724 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 60, NO.
3, MARCH 2013
Maximum Power Point Tracking Employing
Sliding Mode Control
Yoash Levron and Doron Shmilovitz, Member, IEEE
Abstract—A fast and unconditionally stable maximum power
point tracking scheme with high tracking efficiency is proposed for
photovoltaic generators. The fast dynamics and all range stability
are attained by a sliding mode control and the high tracking ef-
ficiency by a maximum power point algorithm with fine step. In
response to a sudden change in radiation, our experiments show
a typical convergence time of 15 ms. This is the fastest conver-
gence time reported to date. In addition we demonstrate stable
convergence all across the photovoltaic curve, from short-circuit
to open-circuit. The theory is validated experimentally.
Index Terms—Maximum power point tracking, MPPT, photo-
voltaic generator, sliding mode control.
I. INTRODUCTION
A T GIVEN atmospheric conditions (mainly dependent
on temperature [1] and insolation level), photovoltaic
(PV) cells supply maximum power at a particular operating
point—the maximum power point (MPP). Unlike conventional
(fueled) power sources, it is desirable to operate PV systems at
their MPP [1]–[13]. However, the MPP locus varies over a wide
range, depending on a PV array’s temperature and insolation
intensity [1], [2]. Instantaneous shading conditions and ageing
of PV cells also affect the MPP locus. The problem is further
complicated should the load’s electrical characteristics also
vary [3]. Hence, in order to achieve operation at the MPP, a time
varying matching network which interfaces the varying source
and the potentially varying load is required. The role of this
matching network, called the maximum power point tracking
network (MPPT), is to ensure operation of the PV generator
(PVG) at its MPP, in the face of changing atmospheric condi-
tions and load variations. Typically, the MPPT power stage is
implemented by means of a dc-dc converter at the front end,
most commonly pulse width modulated (PWM). In the case
of standalone PVG this PWM controlled dc-dc converter may
constitute the entire power stage, whereas in the case of a grid
connected PVG, it would be followed by a dc-link capacitor
and an inverter.
Since the matching network must be adaptive [2], [3], the
MPPT network must include a self-tuning mechanism which
governs the power stage and drives the system to operate at the
MPPT (eventually through the converters duty cycle). Many
Manuscript received November 17, 2011; revised February 14, 2012; ac-
cepted March 11, 2012. Date of publication February 13, 2013; date of current
version February 21, 2013. This paper was recommended by Associate Editor
E. Alarcon.
The authors are with the School of Electrical Engineering, Tel-Aviv Univer-
sity, Tel Aviv, 69978, Israel (e-mail: shmilo@eng.tau.ac.il).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TCSI.2012.2215760
1549-8328/$31.00 © 2013 IEEE
MPPT algorithms have been proposed [4], some exhibit better
performance in steady state [4], [5], while others are superior
during transitions [6]–[15]. The “perturb and observe” (P&O)
and the Hill-Climbing algorithms are probably those most
widely used. The MPPT algorithm operation principle is quite
similar with both, the voltage and current at the PVG (i.e., at
the power stage input) are sensed and the power is calculated.
Then, the MPP is sought iteratively. These algorithms imply
a tradeoff in choosing the increment value by which the con-
trolled parameter, such as duty cycle or reference voltage, is
adjusted; small values decrease the losses in steady state due
to small perturbations around the MPP, while large values
improve the dynamic behavior in situations involving quickly
changing irradiation conditions or load characteristics, such
as in portable applications [6]–[8]. Recently, effort was made
to implement MPPT algorithms with adaptive step size. For
instance, [6] suggested a sophisticated algorithm where an
intermediate control variable— —was employed. Reference
[6] showed that in a first phase may be tracked rather than
the power, but the fast tracking in the first phase has not com-
promised tracking accuracy in steady state. In order to attain
high steady-state tracking-efficiency, a second phase consisting
of a conventional MPPT scheme was required.
Sophisticated algorithms present two main shortcomings: in-
creased computational load which may require costly hardware,
and a slow dynamic response. The latest is an intrinsic limitation
resulting from the iterative search nature requiring successive
sampling of voltage and current. The time intervals between al-
gorithm iterations should be short to allow faster tracking, but on
the other hand they must be longer than the settling time of the
PV current and voltage for reliable signal measurement. There-
fore, the analysis of a PV system dynamics is required to deter-
mine the P&O time intervals and controller parameters [9].
Employing a converter that is optimized for interfacing with
the PVG has an important effect on the dynamics and stability
of the system [10]–[13]. The analysis presented in [13] shows
that commonly employed voltage fed converters are actually not
suitable for PVG due to its nonlinear characteristics. This in-
compatibility may result in reduced performance and even in-
stability.
Based on the PV module characteristics, the MPP locus may
be approximated by a linear relation [11], [12]. Thus, a linear
controller was designed which drives the PVG to its approx-
imate MPP. Hardware implemented, this action is performed
much faster than the MPPT algorithm. The MPPT algorithm
may then apply small steps to drive the operation point to the
exact MPP.
It has become widely accepted that sliding mode control is
very adequate for controlling switched mode converters [16],
LEVRON AND SHMILOVITZ: MAXIMUM POWER POINT TRACKING EMPLOYING SLIDING MODE CONTROL 725

