, and further extended to deal with time-invariant ject to disturbance may be generally described by
plant uncertainties based on delay estimation (Lechappe, Rou- x(k + 1) = [A + ∆A(k)]x(k)
{
quet, Gonzalez, Plestan, Leon, Moulay, & Glumineau, 2016). By +[B + ∆B(k)][u(k − h) + d(k)], (1)
introducing a linear tracking differentiator, an enhanced distur- y(k) = Cx(k),
bance rejection design for continuous-time LTI systems with in-
put delay was developed in Sanz, Garcia, and Albertos (2016) where x(k) ∈ Rn , u(k) ∈ Rm , y(k) ∈ Rl denote the system state,
based on state feedback. However, it remains open as yet to input and output, respectively. The input delay is denoted by h
extend the above-mentioned methods to anti-disturbance pre- with h ≥ 1. A ∈ Rn×n , B ∈ Rn×m and C ∈ Rl×n are system matrices
dictor design for input-delayed systems with time-varying plant with appropriate dimensions, ∆A(k) and ∆B(k) denote system
uncertainties. uncertainties estimated by [∆A(k) ∆B(k)] = E ∆(k)[FA FB ], where
Considering that the plant state information is usually not E ∈ Rn×l1 , FA ∈ Rl2 ×n and FB ∈ Rl2 ×m are known matrices,
available in industrial applications, this paper proposes a novel ∆(k) ∈ Rl1 ×l2 is a time-varying matrix with ∆T (k)∆(k) ≤ Il2 .
output feedback control design for input-delayed systems with Without loss of generality, the unknown disturbance d(k) is
time-varying uncertainties subject to load disturbance that is not described by the following exogenous system (Yao & Guo, 2013),
bounded by the H2 norm. The main contributions include:
ω(k + 1) = S ω(k) + M δ (k),
{
(i) A novel predictor, named state & disturbance observer– (2)
d(k) = H ω(k),
predictor (SDOP), is proposed based on only output mea-
surement to improve disturbance rejection performance where S ∈ Rnω ×nω , H ∈ Rm×nω and M ∈ Rnω ×mω are known
for input-delayed systems with time-varying uncertainties, matrices, and δ (k) ∈ Rmω denotes additional perturbations or
compared to the recently developed methods (e.g. Lechappe uncertainties to the exogenous system, which is assumed to be
et al., 2015; Sanz et al., 2016, 2017) that need full system energy-bounded and belongs to l2 [0, +∞).
state measurement together with absence of time-varying
uncertainties. Accordingly, the adverse effect from various Remark 1. The commonly encountered disturbance types in
disturbances commonly encountered in engineering prac- engineering practice could be described by system in (2), i.e., con-
tices such as constant, asymptotically stable, step, ramp, or stant, asymptotically stable, or step type disturbance by letting
harmonic type could be asymptotically eliminated; S = 1 and H = 1, time-varying disturbance of the (r − 1)-th order
10 S. Hao, T. Liu and B. Zhou / Automatica 104 (2019) 8–16

specified in (2) for disturbance rejection. Note that the proposed


SDOP can simultaneously predict the ‘delay-free’ system state and
disturbance, compared to the developed state observers in the
existing references (e.g. Mazenc & Malisoff, 2017b; Sanz et al.,
2017; Zhou et al., 2017) that could not predict the disturbance
dynamics.
By reformulating the system description in (1) with the ex-
ogenous system in (2), an augmented description of the nominal
system with ∆A(k) = 0, ∆B(k) = 0 is given by
xg (k + 1) = Ag xg (k) + Bg u(k − h) + Mg δ (k),
{
(6)
y(k) = Cg xg (k),
Fig. 1. Proposed control scheme by SDOP.
where xg (k) = [xT (k) ωT (k)]T , and
[ ] [ ] [ ]T [ ]
A BH B CT 0
polynomial in time by letting S = e S1 Ts
and H = [1 01×(r −1) ],
Ag = , Bg = , Cg = , Mg = .
0 S 0 0 M
harmonic disturbance with identified frequency ωc by letting S =
eS2 Ts and H = [1 0], where A full-order SDOP based on only output measurement is proposed
as
ωc
[ ] [ ]
0(r −1)×1 Ir − 1 0
S1 = , S2 = , ẑ(k + 1) = Ag ẑ(k) + Bg u(k) + L[Cg ẑ(k − h) − y(k)], (7)
0 01×(r −1) −ωc 0

and Ts is the sampling period. More details about the disturbance where ẑ(k) = [x̂T (k) ω̂T (k)]T is a h-step prediction of xg (k).
expression can be found in Su, Chen, and Yang (2016). Note that x̂(k) and ω̂(k) are predictions of the system state and exogenous
M may be specified as an identity matrix if no prior knowledge system state, respectively. L is the SDOP gain to be determined.
or identification information could be used. Define the state & disturbance prediction error by

The control task is to design a state & disturbance prediction e(k) = ẑ(k − h) − xg (k). (8)
based control scheme based on only the output measurement,
Based on (7) and (8), we can derive
such that the input-delayed system in (1) under control could
eliminate the disturbance influence while maintaining robust e(k + 1) = Ag e(k) + LCg e(k − h) − Mg δ (k). (9)
stability against time-varying plant uncertainties.
For study, the following assumption and lemmas are given for
control design and robust stability analysis.
Remark 2. For the delay-free nominal system in (1), i.e., h = 0,
Assumption 1 (Guo & Chen, 2005). (i) (A, B) is controllable and the error dynamics in (9) is reduced to
(C , A) is observable; (ii) (S , BH) is observable. e(k + 1) = (Ag + LCg )e(k) − Mg δ (k). (10)

Lemma 1 (Gu, Kharitonov, & Chen, 2003). Given matrices X and Based on Assumption 1, there exists a matrix L such that system
Y with appropriate dimensions, the inequality below holds for any (10) is asymptotically stable in the presence of δ (k). Accordingly,
scalar ε > 0 and the matrix Θ satisfying Θ T Θ ≤ I with appropriate the error dynamics in (9) could be stabilized if the input delay
dimension, h is sufficiently small, as discussed in Najafi et al. (2013). Note
that the minimum of h in discrete-time domain is related to the
X Θ Y + Y T Θ T X T ≤ ε XX T + ε −1 Y T Y . (3)
sampling period.

