Design of Fractional-order Multivariable Extremum Seeking with
Newton Algorithm*
Chun Yin
1
, Member, IEEE, Binyang Hu
1
, Sara Dadras
2
, Member, IEEE,
Hadi Malek
2
, Senior Member, IEEE, Jianhong Xue
1
, and Xuegang Huang
3
Abstract— This paper develops a fractional-order multivari-
able extremum seeking control with Newton algorithm (FO
ESC-NA) for multi-input optimization problem. The proposed
FO ESC-NA is investigated to improve the convergence rate
and enhance the control accuracy. The influence of unknown
Hessian matrix on the convergence speed existed in conventional
methods is effectively eliminated in the proposed FO ESC-NA.
The proposed FO ESC-NA can obviously guarantee the faster
convergence of the extremum seeking scheme to improve the
search efficiency of the extrema by adjusting the fractional-
order. Simulation results illustrate the effectiveness and advan-
tages of the proposed FO ESC-NA.
I. INTRODUCTION
Extremum seeking control (ESC) has been considered
as a useful method to maintain the output at the extrema
level through manipulating the plant input [1], [2]. Since the
achievement of the first stability proof of ES scheme [3], the
significant developments in theory have occurred over the
last decade [4]–[8], meanwhile, ESC has also been applied
in many engineering fields, such as antilock braking systems
[9], maximum power point tracking [10], hybrid lighting
system [11], aerospace application [12].
ESC is a model-less real-time optimization control ap-
proach for nonlinear dynamic problems. Many relative in-
vestigations have been conducted to improve the extremum
seeking scheme. In [13], the proposed ESC with a time
function that is monotonically decreasing to regulate the
perturbation signal amplitude, could enlarge the searching
arguments to reduce the possibility that the local extremum
is realized. A newton-based ESC was bringed by estimating
the second derivative of the output function in [14]. Liu
et al. [15] investigated stochastic averaging principle and
provided the performance analysis of stochastic ESC. The
effect of different perturbation signal on the performance of
ESC was investigated by [16]. The numerical optimization
with state regulator was introduced by [17]. Wang et al. [18]
designed a new ESC scheme by reducing the perturbation
signal amplitude to eliminate the steady-state oscillation. The
FO ESC and its applications was developed by investigating
*This work was supported by National Basic Research Program of China
(Grant No. 61503064 and 61573076).
1
School of Automation Engineering, University of Electronic Sci-
ence and Technology of China, Chengdu 611731, P. R. China
(yinchun.86416@163.com, chunyin@uestc.edu.cn)
2
Electrical and Computer Engineering Department, Utah State Uni-
versity, Logan, UT 84321, United States (s
dadras@ieee.org,
hadi.malek@yahoo.com)
3
Hypervelocity Aerodynamics Institute, China Aerodynamics Re-
search & Development Center, Mianyang 621000, P. R. China
(emei-126@163.com).
FO operators [19], to enhance the control performance of
ESC. Note that all above contributions are just applied in
single-input-single-output (SISO) problems.
In reality, considering numerous complicated optimization
problems, most of the methods above-mentioned cannot
be applied for the problem of multivariate optimization
with multi-input extremum-seeking control scheme. How-
ever, there exist limited researches and reports about these
extremum seeking scheme for multi-input dynamical systems
[20]–[25], to our best knowledge. Both the gradient-based
and newton-based ESC are considered to be the possible
solutions to these multivariable problems (i.e. multi-input-
single-output (MISO) problems) [20], [22]. All of them have
achieved the control purpose that multivariate optimization
are guarantied in the extremum seeking scheme for MISO
systems. However, due to the wide range of engineering
applications, there is still room among researchers and sci-
entists for improvements in convergence rate and control
accuracy for multivariate ESC.
In this paper, an FO ESC-NA is designed for multi-
input optimization problem. The proposed FO ESC-NA is
discussed to enhance the convergence rate and the control
accuracy. Furthermore, the proposed FO ESC-NA effectively
eliminates the influence of unknown Hessian matrix on
the convergence speed existed in conventional methods.
The faster convergence of the proposed FO ESC-NA can
improve the search efficiency of the extrema by adjusting
the fractional-order. Simulation results illustrate the potential
benefits of the proposed FO ESC-NA.
II. FRACTIONAL ORDER EXTREMUM SEEKING CONTROL
WITH NEWTON ALGORITHM
Consider a general MISO nonlinear system
d
dt
x = η(x, u), y = µ(x), (1)
in which x ∈ R
m
, y ∈ R and u ∈ R
n
represent the state,
output and input respectively.η : R
m
× R
n
→ R
m
and µ :
R
m
→ R express smooth function. Consider the controller
u = α(x, ϑ) with ϑ ∈ R
n
. Furthermore, the system (1)
is further expressed by ˙x = η(x, α(x, ϑ)). The following
assumptions will be adopted.
Assumption 1. η(x, α(x, θ)) = 0 can be guaranteed, if
and only if x = l(θ) involving χ : R
n
→ R
m
.
Assumption 2. The equilibrium x = l(θ) of ˙x =
η(x, α(x, θ)) is locally exponentially stable, θ ∈ R
n
.
2018 Annual American Control Conference (ACC)
June 27–29, 2018. Wisconsin Center, Milwaukee, USA
978-1-5386-5428-6/$31.00 ©2018 AACC 1689