You are on page 1of 9

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. , NO.

, DATE YEAR

Perturb and Observe versus Incremental


Conductance MPPT Algorithms
Thomas Bennett, Ali Zilouchian and Roger Messenger

IV Curve

Current

AbstractThe Perturb and Observe (P&O), and Incremental


Conductance (IncCond) algorithms used for Maximum Power
Point Tracking (MPPT), are perhaps the two most popular
methods in use. The latter method is typically claimed as the
preferred algorithm, but this is not well justified. In this paper,
the reason for the differences between the two closely related hill
climbing methods is shown mathematically. Their responses to
both steady, and changing, irradiance conditions are shown. The
effects of sampling for the algorithm are also addressed
Index TermsMaximum Power Point Tracking (MPPT), Photovoltaic (PV) System, Perturb and Observer, Incremental Conductance

I. I NTRODUCTION
Voltage

dP
di
=
v+i=0
dv
dv
Thus, there is a maximum power value. See Figure 1b for the
power curve of the same PV module.
The problem now lies in trying to operate at the maximum
power point (MPP) (Vm , Im ). Any point on the curve can be
obtained if one had a load that matched it, i.e. RL = vi . The
m
ideal load would then be Rm = VIm
. Unfortunately, due to
T. Bennett is a phd candidate at Florida Atlantic University
Manuscript received Month Day, Year; revised Month Day, Year.

(a) An IV curve of a typical PV module.


Power curve

Power

Aximum Power Point Tracking (MPPT) refers to the


process of maintaining the maximum power output of
an energy source, when its power output changes in time [1].
A PV module can increase its output greatly, when coupled
with a converter that uses MPPT. Like a battery or fuel cell,
a PV module has an IV curve of different current-voltage
pairs where it can operate for a fixed environment (i.e. fixed
irradiance and temperature). In fuel cells and batteries, using
a higher current relates to using up more of the chemical fuel,
and so maximizing output does not coincide with optimizing
(increasing the efficiency of) the system. This is not so for
a PV module; having the maximum output power is desired,
since only energy from the sun gets used (otherwise it would
be converted to heat or reflected). Consider a typical IV curve
of a PV module shown in Figure 1a (effects due to partial
shading will not be discussed in this paper). The power is zero
when either the voltage or current is zero (i.e. at open-circuit
voltage or short-circuit current). The function is concave, zero
on the endpoints, and positive. At some point, the percent
increase in voltage is equal to that of the percent decrease in
current. That is:
di
dv
=
i
v
Or equivalently:

Voltage

(b) The Power curve of the same PV module.

Fig. 1: An IV and power curve for a typical PV module.

changing atmospheric conditions, the IV curve, and hence the


maximum power point (MPP) changes; increasing irradiance
mostly causes the graph to shift up, and increasing temperature
mostly causes the graph to shift to the left. Consequently, Rm
needs to be variable. In order to track the MPP, a converter
with a controllable duty cycle is typically used (either DC/DC
or DC/AC). The output voltage is typically fixed (or nearly
so), allowing the input (PV) voltage to be changed by the duty
cycle of the attached converter. However, trying to operate the
PV array at Vm presents some challenges. In particular, the IV
curve is not only changing, but it is unknown, therefore Vm

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. , NO. , DATE YEAR

is not known. This is the problem for MPPT.