Lemma 2 (Seuret, Gouaisbaut, & Fridman, 2015). Given a symmetric Define the H∞ performance from δ (k) to e(k) by
positive definite matrix Z , any sequence of discrete-time variables
denoted by x in I[−h, 0] → Rn , the following inequality holds ∥e(k)∥ < γ ∥δ (k)∥. (11)

k−1 [ ]T [ The following theorem is given for the error dynamics in (9) in
1 ϑ1 ϑ1
][ ]
∑ Z 0
ζ T (i)Z ζ (i) ≥ , (4) terms of the above H∞ performance.
h ϑ2 0 3α (h)Z ϑ2
i=k−h
Theorem 1. The error dynamics in (9) is asymptotically stable with
where ζ (i) = x(i + 1) − x(i), and
a robust H∞ performance level γ , if there exist matrices P > 0, Q >
ϑ1 = x(k) − x(k − h), 0, R > 0, and L such that the following matrix inequality holds
k ⎡ T
−G1 PG1 + Q̄ − GT3 R̄G3 GT2 hGT4 Ī T

2 ∑
ϑ2 = x(k) + x(k − h) − x(i), ∗ −P −1 0 0 ⎥
h+1 (5) ⎦ < 0, (12)

i=k−h
∗ ∗ −R−1 0

, if h > 1, h ∈ Z+ ;
{ h+1
∗ ∗ ∗ −γ Iη
α (h) = h−1
1, if h = 1, h ∈ Z+ .
where Q̄ = diag(Q , −Q , 0, −γ Imω ), Ī = [Iη 0 0 0], R̄ = diag
(R, 3α (h)R), G4 = [Ag − Iη LCg 0 − Mg ], η = n + nω , and
3. Anti-disturbance control design by SDOP
[ ] [ ]
Iη 0 0 0 I − Iη 0 0
A novel anti-disturbance control scheme is proposed as shown G1 = , G3 = η ,
− Iη 0 (h + 1)Iη 0 Iη Iη −2Iη 0
in Fig. 1, where the dot-line box outlines the real plant with [ ]
Ag LCg 0 −Mg
input delay, the dash–dot-line box includes the proposed SDOP, G2 = .
0 − Iη (h + 1)Iη 0
K is an observer-based state feedback controller, and H has been
S. Hao, T. Liu and B. Zhou / Automatica 104 (2019) 8–16 11

Proof. Firstly, we define the following Lyapunov–Krasovskii func- objective function,


tional candidate for δ (k) = 0, ∞ [
∑ ]
∑3 J = γ −1 eT (k)e(k) − γ δ T (k)δ (k) .
V (k) = Vi (k), V1 (k) = ξ T (k)P ξ (k), k=0

i=1 Under zero initial condition, we have


k−1 ∞ [
∑ ]
eT (i)Qe(i), σ (i) = e(i + 1) − e(i),

V2 (k) = J ≤ γ −1 eT (k)e(k) − γ δ T (k)δ (k) + ∆V (k)
i=k−h k=0
−1 ∑
∑ k−1 ∞
∑ [
V3 (k) = h σ T (i)Rσ (i), ≤ φ T (k) GT2 PG2 − GT1 PG1 + Q̃
j=−h i=k+j k=0

+h2 GT4 RG4 − GT3 R̄G3 φ (k),


]
∑k−1
where ξ (k) = [eT (k) T T
i=k−h e (i)] . For abbreviation, we also
define where Q̃ = diag(Q + γ −1 Iη , −Q , 0, −γ Imω ) and φ (k) =
1 ∑ k [φ̂ T (k) δ (k)]T . By using the Schur complement, it follows that
T

