1684 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 22, NO.

5, SEPTEMBER 2014

Power Optimization and Control in Wind Energy

Conversion Systems Using Extremum Seeking
Azad Ghaffari, Miroslav Krstic, Fellow, IEEE, and Sridhar Seshagiri

Abstract Power optimization and control for grid-coupled and a matrix converter (MC). The MC, which is a replacement
wind energy conversion systems (WECS) has been extensively for the conventional rectifierinverter combination (acdcac),
studied for variable speed wind turbines. However, existing features no energy storage components, has bidirectional
methods widely use model-based power optimization algorithms
in the outer loop along with linear control techniques in the power flow capability and controllable input power factor [37].
inner loop. The transient performance of this combination is It connects the IG to the power grid, and along with the
dependent on the systems operating point, especially under fast presented control/optimization design, steers the WT to its
varying wind regimes. We employ extremum seeking (ES) in the maximum power point (MPP) by controlling the electrical
outer loop, which is a nonmodel-based optimization approach, to frequency and voltage amplitude of the stator of IG, which
perform maximum power point tracking, i.e., extract maximum
power from WECS in their subrated power region. Since the in turn leads to a variation in the turbine speed. It also assists
convergence rate of the ES design may be limited by the speed of in voltage regulation or power factor correction by controlling
the system dynamics, we also design a nonlinear controller, based the reactive power transfer to the grid.
on the field-oriented control concept and feedback linearization, A design for a system similar to the one we consider here
that yields improvement in convergence rate by two orders has been presented in [5], and is based on a speed-sensoreless
of magnitude. The outer ES loop tunes the turbine speed to
maximize power capture for all wind speeds within the subrated power signal feedback (PSF) algorithm. The speed-sensoreless
power operating conditions. The inner-loop nonlinear control PSF algorithm uses lookup table values that are dependent on
maintains fast transient response through a matrix converter, the system model and parameter values. In addition, accuracy
by regulating the electrical frequency and voltage amplitude of of the method depends on the accuracy and resolution of the
the stator of the (squirrel-cage) induction generator. Simulation data obtained for the lookup table. Furthermore, the control
results are presented to show the effectiveness of the proposed
design. design employs Jacobian linearization, and uncertainty in the
system dynamics and/or variations in the working conditions
Index Terms Adaptive systems, nonlinear control systems, cause the system to move away from its MPP. Another
power control, wind power generation.
method based on fuzzy logic principles and four-leg-improved
MC model, used for performance enhancement and efficiency
I. I NTRODUCTION optimization, is presented in [20]. Model-dependent designs

A VARIABLE speed wind turbine (WT) generates power have the drawback that the optimization algorithm and con-
in two different regions, subrated power region and rated troller need to be redesigned carefully for each WECS. To
power region. In the subrated power region, the maximum overcome these difficulties, we present an extremum seeking
achievable turbine power is a function of the turbine speed (ES) algorithm, which is: 1) nonmodel based and 2) with
at any given wind velocity. To achieve maximum power easily tunable design parameters [2], [3], [19], [30][33].
point tracking (MPPT), i.e., extract maximum power, an Furthermore, ES shows promising results for a wide variety
optimization algorithm is needed [5], [8], [13], [15], [17], of applications [6], [10], [11], [16], [23], [34]. ES designs
[20], [26][28], and is often used in conjunction with a con- for MPPT of WECS are also presented in [8], [17], and
troller that guarantees other closed-loop desired performance [26], but differ from the design in this paper in several
specifications. In this paper, we focus on the optimization respects, including assumptions on the system model, transient
and control of a wind energy conversion system (WECS) performance, and performance robustness.
composed of a WT, a squirrel-cage induction generator (IG), With the exception of [5], none of the previous works on the
power extraction have focused on the transient performance,
Manuscript received August 2, 2013; revised November 11, 2013 and and the model in [5], unlike ours, is based on linearization and
January 15, 2014; accepted January 21, 2014. Manuscript received in final is highly model dependent. While the ES design we present
form January 25, 2014. Date of publication February 19, 2014; date of current
version July 24, 2014. Recommended by Associate Editor A. G. Aghdam. alleviates this problem, a requirement for the design is that
A. Ghaffari and M. Krstic are with the Department of Mechanical and its dynamics be slower than that of the WECS. While the
Aerospace Engineering, University of California at San Diego, La Jolla, WECS system is stable by itself, its linearization has slow
CA 92093-0411 USA (e-mail: [email protected]; [email protected]).
S. Seshagiri is with the Department of Electrical and Computer Engineering, poles, which therefore limits the convergence rate for the
San Diego State University, San Diego, CA 92182-1309 USA (e-mail: ES algorithm. To improve the transient response, we propose
[email protected]). an inner-loop IG control based on field-oriented control (FOC),
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. the elements of which can be found in [21] and [24]. For
Digital Object Identifier 10.1109/TCST.2014.2303112 the ES design, the turbine speed is considered as the variable
1063-6536 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
GHAFFARI et al.: POWER OPTIMIZATION AND CONTROL IN WECS 1685

Fig. 1. WECS including WT, gear box, IG, and MC.

Here, the generator operates at below rated power. The the-

oretical shape of this curve reflects the basic law of power
production, where power is proportional to the cube of the
wind speed. In Region III, the power output is limited by the
turbine; this occurs when the wind is sufficient for the turbine
to reach its rated output power. Region IV is the period of
stronger winds, where the power in the wind is so great that
it could be detrimental to the turbine, so the turbine shuts
down [14].
The wind power available on the blade impact area is
defined as
Fig. 2. Typical power curve of WT including four operating regions.
1
parameter to tune MPP. The turbine power is the cost function Pw = AVw3 , A = R 2 (1)
2
for the ES algorithm, and electrical frequency and voltage where R is the blade length and Vw is wind speed. For
amplitude of the stator of IG are controlled through the MC Region II MPPT, assuming zero blade pitch angle, the turbine
to reach desired closed-loop performance. As a result of power is related to the wind power as
including the inner loop, the overall design has faster response
time, and furthermore magnetic saturation of the IG is avoided. Pt = t Tt = C p (Vw , t )Pw (2)
In comparison with model-based designs, ES better handles
where Tt is the rotor torque, t is the turbine speed, and C p
model uncertainty in the turbine power map, resulting in
is the nondimensional power coefficient, which is a measure
improved power extraction. To the best of our knowledge, this
of the ratio of the turbine power to the wind power. The
is the first work in the literature that combines the MPPT with
power coefficient is a function of wind and turbine speed.
nonlinear control design that has good performance robustness
The theoretical limit for C p is 0.59 according to Betzs law,
to uncertainty, and faster transient performance, allowing for
but its practical range of variation is 0.20.4 [20]. The power
power tracking under rapidly varying wind conditions.
coefficient has been approximated numerically in several ref-
The rest of this paper is organized as follows. An intro-
erences, e.g., [1] and [29]. For simulation purposes, we select
duction to modeling of the WECS with concentration on the
one of the most common equations used for power coefficient
squirrel-cage IG dynamics in stationary reference frame and
as follows:
the MC is discussed in Section II. Our nonlinear controller
Vw
design is discussed in Section III, and the ES algorithm in 151 R 13.635
Section IV. Simulation results to verify the effectiveness of C p (Vw , t ) = 0.73 ! t ". (3)
Vw
exp R 0.003
the proposed scheme are presented in Section V, and our t