There are numerous methods that address the issue of
operating the PV module at the MPP. The open voltage method
measures the open circuit voltage and operates at a voltage
around 76% of that value [2]. This is based on the observation
that the MPP voltage is almost directly proportional to the
open circuit voltage. The short circuit method is similar, but
uses current instead, operating near 85% of Isc [3]. Fuzzy
controllers may also be used [4], [5]. However, the hill
climbing methods; P&O and IncCond, appear to be the most
common in the literature, and are what are discussed in this
paper.
II. S OME I NCONSISTENCIES
Before getting to the comparison, it seems worth mentioning
a couple inconsistencies that appear in the literature. This supports the notion that the difference between the two algorithms
isnt as well known as it should be.
Many papers claim that the P&O method oscillates at steady
state, even though it seems this was never really an issue. A
1983 paper [6] shows it can be made to converge at steady
state, as well as being acknowledged by [7], the paper that
introduces the IncCond method. However, many articles state
that the P&O method oscillates at steady state [8], [9]. It is
not necessarily wrong to state that the P&O method oscillates
at steady state. However, the implication is that this is not
a problem with the standard IncCond algorithm. However, it
is also an issue with the IncCond in its original form, as is
sometimes acknowledged [10]. If one looks at both of the
flowcharts (see figures 2 and 4), it is easy to see that neither
algorithm, in their most basic form, change the value of V .
Another inconsistency with these methods comes in the
discussion of the needed sensors. For example, [11] states 4
sensors are needed for the IncCond method, which is more
than needed for the P&O method, but they dont expand on
this claim. In [12] it is stated that the same number of variables
are measured in both IncCond and P&O. In [9] they just say
A disadvantage of the INC algorithm, with respect to P&O is
in the increased hardware and software complexity; moreover,
this latter leads to increased computation times and to the
consequent slowing down of the possible sampling rate of
array voltage and current.
It would seem that the same number of sensors would be
required for both methods. Current and voltage need to be
measured for both of the algorithms to work, as seen in the
flowcharts. It is true that the IncCond algorithm is slightly
more complex, but this isnt really much of an issue since most
microcontrollers should have plenty of memory to hold either
algorithm. Also, as will be seen later, the MPPT algorithm
should not be run at a very high speed anyway, and so a
slight increase in computation time should not be an issue. The
frequency of the switching, and perhaps even the sampling,
will likely be much faster than the actual running of the
algorithm.
In this paper, the algorithms will be discussed in more detail
than is normally given. The IncCond method will be seen to
be slightly better than the P&O method, and the reason for
this will be mathematically justified.

III. T HE MPPT C ONTROL A LGORITHMS


1) Perturb and Observe: The P&O method gets its name
from how it works. The algorithm will change (perturb) the
voltage of the PV panel (by changing the duty cycle), and
then measure (observe) how the power changes. If the power
increases, the voltage will continue to be changed in this
direction. If a change in voltage causes a decrease in power,
the voltage will then be changed in the other direction. A
condensed P&O algorithm is shown in Figure 2.1
Measure voltage and current
v(n), i(n)

YES

p(n)-p(n-1)=0?

NO

p(n)>p(n-1) and v(n)>v(n-1)


OR
p(n)<p(n-1) and v(n)<v(n-1)

NO

YES
Vset=Vset+V

Vset=Vset-V

Return

Fig. 2: Perturb and Observe MPPT algorithm.


The P&O method may sometimes move in the wrong
direction. To illustrate this for a fixed IV curve (the case
of changing conditions will be considered later), consider a
scenario in which the voltage is increasing. Suppose going
from v1 to v2 caused the power of the system to increase, and
moving to v3 decreases the power. This could happen in two
ways, as illustrated in the figures (3a) and (3b). Either v2 < Vm
or v2 > Vm . In the scenario in Figure 3b, the voltage should
be decreased, not increased, from v2 , because Vm has been
passed. Once it moves to v3 , it will then, without a decreasing
voltage step size, move back to v2 , which is at a higher power,
and so on again to v1 . For the scenario in Figure 3a, when the
algorithm goes from v3 to v2 it should then turn around, but it
will go to v1 instead. This does not seem too terrible. In either
case, the algorithm oscillates around 3 points. The bigger issue
is how tracking occurs during changing irradiance conditions.
The IncCond method is supposed to have better behavior in
this regard. This will be discussed more later.
2) Incremental Conductance: In [7], the authors state that
the P&O method has a problem at steady state; namely that
it continually oscillates around the maximum power point,
since the voltage is always being changed. However, they
acknowledge that this can easily be remedied by decreasing
the perturbation step size. This paper states that the main
problem with the P&O method occurs during rapidly changing
atmospheric conditions. The problem is that, for example, a
1 The open source program Dia was used to make the flowcharts in this
paper.