µ(k) ≜ e(i), J < 0 in terms of LMI condition in (12), which guarantees the
h+1 robust H∞ performance level (11). This completes the proof. □
i=k−h
]T
φ̂ (k) ≜ eT (k) µT (k) .
[
eT (k − h) Note that the inequality in (12) is not a strict LMI due to
the terms P −1 and R−1 . A widely used cone complementarity
It follows that
[ ] [ ] linearization (CCL) algorithm (Ghaoui, Oustry, & AitRami, 1997)
e(k) Iη 0 0 can be adopted to solve the nonlinear matrix inequality.
∑k−1 = φ̂ (k) ≜ Ĝ1 φ̂ (k),
i=k−h e(i) − Iη 0 (h + 1)Iη Based on the predicted system state and disturbance, the
[ ] [ ] following control law is designed,
e(k + 1) Ag LCg 0
∑k = φ̂ (k) ≜ Ĝ2 φ̂ (k).
i=k+1−h e(i)
0 − Iη (h + 1)Iη u(k) = K x̂(k) − H ω̂(k), (13)
Taking the forward difference of V (k) yields where K is an observer-based state feedback controller that could
be designed independent of the SDOP design.
∆V1 (k) = φ̂ (k)[ T
ĜT2 P Ĝ2 − ĜT1 P Ĝ1 ]φ̂ (k),
∆V2 (k) = φ̂ (k)Q̂ φ̂ (k),
T 4. Anti-disturbance control design by SSDOPs
k−1
∑ In the presence of long input delay, the SDOP in (7) may not
∆V3 (k) = h2 φ̂ T (k)ĜT4 RĜ4 φ̂ (k) − h σ T (i)Rσ (i),
be capable of guaranteeing no prediction error. Motivated by the
i=k−h
observer design approach in Liu and Zhou (2017), Najafi et al.
where Ĝ4 = [Ag −Iη LCg 0] and Q̂ = diag(Q , −Q , 0). By Lemma 2, (2013) and Zhou et al. (2017), another control scheme is proposed
we obtain to cope with this problem as shown in Fig. 2, where the proposed
SSDOPs is outlined by the dash–dot-line box. For any given h, h̄ ∈
k−1 [ ]T [
θ1 θ1 Z+ , there exist r and ∆h ∈ Z+ with ∆h < h̄ such that
][ ]
∑ R 0
−h σ T (i)Rσ (i) ≤ − ,
θ2 0 3α (h)R θ2
i=k−h h = r h̄ + ∆h. (14)
where The SSDOPs are established by cascading r + 1 coupled SDOPs as
follows,
θ1
[ ] [ ]
e(k) − e(k − h)
=
θ2 2
∑k ⎧
e(k) + e(k − h) − h+1 i=k−h e(i) ⎪ ẑ0 (k + 1) = Ag ẑ0 (k) + Bg u(k)

+ L0 [ẑ0 (k − h̄) − ẑ1 (k − h̄ + ∆h)],
[ ] ⎪
I −Iη 0

= η φ̂ (k) ≜ Ĝ3 φ̂ (k).



Iη Iη −2Iη ẑi (k + 1) = Ag ẑi (k) + Bg u(k − (i − 1)h̄ − ∆h)


(15)
Then, we have ⎪
⎪ + Li [ẑi (k − h̄) − ẑi+1 (k)], i ∈ I[1, r − 1],

Ag ẑr (k) + Bg u(k − (r − 1)h̄ − ∆h)
( ⎪
∆V (k) ≤ φ̂ T (k) ĜT2 P Ĝ2 − ĜT1 P Ĝ1 + Q̂ + h2 ĜT4 RĜ4


⎪ ẑr (k + 1) =


) + Lr [Cg ẑr (k − h̄) − y(k)],
− ĜT3 R̄Ĝ3 φ̂ (k).
where Li , i ∈ I[0, r ] are the observer gains to be determined.
By using Schur complement to the LMI condition in (12), we have
Remark 3. If ∆h = 0, i.e., h = r h̄, the dynamics z0 (k) is the
Π1
[ ]
ĜT2 P M̂g − h2 ĜT4 RMg same as that of z1 (k). Note that h̄ should be chosen to ensure
< 0,
∗ M̂gT P M̂g + h2 MgT RMg − γ Iη feasibility of the matrix inequality condition in (12) with respect
to a specified γ .
where Π1 = ĜT2 P Ĝ2 − ĜT1 P Ĝ1 + Q̂ + h2 ĜT4 RĜ4 − ĜT3 R̄Ĝ3 + γ −1 Î T Î ,
Define the following sequential prediction errors (i ∈ I[1, r −
M̂g = [−MgT 0]T and Î = [Iη 0 0]. 1]),
Therefore, it follows from the above inequality that
e0 (k) = ẑ0 (k − r h̄ − ∆h) − ẑ1 (k − r h̄),
ĜT2 P Ĝ2 − ĜT1 P Ĝ1 + Q̂ + h2 ĜT4 RĜ4 − ĜT3 R̄Ĝ3 < 0.
ei (k) = ẑi (k − (r − i + 1)h̄) − ẑi+1 (k − (r − i)h̄), (16)
Hence ∆V (k) < 0, which implies that the error dynamics in (9)
er (k) = ẑr (k − h̄) − xg (k).
is asymptotically stable for δ (k) = 0. ∑r
In the sequel, we will prove the H∞ performance level is Then, we have ẑ0 (k − h) − xg (k) = i=0 ei (k). If all the above error
satisfied for δ (k) ̸ = 0. To this end, we define the following vectors converge to zero, ẑ0 (k) = [x̂T0 (k) ω̂0T (k)]T could predict the
12 S. Hao, T. Liu and B. Zhou / Automatica 104 (2019) 8–16

Remark 4. It can be easily verified that if the matrix inequality


in (20) holds, so does the matrix inequality in (19) by letting
Li = Lr Cg , i ∈ I[0, r − 1], P1 = P2 , Q1 = Q2 and R1 = R2 . Note
that the left-hand side of (20) is a submatrix of that in (12). If a
feasible solution of (12) can be computed for a given h, then a
feasible solution of (20) can also be easily found by letting h̄ = h.
Hence, the input delay is allowed to be longer if the number of
SSDOPs is taken sufficiently large such that the partitioned delay
h̄ in (14) could ensure the matrix inequality in (12) feasible.
For δ (k) ̸ = 0, the robust ∑ H∞ performance from δ (k) to
r
the total estimation errors, i.e., i=0 ei (k), should be optimized.
In Najafi et al. (2013, 2014), the augmented error dynamics
[eT0 (k) eT1 (k) . . . eTr (k)]T is adopted to determine the optimal H∞
performance. However, the admissible input delay cannot be
arbitrarily large, and the computation burden is dramatically
Fig. 2. Proposed control scheme by SSDOPs.
increased as the number of SSDOPs increases owing to high
dimension of the augmented error system. Therefore, the fol-
lowing suboptimal H∞ performance is proposed to reduce the
future augmented system state xg (k + h), and correspondingly, computation burden by a recursive computation manner,
x̂0 (k) and ω̂0 (k) could predict the future system state x(k + h)
∥er (k)∥ < γr ∥δ (k)∥,
and the exogenous system state ω(k + h), respectively. Hence, the (22)
dynamics of sequential prediction errors are given by ∥ei−1 (k)∥ < γi−1 ∥ei (k − h̄)∥, i ∈ I[1, r ].