[17]. This control exhibits high stability and fast dynamics (typ-
ically superior to linear controllers), and in many cases is simple
to implement (depending on the sliding surface). In particular,
canonical power processors suitable for interfacing PVGs were
facilitated in [18].
In contrast to PWM based MPPT, we propose a sliding-mode
controlled MPPT. Such a controller presents two major advan-
tages: first, by a proper choice of the switching surface, the
response to variations in radiation is accelerated by an order
of magnitude. In addition, the sliding-mode control facilitates
operation as either a voltage-source or a current source, thus,
it guarantees stability all across the photovoltaic curve—from
short-circuit to open-circuit. The implementation of the con-
trol is simple and requires inexpensive hardware. This work Fig. 1. Dual loop MPPT.
validates these results by theory, simulations, and experimen-
tation. This work may be viewed as an extension of the work by
Bianconi et al. [19]. This group suggested an MPPT controller
based on an inner sliding-mode loop, which uses the input ca-
pacitor current as the main state variable. By comparison, in this
work the inductor’s current is taken as the main state variable.
This approach is generally more robust and stable, as the en-
ergy stored in the converter is a direct outcome of the inductor’s
current. Consequently, the converter is inherently over-current
protected, even during start-up or transient conditions, and is
stable over a very wide range of voltages and currents. This
is confirmed by the theoretical stability analysis, presented in
Section V. Moreover, by combining the inductor’s current with
the input voltage, it is possible to design an optimal switching
surface, which results in faster transient response. These princi- Fig. 2. Sliding mode based MPPT implementation.
ples are validated both experimentally and theoretically.
the MPPT reaction. This allows using a small step increment
II. PRINCIPLE OF OPERATION in the MPPT algorithm (resulting in a high tracking efficiency
The PVG i-v characteristics are nonlinear. In addition, the in steady state) since most of the tracking is carried out by the
power to voltage (or current) relation is non-monotonic and ex- inner loop, and the MPPT requires very few iterations (as in our
hibits a maximum. Therefore, tracking of the MPP necessitates experiments 1 or 2).
a nonlinear, and in most cases iterative, technique, typically in-
volving sampling of the PVG voltage and current. The time in- III. CONTROL DESIGN AND IMPLEMENTATION
tervals between successive samples must not be shorter than the In typical dc-dc converter applications, it is desirable to reg-
converter settling time, otherwise the sampled data would be ulate the output voltage to a constant value and the switching
false. The converter interfacing the PVG is at least a second surface is chosen according [16], [17]. In other cases, such as
order system consisting of reactive components with signifi- sliding-mode controlled inverters, the output voltage needs to
cant values. This results in long settling times and consequently follow a sinusoidal shape resulting in a time varying switching
in long intervals between the MPPT’s successive samples [9]. surface [20]. Similarly, in the case of gyrators [18]–[23], an
Consequently, as the MPP search requires multiple iterations, adaptive surface is required. For MPP tracking, we define the
the MPPT convergence speed is limited and may be regarded as switching surface by a linear combination of the PVG voltage
a low bandwidth control loop. This is an intrinsic limitation of and current.
the system which cannot be overcome by a faster processor in The system is illustrated in Fig. 2. Assuming a large output
the MPPT unit. The transient response is improved by adding capacitor (or battery load) the converter output voltage may be
an inner control loop with a much wider bandwidth. Typically assumed to be semi-constant. Thus, the PVG voltage and cur-
the inner controller regulates the PVG voltage [2]–[5] or current rent constitute all the converter’s state variables, resulting in the
[9]–[13], see Fig. 1. switching surface given by (1).
With the proposed system, the MPPT algorithm unit provides
the sliding mode inner controller with an adaptive reference,
which results in an adaptive switching surface. This in a way (1)
resembles sliding mode based inverters [20]. The inner feed-