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. , NO. , DATE YEAR

Power Curve

Measure voltage and current


v(n), i(n)

195

di=i(n)-i(n-1)
dv=v(n)-v(n-1)

194.5
Power

v3

NO
YES

194

YES

dv=0?

v2

di/dv=-i/v?
NO

v1
YES

NO

di/dv>-i/v?

193.5
24

24.5

25

YES

di>0?

NO
Vset=Vset+V

YES

di=0?

NO

Vset=Vset-V

Vset=Vset-V

Vset=Vset+V

25.5

Voltage

Return

(a) v2 < Vm

Fig. 4: Incremental Conductance MPPT algorithm.

Power Curve
195

Power

194.5

given later will be better understood. In particular, a model for


the PV module, as well as a converter are given. It is to this
PV-converter system that the MPPT algorithms will be applied.
For more details on the PV system setup in this section, one
may refer to [13], [14].

v2
v1

A. Model of PV module

v3

194

Perhaps the most widely used model for the PV module is


given in Figure 5. This is the single diode model.
193.5
24

24.5

25

25.5

Voltage

Rs

(b) v2 > Vm

Fig. 3: Increasing power, and being left of MPP and being


right of MPP.

decrease in power may be due to a decrease in irradiance,


rather than because the operating point moved further from
the MPP. Suppose, for example, that the voltage is increased,
but is still to the left of the MPP. Then it should continue
to increase. However, due to a sharp decrease in irradiance,
the algorithm incorrectly decreases the voltage. (This will be
illustrated later.) To remedy this, they introduce the IncCond
method, illustrated in Figure 4.
The IncCond method claims to make use of the fact that
dP/dv = 0 at the MPP, dP/dv > 0 to the left of the MPP,
and dP/dv < 0 to the right of the MPP. Using dP/dV =
d(IV )/dV = I + V dI/dV , inequalities in terms of i and v
are obtained. The derivatives are approximated numerically by
sampling quickly enough.
IV. M ODELING THE PV SYSTEM
Before comparing the algorithms, a brief discussion of the
PV system is given. This is so that the origin of the figures

Ig

Rp
Fig. 5: A circuit representation of a PV module.

The equation of the circuit in Figure 5 is:


i = Ig Is (e

v+iRs
a

1)

v + iRs
Rp

(1)

In this equation, i is the PV current, v is the PV voltage,


Rs is the series resistor, Rp is the parallel resistor, Ig is the
light-generated current, Is is the diodes saturation current,
and a = AkT /q, where A is the diode ideality factor, k is
Boltzmanns constant, T is the temperature, and q is the charge
of an electron.
For simulation purposes, the no resistor model (NRM)
(Rs = 0, Rp = ) is adequate [13]. The module that will
be used for the simulations in this work is shown in Table
I. Its parameter values for the single diode model with no
resistors (NRM) is given in Table II.

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. , NO. , DATE YEAR

Model
BP-SX3195

Voc (V )
30.7

Isc (A)
8.6

Vm (V )
24.4

Im ()
7.96

BP NRM with Buck Converter and control


26

TABLE I: PV module used for testing.


Is (A)
2.7307e-5

a(V )
2.4249

Rp ()

TABLE II: Parameter values for BP PV model.

B. Model of PV system

v2 = v1 + v d2 = vout /

vout
+ v
d1

L
v

24
0

10

20

25

30

(a) PV voltage output.

BP NRM with Buck Converter and control


195

(2)

iL
Vout

194.8

194.6

194.4

194.2

P&O

Fig. 6: The PV model connected to a buck converter.


