[IEEE 2013 IEEE 14th Workshop on Control and Modeling for Power Electronics (COMPEL) - Salt Lake...
Transcript of [IEEE 2013 IEEE 14th Workshop on Control and Modeling for Power Electronics (COMPEL) - Salt Lake...
Implementation of Dithering DigitalRipple Correlation Control for
PV Maximum Power Point Tracking
Christopher Barth, Robert C.N. Pilawa-Podgurski
UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN
URBANA, ILLINOIS 61801
EMAIL: [email protected]
Abstract—This work demonstrates a method for rapid max-imum power point tracking in photovoltaic applications usingdithered PWM control. Constraints imposed by power consump-tion, cost and component size, limit the available PWM resolutionof a power converter, and may in turn limit the MPP trackingefficiency of the PV system. PWM dithering can be used toimprove average PWM resolution. However, additional trackingdelays are also introduced. By applying digital dithering ripplecorrelation control to the ripple caused by PWM dithering itis possible to achieve high tracking speed in addition to highPWM resolution. We present a theoretical derivation of theprinciples behind DDRCC, as well as experimental results thatshow excellent tracking speed and accuracy with basic hardwarerequirements.
I. INTRODUCTION
The need for maximum power point tracking (MPPT) arises
from the non-linear I-V curve of photovoltaic panels, and the
fact that panel characteristics change with insolation, tempera-
ture and age. Effective MPPT requires adequate measurement
resolution of the power flowing into the converter as well as
sufficient resolution on converter control. The control resolu-
tion of a switching power converter is determined by the PWM
resolution of the controller. Many modern systems depend on
microprocessor-based controllers that employ digital counter-
based PWM. In these implementations, the PWM resolution
of the converter is proportional to the clock speed of the
microcontroller and inversely proportional to the switching
frequency of the converter. Because microcontroller clock
speed may be limited by hardware constraints in low cost,
low power applications and current trends toward integration
often require smaller passive components and accompanying
higher switching speeds, there are instances when the available
PWM resolution becomes a constraining factor in efficient
MPPT operation. Moreover, for high efficiency operation,
control losses must be kept low. In low power applications
the power consumed by a high-frequency microcontroller can
be significant [1], [2].
This work was supported in part by the Illinois Center for a Smarter ElectricGrid, through Illinois Department of Commerce and Economic Opportunity(DCEO) under grant 12-197001
In these cases the technique of PWM dithering can be used
to increase the effective resolution of the converter without
increasing digital logic frequency. By alternating between
adjacent native PWM ratios at a predetermined dithering fre-
quency, an intermediate average PWM ratio can be obtained.
Unfortunately, the use of dithering imposes additional delay on
the update speed of the MPPT algorithm as both the switching
ripple and the dithering ripple must be adequately averaged in
order to achieve a high resolution measurement. This creates
a trade-off between PWM resolution and tracking speed.
Ripple correlation control (RCC) has been shown to be an
adaptable approach for system optimization in the presence
of a sustained ripple [3]. Further work has demonstrated that
RCC is exceptionally well suited for stable digital implemen-
tation [4]. This work builds on the theoretical introduction of
dithering digital dipple correlation control (DDRCC) [5] and
demonstrates that DDRCC can be applied to the system ripple
created by PWM dithering to achieve excellent PV MPPT
speed and convergence in MPPT systems which use dithered
PWM control. Additionally, by applying RCC to the average
ripple caused by dithering, DDRCC avoids the limitations
placed on switching frequency in RCC controlled converters
[6].
II. DITHERING FOR INCREASED EFFECTIVE RESOLUTION
As mentioned previously, precise MPPT requires good
converter control resolution and by extension, finely adjustable
PWM signal. The duty ratio resolution of a counter-generated
PWM signal is limited by the clock speed of the microcon-
troller as shown in (1). For example, an 8 MHz counter clock
driving a 250 kHz PWM signal results in a resolution limit of
3.132%.
FPWM
Fclk= ∆Dmin (1)
One well-known technique to circumvent this limitation
is to dither between the available duty ratios at appropriate
intervals, in order to generate a PWM signal with an average
value between the duty ratios available in hardware [7]. The
978-1-4673-4916-1/13/$31.00 ©2013 IEEE
0% 20% 40% 60% 80% 100%
20%
40%
60%
80%
100%
Desired Duty Ratio
Available Duty Ratio
Native resolution
Dithered resolution 52%
53%
54%
55%
Δ Dmin Native = 3.13%
Δ Dmin Dither = 0.079%
Fig. 1. Measurements of average duty ratio resolution with both native PWMand dithering.
