Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
-XO\+HIHL&KLQD
7071
In order to apply the FFDL-MFAC scheme, equation (2) is where I (k ) [I1 (k ),, IL (k ), IL 1 (k ), , IL Lu (k )]T ,
y y y
discretized by first-order Euler’s method
'y(k 1) y(k 1) y(k ) , 'H (k ) H (k ) H (k 1) ,
,
2 2
T k u (k ) I (k ) d b , and both Ly and Lu are called pseudo orders.
y(k 1) 2 y(k ) y(k 1) u T 2g (3)
m y 2 (k )
According to Lemma 1, the magnetic levitation system (3)
where T denotes the sample time. can be transformed into the following FFDL data model,
where u(k ) R and y(k ) R are the control input and the 3.2 Control Scheme Design
system output at time instant k , respectively. n y and nu are
In this subsection, the model-free adaptive control scheme
two unknown integers. f () is an unknown nonlinear is designed based on the FFDL data model (7).
function.
Consider the following cost function of control input
L L
Denote H (k ) R y u as a vector consisting of all control
2 2
input signals within a input-related moving time window J (u(k )) yd (k 1) y(k 1) O u(k ) u(k 1) , (8)
[k Lu 1, k ] and all system output signals within a
output-related moving time window [k Ly 1, k ] , where O ! 0 is a weighting factor, yd (k 1) is the reference
signal.
H (k ) [ y(k ),, y(k Ly 1), u(k ),, u(k Lu 1)]T , (5)
Substituting (7) into (8), differentiating (8) w.r.t. u (k ) ,
and solving it, yield the control algorithm as follows,
where two integers Ly and Lu ( 0 d Ly d ny , 1 d Lu d nu )
are called pseudo orders of the system in this paper. U1IL 1 (k ) yd (k 1) y (k )
y
u (k ) u (k 1) 2
Assumption 1 The partial derivatives of f () with respect O IL 1 (k ) y
IL 1 (k )¦ UiIi (k ) 'y (k i 1)
y
Assumption 2 System (4) satisfies generalized Lipschitz i 1
2
(9)
condition, that is, O IL 1 (k ) y
Ly Lu
y(k 1) y(k ) d b H (k ) H (k 1) , (6)
IL 1 (k )
y ¦
i Ly 2
UiIi (k )'u (k Ly i 1)
holds for any H (k ) H (k 1) z 0 . b is a positive 2
,
O IL 1 (k )
constant. y
The following Lemma presents Full Form Dynamic where 0 Ui d 1, i 1, 2,, Ly Lu is the step factor, which is
Linearization method of system (4) in detail. added to make the control algorithm more flexible.
[12]
Lemma 1 For nonlinear system (4) satisfying Assumption Since PG I (k ) is unknown, we present the control
1 and Assumption 2 and any fixed 0 d Ly d ny , 1 d Lu d nu ,
algorithm as
L L
then there must exist a time-varying vector I (k ) R y u ,
called Pseudo Gradient (PG), such that the system (4) can be
transformed into following FFDL data model,
'y(k 1) I T (k )'H (k ),
7072
U1IˆL 1 (k ) yd (k 1) y (k )
y
U1IˆL 1 (k ) yd (k 1) y (k )
y
u (k ) u (k 1) 2
u (k ) u (k 1) 2
O Iˆ Ly 1 (k ) O IˆL 1 (k )
y
Ly Ly
Ly Lu Ly Lu
IˆL 1 (k )
y ¦ UiIˆi (k )'u (k Ly i 1) IˆL 1 (k )
y ¦ UiIˆi (k )'u (k Ly i 1)
i Ly 2 i Ly 2
2
. 2
.
O IˆL 1 (k )y
O IˆL 1 (k ) y
where Iˆi (k ) is to learn parameter Ii (k ) and updated in where H is a small positive constant, Iˆ(1) is the initial value
terms of the optimal solution of the following criterion index of Iˆ(k ) .
function
2
Remark 1 From (13)-(15), we can see that the FFDL-MFAC
J (I (k )) y (k ) y (k 1) I T (k )'H (k 1) scheme only utilizes the online measurement I/O data of the
2
(11) closed loop controlled system, and does not explicitly or
P I (k ) Iˆ(k 1) , implicitly include any information on the system dynamic
model.
where P ! 0 is a weighting factor.