194
0

V. T ESTING AT A FIXED IRRADIANCE AND TEMPERATURE


To begin with, a number of comparisons will be made at
standard test conditions (STC). In other words, the MPP is
fixed, and so the algorithms need only converge to a point on
a fixed IV curve, rather than deal with changing conditions,
and so a changing MPP.
A. P&O vs IncCond
It is easy to see that dP
dv > 0 is equivalent to di/dv > i/v,
as shown in Equation 3 (the cases where > is replaced with
< and = are the same):
dP
d(iv)
di
di
i
=
=v
+ i
>
(3)
dv
dv
dv
dv
v
Since the P&O algorithm looks at dP/dv > 0 and the
IncCond looks at di/dv > i/v, they appear to be equivalent.
The P&O algorithm checks on increase (or decrease) in power
relative to changes in voltage, and the IncCond method checks
whether change in current over change in voltage (i.e. a
change in conductance; which is where the name of the
algorithm likely comes from) is greater (or less or equal)
than the negative of current over voltage. However, when the
two algorithms were compared, their behavior is seen to be
different. This comparison can be seen in Figure 7.
0<

15
time (s)

i
Rs
C

IncCond

PV Power

Rp

25

24.5

In this paper a buck converter, with a fixed voltage output,


will be used. This circuit is shown in Figure 6. In the figure,
the two PV resistors were kept, though they are not used in
the simulation as discussed previously. Using the equation of
the buck converter, v = vout
d , where d is the duty cycle, v
is the PV voltage, and vout is the fixed output voltage, one
can obtain Equation 2 as a means of changing the PV module
voltage for use with the MPPT algorithms.

Ig

P&O

25.5

Rs ()
0

PV Voltage

Ig (A)
8.6

10

15
time (s)

20

25

30

(b) PV power output.

Fig. 7: Results of P&O and IncCond methods using a fixed


voltage step size of 0.1 volts.

The original paper [7] does not explain the differences (nor
any other papers that were considered in this research), but
an explanation will be given presently. Suppose the P&O
algorithm moved from position 1, to position 2 (v1 to v2 ).
Then that algorithm is comparing the powers, P2 vs P1 , at
voltages 2 and 1. More exactly, one is comparing v2 i2 to v1 i1 .
Suppose one wishes to see if the change was positive. That is,
the following is checked:
v2 i2 v1 i1 > 0
Compare this to the IncCond case, where the following inequality is checked:
di
i2 i1
i2
=
>
2i2 v2 i1 v2 i2 v1 > 0
dv
v2 v1
v2
The fact that the derivatives are not being used, but rather
numerical methods, is what makes the algorithms different.
In the original paper [7], it is stated Hence, the PV array

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. , NO. , DATE YEAR

v1 i1 = v2 i2 ,

v2 > v1

Now, looking at the power curve, it should be clear that v2


must be to the right of the MPP, and v1 to the left (remember,
at this point STC is assumed). However, the P&O algorithm
does not know this. What about the IncCond method? For it to
work better, it would be necessary that i/v < i/v. Some
simple algebra (requiring the substitution of v1 i1 = v2 i2 ),
yields equation 4, which is clearly true.
i2
i2 i1
< (v1 v2 )2 > 0
(4)
v2 v1
v2
For the other case (v2 < v1 ) the math is quite similar. Now,
v2 v1 is negative, so when multiplying be sure to flip the
inequality sign, and the result is the same. This explains the
slight difference between P&O and IncCond. This is a strong
equality. Hence it is possible for a situation like that illustrated
in Figure 3b to move left, instead of right, at v2 , even if the
power is greater at v2 . This explains why the IncCond turns
around quicker than the P&O algorithm in Figure 7. Note,
that the IncCond method does not always turn around when
it should, but only better than P&O.
Consider Figure 8. The power at v2 is more than that at
v1 , and so the P&O algorithm will perturb once more to the
right (which in this case would cause it to go far off the plot though it should be stated that a 3 volt step might be a bit larger
than normal). What about the IncCond algorithm? It is easy
i
i
= 0.293 and vi22 = 0.283. Hence v
<
to calculate v
i2
v2 , and so the voltage will move left. Assuming the step size
doesnt change, it will go back to v1 . Since, vi11 = 0.358, it
will then move back to the right. In other words, the IncCond
will bounce around v1 and v2 , whereas the P&O algorithm
will bounce around v1 , v2 , and v3 (not shown) causing more
power loss. It is important to note that if v2 were slightly
smaller, but still at a larger power, then IncCond would have
failed just as P&O did.