Time
Scaled Amplitude
PWM S ignal
Panel Current
Panel Voltage
t = RdTd
t = 0 t = Td
Fig. 2. Simulation of dithering voltage and current waveforms resulting fromchanges in PWM duty ratio.
percentage of time the converter runs at either duty ratio is ad-
justed to generate an average duty ratio between the two native
values of the hardware. In theory this approach generates an
arbitrarily fine PWM resolution as shown in (2). In practice
the increase in resolution is limited by the input capacitance
of the converter and the maximum acceptable ripple. For the
PWM generator described above, averaging over 40 cycles
allows for an average resolution of 0.078%. Figure 1 shows
measurements of the duty ratios attainable with both native
resolution and dithering for an MSP430G2553 microcontroller
running at an 8 MHz clock speed and generating a 250 kHz
PWM signal.
∆Dmin
N= ∆Ddithered (2)
As the MPPT converter dithers between native duty ratios,
the average current and voltage of the panel will change
approximately linearly in time, with the current rising and
voltage falling over periods of high duty ratio, and the reverse
over periods of low duty ratio. This ripple in the average I
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.80
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Current [A]
Power [W]
A BP*
I*
Fig. 3. Power vs. current characteristic for PV panel under illumination inexperimental setup.
and V of the PV panel can be used as a means of maximum
power point tracking and will be explored in this work.
Figure 2 shows a simulation of the input current and voltage
of a buck converter when it is used as a PV power converter
with dithering applied. The dithering period is denoted as Td,
and extends from the time at which the converter shifts to the
native duty ratio which is just above the desired average duty
ratio until the converter has finished running at the ratio which
is just below the desired average duty. The native duty ratio
above the desired duty ratio can be referred to as the “high”
duty ratio, and the one below as the “low” duty ratio The
dithering ratio, denoted as Rd marks the point in the dithering
period at which the converter transitions from the high to the
low duty ratio. By adjusting the dithering ratio (not to be
confused with the duty ratio) the converter is able to vary
the average PWM duty ratio between the high and low native
values.
III. PRINCIPLES OF DDRCC
The principle of ripple correlation control has been pre-
sented in both analog and digital form [3], [4]. With some
modifications, the concept of RCC can be extended to an
average basis for application to the ripple resulting from
converter dithering. The main points of the derivation will be
repeated here to illustrate the approach.
Extracting maximum power from a PV array requires the
panel to be operated at some optimal value of current and
voltage. Since I and V are interdependent, panel power can
be plotted as a function of either variable. Figure 3 shows the
power extracted from a 20 W, 9.5 V, solar panel in the lab over
a range of currents [8]. It is clear that power is maximum at a
value of I* for which the derivative of power with respect to
current is zero. For portions of the power vs. current graph to
the left of I*, for instance the point marked A, the derivative
of power with respect to current is positive. Conversely, the
derivative of power with respect to current is negative for all
points on the curve with I greater than I*, such as at point
B. These relationships are presented in (3) and (4).
dP
dI> 0 =⇒ I < I∗ (3)
dP
dI< 0 =⇒ I > I∗ (4)
The power converter input current (and thus the PV panel
current) is proportional to the duty ratio which makes it
possible to traverse the power curve of Fig. 3 by controlling
the duty ratio. For a buck converter, equations (3) and (4)
can be restated as shown in (5) and (6) to determine the
direction of the change in duty ratio required to reach the
panel maximum power point. The direction of inequalities will
change depending upon the type of converter used.
dP
dI> 0 =⇒ D < DMPP (5)
dP
dI< 0 =⇒ D > DMPP (6)
For a buck converter, panel current increases as the duty
ratio increases, and the sign of the required change in duty
ratio is the same as the sign of dPdI
. Therefore, integral control
with a suitable scalar gain can be applied over time to drive
the average converter duty ratio to DMPP as shown in (7).
The duty ratio will be adjusted until dPdI
is equal to zero.
This relationship holds as long as I(D) is monotonic and the
average duty ratio, D, is adjustable in time [4].
