Sie sind auf Seite 1von 6

&VSPQFBO$POUSPM$POGFSFODF

&$$

"VHVTUo4FQUFNCFS ,BSMTSVIF (FSNBOZ

An Adaptive PassivityBased Controller for a Power Factor


Precompensator
yG. Escobar, D. Chevreau, yR. Ortega and E. Mendes

Lab. des Signaux et Systemes


y
Lab. de Genie Elec. de Paris
CNRS-SUPELEC URA CNRS 0127-SUPELEC
91192 Gif sur Yvette, France 91192 Gif-sur-Yvette, France
E-mail: escobarrortega@lss.supelec.fr E-mail: mendes@lgep.supelec.fr

Keywords: Passivity, nonlinear feedback, power elec- considerable pollution introduced in the line source. They
tronics, energy conversion. are commonly addressed as Power Factor Precompen-
sators PFP, since they are mainly used to guarantee
a near the unity power factor functioning, thus appearing
Abstract as a rst step in a power source.
In fact, the feedback control objective in PFP's is
We develop in this paper a passivitybased control scheme twofold. First, it is required for the feedback control policy
for the feedback regulation of a class of switched power to achieve a nearly unit power factor at the input mains of
converters commonly addressed as power factor precom- the converter and secondly, to achieve ecient load volt-
pensators. In such applications, aside from load voltage age regulation to a desired constant level. By dening a
regulation to a prespecied constant level, a vital addi- reference signal tracking problem on the input current of
tional control objective consists in keeping the input power the converter, the power factor can be made very close to
factor close to unity. By forcing the converter input cur- unity as long as the tracked signal is in phase with the
rent to follow a sinusoidal reference signal which is in phase input voltage.
with the supplied voltage, one may approximately satisfy The topology of the PFP circuit studied in the present
both control objectives within a single power stage. A fur- work consists in a complete bidirectional bridge of
ther complication in our design is to consider the output transistordiode arrays, which contrast with the classi-
load, usually modeled by a resistance, as unknown. cal topology usually composed of a diode bridge associ-
ated with a boost chopper. In 2
the advantages of a
1 Introduction full bridge over a diode based boost type PFP are shown,
basically the authors present a comparative study of the
zero crossing behavior of the input current signal in both
The regulation of switched power converters is quite an topologies, from this, they observe that while the former
active area of research, both in the power electronics area doesn't seem to have any problem, the diode based topol-
3, 4
and in automatic control theory 5, 7
. This is due ogy presents a slight deformation in the signal around the
to the fact that power converters are, generally speak- zero crossing.
ing, an ubiquitous power source whose applicability ranges In this article, we propose a nonlinear feedback con-
from electrodomestics and digital computers to industrial troller whose closed loop performance achieves both con-
electronics and sophisticated communications equipment. trol objectives. The controller design follows the ideas of
From the theoretical viewpoint, they also constitute an passivitybased control PBC with total energy shaping
interesting class of discontinuous nonlinear systems regu- ES advocated in 7
. In order to estimate the output
lated by means of a commanded switch position function. load resistance adaptation is added to the basic scheme.
These feature makes switched power converters attractive This feature allows us to avoid the use of an extra output
for both theoretical and practically oriented studies. current sensor, thus making its practical implementation
In this paper we are interested in the problem of per- more attractive.
formance enhancement of ACtoDC converters via non- The laboratory experiments showed that the losses in
linear control technics. ACtoDC converters are simple the input parasitic resistances strongly aect the perfor-
circuit that alleviates the main drawbacks found in DC mance of the system. Thus special care is taken to con-
toDC converters, namely lowinput-power factor and the sider this eect in our algorithm without further compli-
This work was partially supported by the Consejo Nacional de cations in the controller design. We observed that with a
Ciencia y Tecnologa of Mexico. simple modication in our original algorithm the experi-

