You are on page 1of 5

Control of Non-affine Systems Using the State-dependent Riccati Equation (SDRE)

Saeed Rafee Nekoo1 and Behdad Geranmehr2


School of Mechanical Engineering,
Iran University of Science and Technology (IUST)
Tehran, Iran
1
rafee@iust.ac.ir ; 2 behdad.geranmehr@gmail.com

AbstractThe purpose of this paper is to present the


application of the state-dependent Riccati equation (SDRE)
method for controlling a class of non-affine nonlinear systems.
By mentioning a class of non-affine system, a system with
nonlinear states and control inputs is meant. The non-affine in
control systems have been secluded in contrast to affine in
control systems, especially in the SDRE field. The SDRE
method is one of the approaches for solving the Hamilton
JacobiBellman (HJB) equation as a closed loop nonlinear
optimal control method. In this work, the SDRE is solved with
exact solution (ES) method and online control update (OCU)
formulation. The proposed algorithms are verified by three
examples which are selected from previous works. The
effectiveness of the SDRE is demonstrated through numerical
simulations and comparisons with previous works which the
SDRE brilliantly outperforms the other techniques in presence
of external disturbance.
Keywords-SDRE; non-affine nonlinear system; exact
solution; online control update; nonlinear optimal control

I.

INTRODUCTION

In this paper, the state-dependent Riccati equation


(SDRE) for controlling of non-affine nonlinear systems is
used. The SDRE extends quadratic regulator problem for
nonlinear case by allowing the matrices involved to be
functions of the state variables, and possibly the controls as
well. Pearson [1] proposed the SDRE method, then,
developed by Wernli and Cook [2]. The SDRE method is
discussed in details by Cloutier et al. in [3]. Nonlinear
controller design methods based on estimating the solution
of the HamiltonJacobiBellman (HJB) equation can be
found in a comprehensive review article in [4].
There are good literatures available on adaptive, neural
networks and fuzzy adaptive and etc. in the field of control
for non-affine nonlinear systems which can be categorized in
two major groups. The first one has worked on the nonaffine nonlinear problems, which in them it has been usually
assumed a non-affine system in state variables due to
uncertainty and affine in control inputs [5-7].

Some of researchers, second group, assumed non-affine


nonlinear systems as non-affine in control input problems [815]. For example, in the first case, Wang et al. used an
adaptive T_S fuzzy-neural control methods based on on-line
identification algorithm for the virtual linearized system and
put a significant emphasis on the robust tracking controller
design using an adaptive scheme for the general unknown
systems [7]. In addition, the proposed projection update law
used to tune the adjustable parameters to prevent parameter
drift.
In the second case, Gao et al. developed time scale
separation based fuzzy adaptive control method for a class of
non-affine nonlinear systems [12]. Hovakimyan et al.
presented the design methodology for approximation of
dynamic inversion of multi-input non-affine systems using
time-scale separation [10]. The control signal is sought as a
solution of a fast dynamical equation that satisfied the
conditions of Tikhonovs theorem from singular
perturbations theory.
Since there are many applications where the control
inputs are nonlinear such as chemical reactions and PH
neutralization, aircraft dynamics and so on, here in this
article, second case form is considered for analysis and
control via SDRE method. Beeler et al. proposed SDRE for
the non-affine nonlinear problem [4]. In order to illustrate the
effectiveness of the control scheme, proposed in this paper,
numerical and exact solutions have been conducted based on
[4]. The results of the illustrative examples show that the
SDRE control scheme outperform the task and is adaptable
to the non-affine nonlinear systems.
In section II of this paper a brief description of the SDRE
in the context of non-affine nonlinear system is given. The
three examples to illustrate the potential of the SDRE
technique for the design of nonlinear feedback controllers are
provided for non-affine nonlinear system in Section III.
Finally, overall conclusion is presented in section IV.
II.

PROBLEM STATEMEN

In this section the optimal control problem is formulated


with the assumption that all state variables are available for

2538

providing closed loop feedback. The SDRE approach


involves manipulating the dynamical equations

x f ( x) g ( x, u)

A. Exact Solution
Consider the tracking problem for the scalar nonlinear
system which is given by

x 0.5x tanh(u 3) tanh(u 3) 0.01u

where the states are x ,the vector


n

f ( x)
n

and g ( x, u) . In the state-dependent


coefficient (SDC) form the following equation is obtained
n

x A( x) x B( x, u)u

with initial condition x(0) 0.5 which is selected from


[8]. The numerical simulation is performed applying the
following reference input to be tracked by the controller

A( x) : n nn , the input vector is u m and


B( x, u) : n nm . The aim of this method is to

where

1
1.5

1 exp(t 8) 1 exp(t 15)

1
0.5.
1 exp(t 30)
r (t )

minimize the following integral:

In this example for providing the SDC parameterization,


Taylor series expansion of tanh(.) with three terms is used.
The gain of the controller via exact solution is attained:

1
xT Qx u T Ru dt.