⎪ ei (k + 1) = Ag ei (k) + Li ei (k − h̄) Recursive computation of the SSDOPs gains and optimization of
the above sub-optimal H∞ performance can be conducted by

− Li+1 ei+1 (k − h̄), i ∈ I[0, r − 2],



solving the matrix inequality conditions given in the following


er −1 (k + 1) = Ag er −1 (k) + Lr −1 er −1 (k − h̄) (17) corollary.
− Lr Cg er (k − h̄),




Corollary 1. Given the input delay h̄, the sequential error dynamics


er (k + 1) = Ag er (k) + Lr Cg er (k − h̄) − Mg δ (k).

in (17) is asymptotically stable with a recursive H∞ performance
Owing to an upper triangular form of the sequential error dy- level γr −i , i ∈ I[0, r ], if there exist matrices Pr −i > 0, Qr −i >
namics in (17), it is concluded that asymptotical stability of the 0, Rr −i > 0 and Lr −i , such that the following recursive matrix
sequential error dynamics in (17) is equivalent to the following inequalities hold
system with δ (k) = 0,
Ψr −i < 0, (23)
ei (k + 1) = Ag ei (k) + Li ei (k − h̄), i ∈ I[0, r − 1],
{
(18) with the obtained Lr −i ,
er (k + 1) = Ag er (k) + Lr Cg er (k − h̄).
Ψr −i−1 < 0, (24)
Asymptotic stability conditions for the separable error dynamics
in (18) are given below. where Ī = [Iη 0 0 0], Q̄i = diag(Qi , −Qi , 0, −γi Imω ), R̄i =
diag(Ri , 3α (h̄)Ri ), i ∈ I[0, r ],
Theorem 2. Given the input delay h̄, the sequential error dynamics
Ψi , 1
⎡ ⎤
G̃T2,i h̄G̃T4,i Ī T
in (18) is asymptotically stable, if there exist matrices Pj > 0, Qj >
0, Rj > 0, j = 1, 2 and Li , i ∈ I[0, r ], such that the following two
⎢ ∗
Ψi = ⎣
⎢ −Pi−1 0 0 ⎥ ⎥,
1
matrix inequalities hold ∗ ∗ − R− i 0 ⎦
∗ ∗ ∗ −γi Iη
Ξ1
⎡ ⎤
ḠT2,i h̄ḠT4,i
Ψi,1 = − G̃T1 Pi G̃1 + Q̄i − G̃T3 R̄i G̃3 ,
⎣∗ −P1−1 0 ⎦ < 0, (19) [ ] [ ]
−R− 1 Iη 0 0 0 I − Iη 0 0
∗ ∗ 1 G̃1 = , G̃3 = η ,
−Iη 0 (h̄ + 1)Iη 0 Iη Iη −2Iη 0
Ξ2
⎡ ⎤
ḠT2,r h̄ḠT4,r
[ ]
Ag Lr C g 0 −Mg
G̃2,r = ,
⎣∗ −P2−1 0 ⎦ < 0, (20) 0 − Iη (h̄ + 1)Iη 0
1
∗ ∗ −R−
2
[
Ag Lr −1 0 −Lr Cg
]
G̃2,r −1 = ,
where i ∈ I[0, r − 1], Ξj = −ḠT1 Pj Ḡ1 + Q̄j − ḠT3 R̄j Ḡ3 , j = 1, 2, and 0 −Iη (h̄ + 1)Iη 0
[ ]
Ag Li 0 −Li+1
, i ∈ I[0, r − 2],
[ ] [ ]
Iη 0 0 A Li 0 G̃2,i =
Ḡ1 = , Ḡ2,i = g , 0 −Iη (h̄ + 1)Iη 0
− Iη 0 (h̄ + 1)Iη 0 −Iη (h̄ + 1)Iη
[ ] [ ] G̃4,r = [Ag − Iη Lr Cg 0 − Mg ],
Iη − Iη 0 A Lr Cg 0
Ḡ3 = , Ḡ2,r = g , G̃4,r −1 = [Ag − Iη Lr −1 0 − Lr Cg ],
Iη Iη −2Iη 0 −Iη (h̄ + 1)Iη
Ḡ4,i = [Ag − Iη Li 0], Ḡ4,r = [Ag − Iη Lr Cg 0], G̃4,i = [Ag − Iη Li 0 − Li+1 ], i ∈ I[0, r − 2].

Q̄j = diag(Qj , −Qj , 0), R̄j = diag(Rj , 3α (h̄)Rj ).