conclusion is presented in Section VI. Power coefficient (3) depends on the turbine speed, which
can be used for power control and optimization. The MPPT
II. W IND E NERGY C ONVERSION S YSTEM algorithm in subrated power region should be able to govern
A schematic diagram of a WECS including WT, IG, and the WT to its MPP regardless of the variations of the wind
MC is shown in Fig. 1. WTs work in four different regions, as speed. As shown in Fig. 3, the maximum value of the power
shown in Fig. 2. In Region I, the wind speed is too low for the coefficient happens at different turbine speeds when wind
turbine to generate power. Region II, also called the subrated speed is varying, but the maximum value stays at the same
power region, lies between the cut-in speed and rated speed. level of C p .
1686 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 22, NO. 5, SEPTEMBER 2014

where r is the angular electrical frequency of the rotor of

IG, t is WT angular position, r is the electrical angle of the
rotor of IG, p is the number of pole pairs of the IG, n is the
gearbox ratio, Tt is the turbine torque generated by the turbine
power, and TL is the load torque created by the spring-damper
model of the shaft
# $
r
TL = K s + B t (7)
pn
where K s is the stiffness coefficient of the spring and B is
the damping ratio. The generator rotor angular speed equals
r / p. The numerical values of the parameters are given in
Table I (see Appendix).
Squirrel-cage IGs are relatively inexpensive, robust, and
Fig. 3. Variation of turbine power coefficient versus turbine speed for require a little maintenance. When operated using vector
different wind speeds where = 0. The maximum value of the power
coefficient is C p . control techniques, fast dynamic response, and accurate torque
control is obtained [9]. From [18], the (, ) model equations
for the squirrel-cage IG are as follows:
d v
i = a0i + a1 + a2 r + (8)
dt Ls
d v
i = a0i a2 r + a1 + (9)
dt Ls
d
= a3 i a4 r (10)
dt
d
= a3 i a4 + r (11)
dt # $
d p TL
r = Te (12)
dt J n
where i and i are stator currents, and are rotor flux
Fig. 4. Variation of the turbine power versus turbine speed for different
linkages, v and v are stator voltages, and the electromagnetic
wind speeds where = 0. The MPP moves on C p Pw curve, which shows torque generated by the IG is
the characteristic of the subrated region of WECS.
3 Lm % &
Te = p i i (13)
2 Lr
From (1)(3), and considering the fact that the blade pitch where L m is the mutual inductance, L s = L ls + L m is stator
angle is zero, we have inductance, L r = L lr + L m is rotor inductance, and =
Vw
0.09 1 L 2m / (L r L s ). The numerical values of the parameters are
Pt (Vw , t ) = 55.115 A !Rt " Vw3 (4) defined in the Appendix.
Vw
exp Rt 0.003 Remark 1: Since MPPT in subrated power region is the
which shows that the power captured by the WT is defined main focus of this paper and, as shown in Fig. 2, the turbine
by the wind speed, Vw , and the turbine speed, t . However, power curve stays above zero, then (2) implies a nonzero
the wind speed is a disturbance input and we can manipulate turbine speed, t = 0, for Vcutin < Vw < Vrated , which
the turbine speed to govern the turbine power to its MPP in also leads to 2 + 2 = 0.
subrated region. The variation of turbine power versus turbine As shown in Fig. 1, the generator is connected to the
speed is shown in Fig. 4 for different wind speeds. From (4) ac grid through an MC, which includes nine bidirectional
and as shown in Fig. 4 under a constant wind speed, the switches operating in 27 different combinations. MCs provide
relevant power curve has a unique MPP, which is defined by bidirectional power flow, sinusoidal input/output currents,
a specific turbine speed. In addition, the MPP moves on a and controllable input power factor. Due to the absence of
third-order curve, which defines the maximum power captured components with significant wearout characteristics (such as
by the WT. electrolytic capacitors), MC can potentially be very robust
As shown in Fig. 1, the WT shaft is modeled as a spring and reliable. The amount of space saved by an MC, when
damper. The dynamic equations of the turbine, the shaft, and compared with a conventional back-to-back converter, has
the gearbox are been estimated as a factor of three. Therefore, due to its small
size, in some applications, the MC can be embedded in the
d r r machine.
= t , = t (5)
dt pn pn Furthermore, there is no intrinsic limitation to the power
d 1 Pt (Vw , t ) rating of an MC [9]. Therefore, we use MCs instead of
t = (Tt TL ) , Tt = (6)
dt Jt t conventional back-to-back converters. The model for MCs that
GHAFFARI et al.: POWER OPTIMIZATION AND CONTROL IN WECS 1687