*4#/ 
mental results were improved substantially. terms averaged function and DCcomponent are used in-
distinctly along the present text. We will use capitalized
letters to refer to these variables and in some cases the
2 Problem formulation averaging function hiDC will be explicitly employed. No-
tice that the system described by 1 and 2 can be seen
Consider the circuit of a full bridge boost type PFP shown as the interconnection of two subsystems having dierent
in Fig. 1. This circuit is composed in its main part by time constants. Actually 2 acts as an output lowpass
a complete two legs bidirectional bridge. The switches lter whose time constant is much slower than 1, thus
formed by a diode in parallel to a transistor are controlled the harmonics content of x2 may be neglected in practice.
in its gate by a switching signal which denotes the switch The control objectives are:
position function taking values in the nite set f,1 1g.
Moreover, the switches are working in a complementary 1. The DCcomponent of the output x2, i.e., X2 should
way as shown in the circuit, where is used to denote be driven to some constant desired value2 Vd  E ,
the complement of , i.e., when = 1 then = ,1, and i.e., X2 = Vd .
viceversa. 2. To enhance the eciency in these converters, it is de-
sirable to operate the circuit with a power factor close
L r to one. This is achieved ensuring that, in steady
δ δ state, the inductor current x1 follows a sinusoidal sig-
vi x1
+ + nal with the same frequency and phase as the acline
C R x2 voltage source, i.e., x1 = Id sinwt, where the value
- - of Id should be computed in order to accomplish the
δ δ regulation objective above.
There are two important obstacles to achieve the control
objectives: rst, it is known that the zero dynamics of the
Figure 1: Full bridge PFP Boost circuit system associated to the output voltage is unstable 7.
To overcome this problem we regulate the desired output
The model describing the averaged behavior of the cir- to its equilibrium point by means of indirectly stabilizing
cuit is given by the inductor current. Second, even though the system
parameters L and C are well known, the output load,
Lx1_ = ,ux2 , rx1 + vi 1 represented here as a pure resistive element R, is unknown.
1
C x_ 2 = ux1 , x2 2
R 3 A solvability condition
where x1 and x2 are the input inductor current and In this section we give a necessary and sucient condi-
the output capacitor voltage variables, respectively, vi = tion for the solvability of the problem. The condition
E sinwt is the voltage of the ACline source, R is the
nominal constant value of the output resistance, r is a re- gives an upper bound on the amplication gain of the
sistance that models the parasitic resistive eects of the PFP, which becomes arbitrarily large for small parasitic
inductor and the impedance associated to the source and resistance r, but may become restrictive in the case of
switches. The switch position function in the circuit is large r. Toward this end, we make a balance between the
substituted, in the averaged representation, by a contin- input active power and the output power in the steady
uous signal u. This signal u, living in a closed interval state. That is, we x the outputs at their desired val-
,1 1, acts as the control input to the system and rep- ues X2 = Vd  x1 = Id sinwt, Id is a constant yet to
resents the duty ratio of a PWM circuit which generates be dened and, neglecting high frequency harmonics, we
the sequence of switching positions . consider only the DCcomponent of these power quanti-
In the actual implementation of the circuit described ties.
above, the output voltage x2, the input inductor current The DCcomponent of the input power in steadystate
x1 and the source voltage vi are signals available for mea-
is given by
surement. To carry out our analysis, we will assume that
the inductor current x1 is accurately described by its rst hIP iDC = hx1vi , rx21iDC = 21 Id E , rId2 
harmonic approximation, i.e., it will be dominated by a while the DCcomponent of the output power is
purely sinusoidal signal. Besides, we will be interested in 2
the DC component or averaged function1 of the output hOP iDC = VRd
capacitor voltage x2 to which we will refer as X2 . The
2 Following the conventionadopted in 7 we will use   to denote
The averaged functionR of a periodic signal f t with period T is external
1 
references while  d stands for some internally generated
dened as f t DC 4
= T1 tt+T f  d

h i signals


Since the transmission of energy from input to output is Notice that this condition is independent to the con-
power conserving, that is hIPi = hOPi , thenDC DC troller design. As a matter of fact, this condition restricts
the physical design of the circuit and is valid for every
1 I E , rI 2  = V 2 3 d proposed controller.
2 d
R d

should hold. If r = 0 we get that


4 Proposed controller
I = 2V
2
In this section we present a controller designed with the
d
d
ER
Otherwise, the equation has two solutions, namely PBC approach referred in 7 as the PBC with total energy
E
r E 2 2V 2 shaping . Following the ideas there, the PBC is designed
for the whole system using as storage function the total
I = 2r  4r2 , rR
d energy of the full system, i.e., H = 2 x21 + 2 x22.
d
L C