back (sliding mode) regulates a linear combination of the PVG where and are the inductor’s current and input voltage, and
voltage and current to attain its operation along a line in close , , and ref define the switching surface. and set the slope
vicinity to the MPP loci (see Fig. 3). This is accomplished in in the PVG i-v plane, and are chosen as non-negative, and ref
a very rapid manner, at least an order of magnitude faster than sets the offset. When (on state), the low transistor is on,
726 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 60, NO. 3, MARCH 2013
Fig. 3. PVG characteristics under different insolations, and the optimal
switching surface (blue) .
and the inductor charges. The current increases, and
increases. When (off state), the low transistor is off, and
the stored energy flows to the load. The current decreases, and
decreases. As a result, is kept in a “switching
band” around . This controller constricts the system’s op-
eration to the sliding switching surface. The sliding mode cir-
cuitry consists of an op-amp and a comparator, which directly
toggles the transistor gates as shown in Fig. 2.
The operating point of the PV module results from the in-
tersection of the switching surface, , with the PV curve
(Fig. 3). To extract maximum power, the switching surface is ad-
justed by the MPPT algorithm, to intersect with the exact MPP at
any given time. This algorithm is implemented in the micro-con-
troller. It samples the input power, and adjusts the switching sur-
face by updating the constant ref, which is fed to the comparator.
The proposed approach is indifferent to the MPPT algorithm se-
lected; we chose P&O due to its simplicity.
IV. OPTIMAL SELECTION OF THE SWITCHING SURFACE SLOPE
Many switching surfaces provide both stability and adequate
dynamics. We choose a linear surface, as in (1), due to the ease
of implementation (a single operational amplifier). The offset
constant, ref, is set by the MPPT controller. However, the slope
of the surface, represented by constants and , is a free pa-
rameter. A good choice of slope can significantly shorten the
MPPT’s convergence time.
In steady state, the switching surface intersects the MPP
point. A change in radiation causes a shift of the MPP point, so
the MPPT should readjust ref to (iteratively) drive the switching
surface to the new MPP. If the new MPP point is already in
close proximity to the switching surface, the required update
of ref is minimized as the new MPP is reached in very few
steps (even if the step size is fine), thus, the MPPT tracking
is accelerated. Therefore, a natural choice of slope is made by
applying LSE (least square estimate) to the set of MPPs which
corresponds to the various radiation levels. The constants and
are chosen so as to minimize the error:
(2)
where are the set of MPPs to be estimated.
Fig. 3 shows a set of MPPs, and the corresponding LSE
switching surface. Fig. 4 shows a conceptual example of the
system trajectory due to an abrupt step of the insolation.
Fig. 4. Conceptual trajectory following a sudden change of insolation. The
system “slides” to the new maximum power point.
V. STABILITY ANALYSIS
The state vector is defined in (3), where is the input
voltage, and the inductor’s current.
(3)
To assure stability, the switching surface must comply with
the converter dynamics. If the switching surface is not prop-
erly designed the system may enter a limit cycle, which can in-
cite harmful jitter and noise. Intuitively, stability is guaranteed
if the state-space trajectories evolve towards the switching sur-
face, . This is visualized in Fig. 5(b), which shows
the system trajectories with stable and instable regions. In a
stable region, the trajectory velocity points toward the switching
surface, while in an instable region, it points away from the
switching surface.
The state space trajectory velocity, , results directly
from the converter dynamics in the “on” and “off” states ac-
cording to (4).
(4)
is the PVG current, is the input capacitance, is the
inductance, and is the output voltage. The criterion for sta-
bility is that “the velocity points toward the switching surface.”
This is the sliding-mode “existence condition,” which is given
by (5):
(5)
(the gradient of ) is normal to the switching surface ,
where and are the constants that define the surface slope.
The stability condition (6) in the “on” state is reached by sub-
stitution. The “or” statement indicates that “on” state stability
LEVRON AND SHMILOVITZ: MAXIMUM POWER POINT TRACKING EMPLOYING SLIDING MODE CONTROL 727