Proof. The proof of this corollary is similar to that of Theorem 1,
(21) and thus is omitted. □

Proof. The proof is similar to that of Theorem 1, and thus is Remark 5. It is easy to see that the separation principle still holds
omitted. □ in Corollary 1, which implies that the input delay (assumed to be
S. Hao, T. Liu and B. Zhou / Automatica 104 (2019) 8–16 13

known) is allowed to be arbitrarily long by increasing the number diag(R, 3α (h̄)R), C¯g = [Cg 0 0 0], Q̄ = diag(Q, −Q, 0, −γ Imω ),
of SSDOPs. η̄ = n + (r + 1)η, G4 = [Ag − Iη̄ Adg 0 Mg ], and
Note that Li , i ∈ I[0, r ] can be solved similarly by using the
[ ] [ ]
Iη̄ 0 0 0 I −Iη̄ 0 0
existing CCL algorithm (Ghaoui et al., 1997). G1 = , G3 = η̄ ,
−Iη̄ 0 (h̄ + 1)Iη̄ 0 Iη̄ Iη̄ −2Iη̄ 0
Similar to the control design with SDOP, the following con- [ ]
Ag Adg 0 Mg
trol law is designed based on the predicted system state and G2 = .
disturbance by SSDOPs, 0 −Iη̄ (h̄ + 1)Iη̄ 0

u(k) = K x̂0 (k) − H ω̂0 (k), (25)


Proof. It follows from a similar proof for Theorem 1 in combina-
where K is an observer-based state feedback controller. tion with Lemma 1, and thus is omitted. □
Correspondingly, the anti-disturbance controller K and Li , i ∈
5. Robust stability analysis
I[0, r ] can be solved from (28) by the CCL algorithm to accommo-
Since the proposed SDOP may be viewed as a special case of date for time-varying uncertainties.
the proposed SSDOPs, robust stability analysis is therefore de-
tailed for using SSDOPs in this section. Applying the control law in 6. Illustration
(25) to the original system in (1) along with the sequential error
dynamics in (17), we obtain the following augmented system Consider an open-loop unstable system with input delay stud-
⎧ ied in Lechappe et al. (2015),
⎨X (k + 1) = [Ag + Eg ∆(k)Fg ]X (k) ⎧ [ ] [ ]
+Adg X (k − h̄) + Mg δ (k), (26) ⎨ẋ(t) = 0 1 x(t) + 0 [u(t − τ ) + d(t)],
y(k) = Cg X (k), −9 3 1

y(t) = 1 0 x(t), x(0) = 1.5 1 ,
⎩ [ ] [ ]
where Cg = [C 0 0 . . . 0], Mg = [0 0 . . . 0 − MgT ]T , Υ =
[K − H ], Eg = [E T 0 0 . . . 0 − E T 0], and where τ is a specified input delay for illustration. There are two
unstable poles located at 1.5 ± 2.598i. By taking the sampling pe-
A + BK BΥ ··· BΥ riod as Ts = 0.1 s for control implementation, the corresponding
⎡ ⎤
0 Ag ··· 0 ⎥ discrete-time system description is obtained as
⎦,

Ag = ⎢ .. .. .. .. ⎥
. . . . 0.9505 0.1149
⎣ ⎧ [ ]
x(k + 1) = x(k)

0 0 ··· Ag −1.0339 1.2952




⎡ ⎤ ⎨
0 0 0 0 ··· 0 0
0.0055
[ ]
⎢0 L0 −L1 0 ··· 0 0 + [u(k − h) + d(k)],
.1149
⎥ ⎪
0
⎢0 ⎪
0 L1 −L2 ··· 0 0
⎥ ⎪

⎥,
⎢ ⎥ ⎪
⎢ ..
Adg = ⎢ .. .. .. .. .. .. ⎥ y(k) = [1 0]x(k), x(0) = [1.5 1],

⎢. . . . . . . ⎥
⎣0 0 0
· · · Lr −1 0 − Lr C g ⎦ where h is the corresponding input delay in discrete-time do-
0 0 0
··· 0 0 Lr C g main. The following three cases are tested.
Fg = [FA + FB K FB Υ FB Υ . . . FB Υ ], Case 1. (Nominal system) The disturbance is assumed to be a
T
X (k) = [x (k) eT0 (k) eT1 (k) ... eTr (k) T]. sinusoidal signal with identified frequency ωc = 0.5 rad/s and
unknown magnitude. The corresponding discrete-time exogenous
Note that for the nominal system (i.e., ∆A(k) = 0 and ∆B(k) = 0), system matrices are written in the form of
the state-feedback controller K and the gains of SSDOPs Li , i ∈
0.9988 0.05
[ ]
I[0, r ], can be separately designed owing to an upper triangular , 0 ,
[ ]
S= H= 1
form of the closed-loop system. However, the separation prin- −0.05 0.9988
ciple could not be applied for the augmented system in (26) in while the additional perturbation δ (k) in the exogenous system
the presence of time-varying plant uncertainties. It is therefore is assumed to zero. By solving the matrix inequality in (12),
necessary to analyze the closed-loop system stability and the
feasible solutions can be found for all h ∈ I[1, 4] by using the
sequential error dynamics together. The following robust stability
proposed SDOP, which means that the input delay is allowed to
theorem is given accordingly in terms of a prescribed H∞ control
be longer if the partitioned delay h̄ belongs to the set I[1, 4].
performance,
Let the initial state of SSDOPs be ẑi (θ ) = [1.5 1 3 0]T for θ ≤
∥y(k)∥ < γ ∥δ (k)∥. (27) 0. For h = 5, corresponding to τ = 0.5 in continuous-time
system, the proposed controller and the gains of SSDOPs with
h̄ = 2 and ∆h = 1 are derived by solving the matrix inequalities
(19) and (20), obtaining K = [−14.1313 − 10.5553], L2 =
Theorem 3. Given the input delay h̄, the augmented system in
[−0.2722 0.0718 − 0.0575 0.0132]T ,
(26) is asymptotically stable with a robust H∞ performance level γ
if there exist matrices P , Q > 0, R > 0, K , Li , i ∈ I[0, r ] and scalar −0.0123 −0.1411 −0.0360 −0.0029
⎡ ⎤
ε > 0 such that the following matrix inequality holds ⎢ 0.6879 −0.5004 −0.0711 −0.0028⎥
Li = ⎣ , i = 0, 1.
−0.0281 0.0039 −0.3357 −0.0463⎦
Ω1 G2T h̄G4T C¯gT F̄gT
⎡ ⎤
0.0047 −0.0012 0.0461 −0.3386
⎢∗ Ω2 0 0 0 ⎥
⎢∗ ∗ Ω3 0 0 ⎥ < 0, (28) Note that the computation time (Dell, Core i7-3770 @ 3.40 GHz)
⎢ ⎥
⎣∗ ∗ ∗ −γ Il 0
⎦ is about 12.5 s. The control results are shown in Fig. 3. The results
∗ ∗ ∗ ∗ −ε Il2 by using the predictor-based control methods given in Lechappe
et al. (2015) and Sanz et al. (2016) based on the controller design
where Ω1 = −G1T PG1 + Q̄ − G3T R̄G3 , Ω2 = −P −1 + ε Ēg ĒgT , formulae given therein are also shown in Fig. 3 for comparison.
Ω3 = −R−1 + ε h̄2 Eg EgT , Ēg = [EgT 0]T , F̄g = [Fg 0 0 0], R̄ = It is seen that improved disturbance rejection performance with
14 S. Hao, T. Liu and B. Zhou / Automatica 104 (2019) 8–16

