Professional Documents
Culture Documents
A Global Maximum Power Point Tracking Scheme Employing DIRECT Search Algorithm For Photo Voltaic Systems
A Global Maximum Power Point Tracking Scheme Employing DIRECT Search Algorithm For Photo Voltaic Systems
v + i
n
S
n
P
R
S
n
S
n
P
R
SH
(1)
where I
ph
is the photocurrent generated by a PV cell, I
rs
is
the reversed saturation current of a diode, A is the ideal factor
with a value from one to two, k is the Boltzmanns constant,
q is the electron charge, R
s
and R
SH
are the series and shunt
resistances, respectively, and i and v are the solar cell current
and voltage, respectively.
When the PV array is under a partially shaded condition, the
PV cells under the same insolation condition can be regrouped
to form new modules [27], [28]. Then, these modules can be
considered to be connected in series and/or parallel to form a
new PV array according to the shading pattern. Assume that
the new conguration consists of N
p
branches in parallel and
a maximum of N
s
modules in series in a branch; the output
current i
a
and voltage v
a
can be described as
i
a
=
N
p
b=1
i
b
(2)
v
a
=
N
S
j=1
v
j
(3)
where i
b
is the current in the bth branch and v
j
is the voltage
across the jth module in a branch.
B. Lipschitz Characteristics of PV Function
Multiplying (2) by (3) yields the power p
a
, and then per-
forming partial differentiation with respect to v
a
yields the
following:
p
a
v
a
= i
a
+ v
a
i
a
v
a
v
a
[a, b] (4)
where
i
a
v
a
=
N
p
b=1
N
S
j=1
v
j
i
b
1
v
a
[a, b]. (5)
Consider the jth module in the bth branch; it can be veried that
v
j
i
b
=
1 +
q
kTA
I
RS
R
S
e
K
_
v
bj
+i
bj
n
S_bj
n
P_bj
R
S
_
+
R
S
R
SH
qn
P_bj
kTAn
S_bj
I
RS
e
K
_
v
bj
+i
bj
n
S_bj
n
P_bj
R
S
_
+
n
P_bj
n
S_bj
R
SH
.
(6)
From (2)(6), P
a
/V
a
exists and is bounded by a maximum
value M. By the mean value theorem, for every v
1
, v
2
[a, b],
there exists c [v
1
, v
2
] such that
|p(v
1
) p(v
2
)|
|v
1
v
2
|
=
p
v
a
c
= p
(c), c (v
1
, v
2
). (7)
As p
I
sc_min
=
V
oc_max
I
sc_min
(14)
where and are the scaling factors for the estimated V
oc_min
and I
sc_max
, respectively. Their values are less than one and
determine the range of the estimated interval.
Depending on the dcdc converter topology, the duty cycle
boundaries D
min_abs
and D
max_abs
can be estimated accord-
ingly. If the buckboost topology is implemented, with the
introducing of the converter efciency
bb
as that in [29], the
relationship between the reected equivalent load at the PV
array output R
pv
and R
load
[3] becomes
R
pv
=
bb
_
1 D
D
_
2
R
load
. (15)
Rewriting (15), we have
D =
bb
R
load
_
R
pv
+
bb
R
load
. (16)
From (16)
D
min_abs
=
_
bb
R
load_min
_
R
pv_max
+
_
bb
R
load_min
(17)
D
max_abs
=
_
bb
R
load_max
_
R
pv_min
+
_
bb
R
load_max
. (18)
In some situations, the initial tracking, for example, there
may be insufcient information obtained from the present IV
characteristic, and the absolute sampling interval is feasible
for tracking the GP successfully. However, much informa-
tion can be obtained from previous sampled points in most
cases, and V
oc
or I
sc
can be estimated. Thus, by replacing
V
oc_max
, V
oc_min
with V
oc
and I
sc_max
, I
sc_min
with I
sc
in
(13) and (14), the specic sampling interval (D
min
, D
max
) can
be estimated smaller than (D
min_abs
, D
max_abs
). The smaller
sampling interval will lead to an increase in convergent speed.
B. Partially Shaded Condition Identication
Under partially shaded condition, it is observed that the
IV curves have multiple stairs while the PV curves are
characterized by multiple peaks [27]. Assume that DIRECT
has successfully moved the OP to the MPP D
m1
in previous
steps, as shown in Fig. 3(a). When a partially shaded condition
happens, the OP will move from 1 to 2 due to the change of the
IV curve. The movement results in a reduction of the tracked
power even though the duty cycle is not changed. To track a
new MPP, DIRECT initializes the sampling interval and takes
three new samples D
i
(i = 1, 2, 3) to identify whether the PV
arrays are under uniform or partially shaded conditions.