Power Curve
200
X: 24.84
Y: 194.6

195

Power

terminal voltage can be adjusted relative to the [MPP] voltage


by measuring the incremental and instantaneous array conductances (dI/dV and I/V, respectively).... The derivatives were
approximated using the differences between the voltages and
currents at positions 1 and 2. Many other papers seem to reuse
this explanation, as if P&O and IncCond were not equivalent
in the differential form. One should use i/v rather than
the derivative notation, since it is the difference equation form
that sets the two algorithms apart, not the differential form.
Now that the equations have been seen to appear differently,
it will be shown mathematically that the IncCond is better in
turning around once the MPP is passed.
To do this, a comparison is made for the cut off case of the
P&O algorithm; the case where v1 6= v2 , but P1 = P2 . It is
this case where the algorithm is at the threshold of moving
the voltage left or right. Suppose that v2 > v1 , as the other
case is similar. If the voltage were a pinch more to the left,
power would have increased, and so voltage would have been
increased after v2 ; if it were a pinch to the right, power would
have decreased, and so voltage would have been decreased
after v2 . How does the IncCond algorithm fare for this case?
In this case,

190

X: 26
Y: 191.4

vm

X: 23
Y: 189.5

v2
v1

185

180

175
22

23

24

25
Voltage

26

27

28

Fig. 8: An example of IncCond behaving better than P&O.

It should also be mentioned, though it wont be proven here,


that the IncCond method does not turn around too early. For
example, if Vm > v2 > v1 , then v3 will be greater than v2 .
The algorithm wont cause the voltage to turn around before
getting to the MPP.
See Figure 7 for how the two compare using a step size of
0.1 volts. Notice that the IncCond voltage doesnt go as high
above Vm as the P&O voltage does at steady state. It should be
noted that it took a handful of simulations to get good results
like these. The original results obtained were used a step size
of 0.3 volts, and an average sampling method (discussed later)
to obtain the measured value. The results of this are shown in
Figure 9. In this case the two algorithms were different, but
just as bad, with the P&O going one more voltage to the right
of the MPP than the IncCond method, but the IncCond method
going one more to the left. This is likely due to an issue with
the measured sample (some transient affecting the results). It
is not because the IncCond method behaves more poorly than
the P&O when decreasing the voltage. In the model used in
this work, the transient behavior is more pronounced than it
would be in practice. Some current work is being done by the
author showing how converter losses bring down the transient
drastically. Hence, with a real system, and smarter sampling,
this shouldnt be an issue. The other tested cases had identical
results for the P&O and IncCond methods. So, the advantage
of the IncCond method is not that pronounced.
VI. T ESTING WITH CHANGING CONDITIONS
There are many papers that compare algorithms [3], [7]
[9], but the method of comparison is usually not well-defined,
and there is no standard method for comparison. Recently, a
standard method for testing the algorithms was presented [15].
This proposal suggests the irradiance input shown in Figure
10. Temperature changes are relatively slow, and so the effects
of these changes are not considered.
The times when it changes are at the following seconds:
60,140,200,204,264,266,386,388,448,452,512,592; starting at
0 and ending at 652.

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. , NO. , DATE YEAR

BP NRM with Buck Converter and control

Ropp Input

30

1100
1000
Irradiance (W/m2)

28

PV Voltage

P&O
26

24

900
800
700
600
500
400

IncCond

22

300
200

20
0

10

15
time (s)

20

25

30

(a) PV voltage output.