Remark 2 FFDL-MFAC scheme has a L y Lu
Using the optimal condition and matrix inversion lemma, dimensional vector Iˆ(k ) to tune on-line, whilst the pseudo
the optimizer of (11) gives
orders Ly and Lu are also adjustable. For an unknown
Iˆ(k ) Iˆ(k 1) system, the pseudo orders Ly and Lu can be set to be the
K'H (k 1) y (k ) y (k 1) IˆT (k 1)'H (k 1) (12)
,
approximations of n y and nu , or just chosen as small as
2 possible in order to get a simple controller with low
P 'H (k 1)
computational burden. When Ly 0 and Lu L ,
where 0 K 2 is a step factor. FFDL-MFAC scheme (13)-(15) becomes PFDL-MFAC
scheme [13]; and when Ly 0 and Lu 1 , FFDL-MFAC
By integrating control algorithm (10) and PG estimation scheme (13)-(15) becomes CFDL-MFAC scheme [13].
algorithm (12), FFDL-MFAC scheme is constructed as Comparing with CFDL-MFAC scheme and PFDL-MFAC
follows, scheme, FFDL-MFAC scheme has more adjustable degrees
of freedom and more design flexibility by introducing more
Iˆ(k ) Iˆ(k 1) step factors as U1 , U2 , U Ly Lu .
K'H (k 1) y (k ) y (k 1) IˆT (k 1)'H (k 1) , 4 Simulations
2
P 'H (k 1) (13)
In this section, the comparison simulations are provided to
Iˆ(k ) Iˆ(1) if Iˆ(k ) d H or 'H (k ) d H or
illustrate the effectiveness of the FFDL-MFAC scheme. The
sign IˆLy 1 (k ) z sign IˆLy 1 (1) . parameters of the magnetic levitation system in this paper are
given in Table 1.
m 22 g x0 20.00 mm
i0 0.6105 A r 12.5 mm
2.3142e-004
k N 2450
Nm2/A2
R 13.8 : g 9.8 m/s2
7073
For comparison with MFAC controller, PID controller is
taken as the control group. Matlab S-function builder is used
to build the MFAC controller and PID controller with the
discrete sample mode and sample time being 0.003s.
Fig. 1: Simulation Diagram using PID and MFAC Fig. 2: The output response of step signal using PID controller and
MFAC controller
The control algorithm of PID is:
Table 2: RMS of the PID Controller and MFAC Controller
T k
T ½
u (k ) k p ®e(k ) ¦ e(i ) d [e(k ) e(k 1)]¾ (16)
Ti T Control algorithm RMS
¯ i 0 ¿
PID 6.555h
where e(k ) yd (k ) y(k ) , yd (k ) denotes the reference
signal, k p denotes the proportional coefficient, Ti denotes MFAC 3.303h
the integration time constant, Td denotes the derivative time Case 2: with output disturbance
constant, and T denotes the time step.
In this case, the disturbance signal
In order to compare the efficiency of MFAC and PID 0, k d 0.6 s
control, we take the root mean square (RMS) as an index °
d (k ) ®0.005, 0.6s k d 0.609 is added to magnetic
which is given by: °0, k ! 0.609
¯
N levitation system. The simulation results are shown in Fig. 3
eRMS ¦ e( k )
k 1
2
N (17) and Table 3. We can see that the FFDL-MFAC scheme
guarantees the BIBO stability.
After several trial and error runs, the best PID parameters
are set to be
method.
MFAC 3.917h
7074
Case 3: With output disturbance and measurement noise [3] M. Y. Chen, M. J. Wang, and L. C. Fu, Modeling and
controller design of a maglev guiding system for application
In this case, a uniformly distributed measurement noise on in precision positioning, IEEE Trans. on Industrial
interval (Ѹ 0.0005, 0.0005) is also added to the output y (k ) . Electronics, 50 (3): 493̢506, 2003.