The analysis based on the observation of a single stair in
a multistair IV curve can always be considered as part of
the IV curve of a module under uniform condition, e.g.,
the gray area in Fig. 3(a). Thus, under a partially shaded
condition, I
D1
would be smaller than I
MPP
(I
MPP
I
D3
) of
the IV curve representing the rst stair, and V
D2
would also be
larger than V
oc
(V
oc
> V
D1
) of the IV curve representing the
middle stair. In other words, I
Di
and V
Di
(i = 1, 2, 3) are well
separated, which means that both (19) and (20) are satised
I
D3,1
I
D3
=
I
D3
I
D1
I
D3
0.1 (19)
V
D1,2
V
D2
=
V
D2
V
D1
V
D2
0.2. (20)
The criterion values 0.1 and 0.2 are chosen based on the
observation that I
MPP
and V
MPP
are about 90% and 80% of
I
sc
and V
oc
of a single IV curve, respectively [12].
If the IV curve has more than three stairs, any two consec-
utive stairs can always be considered as under one larger stair,
as shown in Fig. 4(b). Thus, (19) and (20) easily hold true in
those cases. Fig. 4(a) shows that (19) and (20) also hold true, in
general, for two-stair IV curve cases.
In the case that only (19) or (20) is satised, DIRECT would
need to take further samples to identify the insolation condition.
Fig. 5(a) shows cases when (19) is not satised, which means
that I
D3
is close to I
D1
. Since V
D3
< V
D1
, P
D3
< P
D1
. Thus,
either sample at D
1
or D
2
is the maximum among the rst
three samples. Assume that a sample at D
i
(i is one or two) is
maximum; according to the denition of POIs, DIRECT would
divide interval D
i
into three subintervals and take two new
samples D
i2
and D
i3
at the new middle points, as shown in
Fig. 5(a). Again, with information given by D
1
, D
i2
(or D
i3
),
and D
2
, the insolation condition can be identied properly.
Similar process is applied for cases when only (20) is not
satised.
In general, DIRECT has high chance to identify a partially
shaded condition if there is at least one of the rst three
samples taken in a different stair from others. This condition
3460 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 57, NO. 10, OCTOBER 2010
Fig. 3. (a) OP movement. (b) First three samples illustration.
Fig. 4. Partially shaded condition identication in (a) two-stair IV curves and (b) more-than-three-stair IV curves.
Fig. 5. IV curve has two stairs.
always happens if the initial sampling interval is estimated large
enough. However, if the rst three samples are taken in the
same stair, as shown in Fig. 5(b), DIRECT would identify the
condition as a uniform case and switch to the locally biased
strategy for faster tracking speed. Since the area under the stair
that the rst three samples are taken in is always larger than
others if they present, DIRECT still can reach the GP.
C. Globally Biased Strategy of DIRECT Algorithm
When the PV arrays are under a partially shaded condition,
the globally biased form of the DIRECT algorithm that focuses
on exploring a larger POI is implemented. In the second itera-
tion, DIRECT triply divides POI Dj [j = 1 in the case shown
in Fig. 6(a)] and then samples power values at two new middle
points D
J3
and D
J2
, marked by a cross sign in Fig. 6(a). At
the end of the second iteration, DIRECT records the presence of
ve intervals which can be categorized into two types of interval
length, one-third and one-ninth of the original interval length, as
shown in Fig. 6(b). In Fig. 6, the arrows illustrate the movement
of Dj when it changes from representing an old larger interval
that has been triply divided to representing a smaller interval.
According to (11) and (12), both interval D
2
and D
12
are POIs.
However, the globally biased DIRECT only chooses the one
with the largest interval length, i.e., D
2
, for further sampling in
the third iteration, as shown in Fig. 6(c). This results in six small
NGUYEN AND LOW: GLOBAL MPPT SCHEME EMPLOYING DIRECT SEARCH ALGORITHM FOR PHOTOVOLTAIC SYSTEMS 3461
Fig. 6. Iterations of DIRECT in global-biased form.
intervals and one larger interval, interval D
3
, whose length is
one-third of the original length as shown in Fig. 6(d).
In the fourth iteration, there are two POIs theoretically.
However, DIRECT still divides the one with the larger length,
interval D
3
as shown in Fig. 6(e). The dividing results in nine
intervals whose lengths are all equal to one-ninth of the original
length. Depending on the known parameters available, when
estimated as described in Section IV-A, the initial sampling
intervals length ranges from 0.3 to 0.5. Then, the length of
those nine intervals ranges from 0.03 to 0.05. In most appli-
cations, this value is sufciently small enough so that some
sampling points have a near approach to the GP. In other words,
the intervals that they represent contain the GP. Thus, if two
samples that have the highest powers are consecutive, D
1,1
and
D
1,2
in Fig. 6(e), the GPs region is considered to be found
and falls in the POI. To reach the GP, DIRECT switches to
locally biased form. However, if those two sampling points are
not consecutive, this indicates that the system may have two
GPs or one local MPP whose value is closed to the GP. In this
case, globally biased DIRECT would continue for two more
cycles. This means that DIRECT will further divide and sample
the single POI, interval D
1,2
, in the fth iteration. As a result,
there are two types of interval length again: 1/9 and 1/27 of the
initial sampling intervals length. In the sixth iteration, DIRECT
will further sample the POI that has the larger length (1/9 of
the initial sampling intervals length) to continue exploring the
searching range. Then, DIRECT changes to locally biased form
to focus on reaching the GP.