Solving for the current :

(9)

Then, combining (7) and (9):

(10)

Equation (10) is the eventual stability condition, which defines

a “stability region” in state-space, see Fig. 5.
Stable points, i.e., points that are attracted toward the
switching surface, are those that satisfy (10). Stability is guar-
anteed as long as the trajectories are bounded within this region.

VI. SWITCHING FREQUENCY

As a first order analysis, the switching surface was assumed
to be a line (of zero width). Switching occurs over a theoret-
ical line, , as in (1). However, in practice, the switching
surface must have a small but finite width, to prevent chattering.
This width is denoted by , and the actual switching borders are
given by (11). is set by adjusting the positive feedback resistor
across the comparator (Fig. 2). The switching frequency is di-
rectly associated with the width, , thus, this width must be well
chosen to assure an adequately low switching frequency and
a resulting high efficiency, under all possible operation points.
Equation (12) relates the switching frequency, , to the width
. Both sides of the equation equal the switching period, and
are therefore equal to each other.

(11)

(12)

Applying the gradient of S, and the state space velocity of (4)

Fig. 5. (a) Stability region and optimal switching surface in the state-space. (b)
Typical system trajectories. and (5), results in an explicit expression for .

applies only if . Thus, if , the condition is irrele-

vant.
(13)

The following analysis is limited to steady-state conditions,

(6) after convergence to the MPP. At this operation point, some
approximations can be assumed, resulting in a much simpler
Solving for the current , results in (7), which is an explicit sta- expression for :
bility criterion for the “on” state:

(7)

(14)
Similarly, (8) is the “off” state stability condition.

Rearranging (14) yields (15).

(15)
(8)
728 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 60, NO. 3, MARCH 2013

Fig. 6. Experimental setup.

This last expression presents the switching frequency, , in

steady-state, with respect to the MPP of the PV panel. With a
larger width , the frequency decreases.

VII. EXPERIMENTAL RESULTS

A. System Structure
The experimental setup (Fig. 6) includes a PV emulator,
a synchronous boost converter, gate drive circuitry, sensors,
sliding-mode controller implementation, and a micro-con-
troller for MPPT execution. The PV emulator is implemented
by means of a laboratory power supply operated in the cur-
rent limit mode, set to 3.5 A, in parallel connection with a
diodes’ string. The diodes string consists of 30 diodes of type
“MUR820” connected in series. An electronic load, config-
ured as a current source, is connected in parallel to the power
supply and diodes, thus operating as a programed current sink.
Radiation changes are emulated by stepping the current sink,
in the range of 0.5 A ( , 1 sun) to 3.5 A ( ,
0 sun). The emulated PV curves are measured, as shown in
Fig. 3. It may be noted that the emulated PV characteristics are
similar to those exhibited by actual PVGs. Thus, this emulation
method is quite effective for verifying the MPPT & the sliding
mode controller performance. The boost converter consists of
an inductor of 200 ( saturation, series resis- Fig. 7. MPPT algorithm used with the sliding-mode controller. This being the
tance), an input capacitor of 100 , and an output capacitor ordinary “perturb & observe” (P&O) algorithm, outputting ref to the D/A.
of 200 . The load is resistive, 28 . Two complementary
N-ch FETs of type “STP60NF10” are used. Each FET (high (AD8605) and a fast comparator (MAX941). The MPPT is im-
and low side) is driven by floating gate circuitry, based on plemented on a dsPIC30F2020 evaluation board (microchip). It
the “MIC4423,” 3 A gate driver. A delay circuit synchronizes runs a simple perturb and observer (P&O) MPPT algorithm (see
the on/off transitions of the high and low FETs, to prevent Fig. 7). The micro-controller drives an external D/A (AD5060),
shoot-through currents. Turn-off is immediate, and turn-on is which outputs ref back to the sliding-mode control.
delayed by 20 nS A high-side current sensor (MAX4173) is We also run a “PWM based MPPT” test mode, in which the
positioned in series with the inductor on the input side. micro-controller generates a PWM signal with a varying duty-
The sensed input voltage and inductor’s current are fed to cycle, which is fed directly to the gate drive (no sliding-mode).
both the sliding-mode controller and the micro-controller in- This being the ordinary “PWM” based MPPT. We use it as a
ternal A/D. The sliding-mode circuitry is based on a fast op-amp test case to compare the sliding-mode performances.
LEVRON AND SHMILOVITZ: MAXIMUM POWER POINT TRACKING EMPLOYING SLIDING MODE CONTROL 729

Fig. 9. Stability verification. Convergence from an initial “open-circuit”:

the current climbs monotonically. Yellow trace -voltage (5 V/div), pink
trace—output power (10 W/div), Blue trace -current (0.5 A/div), Time base:
40 ms/div.