we use is based on [4], [12], and [35]. The input phase voltage where

of MC, v i = [v A v B v C ]T, which is connected to the ac grid, Vom 3
is given by m= , 0m (24)
Vim 2
' # $ # $(
2 2 T is the MC output-to-input voltage gain and
v i = Vim cos i cos i cos i + (14) % & % &
3 3 2 2
%cos 1 2 & cos 1 3 cos %1 + 3 &
where Vim is the peak value of the input voltage amplitude T1 =cos %1 + 3 & 2
(25)
and %cos 1 2 & cos 1 3
) t cos 1 2 3 cos 1 + 3 cos 1
% & % &
i = i d (15) 2 2
0 %cos 2 2 & cos %2 2 3 & cos 2 + 3
T2 =cos %2 3 & cos 2 + 3 cos
% 2 &
(26)
is the input electrical angle, where i = 2 f i is the input 2 2
cos 2 + 3 cos 2 cos 2 3
electrical frequency of the MC. In this case, due to the
inductive nature of the IG, the output phase current can be where 1 = i o and 2 = i + o . The solution in (22)
assumed sinusoidal, and hence given by yields i = o , giving the same phase displacement at the
input and output ports, whereas the solution in (23) yields
% (o + o )2 &
cos
i = o , giving reversed phase displacement. Combining
i o = Iom cos %o + o 3 & (16) the two solutions provides the means for input displacement
2
cos o + o + 3 factor control [4], [35]
where Iom is the peak value of the output current amplitude,
S = cS1 + (1 c)S2 , 0 c 1 (27)
o is the load displacement angle at the output frequency
o = 2 f o , and where c controls the power factor at the grid side. Reactive
) t power control is not the focus of this paper, so we simply
o = o d (17) set c = 0.5 to obtain unity displacement power factor at the
0
MC input terminals. However, this does not result in a unity
is the output electrical angle. displacement power factor at input source terminals.
Output voltage is v o = [v a v b v c ]T. It is the job of the Considering zero losses in MC, the input and output powers
MC to create local-averaged sinusoidal output phase voltage are equal and related as follows:
(the stator voltage of IG) and input phase current (the ac grid
3
current) Pi Vim Iim cos i (28)
' # $ # $( 2
2 2 T 3
vo = Vom cos o cos o + cos o + (18) Po Vom Iom cos o . (29)
3 3 2

% (i + i ) &
cos From (24) and equating (28) and (29), we have
ii = Iim cos %i + i 2
3 &
(19) cos o
cos i + i + 2 Iim = m Iom . (30)
3 cos i
where i is the input displacement angle, which controls the We perform our design in the (, ) framework. We need to
power factor at the grid connection. Output electrical fre- transform the output voltage to the stationary frame and calcu-
quency, o , and voltage amplitude, Vom , are actuated through late the supplied current to the grid based on the stator current.
MC to achieve desired closed-loop performance. More details The local-averaged voltage of the stator is vo . Three-phase
about MC modeling can be found in [4], [35], and [12]. variables are transformed to two-phase stationary frame [25]
We are interested in finding modulation matrices such that ' ( . / ' (
v 1 0 0 Vom cos o
vo = Sv i (20) = 0 1 1 , vo = . (31)
v 3 3
Vom sin o
ii = S T i o . (21)
Some designs use v and v as inputs instead of o and Vom .
The solutions to the modulation problem should satisfy: We remind the reader that the output electrical angle and the
1) restrictions on the duty cycle of the MC switches that peak amplitude of the output voltage can be calculated from
prevent short circuit of the input sources and open circuit of the v and v as
inductive load; 2) sinusoidal output voltages with controllable 0
frequency and magnitude; 3) sinusoidal input currents; and Vom = v 2 + v 2 (32)
4) desired input displacement power factor [4]. There are two # $
v
basic solutions, which satisfy 1)3) o = arctan . (33)
v
111 To calculate the local averaged current supplied to the grid,
1 2
S1 = 1 1 1 + mT1 (22) we use (21), where
3 111 3

1 0 ' (
111 i
1 2 i o = 12 23 . (34)
S2 = 1 1 1 + mT2 (23) i
3 111 3 12 23
1688 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 22, NO. 5, SEPTEMBER 2014

As shown in Fig. 5, one can manipulate stator voltage ampli-

tude, Vom , and its frequency, o , through the MC to obtain
the desired closed-loop performance for WECS. Referring to
[21], [22], [24], and employing FOC idea, we introduce an
integrator and an auxiliary input, u 2 , to achieve inputoutput
decoupling in WECS dynamics. Using (35)(42) and one step
of integration in front of Vom , the extended equations of WECS
are introduced as follows:
x = f (x) + g1 u 1 + g2 u 2 , x R9 , u R2 (43)
where
Fig. 5. Block diagram of WECS.
a0 x 1 + a1 x 3 + a2 x 4 x 7 + x6cos Ls
x5
x sin x
a0 x 2 a2 x 3 x 7 + a1 x 4 + 6 L s 5
Denoting (5), (6), (8)(12), (17), and (31), we summarize
a3 x 1 a4 x 3 x 4 x 7
the state-space dynamics of WECS as follows:
a3 x 2 a4 x 4 + x 3 x 7
d cos o
0
i = a0 i + a1 + a2 r + Vom (35) f (x) = (44)
dt Ls 0 ! "
d sin o
a (x 2 x 3 x 1 x 4 ) a6 x 8 a7 x 9 x7
i = a0 i a2 r + a1 + Vom (36) 5 pn
dt Ls x7
d x 9 pn
= a3i a4 r (37) ! "
x7 Tt
dt a9 x 9 pn a8 x 8 Jt
d
= a3i a4 + r (38) g1 = [0 0 0 0 1 0 0 0 0]T (45)
dt
d g2 = [0 0 0 0 0 1 0 0 0] T
(46)
o = o (39)
dt # $ 3 4T
d 2
3 p Lm % & pK s pB r t , u 1 is the
where x = i , i , , , o , Vom , r , ,
r = i i t (40) electrical frequency of the stator, o , and u 2 is an auxiliary
dt 2L r J nJ nJ pn
d r input (voltage amplitude rate), which generates the voltage
= t (41) amplitude of the stator. The constant parameters are defined
dt pn # $
d Pt (Vw , t ) Ks B r in the Table II (see Appendix).
t = t (42) From (4) and Fig. 4, we know that the turbine speed
dt Jt t Jt Jt pn
controls the power generation. In addition, we are interested
where o and Vom are actuated by the MC. The wind speed in decoupling the rotor flux and electromagnetic torque to
Vw is an unknown disturbance that determines the MPP level obtain the benefits of FOC. For these reasons, we introduce
of the WT. A block diagram of WECS that highlights the turbine speed, y1 = t , and flux amplitude, 1 = ||2 , as
structure of the state-space model (35)(42) is shown in Fig. 5. measurable outputs. For future analysis, we assume that the
Assuming that the blade pitch angle is zero, we can employ power coefficient and wind speed function satisfy following
the turbine speed, t , for MPPT of the WT for wind speeds assumption.
between cut-in and rated wind speed. Assumption 1: The power coefficient C p (t , Vw ) and wind
speed function Vw (t) are bounded C 3 functions with bounded
III. C ONTROLLER D ESIGN derivatives. Hence, the mechanical torque, Tt , is a bounded C 3
In many motor drive systems, it is desirable to make the function with bounded derivatives.
drive act as a torque transducer wherein the electromagnetic Based on the selected outputs and having Assumption 1
torque can nearly instantaneously be made equal to a torque satisfied, we apply feedback linearization with the following
command. In such a system, speed or position control is change of variables to WECS dynamics:
dramatically simplified because the electrical dynamics of the
drive become irrelevant to the speed or position control prob- y1 = ,1 (x) = x 9 (47)
# $
lem. In the case of induction machine drives, such performance x7 Tt
y2 = L f ,1 (x) = a9 x9 a8 x 8 (48)
can be achieved using a class of algorithms collectively known pn Jt
as FOC [18]. 0 Tt Tt
y3 = L2f ,1 (x) = b0 q +b1 L f ,1 (x)+b2 x 8 +b3 (49)
When flux amplitude, 2 + 2 , is regulated to a constant Jt Jt
reference value, and considering the fact that the dynamics y4 = L3f ,1 (x)
of t are considerably slower than the electrical dynamics,
= b4 L2f ,1 (x) + b5 L f ,1 (x) + b6 x 8
we can assume that the dynamics are linear, but during flux ! "
transient, the system has nonlinear terms and it is coupled. b0
x 6 q x 7 b7 ,2 (x) + b8 L f ,2 (x)
This method can be improved by achieving exact inputoutput Ls
decoupling and linearization via a nonlinear state feedback that Tt Tt Tt
is not more complex than the conventional FOC [24]. +b9 + b10 (50)
Jt Jt Jt
GHAFFARI et al.: POWER OPTIMIZATION AND CONTROL IN WECS 1689