D = k
∫
dP
dIdt (7)
Since the derivative of (7) is not easily measurable, this
expression can be restated by multiplying by a positive con-
stant. Since measurement of power and current are routinely
performed in the time domain, the term ( dIdt
)2 can be used to
restate (7) in terms of derivatives with respect to time as shown
in (8) [4]. Note that ( dIdt
)2 is not only positive, but also constant
with respect to the power vs. current relationship depicted in
Fig. 3. Hence it does not effect the ability of (7) to drive the
duty ratio to DMPP .
D = k
∫
dP
dI
dI
dt
dI
dtdt = k
∫
dP
dt
dI
dtdt (8)
A deeper intuition for (8) and the operation of DDRCC in
general can be seen by considering the relationship depicted
in Fig. 3 as power and current are changed over time. As
discussed, an increase in the panel current for I < I* will
result in an increase in power. With prime denoting the new
value and nought the original, this can be stated as Po < P ′ as
long as Io < I ′ and Io < I*. Conversely, Po > P ′ for Io > I*
also holds. These relations can be combined to determine the
direction of change in I required, and hence the direction of
change in D required to reach the maximum power point [6].
dI
dt
dP
dt> 0 =⇒ I < I∗ =⇒ D < DMPP (9)
dI
dt
dP
dt< 0 =⇒ I > I∗ =⇒ D > DMPP (10)
The time-based approach in (9) and (10) therefore yields (8)
directly, however the underlying relationship between power
and current which is the basis for DDRCC is somewhat
obscured by the introduction of time derivatives.
Equation (8) can be simplified by considering that the aver-
age panel current and voltage change approximately linearly
as the converter dithers. Noting the waveforms in Fig. 2, it
is justifiable to approximate dIdt
as a positive and a negative
constant over the region of high and low duty ratio respectively
as shown in (11) [4].
dI
dt=
S+ mod(t, Td) ∈ [0, RdTd)S− mod(t, Td) ∈ [RdTd, Td)
(11)
By making use of (11) and taking the initial duty ratio into
account, the integral in (8) can then be rewritten as the sum
of two integrals each multiplied by the corresponding constant
current slope as shown in (12). Symbolic evaluation then leads
to (13).
D(t) = D(0) + kS+
∫ RdTd
0
P dt+ kS−
∫ Td
RdTd
P dt (12)
D(t) = D(0)+kS+(P (RdTd)−P (0))+kS−(P (Td)−P (RdTd))(13)
When the converter is operating in equilibrium, the variation
in average values of input voltage, current, and power will all
be periodic over the dithering cycle. This means that (14) and
(15) will hold.
I(0) = I(Td) (14)
P (0) = P (Td) (15)
In order for this to be true, the product of the rise time
of the average current and the positive slope of the average
current must be equal to the product of the period over which
the current is decreasing multiplied by the rate of decrease. Or
equivalently, ∆t ∗ dIdt
must be equal on the falling and rising
slope. Because the period of rising current is given as TdRd,
and that of the falling current is Td(1 − Rd), it is possible
to solve for the negative slope of the current in terms of the
dithering ratio (Rd) and the positive slope as shown in (16).
S+Rd = S−(Rd − 1) =⇒ S− =−S+Rd
1−Rd
(16)
Equations (15) and (16) can then be used to simplify (13)
into (17).
D(t) = D(0) +kS+
1−Rd
(P (RdTd)− P (0)) (17)
Depending upon the resolution of the dithering process,
and the constant gain (k) used, there may be no discernible
difference between using the magnitude of the difference in
(17) and only the sign of the difference with a constant
gain. With an appropriate k′ value, (18) can be considered
functionally equivalent to (17).
0.01 0.012 0.014 0.016 0.018 0.02 0.022
15.63%
18.75%
21.88%
25%
Time [sec]
Increasing Average Duty
3−Way TransitionInstaneous Duty Ratio
0.0165 0.017 0.0175 0.018 0.0185
15.63%
18.75%
21.88%
25%
Increasing Average Duty
3−Way Transition
Fig. 4. Three-Way dithering process. Left plot shows instantaneous duty in time as average duty is progressively increased. Right plot shows closeup ofthree-way transition used to maintain ripple.