Fig. 3. Control results for h = 5 and h̄ = 2.

Fig. 5. Control results for h = 10 and r = 5.

Table 1
Computation time with respect to r for h̄ = 2 in Case 3.
No. of SSDOPs r =1 r =2 r =3 r =4
Computation time (s) 205 389 4349 48698

of a larger input delay. In contrast, a larger steady-state output


error is resulted by using the control methods given in Lechappe
et al. (2015) and Sanz et al. (2016). Fig. 5(c) shows the sequential
prediction errors of system state and disturbance by the proposed
method, well demonstrating the asymptotic stability.
Case 3. (Time-varying uncertainties) The disturbance is assumed to
be a combination of a step type disturbance with unknown mag-
Fig. 4. Control results for h = 5 and h̄ = 2 with respect to the mismatched nitude and a sinusoidal disturbance as in case 1. The correspond-
disturbance frequencies. ing discrete-time exogenous system matrices can be written as
0.9988 0.05 0
[ ]
−0.05 0.9988 0 , 1 .
[ ]
S= H= 1 0
no steady-state output error is obtained by the proposed control 0 0 1
method, compared to the control methods given in Lechappe
et al. (2015) and Sanz et al. (2016) that could not eliminate An additional perturbation in the exogenous system is assumed
the steady-state output error. It should be noted that a larger to be δ (k) = sin(k)/(1 + k) along with a perturbation matrix
oscillation in the transient response by the proposed method M = [0 0 0.1]T . The uncertain system matrix ∆A(k) is composed
arises from the initial prediction errors of state and disturbance of E = [0 0.03]T , FA = [0.1 0.2], FB = 0.05 and ∆(k) = sin(k).
based on only output measurement, whereas the control methods With h̄ = 2 and γ = 200, the computation time for solving the
in Lechappe et al. (2015) and Sanz et al. (2016) were based on full matrix inequality condition in (28) by the CCL algorithm is listed
system state measurement for feedback control. If the real distur- in Table 1 with respect to the number of SSDOPs, which illustrates
bance frequency (ωr ) is not matched with the identified frequency that the computation burden dramatically increases with respect
(ωc = 0.5 rad/s), the control results are shown in Fig. 4, demon- to the value of r when the separation principle is not satisfied
strating good robustness with respect to an inexactly estimated any more. For illustration, the anti-disturbance controller and the
disturbance frequency. gains of SSDOPs with r = 3, i.e., h = 6, are given below,
Case 2. (Long input delay) The input delay is assumed to be twice K = [−1.2189 − 5.6011],
larger, i.e., h = 10, the disturbance is considered as the same with 0.0103

−0.1655 −0.0371 −0.0039 −0.0618

Case 1. Since h̄ = 2 is taken for r = 5, the gains of SSDOPs ⎢ 1.0638 −0.6299 −0.1617 0.0582 −0.1819⎥
are taken the same as Case 1, i.e., L5 = [−0.2722 0.0718 − L1 = ⎢ 0.0823

−0.0072 −0.2583 −0.0433 −0.0593⎥ ,

0.0575 0.0132]T , ⎣−0.0437 0.0032 0.1168 −0.3421 −0.0731⎦
−0.0123 −0.1411 −0.0360 −0.0029 0.1084 −0.0213 −0.0393 0.0002 −0.3870
⎡ ⎤
⎢ 0.6879 −0.5004 −0.0711 −0.0028⎥ ⎡
−0.0627 −0.0503 0.0009 −0.0015 −0.0115

Li = ⎣ , i ∈ I[1, 4].
−0.0281 0.0039 −0.3357 −0.0463⎦ ⎢ 0.6417 −0.3616 −0.0274 0.0028 −0.0345⎥
0.0047 −0.0012 0.0461 −0.3386 L2 = ⎢ 0.0707