To explain the rational of the design procedure we will


which are real if and only if consider rst the case when R is known. An adaptive
V  R
r scheme will later be added. As will become clear, even in
E
d

8r 4 the known parameters case a full analysis of the nonlinear


closed loop system is extremely dicult. Nevertheless,
holds. adding a practically reasonable assumption, the stability
In this case, both solutions are positive and both seem analysis can be completed.
to be physically meaningful. We select the smallest one,
i.e.,
E
r E 2 2V 2 4.1 Controller design. The known param-
I = 2r , 4r2 , rR
d 5 d
eter case
since its value converges towards the right value when r Following 7 we implement a copy of 1, 2 which is com-
goes to zero. Moreover, it has the advantage of less power monly addressed as auxiliary dynamics, and add damping
consumption. on the errors
Of course, the same condition 4 may be obtained by
studying the steady state of the dierential equations 1, Lx_ 1 = ,ux2 + v , rx1 + R1x1 , x1  10
d d i d

2. As before, we x the states at their desired values C x_ 2 = ux1 , R1 x2 + R,2 1 x2 , x2  11
x2  X2 = V  x1 = I sinwt and consider only the
d d
d d d d

DCcomponent of the equation 2. This yields the fol- where R  R,1 2 IR are design parameters, x  x
lowing set of equations 1 2 + 1
are auxiliary variables. The total energy for this system
2 d d

wLI coswt = ,rI sinwt + E sinwt , uV 6 is H = 2 x21 + 2 x22 .


d d d d
L C

The idea behind the construction of this auxiliary sys-


d d

huI sinwti = VR 7 tem is to reshape the total energy assigned to the error
d

dynamics established between the system and the auxil-


d DC

We multiply 6 by I sinwt and extract the DC iary dynamics, and add some damping via R1 and R,2 1 .
d

component The error dynamics is obtained by subtracting 10, 11


from 1 2 as follows
h, wLI 1 , coswt i =
2

2
d
sin2 wt , rI 2
, EI  2 Lx~_ 1 = ,ux~2 , R1x~1
d d DC

= V huI sinwti
d d 8
DC
C x~_ 2 = ux~1 , R1 x~2 , R,2 1 x~2 12
direct substitution of 7 in 8 yields
where x~1 = 4 x , x , x~ = 4x ,x .
1 I E , rI 2  = V 2 1 1 2 2 2 d d

The new assigned energy will take the form of an incre-


d

2 d
R d

which coincides with 3 above. mental energy H = 2 x~21 + 2 x~22. Taking its time derivative
r
L C

From this result we can state the following proposition. along the trajectories of 12 gives
Proposition 1 Consider the system 1, 2. A neces- H_ = ,R1 x~21 , R,1 + R,2 1~x22  ,H
r r

sary and su cient condition for the existence of a steady


minf 1 ,1 + ,1 g
state regime X2 t = V  x1t = I sinwt is that the where  = 2 maxf g 2 . Hence, x~ will converge
R R R
d d LC

ampli cation gain satis es the upper bound to zero exponentially fast. Remark that this means only
V  R
r that x converges exponentially to the auxiliary variables
E
d

8r 9 x . We now have to compute x so that x~ ! 0 ensures


d d

the control objective.