D(Td) = D(0) + k′sgn(P (RdTd)− P (0)) (18)
Equation 18 is the essence of DDRCC. By sampling the
power of the panel at the beginning of the dithering cycle
(P (0)) and at the transition in duty ratio (P (RdTd)), and
then adjusting the duty ratio accordingly, the converter will
iteratively converge to the optimal average duty ratio for MPP
operation.
Because DDRCC makes use of the dithering frequency
instead of the switching frequency to perform MPPT, the
DDRCC dithering frequency/tracking frequency can be cus-
tomized at design time or adaptively during operation.
DDRCC is well-suited for mobile applications, because the
tracking speed of the converter can be a respectable percentage
of the switching frequency, and the fact that the dithering
strategy is most applicable to low-power applications, and low-
cost control hardware.
IV. HARDWARE IMPLEMENTATION
In traditional dithering approaches, when the desired duty
ratio is equal to a native value, the converter would typically
stop dithering and run at a constant duty ratio. Since ripple is
essential for DDRCC operation, the dithering approach must
be modified to maintain a ripple. In the present implementa-
tion, the required dither is ensured by entering a 3-way mode
in which the instantaneous duty ratio is switched between the
native resolution value needed and the native values above
and below. In practice, the three-way mode must be extended
beyond the single desired duty ratio at which it is required
to maintain ripple, in order to maintain the continuity of the
average signal. When the average duty ratio has increased
beyond the transition region, the system will return to two-
way dither. This sequence of increasing the dithering ratio and
then using a 3-way pattern repeats as the average duty ratio
is increased. The transition to three-way mode and the timing
of transitions during the three-way mode are chosen such that
the average duty ratio increases linearly.
TABLE ICOMPONENT VALUES FOR CONVERTER
Device Value/Model Manufacturer
Microcontroller MSP430G2553 Texas Instruments
Op amp LMP8602 Texas Instruments
DrMOS FDM6704V Fairchild
Rsense 0.05 Ω Vishay
Lbuck DR127, 22 µH Cooper Bussman
Rdiv1 7.5 kΩ Yageo
Rdiv2 2.4 kΩ Yageo
C1 2 nF Taiyo Yuden
Cin 44 µF Taiyo Yuden
Cout 20 µF Taiyo Yuden
R1, R2 8.2 kΩ Yageo
C2, C3 2.2 µF Velleman
Rload 1.4 Ω
Solar Panel
−+
Op Amp
R2R1
C1
C2
Microcontroller
DrMOS
C3
Rsense Lbuck
Cout
Rdiv1
Rdiv2
Cin
Rload
Fig. 5. Schematic of the buck converter used for DDRCC verification.
Figure 4 shows a simulated plot of the instantaneous PWM
duty ratio used by the converter micrcontroller as the dithering
ratio (Rd) is increased over time [9]. The y-axis of Fig. 4 gives
the instantaneous duty ratio of the converter. In this example
the PWM counter is running at 8 MHz, and the converter at
a switching speed of 250 kHz. Therefore, there are 32 clock
pulses per PWM period, and a native duty ratio resolution of
3.13%. However, by adjusting the amount of time the converter
HP6060BMagna-PowerPQ375-27
Buck Converter
Electronic LoadDC Power Supply
Agilent 34410A
Agilent 34410A
V DC
I DC
Rload
Fig. 6. Converter test setup.
0 0.5 1 1.5 2 2.50
2
4
6
8
10
12
Panel Voltage [V]
Time [S]0 0.5 1 1.5 2 2.5
0
0.5
1
Panel Current [A]
Panel Current
Panel Voltage
Fig. 7. Current and voltage during MPP convergence.
operates at the native duty ratios of 18.75% and 21.88%, the
average duty ratio can be set at an intermediate value. In
Fig. 4, the converter is averaging over 40 PWM switching
cycles. Therefore, as seen in (2) the average effective duty
ratio resolution is increased to 0.078%.
Figure 5 shows a schematic of the buck converter used
to demonstrate DDRCC, and component values are given in
Table I. DDRCC control for the buck converter was imple-
mented on an MSP430G2553. The G2553 microcontroller is
part of Texas Instruments’ Value Line series and is available in
quantities of 2000 for approximately $1.00 USD. It was chosen
to demonstrate the minimal hardware required to perform
DDRCC.