Fig. 8. Convergence comparison of sliding-mode MPPT and PWM based

MPPT following an insolation step. (a) Sliding-mode, reaction to a positive
step. (b) PWM MPPT, reaction to a positive step. (c) Sliding-mode, mode
reaction to a negative step. (d) PWM MPPT, reaction to a negative step. Yellow
trace—input voltage (2 V/div), blue trace—input current (0.5 A/div), pink
trace—output power (10 W/div). Time base is 40 ms/div.

B. Controller Design Procedure

The optimal switching surface is selected by applying (2) to
the MPP loci, see Fig. 3. The resulting slope (normalized by ) is
, . ref is set in real-time by the MPPT algorithm.
Its theoretical optimal value, according to (2) is
The switching surface width, , is designed for a switching fre-
quency of . Applying (15), with
, results in . A graphical stability analysis
with these values is presented in Fig. 5, showing stability all
across the PV curve, with adequate margins. The actual imple-
mentation is shown in Fig. 6.
C. Measurements
The sliding-mode MPPT, with optimal switching surface, is
compared with an ordinary “PWM based MPPT.” We use the
PV emulator to generate a sudden step in input “radiation,” and
measure the step response of both controllers. Fig. 8 shows the
results. The PV emulator’s current source is switched in-
stantaneously from 1.5 A to 3.0 A to emulate a 100% positive
insolation step (on the left), and from 3 A to 1.5 A to emulate
a 50% negative insolation step (on the right). The MPPT algo-
rithm step size is for the sliding mode MPPT, and
2% for PWM based MPPT. The step time is identical for both
controllers—20 ms per step. The sliding mode MPPT is seen to
converge within 15 ms, whereas the PWM based MPPT con-
verges within 210 ms. It should be noted that the power shown
in Fig. 8 is the converter’s output power while the PVG’s power
rises even faster, see Fig. 11 for a comparison of the input and
output powers following an insolation step. In this experiment,
the current source is switched instantaneously from 1.5 A to 3.0
A and the input power is seen to rise within 5 ms.
Figs. 9 and 10 display the results of the sliding mode sta-
bility tests. The system is initialized far from the MPP, in order
to cause it to stroll across the PV curve, thus testing stability
through a variety of operation points. In Fig. 9, the system con-
.
Fig. 10. Stability verification. Convergence from an initial “short-circuit”: The
current is steady while the voltage climbs monotonically. Blue trace -current
, (0.2 A/div), pink trace—output power (10 W/div), yellow trace -voltage (5
V/div), Time base is 100 ms/div.
verges from an initial “open-circuit.” The initial state is deter-
mined by setting an initial high ref value of 19.4 (compared to
about 13.7 at MPP). This initializes the system on the right side
of the PV curve, where the PV module exhibits voltage source
characteristics. The initial current is low (0.85 A) and converges
to the MPP (2.73 A). The voltage is almost steady throughout
the transient. In this experiment, ref step and the
MPPT step time was 20 ms.
Fig. 10 shows the convergence from an initial “short-circuit.”
ref is initially low , setting the system on the left side
of the PV curve, where the module operates as a current source.
The initial voltage is low (3.6 V) and converges to the MPP
(20.2 V). The current is seen to be nearly constant throughout
the transient. As in the previous experiment, step and
the MPPT step time was 20 ms.
VIII. DISCUSSION
This work presents a sliding-mode control based MPPT ap-
plying an optimally matched switching surface. The sliding-
mode MPPT results in two major advantages over PWM based
MPPT: fast tracking in response to a radiation change, and sta-
bility across the entire photo-voltaic curve.
A. Fast Tracking
With sliding-mode, the tracking response to a radiation
change is highly accelerated, as seen in Fig. 8. This was
730 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 60, NO. 3, MARCH 2013