100
0

100

200

300
400
time (s)

500

600

700

Fig. 10: Changing Irradiance Input for MPPT testing.

BP NRM with Buck Converter and control

BP NRM with Buck Converter and P&O method

195
30
194

193

PV Voltage

PV Power

25

IncCond

P&O
192

20
15
10

191
5
190
0

10

15
time (s)

20

25

30

0
0

100

200

(b) PV power output.

300
400
time (s)

500

600

(a) Voltage.

Fig. 9: Results of P&O and IncCond methods using a fixed


voltage step size of 0.3 volts.

BP NRM with Buck Converter and P&O method


200

A. P&O vs IncCond
150
PV Power

In the paper introducing the IncCond method, [7], it was


claimed that this algorithm was developed to deal with poor
tracking of the P&O algorithm during changing irradiance
conditions. One can show mathematically that the IncCond
is better during changing conditions. The mathematics are the
same as in the STC case, except this time v1 is on IV curve 1,
and v2 is on IV curve 2, where, assuming conditions changed
during this time, the curves are not the same. However, as
stated before, the differences arent that great. The results of
the two algorithms are given in figures 11 and 12, where a step
size of 0.3 volts was used. The steady state behavior appears
the same. The only noticeable difference appears to be around
500 seconds, where the IncCond is at a slightly lower voltage
during the first part of the downward ramp.
The P&O algorithm had an efficiency of 98.81%, and the
IncCond algorithm gets 98.85% efficiency. The efficiency is
calculated starting at 60s, since initial conditions should not

100

50

0
0

100

200

300
400
time (s)

500

600

(b) Power.

Fig. 11: Results of P&O algorithm with Ropp input.

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. , NO. , DATE YEAR

count against the algorithms used (since the algorithms dont


have a standard for how they start). It is seen that the IncCond
method is better, but not substantially so. Still, if there is no
disadvantage to using the IncCond method in place of the P&O
method, it should be done.
BP NRM with Buck Converter and IncCond method
30

PV Voltage

25
20
15
10
5

VII. R EMARKS ON SAMPLING


In the description of the algorithms, little is said about
how the sampling is done. Papers claim to be comparing
measured voltage and current values to the prior measured
values, but not how these values are obtained. In the above,
averages were used, and were a second apart. In particular, 0.5
seconds was given for the transient to end, and then voltages
and currents were recorded for 0.5 seconds at 1kHz. These
values were then averaged. (Originally, checking the standard
deviation was also done to make sure one was at steady state
before changing the voltage, but this proved not to work well
during changing irradiance conditions.) In this section, only
one sample will be used to calculate the voltage and current
used within the algorithm. This will make use of voltage and
current values sampled at 10ms, 100ms, and 1s; with the
algorithm running after each sample. The results for the P&O
algorithm (the IncCond are similar) are shown in Figure 13
for steady irradiance conditions.
BP NRM with Buck Converter and control

0
0

100

200

300
400
time (s)

500

30

600

(a) Voltage.

1000ms

28

10ms

BP NRM with Buck Converter and IncCond method


PV Voltage

200

PV Power

150

26

24

22
100

100ms

Vm
20
0

50

8
time (s)

10

12

14

Fig. 13: Results of P&O method using one voltage/current


sample.
0
0

100

200

300
400
time (s)

500

600

(b) Power.

Fig. 12: Results of IncCond algorithm with Ropp input.


Notice that the algorithms move in the wrong direction near
the beginning. This is due to the fact that increased irradiance
might give an increase in power, even if moving in the wrong
direction. In other words, the increase in energy due to increase
in irradiance is greater than the decrease in energy due to
moving away from Vm . Even though the IncCond is better at
recognizing when it passes the MPP, it is not perfect. Both
also fail to track the decreasing ramp that occurs around 500s.
This is because the power is decreasing due to irradiance
regardless of what direction you move in (this is dependent
on slope of ramp, as well as step size chosen). Another issue,
as stated previously with regards to the unmodified version
of these algorithms, is the oscillation at steady state (for both
algorithms, not just P&O!).