1 = ,2 (x) = x 32 + x 42 (51) 2 = 3 (75)

2 = L f ,2 (x) = 2a3 d 2a4 ,2 (x) (52) 2a3 q 2a3 d
3 = G 2 + x6u 1 + u2 (76)
3 = L2f ,2 (x) Ls Ls
y a T
= b12 ,2 (x) b11L f ,2 (x) . = 2 8. t (77)
# $ a9 a9 a9 Jt
2 Tt Tt a3
#
Tt Tt
$
+ b13 x 7 L f ,1 (x)b1L f ,1 (x)b2 x 8b3 + = r + y3 b 1 y2 b 2 . b 3 + (78)
Jt Jt b0 1 Jt Jt
2a 3
+ 2a32i s + x 6 d (53) where (77) and (78) are zero dynamics of the system and
Ls
. = x8 (54) b6 a8 b6
# $ G 1 = b 4 y4 + b 5 y3 y2 .
x4 a9 a9
= arctan (55) b0 x 6 % &
x3 a3i d a4 q + x 7 d
Ls
where d = x 1 x 3 + x 2 x 4 , q = x 2 x 3 x 1 x 4 , i s = i 2 + i 2 and f 7 (x) (b7 1 + b8 2 ) x 7 (b7 2 + b8 3 )
' ( ' (' ( ...
d cos x 5 sin x 5 x3 b6 Tt Tt Tt Tt
= . (56) + b9 + b10 (79)
q sin x 5 cos x 5 x4 a9 Jt Jt Jt Jt
G 2 = b11 3 + b12 2 + b13 f 7 (x)q
The inverse transformation of (47)(55) is # $
' ( ' (' ( Tt Tt
x1 1 cos sin ,d + b13 x 7 y4 b1 y3 b2 f 8 (x) b3 +
= (57) Jt J
x2 1 sin cos ,q # $ t
2x 6
x 3 = 1 cos (58) + 4a32 a1 d a0i s a2 x 7 q + id
Ls
x 4 = 1 sin (59) 2a3 x 6 % &
# $ + a3i d a4 d x 7 q (80)
Ls
x 5 = arctan (60)

0 where
' ( ' (' (
x 6 = + 2
2 (61) id cos x 5 sin x 5 x1
# $ = . (81)
y2 a8 Tt iq sin x 5 cos x 5 x2
x 7 = pn y1 + + .+ (62)
a9 a9 a9 Jt Defining control signals as follows:
x8 = . (63) v 1 G 1
x6u 1 cos( o ) sin( o ) b0
x 9 = y1 (64) Ls
=

u2 1 sin( ) cos( ) v 2 G 2
where o o 2a 3
2 + 2a4 1 (82)
,d = (65)
2a
# 3 $ and applying another step of change of variables
1 Tt Tt 3 4T
,q = y3 b 1 y2 b 2 . b 3 + (66) z = y1 tref , y2 , y3 , y4 (83)
b0 J Jt
' ( ' ( ' (t 3 % &2 4T
1 cos sin 1d = 1 ||ref , 2 , 3 (84)
= (67)
1 sin cos 1q
we obtain
with
z 1 = z 2 (85)
L s% &
1d = 3 +2a42 2a1a3 1 2a32i s +2a3 (a0 +a4 ),d z 2 = z 3 (86)
2a3
(68) z 3 = z 4 (87)
# 2 3
Ls a a z 4 = v 1 (88)
1q = y4 b1 y3 + 82 y2 + 82 .+b0(a0 +a4),q
b0 a9 a9 1 = 2 (89)
$
a82 Tt Tt Tt 2 = 3 (90)
+ b0 x 7 (,d +a2 1 )+ 2 b3 + . (69)
a9 Jt Jt Jt 3 = v 2 (91)
The change of variables results in the following equations: = z 2 a8 . Tt
. (92)
a9 a9 a9 Jt
y1 = y2 (70) # $
a3 Tt Tt
y2 = y3 (71) = r + z 3 b1 z 2 b2 .b3 + . (93)
b0 1 Jt Jt
y3 = y4 (72) Linear state feedback
b0 d b0 q
y4 = G 1 + x6 u 1 u2 (73) v 1 = k1 z 1 k2 z 2 k3 z 3 k4 z 4 (94)
Ls Ls
1 = 2 (74) v 2 = k1 1 k2 2 k3 3 (95)
1690 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 22, NO. 5, SEPTEMBER 2014