In view of the nonminimum phase obstacle mentioned where we have dened  = VRd44 + w2 L42 Id4 and  =
q
in Section 2, we will control the voltage x2 indirectly via wLI 2 R
regulation of the current x1. Towards this end, the PBC atan 2Vdd2

method proposes to x the output to be controlled, that Notice that the periodic perturbation has now zero
is, to evaluate 10
for x1d = Id sinwt
, Id computed as mean, and the averaged equation becomes
in 5
, and solve for u, from where the following dynamic
_Z = , 2Z~ + 2 pZ~ + z x~2 + R1 , r
x x~1 17

p
controller is nally obtained ~ 1
RC R2C C
u =
1 vi , rx1 , Lx_ 1 + R1x~1 13

x 2d where Z~ , is the average value of z~. Using the same argu-


^
C x_ 2d = ux1 ,  x2d +
1 x~2  x2d0
 0 14
ments as above, and noting that now the perturbation
R2 terms are all exponentially decaying, we conclude that
Here, and throughout the rest of the paper, we assumed Z~ ! 0, henceforth X2 ! Vd as desired.
that x2dt
 "  0. From the previous analysis we con- We summarize our ndings in the following proposition.
clude that x1 ! Id sinwt
. Let us now show that this
happens with internal stability, which amounts to proving Proposition 2 Consider the system described by eqs.
that x2d is bounded. To this end, let us substitute 13
1, 2 in closed loop with the dynamical controller 13,
into 14
, this yields the following expression 14, where the desired current is x1 = Id sinwt
, and Id
is computed by 5.
x2d C x_ 2d = x1vi , rx1 x1 , Lx1 x_ 1 + R1x1 x
x22d
~1 , Assume that x2d is bounded away from zero and Vd sat-
+ xR2d x~2
R is es the solvability condition 9. Then, the variables
2
2 X2 x1  X2d  converge asymptotically toward their desired
With the change of variable z = x22d , the equation above values Vd  x1  Vd .
becomes
p p We can compute in approximated way the steady state
C z_= x1 vi , rx12 +R1 , r
x1 x~1 , Lx1 x_ 1 , 2Rz + R2 z x~2 amplitude of the ripple oscillations produced on z by the
2
15
periodic perturbation  cos2wt,
. For this, assume that
Substituting the terms vi , x1, x_ 1 for their respective the trajectories of the system are already in the manifold
denitions in the expression above and using 3
, we ob- ~1 = x~2 = 0 thus the system is reduced to
x
tain p
C z_ = ,
2z + 2 pz x~ + R , r
x x~ +
2 1 1 1
_
p
z~ = , 2~
z p
RC
,  cos2wt , 
18

R R 2
2 2
+ VRd , VRd cos2wt
, wLI
2
where 
p is used to denote the system response due to
2 sin2wt

d
a periodic nonvanishing perturbation. Since the nominal
The system thus dened consists of an exponentially sta- system is actually a linear stable system then the ampli-
ble actually linear
system C z_ = , 2Rz
perturbed by tude of this signal in steady state is given by
three dierent types of disturbances: s
jjz~ jj = 41R 4V1d ++ ww2RL2 CId2R
4 2 2 4 4
 A linearly bounded term plus a constant multiplied p 19

by an exponentially decaying function. This comes


from thep second and third right hand terms, the
bound z  21 1 + jz j
, and the fact that x~1, x~2 are which inpterms of the original variable x~2pd , is given by
exponentially decaying quantities. jx~2pdj = 2jz~pj.
 A bounded disturbance stemming from the three re-
maining right hand terms. 4.2 Unknown parameter case
Applying Lemma 5.8 in 6 we conclude that z , and In this section we present the adaptive version of the con-
consequently x2d , ultimately enter a residual set, and are
troller previously designed using the PBC approach. In
therefore bounded. this version we will consider the output load resistance
We will now prove that the rst harmonic of x2 actually R as unknown but with upper and lower bounds known.
converges to Vd . Towards this end, consider the change of
For this, an adaptive scheme that keeps the estimate in-
variable z~ = z , V2d , then 15
may be written as
2
side this interval will be used. As we will see in what
pp follows a complete analysis of the nonlinear closed loop
2~
C z~_ = , +
R
z 2
R2
~2 +R1 , r
x1 x~1 ,  cos2wt , 
system
z~ + z x
is extremely dicult. However, under an addi-
tional, practically reasonable, assumption, we can carry
16
out some analysis.