Why do the shorter sampling times yield worse results?


Suppose one is trying to increase the voltage. Then sampling
too soon reads a smaller voltage, due to transient delay, than
what the actual steady state value is. Perhaps now the voltage
is set to the right of the MPP, but the actual value was still
left of the MPP. So, the voltage is increased again; perhaps
a couple more times. Now the voltage is finally sampled to
be to the right of the MPP, and so the algorithm starts trying
to go back. The voltage is decreased, but it is still swinging
up, and so the power decreases. So, now it tries increasing
again, etc. This type of behavior can result in going far past
the MPP before everything is aligned enough to begin coming
back. A slightly slower sampling may cause the values to be
sampled nearer the overshoot, i.e. beyond where the voltage
is set, rather than premature. It may now try to turn around
sooner than it should. Once again resulting in going the wrong
way a bit, though not quite as bad as sampling too soon.
It appears that sampling every second yields the best results.
In fact, those results, using just one sample, yield slightly

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. , NO. , DATE YEAR

better results than the averaging used previously. Averaging


might still be preferable in actual system, where noise could
interfere, depending on sensitivity of equipment used. Perhaps
a smaller interval of averaging could also be considered. The
best approach will depend on the frequency of the switch, as
well as L and C values. Also, as mentioned before, a real
system may have slightly better transient behavior, allowing
for quicker sampling, and running of the MPPT algorithm.
Regardless, it seems that neither of the algorithms will
be operating at a very high frequency (the switch converter
perhaps, and maybe even sampling for measurements, but
not the actual running of the MPPT algorithm). Perhaps
the IncConds slightly increased computational time has a
negligible penalty.
VIII. C ONCLUSION
For non-varying conditions, the two algorithms performed
similarly, and both oscillated around the MPP. The IncCond
method has a slight advantage, and could potentially oscillate
slightly less due to turning around quicker once passing the
MPP. However, for a small step size, the difference would be
fairly negligible. For varying conditions, the IncCond method
also did slightly better for the same reason. However, it does
not seem to have a large advantage to the P&O method.
Assuming there are no added costs in implementing the
IncCond method over the P&O method, it should be the
preferred method. However, if some of the claims that the
IncCond method indeed adds a performance or cost penalty,
then it may not be worth it.