stabilizes this system, which results in the regulation of turbine

speed, t = y1 , to its reference value tref while amplitude of

rotor flux, || = 1 , has converged to its desired value ||ref .
Moreover, the convergence rate of
3 4T
3 = i 2 + i 2 , 2 + 2 , r , o , t , (96)
is exponential, and, in fact, it is independent of the selection
of the reference frame.
Remark 2: The closed-loop system (85)(93) is input
output decoupled and linear. The inputoutput map consists
of fourth-order and third-order systems. This allows for an
independent regulation (or tracking) of the outputs using
control signals (94) and (95). Transient responses are now
decoupled also when ||ref is varied, even independently of
tref . This is an improvement over FOC.
Remark 3: As in FOC, while measurements of (i , i , r )
are available, measurements of ( , ) require installing flux
sensing coils or Hall effect transducers in the stator, which
is not realistic in general purpose squirrel cage machines. Fig. 6. MPPT for a WECS based on P&O using conventional direct FOC.
The abc and its inverse follows from (31) and (34). The dq and
However, ( , ) flux components can be reconstructed by its inverse follows from (81). The flux calculator uses (51) and (55). The
means of an asymptotic observer of reduced order, as shown controllers are proportional-integral.
in [7].
Furthermore, the a-phase voltage equation can be expressed To overcome challenges attached with the conventional
as power control and optimization algorithms and to remove the
d dependence of the MPPT algorithm on the system modeling
v as = Rsia s + as . (97)
dt and identification, we propose ES algorithm, which is a
For steady-state conditions at intermediate to high speeds nonmodel-based real-time optimization technique to MPPT of
wherein the flux linkage term dominates the resistive term in WECS. First, we present ES without the inner-loop control
the voltage equation, the magnitude of the applied voltage is to clarify the advantages of the proposed controller on the
related to the magnitude of the stator flux linkage by closed-loop performance of the system.
The proposed models for power coefficient and turbine
Vs = o 4s (98) power in (3) and (4) are for simulation purposes. In this paper,
we assume that we have access to turbine power measurements
which suggests that to maintain constant flux linkage (to
and we can manipulate the turbine speed through the MC.
avoid saturation), the stator voltage magnitude should be
Furthermore, we do not have a model of the power coefficient
proportional to frequency [18]. Since we regulate the flux to a
or turbine power. However, we know that the turbine power
constant value, the voltage will be proportional to the electrical
map has one MPP under any wind speed, which helps us to
frequency. In the next section, we use our designed controller
present the following assumption.
combined with an ES algorithm to extract maximum power
Assumption 2: The following holds for the turbine power
from WECS under various wind speeds.
map around its MPP for Vcutin<Vw<Vrated (see Fig. 4)
Pt (Vw , t )
IV. MPPT U SING ES (t ) = 0 (99)
t
There are three main MPPT techniques for WECS: wind 2 Pt (Vw , t )
speed measurement (WSM), P&O, and PSF. Measurement of (t ) < 0 (100)
wind velocity is required in WSM method. It is clear that t2
accurate measurement of wind velocity is complicated and where t is the optimal turbine speed.
increases the system cost. Since the P&O method adds delay, Following statement explains actual relation between tur-
it is not practical for medium- and large-inertia WT systems. bine speed and stator electrical frequency.
To implement PSF control, maximum power curve (maximum Remark 4: The torquespeed characteristic of an induction
power versus turbine speed) is required. The maximum power machine is normally quite steep in the neighborhood of stator
is then tracked by turbine speed control [5]. electrical frequency (synchronous speed), o , and so the elec-
Fig. 6 shows a typical block diagram of P&O using direct trical rotor speed, r , will be near the synchronous speed. This
FOC for the IG [9], [36]. To implement FOC scheme, the rotor means that changing the reference value of the turbine speed,
flux magnitude || and its angle are identified by the rotor t , which translates in variation of the electrical rotor speed
flux calculator based on the measured stator voltage (v o ) and eventually results in changing the stator electrical frequency
current (i o ). The turbine speed reference eref is generated by [18]. Thus, by controlling the stator electrical frequency, one
the MPPT scheme. can approximately control the turbine speed or vice versa.
GHAFFARI et al.: POWER OPTIMIZATION AND CONTROL IN WECS 1691

Theorem 1: Consider (105) with constant Vw and Vom

under Assumption 2 and recall that for any fixed electrical
frequency, U = o , (105) is exponentially stable at the nega-
tive slope of the IG torquespeed characteristic, as shown in
[18, Sec. 8.5]. For the system in Fig. 7, there exists a
ball of initial conditions around the point (X, o , g, p)
=
and a such that for
(l(o ), o , 0, Pt ) and constants , ,
all (0, ), (0, ), and a (0, a), the solu-

tion [X (t), o (t), g(t),
p(t)] exponentially converges to an
O( + + a)-neighborhood of that point. Furthermore, Pt (t)
converges to an O( + + a)-neighborhood of Pt .
The turbine power measurement is fed into ES scheme. The
Fig. 7. MPPT for a WECS based on ES without the inner-loop control. optimization parameter for ES without the inner-loop control,
Fig. 7, is the electrical frequency of IG stator, o . Stability of
A schematic diagram of MPPT for WECS with ES without system dynamics is required for convergence of ES algorithm
inner-loop nonlinear control is shown in Fig. 7. Remark 4 to its peak point. It is also required that the ES algorithm works
implies that the power is parameterized by o , which is esti- more slowly than the WECS system dynamics. As previously
mated by ES loop. The other input for WECS that generates mentioned, since WECS in Fig. 7 without the inner-loop
the voltage amplitude has been set to zero, which means the controller shows a slow transient, the entire system has a
stator voltage has a constant peak amplitude. The parameters lengthy convergence process, which results in low power
of the ES loop are defined as follows: efficiency.
6 = 6 (101) We propose to employ the nonlinear control from Section III
to achieve the desired closed-loop performance, including
6 H = 6H = 6H = O() (102) faster response time (high power efficiency), and preventing
6 L = 6L = 6L = O() (103) magnetic saturation. Our proposed ES scheme with the inner-
loop control is shown in Fig. 8. In this case, the reference
k = k = k = O() (104)
inputs of the inner-loop control are tref and ||ref . From
where 6 is a rational number, and are small positive Assumption 2, we know that the MPP is parameterized by the
constants, and 6H , 6L , and k are O(1) positive constants. optimal turbine speed at each wind speed, which is estimated
In addition, a needs to be small. by the ES loop. The other control input ||ref defines the
Stability analysis of the ES without the inner-loop control level of the flux linkage of the rotor, which prevents IG from
follows the same steps as [19]. To start with the proof, we magnetic saturation.
assume that the turbine shaft is rigid. Hence, the WECS Remark 5: From (2) and (4), we observe that the torque on
dynamics is simplified as follows: the WT shaft depends on the wind speed and turbine speed. In
X = F(X, U, Vom , Vw ) (105) addition, to implement the controller in Section III, we need
pnVw
0.09 to have access to the first-, second-, and third-order derivatives
Pt (Vw , X 5 ) = 55.115 A !R X 5 " Vw3 (106) of the torque.
pnVw
exp R X 5 0.003 1) By Assumption 1 and because of fast response dynamics
3 4T of the inner loop and the WT, it is reasonable to assume
where X = i d , i q , d , q , r , U = o , and that (for the inner-loop design shown in Fig. 8) variation
of the wind speed is negligible in comparison with the
a0 X 1 + U X 2 + a1 X 3 + a2 X 5 X 4 + Vom
U X a X a X X + a X
Ls
dynamics of controller system.
1 0 2 2 5 3 1 4 2) Since the convergence time of the estimate of the turbine
F= a X
3 1 a X
4 3 + (U X 5 )X 4
.
(107)
speed generated by ES loop is considerably slower
a3 X 2 (U X 5 )X 3 a4 X 4 than the response time of the controller system, we
p
a5 (X 2 X 3 X 1 X 4 ) + nJ Tt (X 5 , Vw ) can assume that turbine speed reference is constant in
Note that, in this case, t = r /(pn). The IG dynamics are comparison with the fast dynamics of the controller
presented in synchronous frame, namely the frame, which system.
rotates in speed of the stator electrical frequency, U . The equi- Using this observation and from (2) and (4), we can assume
librium, F(X, U, Vom , Vw ) = 0, is parameterized by U, Vom , that the variation of mechanical torque, computed as Pt /t ,
and Vw . We assume that the stator voltage amplitude, Vom , and its derivatives are negligible in comparison with the
and the wind speed are fixed at a constant level, which make dynamics of the controller system. In addition, the inner loop
the equilibrium parameterized only by U , i.e., X e = l(U ). becomes independent of the turbine power map, but it still
Moreover, the Jacobian of IG has left-half plane poles at each relies on the IG dynamics and measurements of the turbine
equilibrium point on the negative slope of the torquespeed power, the turbine speed, t , and angle displacement caused
characteristic, as shown in [18, Sec. 8.5], which makes the by the shaft model, . The ES algorithm generates the estimate
system exponentially stable. Our analytical results for this case of the turbine speed, which is the reference input for the inner
(no inner-loop control) are summarized in the theorem below. loop and maximizes the power generated by the WT, and
1692 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 22, NO. 5, SEPTEMBER 2014