4.2.1 A practical assumption Assume that lower and upper bounds for  are known,
Notice that for the implementation of the proposed con- i.e., c1    c2 . Thus, in order to guarantee that ^ will
troller the knowledge of R is needed, not just in 14, but not leave the interval c1 c2 we estimate this parameter
also for the calculation of Id in 5. Eventhough for the by means of the following adaptive law proposed in 1

rst equation we can get a linear parametrization in terms _


^ =  ^ , c2 ^ , c1 x2d x
~2 c1  ^0  c2 25
of R1 , in the latter expression Id depends nonlinearly on
R , which makes very di cult the development of an es-
1
timator. This complication is due to the existence of the where  2 IR+ is the adaptive gain.
parasitic resistance r. In e ect,2 by neglecting r in the To prove the stability of the closed loop system, consider
the partial Lyapunov function
model, Id simply reduces to 2RE Vd
and an estimator of R1
may be easily implemented.
Hr = x
L 2 C 2
~ 1 ^ , c2
Unfortunately, our experimental evidence has shown 1+ ~
x 2 + log ^
2 2   , c1 +1
that we cannot neglect r without degrading substantially
the performance of the system. We propose then to ex-
tract the voltage losses rx1 from the input source voltage where  =  cc21  0.
4 ,

Its time derivative along the trajectories of 24, 25 is


,

vi and use this as the actual voltage input source to the


system 1, 2 instead of the input voltage source, that given by
is, _
_ r = ,R1 x~21 ,  + R2 1~x22 + ~x2dx~2 , ~ ^
vi = ,rx1 + vi
0 4
20 H ,

 ^ , c2 ^ , c1


Moreover, to facilitate the analysis, we assume that the 26
new resulting voltage vi , as de
ned before, is also a sinu-
0
which is seminegative de
nite, then it is easy to see that
soidal signal with the same phase as the original vi but ~1 x~2 2 L2  L .
x 1

with a di erent amplitude, that is, Proceeding as for the known parameter case, we evalu-
ate 22, 23 in x1d = x1 and solve 22 for u, from where


vi = E sinwt
0 0
21 the following dynamic controller is obtained
This assumption is valid in practice since the dynamics u =
1 v , Lx_ + R x~ 
0 
27
of the subsystem 1 is much faster than the dynamics of x 2d i 1 1 1

the output
lter 2, therefore the current x1 almost imme- C x_ 2d = ux1 , ^x2d + R2 1x

~2 ,
28
diately gets in the phase with vi . In the implementation,
E is computed as E = E , rGpjx1j, where jj stands for
0 0
As before x1 = Id sinwt, where Id in this case is pro-


the absolute value, and Gp, p = dtd , is a lowpass


lter posed to be
4

su ciently slow with unitary gain. Thus, for the sake of Id =


2^Vd2 29
simplicity in the analysis, we assume E is known and con-
0
E 0

stant, therefore r disappear from the model and E will 0

^ and E have replaced 1=R and E , respectively.


replace E in all the instances. We remark that this simple where
0

modi
cation in our algorithm improved substantially our We have that Id 2 L which, together with x~1 2 L ,
1 1

experimental results. implies that x1 2 L . 1

As assumed before x2dt is bounded away from zero,


4.2.2 Controller design this, together with the fact that x~1  ^ 2 L implies that
~_ 1 2 L . From
1

u 2 L which in its turn implies that x


1 1

As before we implement a copy of 1, 2 and add damp- these implications we conclude that x~1 ! 0 as t ! 1.
ing on the errors x~1 and x~2 We know, by construction, that ^ is bounded away from
zero. Moreover, from the previous implications we know
Lx_ 1d = ,ux2d + vi + R1x1 , x1d 
0
22 that u 2 L . Observe that 28 is in fact a
rst order sys-
23 tem which is inputtostate stable because ^ is bounded
1