R EFERENCES
[1] R. Messenger and J. Ventre, Photovoltaics Systems Engineering, 2nd ed.
CRC, 2004.
[2] D.-Y. Lee, H.-J. Noh, D.-S. Hyun, and I. Choy, An improved mppt
converter using current compensation method for small scaled pvapplications, in 18th Annual IEEE Applied Power Electronics Conference and Exposition, 2003.
[3] V. Salas, E. Olias, A. Barrado, and A. Lazaro, Review of the maximum
power point tracking algorithms for stand-alone photovoltaic systems,
Solar Energy Materials & Solar Cells, vol. 90, 2006.
[4] X. Wei and H. Jing, Mppt for pv system based on a novel fuzzy control
strategy, in 2010 International Conference on Digital Manufacturing &
Automation, 2010.
[5] F. Bouchafaa, D. Beriber, and M. S. Boucherit, Modeling and simulation of a g[ri]d connected pv generation system with mppt fuzzy logic
control, in 2010 7th International Multi-Conference on Systems, Signals
and Devices, 2010.
[6] O. Wasynczuk, Dynamic behavior of a class of photovoltaic power
systems, IEEE Transactions on Power Apparatus and Systems, vol.
PAS-102, no. 9, September 1983.
[7] K. H. Hussein, I. Muta, T. Hoshino, and M. Osakada, Maximum photovoltaic power tracking: an algorithm for rapidly changing atmospheric
conditions, Proc. Inst. Elect. Eng., vol. 142, no. 1, January 1995.
[8] D. P. Hohm and M. E. Ropp, Comparative study of maximum
power point tracking algorithms using an experimental, programmable,
maximum power point tracking test bed, in Photovoltaic Specialists
Conference, 2000.
[9] N. Femia, G. Petrone, G. Spagnuolo, and M. Vitelli, Optimization of
perturb and observe maximum power point tracking, IEEE Transactions
on Power Electronics, vol. 20, no. 4, July 2005.
[10] F. Liu, S. Duan, F. Liu, B. Liu, and Y. Kang, A variable step size
inc mppt method for pv systems, IEEE Transactions on Industrial
Electronics, vol. 55, no. 7, July 2008.
[11] C. Hua and C. Shen, Study of maximum power tracking techniques
and control of dc/dc converters for photovoltaic power system, in 29th
Annual IEEE Power Electronics Specialists Conference, 1998.
[12] J. Pan, C. Wang, and F. Hong, Research of photovoltaic charging
system with maximum power point tracking, in The Ninth International
Conference on Electronic Measurement & Instruments, 2009.
[13] T. Bennett, A. Zilouchian, and R. Messenger, Photovoltaic model and
converter topology considerations for mppt purposes, Solar Energy,
vol. 86, 2012.
[14] T. Bennett, Developing a photovoltaic mppt system, Dissertation,
August 2012.
[15] M. Ropp, J. Cale, M. Mills-Price, M. Scharf, and S. G. Hummel,
A test protocol to enable comparative evaluation of maximum power
popint trackers under both static and dynamic irradiance, in IEEE 37th
Photovoltaic Specialist Conference, 2011.

Thomas Bennett Thomas received his bachelors


degree in mathematics from Florida Gulf Coast
University in 2006, and his masters in mathematics
from Florida Atlantic University in 2008. He is set
to finish his Ph.D. in electrical engineering from
Florida Atlantic University in 2012. His current
research is in the areas of renewable energy, power
systems, power electronics and control.

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. , NO. , DATE YEAR

Ali Zilouchian Ali Zilouchian received his Ph.D.


from George Washington University, Washington
DC in 1986. He has been with the College of Engineering and Computer Science at Florida Atlantic
University for the past 26 years. He is currently the
Associate Dean for Academic Affairs, and a Professor in the Department of Computer and Electrical
Engineering and Computer Science.
His recent projects have been funded by Dept.
of Energy, FPL, National Science Foundation, and
the Broward County School district. In addition to
his current research in the area of alternative energy and sustainability, he
has conducted research on the applications of soft computing methodologies
to industrial processes including, desalination processes, oil refineries, jet
engines, model reduction of multivariable systems, and 2-D digital filters.
Dr. Zilouchian is a senior member of IEEE since 1995, and currently
an associate editor of the International Journal of Electrical and Computer
Engineering out of Oxford, UK.

Roger Messenger Roger Messenger received his


Ph.D. in electrical engineering from the University
of Minnesota. He is Professor Emeritus of Electrical
Engineering at Florida Atlantic University where he
taught for 35 years. During his time at FAU, he
worked his way through the academic ranks and also
served in administrative posts for 11 years, including
Department Chair, Associate Dean and Director of
the FAU Center for Energy Conservation.
He is author, along with Dr. Jerry Ventre of the
Florida Solar Energy Center, of the book Photovoltaic Systems Engineering, now in its 3rd edition. Since his retirement from
FAU in 2005, he has been involved in the design of more than 200 PV systems,
ranging from small, stand-alone systems to complex battery-backup, gridconnected systems as well as several systems in the megawatt capacity range.
He is currently serving as Senior Associate/PV Specialist at FAE Consulting
in Boca Raton, FL, where he spends most of his time supervising the design
of PV systems for residential, commercial and government use.

You might also like