V. EXPERIMENTAL RESULTS
Panel illumination for testing was provided by DC powered
halogen lights to eliminate the 120 Hz ripple in panel power
caused by AC illumination. The complete test setup is depicted
in Fig. 6.
The MPP convergence of the DDRCC algorithm was tested
over a range of panel power levels by adjusting the light-
ing intensity on the panel. The panel maximum power was
identified by an automated I-V sweep conducted with an
HP6060B electronic load and Agilent 34410A multimeters
to measure current and voltage. After measuring Pmax, the
converter was started and allowed to converge to the MPP.
Figure 7 shows panel current and voltage as the duty ratio
of the converter is raised to DMPP. In this implementation the
0 0.5 1 1.5 2 2.50
1
2
3
4
5
6
7
Time [S]
Power [W]
Fig. 8. DDRCC algorithm converging to the MPP.
duty ratio of the converter is adjusted each dithering cycle, so
by decreasing the number of PWM cycles per dithering cycle
the convergence speed could be increased. As shown in (2)
this would also decrease the effective control resolution of the
converter. Figure 8 shows a plot of panel power vs. time as
panel power converges to the MPP.
Figure 9 shows the operation of the converter when it is
rippling with a duty ratio below DMPP (left), at DMPP (center),
and above DMPP (right). Arrows on each graph highlight the
logic waveform at the top of the screen which marks the
intervals of high duty ratio. This logic signal returns low when
the converter enters low duty ratio for normal operation, or
mid-duty ratio for three-way operation.
As outlined by (18), the DDRCC algorithm measures con-
verter input power at time t = 0 and t = RdTd corresponding to
the beginning and end of the period of high duty ratio. During
this interval, average panel current flowing into the converter
rises, and average panel voltage falls. If P(0) is less than
P(RdTd) it is clear that the panel MPP exists at some higher
current and average duty ratio, than the present operating point.
This situation is shown in the left plot of Fig. 9. As indicated
by (18) the DDRCC algorithm will raise the average duty ratio
of the converter.
When P(RdTd) is equal to P(0), a point of peak power has
occurred during the interval and the converter is operating
at the maximum power point. This is the case shown in the
center plot of Fig. 9. In this instance, the difference portion of
(18) will evaluate to approximately zero and, with a tolerance
implemented, the average duty ratio of the converter will be
unchanged.
The right plot of Fig. 9 shows the operation of the converter
when the duty ratio is above the MPP. It is evident that P(RdTd)
is lower than P(0), which means that the average value of the
duty ratio must be lowered to reach the maximum power point
as indicated by (18).
In addition to assessing each case from the perspective of
(18), it is also helpful to apply the relationships given by (9)
and (10). In the left plot of Fig. 9 it can be seen that dIdt
dPdt
> 0indicating that the average current must be raised to reach the
MPP according to (9). Conversely in the right plot, dIdt
dPdt
< 0indicating that the average duty must be lowered to operate
At MPP Above MPPBelow MPP
Power
Voltage
Current
t=0 t=RdTd t=Td t=0 t=RdTd
t=Td t=0 t=RdTd t=Td
High Duty
Low Duty
Fig. 9. Oscilloscope screenshot of dithering voltage and current, and power waveforms when the average duty ratio (D) is below, at, and above the MPPduty ratio. Note: DC levels are shifted in each plot for suitable viewing, and current scale on right plot has been increased.
the system at the ideal current, I*, according to (10).
TABLE IIEXPERIMENTAL TRACKING EFFICIENCY
Pmpp [W] DDRCC ηMPPT P&O ηMPPT
5.8755 99.87% 97.55%
4.6368 99.84% 97.89%
3.2238 99.97% 96.82%
2.6085 99.50% 97.04%
1.3404 96.26% 93.18%
A comparison between the tracking efficiency of undithered
perturb and observe (P&O) and dithered DDRCC is shown
in Table II [10]. Tracking efficiency (ηMPPT ) is as defined
in (19). This data is acquired using automated triggering of
Agilent 34410A multimeters to average the power flowing
into the converter for several seconds after the converter has
converged to the MPP. It can be seen that the DDRCC tracking
efficiency is excellent over the range tested, and an average of
2.6 percentage points higher than undithered P&O. Both algo-
rithms were tested at a microprocessor clock speed of 8MHz
and a 250 kHz converter switching frequency. It can be seen
that both methods have a decreased efficiency at 1.3 W of input
power. At this very low power level, MPPT performance is
not limited by PWM resolution, but rather by the resolution of
current sensing. As is to be expected, in this situation DDRCC
fares no better than conventional techniques. Increased light
load tracking efficiency would require improved hardware for
current sensing in this implementation.