Generally, in the rst three iterations, DIRECT takes a total
of nine samples to explore the searching region. To prevent
DIRECT from exploring unpromising regions, the number of
samples can be reduced if the rst three sampling points
D
i
(i = 1, 2, 3) give the following information.
1) V
D3
is less than V
min
(V
min
= 1/3V
oc
in our experi-
ments), which means that D
33
would fall in a region that
global MPP cannot appear. Then, sampling point D
3,3
is
not taken.
2) I
D2
< I
min
(I
min
= 1/3 I
sc
), or V
D2
> V
max
(V
max
=
0.9 V
oc
), which means that D
22
would fall in a region that
global MPP cannot appear. Then, sampling point D
2,2
is
not taken.
3) Inequality (19) is not satised, which means that D
3
and
D
1
fall in the current region of the same peak. Then, D
3,3
and D
3,2
are not taken.
4) Inequality (20) is not satised, which means that D
2
and
D
1
fall in the voltage region of the same peak. Then, D
2,2
and D
2,3
are not taken.
3462 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 57, NO. 10, OCTOBER 2010
Fig. 7. PV characteristic curve under uniform insolation.
D. Locally Biased Strategy of DIRECT Algorithm
1) POI: When PV arrays operate under uniform insolation
condition, the IV curve has only a single step, and the PV
curve has a unique MPP, as shown in Fig. 7.
In this case, the unique MPP always falls in or is very close to
the boundary of POI D
J
, which has the maximum power P
DJ
.
In Fig. 7(b), this is the center subinterval D
1
. This means that
DIRECT does not need to explore other intervals D
i
(i = j)
in later iterations, even though they can be potentially optimal
if they satisfy (11) and (12). In summary, the locally biased
strategy always considers the interval with maximum sampling
power at its center being the only POI.
2) Stopping Criteria and Tracking Speed: The locally bi-
ased strategy of DIRECT reduces the sampling interval to
one-third after iteration if the stopping criterion has not met.
When the change in duty cycle d
cr
reaches a predened
sufciently small value, the search is terminated, and it is
assumed that the maximum point has been reached. The critical
value d
cr
should be chosen to be smaller than the duty cycle
step perturbation d in the P&O or INC algorithm, which can
be evaluated as proposed in [9].
Assume that n is the number of iterations when the search is
terminated;
s
is the sampling time for the system, which can
be evaluated as proposed in [9]; we have
D
max_abs
D
min_abs
3
n
d
cr
. (21)
Then, the maximum tracking speed of the algorithm can be
evaluated as
t
abs
= 3n
S
=
3
ln 3
ln
_
D
max_abs
D
min_abs
d
cr
_
S
.
(22)
Fig. 8 shows the simplied owchart of the DIRECT al-
gorithm. To maintain the OP, the P&O method with small
duty cycle perturbation d
cr
is employed. This increases the
systems tracking efciency in a slightly changing insolation
environment.
Fig. 8. Flowchart of the DIRECT algorithm.
V. EXPERIMENTAL RESULTS
A. System Setup
The operation of the DIRECT algorithm has been evaluated
by experiments. A prototype of the MPPT system shown in
Fig. 9 has been implemented.
According to the comparative study of the converter topolo-
gies for MPPT in a PV system [3], the buckboost dc/dc
converter possesses the characteristics for it to follow the
PV arrays MPP at all times, regardless of the cell tempera-
ture, the solar global irradiation, and the connected load. For
this study, we use a buckboost converter with the following
specications: C = 470 F, C
o
= 220 F, L = 1.5 mH, and
20-kHz switching frequency. As the system should reach the
steady state before another MPPT cycle begins, the sampling
interval is chosen as 0.05 s. To evaluate the effectiveness of
the proposed approach, its performance is compared with that
NGUYEN AND LOW: GLOBAL MPPT SCHEME EMPLOYING DIRECT SEARCH ALGORITHM FOR PHOTOVOLTAIC SYSTEMS 3463
Fig. 9. System block diagram of the PV system.