0.0086 −0.1152 −0.0382 −0.0033⎥ ,

⎣−0.0203 0.0062 0.0082 −0.1373 0.0076 ⎦
The control results are shown in Fig. 5 where the controller
gain is taken the same as in Case 1. It is seen that asymptotic 0.1177 −0.0013 0.0030 −0.0026 −0.1632
disturbance rejection performance is achieved for the presence L3 = [−0.1913 0.2860 − 0.0373 − 0.0033 − 0.0212]T .
S. Hao, T. Liu and B. Zhou / Automatica 104 (2019) 8–16 15

References

Albertos, P., & Garcia, P. (2009). Robust control design for long time-delay
systems. Journal of Process Control, 19(10), 1640–1648.
Artstein, Z. (1982). Linear systems with delayed controls: a reduction. IEEE
Transactions on Automatic Control, 27(4), 869–879.
Cacase, F., & Germani, A. (2017). Output feedback control of linear systems
with input, state and output delays by chains of predictors. Automatica, 85,
455–461.
Du, B., Lam, J., & Shu, Z. (2010). Stabilization for state/input delay systems via
static and integral output feedback. Automatica, 46, 2000–2007.
Fridman, E. (2014). Introduction to time-delay systems. Birkhäuser.
Ghaoui, L. El., Oustry, F., & AitRami, M. (1997). A cone complementarity
linearisation algorithm for static output-feedback and related problems. IEEE
Transactions on Automatic Control, 42, 1171–1176.
Gonzalez, A. (2013). Robust stabilization of linear discrete-time systems with
time-varying input delay. Automatica, 49(9), 2919–2922.
Gonzalez, A., Sala, A., & Albertos, P. (2012). Predictor-based stabilization of
discrete time-varying input-delay systems. Automatica, 48(2), 454–457.
Gu, K., Kharitonov, V., & Chen, J. (2003). Stability of time-delay systems. Boston:
Birkhauser Springer.
Fig. 6. Control results for h = 6 and r = 3 in Case 3. Guo, L., & Chen, W. (2005). Disturbance attenuation and rejection for systems
with nonlinearity via DOBC approach. International Journal of Robust and
Nonlinear Control, 15, 109–125.
Hao, S., Liu, T., & Zhou, B. (2017). Predictor-based output feedback control design
Let the initial states of SSDOPs be ẑi (θ ) = [1.5 1 3 0 1]T for for sampled systems with input delay subject to disturbance. IET Control
θ ≤ 0, the control results are shown in Fig. 6. It is seen that Theory & Applications, 11(18), 3329–3340.
Kojima, A., Uchida, K., Shimemura, E., & Ishijima, S. (1994). Robust stabilization
asymptotic disturbance rejection performance maintains well in of a system with delays in control. IEEE Transactions on Automatic Control,
the presence of the time-varying uncertainties. Note that the 39(8), 1694–1698.
robust stability cannot be guaranteed any more by the control Krstic, M. (2010). Lyapunov stability of linear predictor feedback for time-varying
input delay. IEEE Transactions on Automatic Control, 55(2), 554–559.
methods given in Lechappe et al. (2015, 2016) and Sanz et al. Lechappe, V., Moulay, E., Plestan, F., Glumineau, A., & Chriette, A. (2015). New
(2016) for the presence of the above time-varying uncertainties, predictive scheme for the control of LTI systems with input delay and
which are omitted for brevity. unknown disturbance. Automatica, 52, 179–184.
Lechappe, V., Rouquet, S., Gonzalez, A., Plestan, F., Leon, J., Moulay, E., &
Glumineau, A. (2016). Delay estimation and predictive control of uncertain
7. Conclusions systems with input delay: application to a DC motor. IEEE Transactions on
Industrial Electronics, 63(9), 5849–5857.
Liu, T., & Gao, F. (2012). Industrial process identification and control design:
For input-delayed systems with time-varying uncertainties step-test and relay-experiment-based methods. London: Springer.
subject to load disturbance, a robust predictor-based anti- Liu, Q., & Zhou, B. (2017). Extended observer based feedback control of linear
systems with both state and input delays. Journal of the Franklin Institute,
disturbance control design based on only output measurement
354(18), 8232–8255.
has been proposed to improve system performance in the pres- Mazenc, F., & Malisoff, M. (2017a). Stabilization and robustness analysis for time-
ence of constant, asymptotically stable, step type, ramp type, varying systems with time-varying delays using a sequential subpredictors
polynomial in time, or harmonic type disturbances that are com- approach. Automatica, 82, 118–127.
Mazenc, F., & Malisoff, M. (2017b). Stabilization of nonlinear time-varying
monly encountered in engineering practice. An important merit systems through a new prediction based approach. IEEE Transactions on
of the proposed SDOP is that it could predict the ‘delay-free’ Automatic Control, 62(6), 2908–2915.
state and disturbance simultaneously to facilitate state feedback Najafi, M., Hosseinniab, S., Sheiholeslama, F., & Karimadini, M. (2013). Closed-
loop control of dead time systems via sequential sub-predictors. International
control design and feedforward compensation of the disturbance
Journal of Control, 86(4), 599–609.
effect against the input delay. Moreover, another design of SS- Najafi, M., Sheiholeslama, F., Wang, Q.-G., & Hosseinniab, S. (2014). Robust H∞
DOPs is proposed for systems with long input delay, such that control of single input-delay systems based on sequential sub-predictors. IET
each SSDOP only needs to estimate the future system state and Control Theory & Applications, 8(13), 1175–1184.
Normey-Rico, J., & Camacho, E. (2007). Control of dead-time processes. London:
disturbance in terms of a specified small step size. A recursive Springer.
sub-optimal H∞ design of SSDOPs is given to reduce the computa- Richard, J. (2003). Time-delay systems: An overview of some recent advances
tion burden owing to its independence of the number of SSDOPs. and open problems. Automatica, 39(10), 1667–1694.
Sanz, R., Garcia, P., & Albertos, P. (2016). Enhanced disturbance rejection for
Distinguished from the developed predictor based control designs a predictor-based control of LTI systems with input delay. Automatica, 72,
in the literature where the control history with respect to the 205–208.
delay length requires to be stored for state prediction (not includ- Sanz, R., Garcia, P., Albertos, P., & Zhong, Q. (2017). Robust controller design
for input-delayed systems using predictive feedback and an uncertainty
ing disturbance), the proposed anti-disturbance control method estimator. International Journal of Robust and Nonlinear Control, 27(10),
needs only one observer state in the history for the SDOP or 1826–1840.
r + 1 historical observer states for the SSDOPs, while guarantee- Seuret, A., Gouaisbaut, F., & Fridman, E. (2015). Stability of discrete-time systems
ing asymptotic disturbance rejection performance, no matter if with time-varying delays via a novel summation inequality. IEEE Transactions
on Automatic Control, 60(10), 2740–2745.
there exist time-varying plant uncertainties. Sufficient conditions Smith, O. (1959). Closer control of loops with dead time. Chemical Engineering
guaranteeing robust stability of the (sequential) error dynamics Progress, 53, 217–219.
or the closed-loop system has been established to accommodate Su, J., Chen, W.-H., & Yang, J. (2016). On relationship between time-domain
and frequency-domain disturbance observers and its applications. Journal of
for time-varying uncertainties. An illustrative example from the Dynamic Systems, Measurement, and Control, 138(9), 091013.
literature has well demonstrated the effectiveness and merit of Yao, X., & Guo, L. (2013). Composite anti-disturbance control for Markovian jump
the proposed method in the presence of different disturbances, nonlinear systems via disturbance observer. Automatica, 49(8), 2538–2545.
Zhou, B. (2014). Input delay compensation of linear systems with both state and
long input delay, and time-varying uncertainties. It should be input delays by nested prediction. Automatica, 50(5), 1434–1443.
noted that the extension of the proposed SDOP or SSDOPs to load Zhou, B., Lin, Z., & Duan, G. (2012). Truncated predictor feedback for linear
disturbance completely different from the above commonly en- systems with long time-varying input delays. Automatica, 48(10), 2387–2399.
Zhou, B., Liu, Q., & Mazenc, F. (2017). Stabilization of linear systems with both
countered disturbance types is not trivial, which deserves further input and state delays by observer-predictors. Automatica, 83, 368–377.
exploration in the future.
16 S. Hao, T. Liu and B. Zhou / Automatica 104 (2019) 8–16

