Second level adaption of the characteristic modeling
∗
Huang Huang
Science and Technology on Space Intelligent Control Laboratory
Beijing Institute of Control Engineering
Beijing, 100190, China
hhuang33@163.com
Abstract— This paper addresses the fast adaption of the
characteristic model-based golden-section adaptive control(CM-
GSAC). The CM-GSAC is well-known for its simplicity and
guaranteed transient performance, which depend on the projec-
tion of the identified parameters onto a bounded convex domain.
In order to avoid excessive efforts in calculating those parameter
bounds, a second level adaption procedure is introduced that
determines a convex combination of a set of models. The
weighted values of the unknown parameters are used for control.
By introducing a projection in matrix form, the weighted values
converge in a short period of time. Simulations validates the fast
speed of convergence with projected second level adaption, as
compared to that of the traditional single level adaption method.
Index Terms— Adaptive control, identification, characteristic
modeling
I. INTRODUCTION
The ultimate goals of control could be simpler, faster,
adaptive, and robust. We hope to design a low-order, ideally
linear, controller that can drive the system to the desired
working condition within a short period of time in spite
of all kinds of uncertainties and changing environment. The
Proportional-integral-derivative (PID) controller is a paragon
and thus has taken the dominant position in all kinds of
industry fields. However, with the extension of human ac-
tivities, the PID controller becomes incapable of dealing
with those newly emerged fields or unable to meet the even
higher accuracy, adaptivity, and robustness requirements. The
theoretical research on control algorithm has provided quite a
few fancy results that have been shown powerful in laborato-
ries. Among all those advanced control methods, the model-
reference adaptive control, neural networks, and H
∞
/H
2
robust control are some of the modern control methods that
have been implemented in unmanned aero vehicles(UAVs)
and satellites by NASA, France, and Japan. Quite a few
reports on flight/on-orbit experiments are available, see [1],
[2], [3], [4], [5].
Towards the simple and adaptive objectives, the charac-
teristic model (CM) proposed by Hongxin Wu in 1990s
is an attractive method. The essence of the characteristic
model is to use a low-order discrete time-varying system to
approach a high-order nonlinear/linear system based on the
∗
This work is partially supported by NSFC Grant #61203075 and Grant
#61333008
main features of the plant and the control demands. Rather
than dropping information as in the reduced-order modeling,
it compresses/integrates all the information of the high-
order model into several characteristic parameters. Later the
characteristic model-based golden-section adaptive control
(CM-GSAC) law was further proposed with the prominent
features including the easy implementation and the guaran-
teed stability during parameter identifications, as compared
with other adaptive control laws. The CM is in a simple
linear form by compressing the featured information of the
plant and the control requirements into those characteristic
parameters. In this sense, the CM-GSAC might be a fresh
perspective in the field of adaptive control: creating simplicity
out of complexity.
One important feature of the CM is the analytical parame-
ter bounds that are determined by the sampling time and the
poles of the plant. The unknown parameters are identified
and are projected onto a bounded domain so as to guarantee
stability and speed of convergence. However, although in
analytical form, those bounds are quite difficult to calculate.
Therefore, efforts on avoiding excessive calculation while
maintaining the stability and the fast convergence speed are
quite demanding. This is the motivation of the current paper.
In this paper, following the work in [10], a second level
adaption algorithm is introduced in addition to the traditional
parameter identification. A set of N models is identified using
the least square algorithm without projection. The true values
of the unknown parameters lie within the convex domain
determined by the N models. A second level adaption will
decide the weighted sum of the N models such that the
weighted estimation output of the N models equals the plant
output. The identification of the second level adaption is
projected onto a convex domain through matrix transforma-
tion, which accelerates the speed of convergence. Simulations
show the prominent features of the second level adaption
including fast convergence and easy implementation.
II. T
HE CM-GSAC
The general transfer function for a single-input-single-
output (SISO) linear time invariant (LTI) system is
G(s)=
b
1
s
1
+ b
1
−1
s
1
−1
+ ...+ b
1
s + b
0
s
2
+ a
2
−1
s
2
−1
+ ...+ a
1
s + a
0
(1)
978-1-4673-9104-7/15/$31.00 ©2015 IEEE
Proceeding of the 2015 IEEE
International Conference on Information and Automation
Lijiang, China, August 2015
1833