Fig. 8. ES for MPPT in WECS with the inner-loop control.

governs the system to its MPP, despite high uncertainty in

the turbine power map.
Combination of the controller and WECS includes fast
dynamics and ES algorithm contains slow- and medium-speed
dynamics. The ES algorithm estimates the optimal turbine
speed, tref = t , which can be considered as a constant value
with respect to the fast dynamics of the controller system.
The ES scheme estimates the gradient of the cost function,
Pt , by injecting a small perturbation, a sin(6t), which is
very slow with respect to the dynamics of the controller
system and its amplitude is enough small in comparison with Fig. 9. Variation of wind speed versus time.
t . The high-pass filter removes the dc part of the signal.
The multiplication of the resulting signal by sin(6t) creates in a very low power efficiency. However, we present one
an estimate of the gradient of the cost function, which is simulation that compares the response of the design without
smoothed using a low-pass filter. When t is larger than its the inner loop, as shown in Fig. 7, to our proposed algorithm,
optimal value, the estimate of the gradient, g, is negative and as shown in Fig. 8, which shows the role of the inner loop.
causes t to decrease. On the other hand, when t is smaller In addition, We compare the performance of our proposed
than t , then g > 0, which increases the t toward t . algorithm to the conventional algorithm shown in Fig. 6.
It should be noted that 6 is small enough in comparison with By appropriate selection of the feedback gains in (94) and
the slowest dynamic of the controller system, with an order (95) and using (82), we can obtain the desired closed-loop
less than 10%. response time. Control signals are designed such that the
The analytical results for the closed-loop system with the poles of z-error subsystem (85)(88) and -error subsystems
ES design are summarized in the theorem below, the proof of (89)(91) move to Pz = [550 600 650 700] and
which also follows from [19]. We remind the reader that in P = [570 620 670], respectively. The response time of
the singular perturbation analysis of [19] the dynamics of the closed-loop system is about 20 ms, which is 25 times faster
can be neglected without hurting the proof. than the open-loop system. We select the parameters of the ES
Theorem 2: Consider the feedback system in Fig. 8, which loop as follows: 6 = 100 rad/s, 6 L = 6 rad/s, 6 H = 5 rad/s,
includes the plant (43) under Assumption 1 and 2 with control a = 0.1, and k = 0.004. The amplitude of the perturbation
input (82), where v 1 and v 2 are defined as (94) and (95). Recall function is selected proportional to the turbine speed. Higher
that Remark 5 is in place. There exists a ball of initial condi- values of a reduce the precision of the MPPT, as shown
= ((t ), t , 0, Pt ), and
p)
tions around the point (3, t , g, in Theorem 2.
and a such that for all (0, ),
constants , , (0, ), We show a time frame of 30 s to visualize the differences
and a (0, a),
the solution (3(t), t (t), g(t),
p(t)) exponen- between our proposed algorithm and the two other algorithms,
tially converges to an O( + +a)-neighborhood of that point. properly. Fig. 9 shows the wind regime applied to the WECS.
Furthermore, Pt (t) converges to an O( + +a)-neighborhood The MPPT process is shown in Fig. 10. It is clear from Fig. 11
of Pt . In addition, the magnetic saturation is avoided by that our proposed algorithm stays very close to C p despite fast
selecting a constant reference for the flux amplitude, ||ref . changes in wind speed. The extracted energy by our proposed
algorithm is 2.36% higher than the extracted energy by the
conventional MPPT and FOC. As we expected, the power
V. S IMULATION R ESULT efficiency of the ES design without the inner loop is low and,
As we mentioned earlier, response time of the ES design as shown in Fig. 11, the power coefficient almost always stays
without the inner loop is considerably slow, which results far from C p . We have verified the robustness of our proposed
GHAFFARI et al.: POWER OPTIMIZATION AND CONTROL IN WECS 1693

Fig. 12. Robustness analysis with a 100% increment in the rotor resistor
at time 15 s and back to its nominal value at time 25 s for the proposed
algorithm. Variation of turbine power (solid red line) with perturbation and
(dashed blue line) without perturbation.

Fig. 10. MPPT, (solid red line) our proposed algorithm, (dashed-dotted green
line) ES without inner loop, (dashed blue line) conventional P&O with FOC,
and (dotted black line) maximum power available to the WECS.