Shoulin Hao received his B.S. degree in Mathematics a monograph by Springer. He serves as an Associate Editor of ISA Transac-
and Applied Mathematics, M.S. degree in Operational tions, Systems Science and Control Engineering, an editorial board member
Research and Cybernetics, and Ph.D. degree in Con- of International Journal of Control, a member of the Technical Committee on
trol Theory and Control Engineering all from Dalian Chemical Process Control of IFAC, Technical Committee on System Identification
University of Technology, Dalian, China, in 2010, 2013 and Adaptive Control of the IEEE Control System Society, Technical Committees
and 2018, respectively. He is currently a postdoctoral of Control Theory and Process Control of Chinese Association of Automation.
research fellow with the Institute of Advanced Control
Technology, Dalian University of Technology, Dalian,
Bin Zhou is a Professor of the Department of Control
China. His research interests include industrial pro-
Science and Engineering at the Harbin Institute of
cess control, time-delay systems, and iterative learning
Technology, Harbin, China. He received B.S. degree,
control.
M.S. Degree and Ph.D. degree from the Department
of Control Science and Engineering at Harbin In-
Tao Liu received the Ph.D. degree in Control Science stitute of Technology, Harbin, China in 2004, 2006
and Engineering from Shanghai Jiaotong University, and 2010, respectively. His research interests include
Shanghai, China, in 2006. He was a postdoctoral re- constrained control, time-delay systems, time-varying
search fellow with the Department of Chemical and systems, nonlinear control, multi-agent systems, and
Biomolecular Engineering, Hong Kong University of Sci- control applications in astronautic engineering. In these
ence and Technology, Hong Kong, from 2006 to 2010, areas, he has published over 100 papers in archival
and an Alexander von Humboldt research fellow with journals. He is the author of the book Truncated Predictor Feedback for Time-
the Institute of Process Systems Engineering, RWTH Delay Systems (Springer-Verlag, Berlin Heidelberg, 2014). He is a reviewer for
Aachen University, Aachen, Germany, from 2010 to American Mathematical Review and is an active reviewer for many journals. He
2012. He is a Professor with the Institute of Advanced received the ‘‘National Excellent Doctoral Dissertation Award’’ in 2012 from the
Control Technology, Dalian University of Technology, Academic Degrees Committee of the State Council and the Ministry of Education
Dalian, China. of China. He is currently an Associate Editor of the Conference Editorial Board
His research interests include industrial process identification & modeling, of the IEEE Control Systems Society and an Associate Editor of Asian Journal
in-situ process measurement, control system design, batch process optimization, of Control, Journal of System Science and Mathematical Science and Control and
process monitoring. He has published more than 100 research papers and Decision (In Chinese).

You might also like