The simulation results are shown in Fig. 4 and Table 4. We [4] J. Kaloust, C. Ham, J. Siehling, E. Jongekryg, and Q. Han,
Nonlinear robust control design for levitation and propulsion
can see that MFAC controller has a more strong robustness
of a maglev system, IEE Proc. Control Theory &
with a smaller RMS. Applications, 151(4): 460̢464, 2004.
[5] P. Holmer, Faster than a speeding bullet train, IEEE
Spectrum, 40(8): 30̢34, 2003.
[6] C. T. Lin, and C. P. Jo, GA-based fuzzy reinforcement
learning for control of a magnetic bearing system, IEEE
Trans. on Systems, Man, and Cybernetics, Part B, 30(2): 276
̢289, 2000.
[7] O. S. Kim, S. H. Lee, and D. C. Han, Positioning performance
and straightness error compensation of the magnetic
levitation stage supported by the linear magnetic bearing,
IEEE Trans. on Industrial Electronics, 50(2): 374̢378,
2003.
[8] D. M. Tang, H. P. Gavin, and E. H. Dowell, Study of airfoil
gust response alleviation using an electro-magnetic dry
friction damper, Part 2: experiment. Journal of Sound and
Vibration, 269(3): 875̢897, 2004.
[9] Z. S. Hou, and J. X. Xu, On data-driven control theory: the
Fig. 4: The output response of white noise signal using PID state of the art and perspective, Acta Automatica Sinica,
controller and MFAC controller
35(6): 650̢667, 2009 (In Chinese).
Table 4: RMS of the PID Controller and MFAC Controller [10] Z. S. Hou, On model-free adaptive control˖the state of the art
and perspective, Control Theory & Applications, 23(4):586
Control algorithm RMS ̢592, 2006 (In Chinese).
[11] Z. S. Hou, and W. H. Huang, Model-free learning adaptive
PID 7.880h control of a class of SISO nonlinear systems. Proc of
American Control Conference, Albuquerque, New Mexico,
MFAC 5.580h
343̢344, 1997.
[12] Z. S. Hou, Nonparametric Models and Its Adaptive Control
5 Conclusion Theory. Science press, Beijing, 1999 (in Chinese).
[13] Z. S. Hou, S. T. Jin, A Novel Data-Driven Control Approach
In this study, a FFDL-MFAC control approach is
for a Class of Discrete-Time Nonlinear Systems, IEEE Trans.
presented for magnetic levitation ball system. The main on Control Systems Technology, 19 (6): 1549–1558, 2011.
feature of the approach is that the controller design depends [14] Z. S. Hou, S. T. Jin, Data Driven Model-Free Adaptive
merely on the measured I/O data of a plant without explicitly Control for a Class of MIMO Nonlinear Discrete-Time
using of any information from mathematical model of the Systems, IEEE Trans. on Neural Networks, Special Issue on
controlled plant. Several simulation comparisons between Data-based Optimization, Control and Modeling,
PID and MFAC are made to demonstrate the control 22(12):2173-2188, 2011.
effectiveness of the proposed approach. [15] H. Peng, J. Gao, Multi-model Switching Control Method of
Magnetic Levitation System, Control Engineering of China,
References 18 (2): 206-209, 2011 (In Chinese).
[16] H. K. Liu, W. S. Chang, and X. H. Shi, Adaptive Control of
[1] H. Nakashima. The superconducting magnet for the Maglev Rank-reduced Magnetic Suspension System, Control
transport system. IEEE Trans. on Magnetics, 30 (4): 1572̢ Engineering of China, 13 (5): 410-412, 2006 (In Chinese).
1578, 1994.
[2] J. R. Powell, and G. T. Danby, Maglev vehicles-raising
transportation advances of the ground, IEEE Potentials, 15
(4): 7̢12, 1996.
7075