Fig. 13. Robustness analysis with a 100% increment in the rotor resistor at
time 15 s and back to its nominal value at time 25 s for conventional P&O
with FOC. Variation of turbine power (solid red line) with perturbation and
(dashed blue line) without perturbation.

complexity of the proposed algorithm. Clearly higher power

efficiency is our aim and to this end, we have to sacrifice
Fig. 11. Variation of power coefficient, (solid red line) our proposed
algorithm, (dashed-dotted green line) ES without inner loop, (dashed blue the simplicity in favor of harvesting more energy. Since the
line) conventional P&O with FOC, and (dotted black line) maximum power WECS runs for a long period of time, a small improvement in
coefficient. power efficiency guarantees extracting a higher energy level
and leads to cost reduction of the WECS.
algorithm by adding different amount of perturbation to the
rotor and stator resistance and inductance. We present one of VI. C ONCLUSION
our robustness simulations with a 100% increment in rotor We presented an ES algorithm to extract maximum power
resistance at time 15 s and then back to its nominal value from a WECS for wind speed from cut-in wind speed to
at time 25 s. While as shown in Fig. 12, the performance of rated wind speed. The design employed an inner-loop nonlin-
the proposed algorithms remains unchanged, the conventional ear controller based on field-oriented approach and feedback
MPPT algorithm is not able to attenuate the effect of the linearization technique to control the closed-loop transient
disturbance, as shown in Fig. 13. performance, with respect to which the ES had to be tuned.
The proposed algorithm combines two well-known control Without this inner-loop control, the convergence rate of the
algorithms, namely feedback linearization based on the FOC closed-loop system would be much slower. This optimiza-
concept and ES, to achieve MPPT in a WECS operating tion/control algorithm can readily be extended to other classes
in Region II. Our algorithm provides perfect inputoutput of WECS without major changes. The main parameters that
decoupling and guarantees a larger domain of attraction, which need to be adjusted are the probing frequency and amplitude
increases performance robustness with respect to the system of the perturbation signal. Furthermore, the proposed control
parameters. However, one may question the implementation strategy prevents magnetic saturation in the IG.
1694 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 22, NO. 5, SEPTEMBER 2014

A PPENDIX [9] R. Cardenas, R. Pena, J. Clare, and P. Wheeler, Analytical and experi-
mental evaluation of a WECS based on a cage induction generator fed
TABLE I by a matrix converter, IEEE Trans. Energy Convers., vol. 26, no. 1,
D EFINITIONS OF PARAMETERS AND T HEIR N UMERICAL VALUES pp. 204215, Mar. 2011.
[10] D. Carnevale, A. Astolfi, C. Centioli, S. Podda, V. Vitale, and
L. Zaccarian, A new extremum seeking technique and its applica-
tion to maximize RF heating on FTU, Fusion Eng. Des., vol. 84,
pp. 554558, Jun. 2009.
[11] M. Guay, M. Perrier, and D. Dochain, Adaptive extremum seeking
control of nonisothermal continuous stirred reactors, Chem. Eng. Sci.,
vol. 60, no. 13, pp. 36713681, 2005.
[12] L. Huber and D. Borojevic, Space vector modulated three-phase to
three-phase matrix converter with input power factor correction, IEEE
Trans. Ind. Appl., vol. 31, no. 6, pp. 12341246, Nov./Dec. 1995.
[13] K. E. Johnson, L. Y. Pao, M. J. Balas, and L. J. Fingersh, Control
of variable-speed wind turbines: Standard and adaptive techniques for
maximizing energy capture, IEEE Control Syst. Mag., vol. 26, no. 3,
pp. 7081, Jun. 2006.
[14] S. J. Johnson and C. P. C. van Dam, Active load control techniques
for wind turbines, Sandia Nat. Lab., Albuquerque, NM, USA, Tech.
Rep. SAND2008-4809, 2008.
[15] S. M. R. Kazmi, H. Goto, H.-J. Guo, and O. Ichinokura, A novel
algorithm for fast and efficient speed-sensorless maximum power point
tracking in wind energy conversion systems, IEEE Trans. Ind. Electron.,
vol. 58, no. 1, pp. 2936, Jan. 2011.
[16] N. Killingsworth and M. Krstic, Auto-tuning of PID controllers via
extremum seeking, in Proc. Amer. Control Conf., vol. 4. Jun. 2005,
pp. 22512256.
[17] M. Komatsu, H. Miyamoto, H. Ohmori, and A. Sano, Output maxi-
mization control of wind turbine based on extremum control strategy,
in Proc. Amer. Control Conf., 2001, pp. 17391740.
TABLE II [18] P. C. Krause, O. Wasynczuk, and S. D. Sudhoff, Analysis of Electric
C ONSTANT PARAMETERS Machinery and Drive Systems. New York, NY, USA: Wiley, 2002.
[19] M. Krstic and H.-H. Wang, Stability of extremum seeking feedback
for general nonlinear dynamic systems, Automatica, vol. 36, no. 4,
pp. 595601, 2000.
[20] V. Kumar, R. R. Joshi, and R. C. Bansal, Optimal control of matrix-
converter-based WECS for performance enhancement and efficiency
optimization, IEEE Trans. Energy Convers., vol. 24, no. 1, pp. 264273,
Mar. 2009.
[21] A. D. Luca and G. Ulivi, Dynamic decoupling of voltage frequency
controlled induction motors, in Proc. 8th Int. Conf Anal. Optim. Syst.,
1988, pp. 127137.
[22] A. D. Luca and G. Ulivi, Design of exact nonlinear controller for
induction motors, IEEE Trans. Autom. Control, vol. 34, no. 12,
pp. 13041307, Dec. 1989.
[23] L. Luo and E. Schuster, Mixing enhancement in 2D magnetohydrody-
namic channel flow by extremum seeking boundary control, in Proc.
Amer. Control Conf., 2009, pp. 15301535.
[24] R. Marino, S. Peresada, and P. Valigi, Adaptive input-otput linearizing
R EFERENCES control of induction motors, IEEE Trans. Autom. Control, vol. 38, no. 2,
[1] P. M. Anderson and A. Bose, Stability simulation of wind tur- pp. 208221, Feb. 1993.
bine systems, IEEE Trans. Power App. Syst., vol. 102, no. 12, [25] B. Ozpineci and L. M. Tolbert, Simulink implementation of induction
pp. 37913795, Dec. 1983. machine modelA modular approach, in Proc. IEEE Int. Conf. Electr.
[2] K. B. Ariyur and M. Krstic, Real-Time Optimization by Extremum Mach. Drives, Feb. 2003, pp. 728734.
Seeking Feedback. New York, NY, USA: Wiley, 2003. [26] T. Pan, Z. Ji, and Z. Jiang, Maximum power point tracking of wind
[3] A. Banaszuk, K. B. Ariyur, M. Krstic, and C. A. Jacobson, An adaptive energy conversion systems based on sliding mode extremum seeking
algorithm for control of combustion instability, Automatica, vol. 40, control, in Proc. IEEE Energy Conf., Nov. 2008, pp. 15.
no. 11, pp. 19651972, 2004. [27] L. Y. Pao and K. E. Johnson, Control of wind turbines, IEEE Control
[4] S. M. Barakati, Modeling and controller design of a wind energy Syst. Mag., vol. 31, no. 1, pp. 4462, Feb. 2011.
conversion system including a matrix converter, Ph.D. dissertation, [28] T. Senjyu, R. Sakamoto, N. Urasaki, T. Funabashi, H. Fujita, and
Dept. Electr. Comput. Eng., University of Waterloo, Waterloo, ON, H. Sekine, Output power leveling of wind turbine generator for all
Canada, 2008. operating regions by pitch angle control, IEEE Trans. Energy Convers.,
[5] S. M. Barakati, M. Kazerani, and J. D. Aplevich, Maximum power vol. 21, no. 2, pp. 467475, Jun. 2006.
tracking control for a wind turbine system including a matrix con- [29] J. G. Slootweg, S. W. H. de Haan, H. Polinder, and W. L. Kling,
verter, IEEE Trans. Energy Convers., vol. 24, no. 3, pp. 705713, General model for representing variable speed wind turbines in power
Sep. 2009. system dynamics simulations, IEEE Trans. Power Syst., vol. 18, no. 1,
[6] R. Becker, R. King, R. Petz, and W. Nitsche, Adaptive closed-loop pp. 144151, Feb. 2003.
separation control on a high-lift configuration using extremum seeking, [30] M. S. Stankovic, K. H. Johansson, and D. M. Stipanovic, Distributed
AIAA J., vol. 45, no. 6, pp. 13821392, 2007. seeking of nash equilibria with applications to mobile sensor networks,
[7] A. Bellini, G. Figalli, and G. Ulivi, Analysis and design of a IEEE Trans. Autom. Control, vol. 57, no. 4, pp. 904919, Apr. 2012.
microcomputer-based observer for an induction machine, Automatica, [31] M. S. Stankovic and D. M. Stipanovic, Extremum seeking under
vol. 24, no. 4, pp. 549555, 1988. stochastic noise and applications to mobile sensors, Automatica, vol. 46,
[8] A. I. Bratcu, E. C. Iulian Munteanu, and S. Epure, Energetic optimiza- no. 8, pp. 12431251, Aug. 2010.
tion of variable speed wind energy conversion systems by extremum [32] Y. Tan, D. Neic, and I. Mareels, On non-local stability properties of
seeking control, in Proc. Int. Conf. Comput. Tool, 2007, pp. 25362541. extremum seeking control, Automatica, vol. 42, pp. 889903, Jun. 2006.
GHAFFARI et al.: POWER OPTIMIZATION AND CONTROL IN WECS 1695