verified by comparing the dynamic tracking of the proposed

sliding-mode based MPPT with that of a reference PWM
based MPPT. In order to compare dynamic performance, a
radiation step was emulated. Although such a radiation step
is quite extreme (and maybe not realistic); it provides a good
measure for testing the convergence speed. The PWM based
controller converges within 210 ms, while the sliding-mode
controller converges within 15 ms (14 times faster).Moreover,
the converter’s input power follows the radiation change al-
most immediately, as seen in Fig. 11. It should be noted that,
although the converter’s input power raises in some 5 ms.
(even this is much faster than previously reported PWM based
MPPT performance), it can converge even faster, but we cannot
demonstrate it since the PV emulator’s speed is limited to about
1 A/ms. The fast convergence is attained by an optimal selec-
tion of the switching surface slope (according to (2)). Following
a radiation change, if the switching surface is in proximity to Fig. 11. Input power and output power following an insolation step. (a) Mea-
the new MPP, most of the convergence (typically over 90%) sured results of the proposed system. Blue trace—input power (10 W/div), Pink
trace—output power (10 W/div), Time base: 4 ms/div. (b) Results from Sera et
is performed by the high band-width hardware-implemented al. [15], demonstrating the difference in convergence speed.
inner control loop. The system “slides” along the switching
surface from the previous PV characteristics to the new one,
in close proximity to the new MPP, as shown in Fig. 4. Thus,
values used in the experiment ( , , ,
the output power has risen close to its final value, before the
, , , ). The
MPPT unit (and software) even reacted. Then, convergence
photovoltaic curve, is contained by in the stability re-
to the exact MPP slowly continues by means of the software
gion, verifying the stability of all operation points along the PV
implemented MPPT algorithm. This is seen in Fig. 8(a), where
curve. Figs. 9 and 10 show the experimental stability tests. The
the current still increases by a few percentages following the
system is forced to converge (in software) across the right side
initial response. In comparison, PWM based MPPTs are slower,
and the left side of the PV curve, thus verifying the stability at
because the convergence is performed mainly in software, by
multiple operation points. The system is stable at all the points
the MPPT unit. A comparison of the convergence speed is
on the PV curve, confirming the theoretical stability analysis.
displayed in Fig. 11. The proposed system is compared with a
software based MPPT, using an inner voltage loop, proposed by C. Voltage or Current Feedback
Sera et al. [15]. This MPPT algorithm is optimized for speed,
yet its convergence time is on a scale of seconds. In contrast, Instead of applying the optimal slope, which requires two sen-
our proposed topology converges to maximum power within 15 sors, the sliding-mode control may be implemented by using a
mS. Since they are also fully hardware implemented, extremum single voltage sensor, or a single current sensor. The resulting
seeking based MPPT are also very fast [24], [25]. However, it switching surface is not optimal, and the tracking speed would
seems the sliding mode based MPPT is simpler to implement depend on software. With only a voltage sensor, the slope con-
and requires fewer components. To conclude, the sliding-mode stant is zeroed (1), and with only a current sensor, is zeroed.
based MPPT acts through a hardware implemented inner Fig. 12 shows typical stability regions, which are evaluated by
feedback loop with very high bandwidth, while PWM based (10). With a voltage feedback, the system is quasi-stable, as the
MPPTs converge mainly through the MPPT unit involving instability boundaries overlap the PV curve. However, with a
successive sampling and software based feedback. As a result, current feedback the system is stable at all points, and under
sliding-mode convergence is faster by an order of magnitude. any conditions. This is in compliance with other works [9], [11],
[13].
B. Stability However, it should be noted that it is not recommended to
In contrast to PWM based MPPT which requires careful de- operate the system purely with a current feedback, as it might
sign and accurate knowledge of the PVG and converter parame- loss stability if the target current is set higher than the PV short-
ters in order to ensure stability (and even so the stability is only circuit current. It is best to operate the system with an optimal
attained within a certain range of PVG voltages) [9], [10], [13], slope, combining both the voltage and current feedbacks.
stability is easily attained with the sliding-mode control and it is
D. Switching Frequency
guaranteed over the entire PVG voltage range. After the optimal
slope is estimated, the stability of the corresponding switching A low switching frequency reduces the switching loss, and is
surface should be analyzed. The sliding-mode controller can important for high efficiency. The higher the width, , the lower
operate across the entire photovoltaic curve. Stability is main- the switching frequency, as in (15). We designed the system for
tained when the photovoltaic source is a current source (on the a switching frequency of , resulting in
left) or a voltage source (on the right). This can be confirmed (a lower frequency would have required a larger and bulky in-
per case with the theoretical stability analysis (10), and by plot- ductor). The resulting efficiency was 93% near MPP, at 1 sun
ting the stability regions. Fig. 5 shows the stability region for the . The actual frequency measured 119.9 kHz at
LEVRON AND SHMILOVITZ: MAXIMUM POWER POINT TRACKING EMPLOYING SLIDING MODE CONTROL 731

Fig. 12. Stability regions and typical trajectories with single sensed parameters: (a) a voltage feedback or (b) a current feedback .

scenario is demonstrated by simulation, in Fig. 13. The system

rapidly slides to a point near a new MPP. However it turns out
that in this case the final MPP reached is a local one.
Under heavy shading conditions, the PV module’s i-v curve
may become sporadic. Consequentially, the selection of sliding
surface has no effect on the convergence of the system, and the
transient response will be an outcome merely of the software
based MPPT algorithm, similarly to duty-cycle based systems.