ηMPPT =Pavg
PMPPT
(19)
It should be noted that the circuit requirements for both
algorithms are the same except for the cutoff frequency of
the low-pass filter at the input of the ADC. When applying
DDRCC control, the objective of the low-pass filter is to
remove extraneous switching noise. It is not necessary to
completely filter the switching waveform because the ADC
measurements are triggered by the PWM module and occur
at the same location in each dithering cycle. By sampling at
approximately the same location on each dithering and switch-
ing cycle the switching waveform stays consistent between
samples and the change in the average ripple is observable.
When performing P&O, the low-pass filter is designed such
that it removes both the switching and dithering signal. This
places a delay in the maximum MPPT speed because the
controller must wait multiple dithering cycles in order to take
advantage of the increased resolution provided by dithering.
Because it is common for PV panels with bypass diodes to
have multiple power peaks during periods of partial shading,
a commercial implementation of DDRCC would likely use a
mode switching algorithm in which the converter first adjusts
its duty ratio for a panel voltage of around 70% of Voc and
then converges to the maximum power point using DDRCC.
This would be necessary to prevent the DDRCC algorithm
from converging to one of the lower power peaks instead of
the true global maximum power point [4].
VI. CONCLUSIONS
DDRCC has been shown to be an excellent method of
accomplishing high control resolution and fast MPPT in PV
applications. Using dithering, DDRCC is able to achieve
high control resolution at low controller clock speeds and
high PWM frequencies. By conducting maximum power point
tracking based on dithering ripple, DDRCC is able to achieve
fast convergence and tracking at a significant fraction of the
switching frequency. Instead of averaging the dithering signal
and then perturbing the duty ratio to track the maximum power
point, DDRCC uses carefully timed measurements to track at
the fastest rate achievable in a dithered system.
REFERENCES
[1] R. C. Pilawa-Podgurski and D. J. Perreault, “Sub-module integrateddistributed maximum power point tracking for solar photovoltaic appli-
cations,” in Energy Conversion Congress and Exposition (ECCE), 2012
IEEE, pp. 4776–4783, 2012.[2] S. Qin and R. C. Pilawa-Podurski, “Sub-module differential power
processing for photovoltaic applications,” in Applied Power Electronics
Conference and Exposition (APEC), 2013 Twenty-Eighth Annual IEEE,pp. 101–108, 2013.
[3] P. T. Krein, “Ripple correlation control, with some applications,” in Proc.
IEEE Int. Symp. Circuits and Systems ISCAS ’99, vol. 5, pp. 283–286,1999.
[4] J. W. Kimball and P. T. Krein, “Discrete-time ripple correlation controlfor maximum power point tracking,” IEEE Transactions on Power
Electronics, vol. 23, no. 5, pp. 2353–2362, 2008.[5] C. Barth and R. C. Pilawa-Podgurski, “Dithering digital ripple correla-
tion control for photovoltaic maximum power point tracking,” in Power
and Energy Conference at Illinois (PECI), 2013 IEEE, pp. 36–41, 2013.
[6] T. Esram, J. W. Kimball, P. T. Krein, P. L. Chapman, and P. Midya,“Dynamic maximum power point tracking of photovoltaic arrays usingripple correlation control,” IEEE Transactions on Power Electronics,vol. 21, no. 5, pp. 1282–1291, 2006.
[7] A. V. Peterchev and S. R. Sanders, “Quantization resolution and limitcycling in digitally controlled PWM converters,” in Power Electronics
Specialists Conference, 2001. PESC. 2001 IEEE 32nd Annual, vol. 2,pp. 465–471, 2001.
[8] Solar Power Industries Inc. Panel model: SPI-020M-9.5[9] Plexim electrical engineering software, PLECS
[10] T. Esram and P. Chapman, “Comparison of photovoltaic array maxi-mum power point tracking techniques,” IEEE Transaction on Energy
Conversion, vol. 22, no. 2, pp. 439–449, 2007.