^
C x_ 2d = ux1d , x2d + R2 x2 , x2d 
1 ,

away from zero. Therefore, we can state that x2d 2 L ,


which in its turn implies that x~_ 2 2 L . With this in
1

where ^ denotes the estimate for  = 1=R. 1

Subtraction of 22, 23 from 1, 2 yields the follow- mind, we conclude that x~2 ! 0 as t ! 1.
ing error dynamics The time derivative of x1 , which is used in 27, yields


Lx ~_ 1 = ,ux~2 , R1x~1 x_ 1 =
 2Vd2 h ^ , c ^ , c x x~ sinwt + ^w coswti
2 1 2d 2
~_ 2 = ux~1 , x~2 + ~x2d , R2 1x~2 24 E 0
,
Cx
where we have used 25. It is clear that x_ 1 2 L because 

where ~ = ^ , .
1
4
it depends only on bounded signals.


Remark 1 And alternative way to estimate the parame- References
ter ^ is
^_ = P f, x2d x
~2g 30 1 R. Antonelli and A. Astol . Adaptive output feed-
back stabilization of a class of uncertain continuous
where 2 IR+ is the adaptive gain and Pfg is a projec- stirred tank reactors. In Proc. 14th. IFAC World
tion operator that ensures ^  "  0. Congress, Beijing, China, 1999.
In this case we consider the partial Lyapunov function 2 D. Chevreau and C. Marchand. Pollution har-
~22 + 21 ~2 whose time derivative is given by
Hr = L2 x~21 + C2 x monique du reseau: comparaison de deux re-
dresseurs monophases. In Proc. 9th International
H_ r = ,R1x~21 ,  + R,2 1~x22 + ~x2d x~2 + 1 ~~_ 31 Colloquium on CEM, pages F11F16, Brest, France,
8-11 June 1998.
which is seminegative denite and similar implications can 3 D. Czarkowski and M. K. Kazimierczuk. Energy-
be stated. conservation approach to modeling PWM DC-to-
Notice that the time derivative of x1 is computed now DC converters. IEEE Trans. on Aero. and Elect.
as Syst., 29:10591063, 1993.
= 2EVd0 , 4 M. E. Elbuluk, G. C. Verghese, and D. E. Cameron.
2
x_ 1 x2d x~2 sinwt + ^w coswt 32 Nonlinear control of switching power converters.
Control Theory and Advanced Technology, 54:601
4.3 A necessary condition for existence of 617, 1989.
the control 5 G. Escobar and H. Sira-Ramirez. A passivity based-
As stated in the problem formulation, the control signal sliding mode control approach for the regulation of
u de ned above should be restricted to live in the closed power factor precompensators. In Proc. IFAC NOL-
interval ,1 1. Thus in order to guarantee the existence COS, Enschede, The Netherlands, 1998.
of u, at least in the steady state regime, the following 6 H.K. Khalil. Nonlinear systems. Prentice Hall, 2nd
condition should hold edition, 1996. ISBN 0-13-228024-8.

,1  1 0
E sinwt + wL
2 Vd2
coswt  1 7 R. Ortega, A. Loria, P. J. Nicklasson, and H. Sira-
Vd RE 0 Ramirez. Passivity
based control of Euler
Lagrange
systems. Springer-Verlag, 1998.
where we taken that X2 = Vd  x1 = Id sinwt ^ = 1=R.
The above condition holds whenever
E 02 4Vd2w2 L2
V2
+ R2 E 02  1 33
d

solving for Vd2 yields


p
R2E 02 + RE 02 R2 , 16w2L2
Vd2  8wp 2 L2 34
0
R E , RE
2 2 02 R2 , 16w2L2
Vd2  8w 2 L2 35
Observe that both bounds are real and positive provided
R  4wL. Notice that 34 is implied by 9 while 35
imposes the following lower bound on Vd
p
R2 E 02 , RE 02 R2 , 16w2L2
Vd2  8w 2 L2 36
which is a necessary condition for the existence of u 2
,1 1 in the steady state regime.
2 2 2
In practice the term 4VRd2wE 2L in 33 is relative small.
0

Thus, roughly speaking, the desired voltage Vd is bounded


from below by the input voltage amplitude E 0 .



Das könnte Ihnen auch gefallen