TABLE I
SPECIFICATIONS OF THE 20-W PV PANEL
of the algorithm developed for a partially shaded condition as
proposed in [19]. The experiments were conducted using three
Agilent E4360B solar array simulators (SASs) connected in
parallel as a power source. The SAS is capable of simulating the
IV curves of different arrays under different environmental
conditions if these data points of the IV curve are available
and input to the SAS. The solar array which was used to collect
the actual IV curves is a parallel of three 20-W solar panels
with integrated bypass diodes. The key specications of the
solar panels are shown in Table I. To prevent the systems
voltage drop caused by the shaded panels, blocking diodes were
also installed in each branch. When environmental conditions
were changed, the data points of the new IV curves were
collected so that these IV curves could be built back by the
SAS later in the experiments. Thus, we can reproduce the same
changing scenarios in two experiments implemented with our
proposed algorithm and other algorithm.
Assume that the PV system is operating in the temperature
range of (10
C, 60
C) and insolation range of (0 W/m
2
,
1000 W/m
2
). Moreover, it is assumed that the resistive load is
10 . Then, the absolute sampling interval (0.45, 0.85) can be
estimated using (17) and (18).
B. Partially Shaded Condition Results
The experimental results in Fig. 10(a) and (b) show the track-
ing voltage, current, and power for DIRECT and the algorithm
in [19], respectively. The experiments were conducted with
three consecutive scenarios. In the rst scenario, the SAS gen-
erates the IV curve of the PV array under uniform insolation
condition. This condition was maintained for 1.5 s before it
was changed to partially shaded condition. For partially shaded
condition 2, the IV curve was programmed to have two local
MPPs. After another 1.5 s, the partially shaded condition was
changed such that the IV curve is having three local MPPs.
Fig. 10. Tracking voltage, current, and power. (a) Proposed algorithm.
(b) Algorithm developed in [19].
From the experimental data, the movement of the OP in the rst
three samples for the last two scenarios is shown in Fig. 11(a)
and (b), together with the changing characteristic curves. In
Fig. 11, the thin lines represent the characteristic curves from
the previous scenario, while the bold lines represent the latest
characteristic curves.
The proposed algorithm rst samples three points in the
sampling interval to identify the insolation state, as shown on
the top left in Fig. 10(a). From (19) and (20), the information
obtained from these samples is used to conrm the presence of
3464 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 57, NO. 10, OCTOBER 2010
Fig. 11. Illustration of the OP movement in the experiment. (a) Uniform
insolation to partially shaded condition 2. (b) Partially shaded condition 2 to
condition 3.
the uniform insolation condition. Subsequently, the proposed
algorithm switches to the locally biased mode in the fourth
sampling cycle, as indicated by the vertical arrow in Fig. 10(a).
In Fig. 10(b), the algorithm in [19] rst tracks the nearest local
MPP. It then searches on the left of this local MPP to nd others.
This is illustrated by the gradual increase and then decrease in
the tracking voltage (and power). Since the PV panels are under
uniform insolation, the tracking power drops when the tracking
voltage reduces. The algorithm in [19] detects the presence of
the uniform insolation when the power reduction is larger than
the critical value. Then, it considers the found local MPP as the
GP and set the OP at this local MPP.
In the second scenario, the IV curve changes shape as
the environment changes from uniform insolation to partially
shaded condition. Consequently, the OP shifts from 1 to 1
,
as shown in Fig. 11(a). The movement of the OP results in
a drop of the tracking power. This is indicated by the dotted
circle in Fig. 10(a). This drop in the tracking power activates the
proposed algorithm to take three new samples. The information
derived from the samples veries the presence of the partially
TABLE II
(a) TRACKING PERFORMANCE COMPARISON AT A RATE OF CHANGE
OF 1.5 s. (b) TRACKING PERFORMANCE COMPARISON
AT A RATE OF CHANGE OF 3 s
shaded condition based on (19) and (20). Hence, the globally
biased mode is set to explore the GPs region. This exploration
results in the uctuations of the tracking voltage and current
in Fig. 10(a). At the ninth sampling time, the system nds the
GPs region and switches to local-biased mode to track this GP.
As shown in the PV curve in Fig. 11(a), there is no other
local MPP on the left of the nearest found local MPP. Thus, the
algorithm in [19] keeps searching on the left until it reaches
V
min
, as shown in Fig. 10(b). Then, it searches on the right
to nd other local MPPs. Since the second found local MPP
is higher than the rst, the algorithm continues searching on
the right until it reaches V
max
. The algorithm then considers
the second local MPP as the GP and set it to be the OP. This
searching strategy is illustrated by the experimental results in
Fig. 10(b). In the other two-peak PV curve, the searching
scenario is the same. Thus, this strategy would result in longer
tracking time and larger power loss as compared to the proposed
algorithm.
In the third scenario, the partially shading patterns change
again. This results in the presence of the three local MPPs in
the PV curve, as shown in Fig. 11(b). The OP shifts from
2 to 2