IX. CONCLUSION
This work introduces a sliding-mode based MPPT method. In
comparison to PWM based MPPTs, convergence to maximum
power is accelerated by an order of magnitude. This is accom-
plished by an optimal selection of the switching surface. In ad-
dition, the controller can operate either as a voltage source or a
current source, maintaining stability all across the photo-voltaic
curve. Stability is analyzed using the “stability region” method,
which is easy to comprehend and intuitive.

Fig. 13. System performance under partial shading (simulation results).
MPP, a 20% error in comparison with the theoretical value. As
a first degree approximation, only the MPP voltage affects the
switching frequency. The current at MPP only slightly affects
the switching frequency. Note that this current does not appear
in (15). The MPP voltage changes a little with radiation, thus,
the switching frequency is quite constant under radiation shifts.
E. The Effect of Shading
With partially shaded modules, the estimated sliding-surface
is no longer optimal. However, the system is likely to slide in
close vicinity to a local or global maximum, as a typical shaded
i-v curve usually preserve a maximum power point (local or
global) with voltage similar to the MPP of a non-shaded profile.
As a result, the system will converge fast; however, it may reach
a local maximum point, missing the global maximum. Such a
REFERENCES
[1] P. Maffezzoni and D. D’Amore, “Compact electrothermal macromod-
eling of photovoltaic modules,” IEEE Trans. Circuits Syst. II, Exp.
Briefs, vol. 56, no. 2, pp. 162–166, Feb. 2009.
[2] K. H. Hussein, I. Muta, T. Hoshino, and M. Osakada, “Maximum
photovoltaic power tracking: An algorithm for rapidly changing
atmospheric conditions,” in IEE Proc. Gener., Transm., Distrib., 1995,
vol. 142, pp. 59–64.
[3] Z. Zinger and A. Braunstein, “Dynamic matching of a Solar-Electrical
(photovoltaic) system an estimation of the minimum requirements on
the matching system,” IEEE Trans. Power App. Syst., vol. PAS-100,
no. 3, pp. 1189–1192, 1981.
[4] T. Esram and P. L. Chapman, “Comparison of photovoltaic array max-
imum power point tracking techniques,” IEEE Trans. Energy Convers.,
vol. 22, no. 2, pp. 439–449, Jun. 2007.
[5] D. Shmilovitz, “On the control of photovoltaic maximum power point
tracker via output parameters,” IEE Trans. Power Appl., vol. 152, no.
2, pp. 239–248, Mar. 2005.
[6] S. Jain and V. Agarwal, “A new algorithm for rapid tracking of approx-
imate maximum power point in photovoltaic systems,” IEEE Power
Electron. Lett., vol. 2, no. 1, pp. 16–19, Mar. 2004.
[7] L. Gao, R. A. Dougal, S. Liu, and A. Iotova, “Portable solar systems
using a step-up power converter with a fast-speed MPPT and a parallel
configured solar panel to address rapidly changing illumination,” in
Proc. IEEE APEC, Anaheim, CA, Mar. 2007, pp. 520–523.
732 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS, VOL. 60, NO. 3, MARCH 2013
[8] L. Gao, R. A. Dougal, S. Liu, and A. Iotova, “Parallel-Connected solar
PV system to address partial and rapidly fluctuating shadow condi-
tions,” IEEE Trans. Ind. Electron., vol. 56, no. 5, pp. 1548–1556, May
2009.
[9] M. Sokolov et al., “Small-signal model of photovoltaic power con-
verter for selection of perturb and observe algorithm step time,” in
Proc. 14th Eur. Conf. Power Electron. Appl. (EPE), Sep. 2011, pp.
1–6.
[10] R. F. Coelho, F. M. Concer, and D. C. Martins, “Analytical and experi-
mental analysis of DC-DC converters in photovoltaic maximum power
point tracking applications,” in Proc. 36th Annu. Conf. IEEE Ind. Elec-
tron. Soc. (IECON), Nov. 2010, pp. 2778–2783.
[11] V. V. R. Scarpa, G. Spiazzi, and S. Buso, “Low complexity MPPT