[33] A. R. Teel and D. Popovic, Solving smooth and nonsmoooth mul- Miroslav Krstic (F01) received the Dipl.Ing.
tivariable extremum seeking problems by the methods of nonlinear degree in electrical engineering from the University
programming, in Proc. Amer. Control Conf., 2001, pp. 23942399. of Belgrade, Belgrade, Serbia, and the M.S. and
[34] H.-H. Wang, S. Yeung, and M. Krstic, Experimental application of Ph.D. degrees in electrical engineering from the
extremum seeking on an axial-flow compressor, IEEE Trans. Control University of California, Santa Barbara, CA, USA,
Syst. Technol., vol. 8, no. 2, pp. 300309, Mar. 2000. in 1989, 1992 and 1994, respectively.
[35] P. W. Wheeler, J. Rodriguez, J. C. Clare, L. Empringham, and He holds the Daniel L. Alspach Endowed Chair
A. Weinstein, Matrix converters: A technology review, IEEE Trans. and is the Founding Director of the Cymer Center
Ind. Electron., vol. 49, no. 2, pp. 276288, Apr. 2002. for Control Systems and Dynamics, University of
[36] B. Wu, Y. Lang, N. Zargari, and S. Kouro, Power Conversion and California San Diego (UCSD), San Diego, CA,
Control of Wind Energy Systems. New York, NY, USA: Wiley, 2011. USA. He serves as an Associate Vice Chancellor for
[37] S. Zhang, K. J. Tseng, and T. D. Nguyen, Modeling of AC-AC matrix Research with UCSD. He has co-authored ten books on adaptive, nonlinear,
converter for wind energy conversion system, in Proc. IEEE Conf. Ind. stochastic control, extremum seeking, and control of partial differential
Electron. Appl., May 2009, pp. 184191. equations systems, including turbulent flows and control of delay systems.
Dr. Krstic is a fellow of the International Federation of Automatic Control.
He has held the Russell Severance Springer Distinguished Visiting Profes-
sorship with UC Berkeley, Berkeley, CA, USA, and the Royal Academy of
Engineering Distinguished Visiting Professorship. He serves as a Senior Editor
for the IEEE T RANSACTIONS ON AUTOMATIC C ONTROL and Automatica, an
Editor of two Springer book series, a Vice President for Technical Activities of
the IEEE Control Systems Society (CSS), and a Chair of the IEEE CSS Fellow
Committee. He received the Presidential Early Career Award for Scientists
and Engineers, the National Science Foundation Career Award, the Office
of Naval Research Young Investigator Award, the Axelby and Schuck Paper
Prizes Award, and the first UCSD Research Award from engineering.

Azad Ghaffari received the B.S. degree in electrical Sridhar Seshagiri received the B.Tech. degree in
engineering and the M.S. degree in control engineer- electrical engineering from IIT Madras, Chennai,
ing from the K. N. Toosi University of Technology, India, and the M.S. and Ph.D. degrees in electrical
Tehran, Iran, and the Ph.D. degree in mechanical engineering from Michigan State University, East
and aerospace engineering from the Joint Doctoral Lansing, MI, USA, in 1995, 1998, and 2003 respec-
Program between San Diego State University and tively.
the University of California, San Diego, CA, USA. He joined the Department of Electrical and Com-
His current research interests include demand puter Engineering, San Diego State University, San
response in power systems, extremum seeking and Diego, CA, USA, in 2003, where he is currently an
its application to maximum power point tracking in Associate Professor. His current research interests
photovoltaic and wind energy conversion systems, include nonlinear control with applications to energy
induction machines, power electronics, and sliding mode control. systems, in particular control of power electronic converters.