20

k {2976.85 4.488 u 2.094 u )

This is the performance criterion or quadratic cost


function. The weighting matrix for states is
n

nn

R:

mm

Q:
n

and
.

for

input

controls

it

is

771.60 4.488 2.094 u u } /

Q must be symmetric, positive semi-

u R 1BT kx

[5.263 4.588u u ].
2

definite and R must be symmetric, positive definite. The


control law can be written as

Figure 1 shows tracking performance of r (t ) by the


system state. Control signal is demonstrated in Fig. 2. The
weighting matrices are tuned and selected as Q 100 and

R 2.2 , in a way that close results to [8] obtains. There is


a noticeable error at the beginning which makes a jump at
the commencement of the control signal to track the system.
This difference between the initial condition and the start of
kA AT k kBR1BT k Q 0.

desired trajectory increases the cost of control. In order to


reduce the magnitude of the signal, perfect trajectory is
preferable, the initial condition and start point of trajectory
The SDRE equation can be solved in different ways such
set the same value.
as ES and OCU [4]. The ES method is a good way to solve
Tracking performance
the SDRE equation, but it is applicable for low order and
2
simple systems. In this way, the SDRE provides nonlinear
equations in which some of them are dependent on each
1.5
other. By solving the independent equations and selecting the
positive answers, the gain of the controller is obtained;
1
however, once the dynamics of the system becomes
complicated it becomes necessary to solve the SDRE
0.5
equation with OCU or other numerical approaches. In the
OCU, the control law Eq. (4), and the SDRE in Eq. (5) can
0
be solved repeatedly as the system to update the gain k and
Exact Solution
Desired
input control u . The details of the OCU is expressed in [4].
-0.5
X1

where k is the positive solution of the SDRE as in the


following equation:

III.

ILLUSTRATIVE EXAMPLES

In this section, the following three examples are


considered to illustrate the proposed control method.
2539

10

20
t (sec)

30

Figure 1. State variable in tracking and desired trajecotry.

40

Control input

B. Multi Input Non-affine Nonlinear System


Let the non-affine dynamical system be given by:

30
20

x1 x2

10

2
x 2 x14 0.5u1 0.5u 2

x3 x34 0.1 u1 x12u 2

-10
-20

10

20
t (sec)

30

where x1 , x2 and x3 are state variables, u1 and u 2 are


control inputs [15]. Clearly, the states and inputs are coupled.
The SDC parameterization is given as in the form of:

40

Figure 2. Control input for the first example.

x1 0 1 0 x1 0 0
x2 x3 0 0 x2 0.5u 0.5 u1 .
1

u
1
3
2 2

x
x
3 0 0 x3 3 0.1 0.1x1

Figure 3 and 4 show the perfect tracking of r (t ) by the


system and control input with x(0) 1 , respectively. The
weighting matrices are tuned and selected as Q 100 and
R 1.
Tracking performance

Exact Solution

Simulating the nonlinear system, Figs. 5, 6, and 7 show


the state responses x1 , x2 and x3 to following conditions as

Desired

X1

1.5

x(0) 1.1 0.6 0.7


T

x(0) 1.3 0.6 0.4

x(0) 0.7 1.0 0.4

0.5

10

20
t (sec)

30

40

Figure 3. State variable in tracking and desired trajecotry in perfect


tracking.

1.5

Control input

IC1

IC2

IC3

0.5

X1

0
-2

0
-0.5
-1

-4

-1.5

-6

10

20
t (sec)

30

40

10

15
t (sec)

20

25

30

Figure 5. First state of second example (IC stands for initial condition).

Figure 4. Control input for the first example in perfect tracking.

In Fig. 4 the control signal due to perfect tracking is


obtained less than Fig. 2 with better performance in tracking,
Fig. 3.

2540

C. B(x, u) with External Disturbance


Consider a non-affine system

1
IC

IC

IC

X2

0.5

x1 x2
x 2 0.2(1 e x1x2 )(2 sin( x2 ))

(u e u 1) d d

10

15
t (sec)

20

25

where u is the control input, and d d is the external


disturbance and it is assumed to be a square wave with the
amplitude 0.1 and the period of 2 and this example is
from [10]. The SDC parameterization is given by:

30

Figure 6. Second state of second example.

1
IC

IC

x1 0 1 x1
x 0 0 x
2
2

0.2(1 e x1x2 )(2 sin( x )(2 u / 2)u


2

IC

X3

0.5

-0.5

10

15
t (sec)

20

25

30

Control signals for first initial conditions are given in Fig.


8. The weighting matrices are tuned and selected as
Q 10 I 33 and R I 22 .

matrices are tuned and selected as

R 0.1 .

Q 10 I 22

and

state

2
2

Exact Solution

10

1.5

X1

conditions are selected as x(0) 0 0 . The gain of the


controller via exact solution is attained. The weighting

15
u

In case 1 of example 3 the proposed control law is used


to control the first state of the system to track the reference
signal r 1.5 e 0.5t and in case 2 of example 3 the first
state of the system is to track the reference signal
r sin(0.5t ) cos(t ) . Figures 9 and 10 (case1) and
Figs. 11 and 12 (case 2) show the states of the closed-loop
system. Control signal is given in Figs. 13 and 14. The initial

Figure 7. Third state of second example.