technique exploiting the effect of the PV cell series resistance,” IEEE
Trans. Ind. Electron., vol. 56, no. 5, pp. 1531–1538, May 2008.
[12] M. Sokolov and D. Shmilovitz, “A modified MPPT scheme for ac-
celerated convergence,” IEEE Trans. Energy Convers., vol. 23, pp.
1105–1107, Dec. 2008.
[13] J. Leppaaho and T. Suntio, “Dynamic properties of PCM-controlled
current-fed boost converter in photovoltaic system interfacing,” in
Proc. 14th Eur. Conf. Power Electron. Appl. (EPE), Sep. 2011, pp.
1–10.
[14] S. Singer, R. Giral, J. Calvente, R. Leyva, L. Martinez-Salamero, and
D. Naunin, “Maximum power point tracker based on a loss free resistor
topology,” in Proc. Fifth Eur. Space Power Conf. (ESPC), Tarragona,
Spain, Sep. 21–25, 1998.
[15] D. Sera, R. Teodorescu, J. Hantschel, and M. Knoll, “Optimized
maximum power point tracker for fast-changing environmental condi-
tions,” IEEE Trans. Ind. Electron., vol. 55, pp. 2629–2637, 2008.
[16] G. Spiazzi et al., “Application of sliding mode control to switched-
mode power supplies,” J. Circuits, Syst. Comput. (JCSC), vol. 5, no. 3,
pp. 337–354, Sep. 1995.
[17] L. Martinez-Salamero et al., “Why is sliding mode control method-
ology needed for power converters,” in Proc. 14th Int. Power Electron.
Motion Control Conf. (EPE/PEMC), 2010, pp. S9-25–S9-31.
[18] L. Martinez-Salamero, “Synthesis of canonical elements for power pro-
cessing,” in 6th Int. Multi-Conf. Syst., Signals, Devices, 2009, pp. 1–6.
[19] E. Bianconi, J. Calvente, R. Giral, G. Petrone, C. A. Ramos-Paja, G.
Spagnuolo, and M. Vitelli, “A fast current-based MPPT technique
based on sliding mode control,” in Proc. IEEE Int. Symp. Ind. Electron.
(ISIE), 2011, pp. 59–64.
[20] D. Biel et al., “Sliding-mode control design of a boost-buck switching
converter for AC signal generation,” IEEE Trans. Circuits Syst. I, Reg.
Papers, vol. 51, no. 8, pp. 1539–1551, 2004.
[21] I. S. Uzunov, “Theoretical model of ungrounded inductance realized
with two gyrators,” IEEE Trans. Circuits Syst. II, Exp. Briefs, vol. 55,
no. 10, pp. 981–985, 2008.
[22] R. Y. Barazarte, G. G. González, and M. Ehsani, “Generalized gyrator
theory,” IEEE Trans. Power Electron., vol. 25, no. 7, pp. 1832–1837,
2010.
[23] A. Cid-Pastor et al., “Paralleling DC-DC switching converters by
means of power gyrators,” IEEE Trans. Power Electron., vol. 22, no.
6, pp. 2444–2453, 2007.
[24] R. Leyva, C. Alonso, I. Queinnec, A. Cid-Pastor, D. Lagrange, and
L. Martinez-Salamero, “MPPT of photovoltaic systems using ex-
tremum—Seeking control,” IEEE Trans. Aerosp. Electron. Syst., vol.
42, pp. 249–258, 2006.
[25] C. Cabal, C. Alonso, A. Cid-Pastor, B. Estibals, L. Seguier, R. Leyva,
G. Schweitz, and J. Alzieu, “Adaptive digital MPPT control for pho-
tovoltaic applications,” in Proc. IEEE Int. Symp. Ind. Electron. (ISIE
2007), pp. 2414–2419.
Yoash Levron received his B.Sc. from the Technion,
Haifa, Israel, and an M.Sc. from the Tel-Aviv Univer-
sity, Israel, in 2001 and 2006, respectively. He is cur-
rently pursuing his Ph.D. at Tel-Aviv University. His
research focuses on the control of power processing
systems.
D. Shmilovitz (M’98) received his B.Sc., M.Sc., and
Ph.D. from Tel-Aviv University, Tel-Aviv, Israel, in
1986, 1993, and 1997, respectively, all in electrical
engineering.
During 1986–1990 he worked in R&D for the
IAF, where he developed programmable electronic
loads. During 1997–1999, he was a Postdoctorate
Fellow at the New York Polytechnic University,
Brooklyn. Since 2000, he has been with the Faculty
of Engineering, Tel-Aviv University. His research in-
terests include switched-mode converters, topology,
dynamics and control, and circuits for alternative energy sources and for
powering of sensor networks and implanted medical devices, and general
circuit theory.