Numerical Solution

Desired

0.5

-5

y x1

-0.5
-1

0
0

10

15
t (sec)

20

25

30

10
t (sec)

Figure 9. First state of third example, case 1.

Figure 8. Control input signals for first initial conditions.

2541

15

State
2

nonlinear system. The simulation results were included to


verify the usefulness of the proposed controller.

REFERENCES

X1

[1]

-1

[2]

-2
-3

Exact Solution

Numerical Solution

10

[3]

Desired

15

t (sec)
Figure 10. First state of third example, case 2.
Control input

[4]

15
[5]

10
5

[6]

0
-5

10

[7]

15

t (sec)
Figure 11. Control signal, case 1.
Control input

[8]

30

[9]

20
10

[10]

[11]
-10

10

15

t (sec)

[12]

Figure 12. Control signal, case 2.


[13]

IV.

CONCLUSION

In this paper, state-dependent Riccati equation (SDRE)


for control of non-affine nonlinear system is presented as a
closed loop nonlinear optimal control method. Non-affine
nonlinear system is assumed a system with non-affine in
control nonlinear system. The SDRE solved with two
methods, first, exact solution and, second, online control
update formulation. The three examples provided to illustrate
the effectiveness and potential of the SDRE technique for the
design of nonlinear feedback controllers for non-affine

[14]

[15]

2542

J. D. Pearson, Approximation methods in optimal control, Journal


of Electronics and Control, Vol. 13(5), pp. 453-465, 1962.
A. Wernli and G. Cook, Suboptimal control for the nonlinear
quadratic regulator problem, Automatica, Vol. 11, pp. 75-84, 1975.
J. R. Cloutier, C. N. D'Souza and C. P. Mracek, Nonlinear
Regulation and Nonlinear H Control Via the State-Dependent
Riccati Equation Technique: Part 1. Theory, Proceedings of the First
International Conference on Nonlinear Problems in Aviation and
Aerospace, May 1996.
S. C. Beeler, State-Dependent Riccati Equation Regulation of
Systems with State and Control Nonlinearities, Technical Report
NASA Langley Research Center, National Institute of Aerospace,
NASA/CR-2004-213245, 2004.
C.-H. Lee, J.-C. Chien, H.-H. Chang, C.-T. Kuo, and H.-H. Chang,
Direct Adaptive Backstepping Control for a Class of MIMO Nonaffine Systems Using Recurrent Neural Networks, Proceedings of
the International Multi Conference of Engineers and Computer
Scientists, Vol. 1, March, 2009.
S. M. Hoseini and M. Farrokhi, Observer-Based Stabilization of
Some Nonlinear Non-Minimum Phase Systems Using Neural
Network, International Journal of Modeling, Identification, and
Control, 2010.
W.-Y. Wang, Y.-H. Chien, Y.-G. Leu, and T.-T. Lee, Adaptive T_S
fuzzy-neural modeling and control for general MIMO unknown
nonaffine nonlinear systems using projection update laws,
Automatica, Vol. 46 (10), pp. 852863, 2010.
E. Lavretsky and N. Hovakimyan, Adaptive Dynamic Inversion for
Nonaffine-in-Control Systems via Time-Scale Separation Part II,
Proceedings of the American Control Conference, June 2005.
N. Hovakimyan, E. Lavretsky and C. Cao, Dynamic Inversion of
Multi-input Nonaffine Systems via Time-scale Separation,
Proceedings of the American Control Conference, June 2006.
W.-Y. Wang, I.-H. Li, L.-C. Chien, and S.-F. Su, On-line Modeling
and Control via T-S Fuzzy Models for Nonaffine Nonlinear Systems
Using A Second Type Adaptive Fuzzy Approach, International
Journal of Fuzzy Systems, Vol. 9 (4), September 2007.
J. Teo and J. P. How, Equivalence between Approximate Dynamic
Inversion and Proportional-Integral Control, Proceedings of the 47th
IEEE Conference on Decision and Control, pp. 2179-2183, 2008.
J. Xiang, Y. Li. W. Wei, Stabilization of a class of non-affine
systems via modelling error compensation, IET Control Theory
Appl., Vol. 2 (2), pp. 108116, 2008.
G. Bartolini, E. Punta, T. Zolezzi, Reduced-Order Observer for
Sliding Mode Control of Nonlinear Non-Affine Systems,
Proceedings of the 47th IEEE Conference on Decision and Control,
December 2008.
D. Gao, D. Lu and Z. Sun, Fuzzy Adaptive Control for a Class of
Non-Affine Systems via Time Scale Separation, Proceedings of the
World Congress on Engineering and Computer Science Vol II,
October 2012.
S.-H. Tsai, An Improved Fuzzy Modeling Method for a Class of
Multi-Input Non-affine Nonlinear Systems, J Optim Theory Appl
(Article in press), Doi 10.1007/s10957-012-0177-4, 2012.