International Conference on Renewable Energies and Power Quality

European Association for the Development of Renewable Energies, (ICREPQ’10)

Environment and Power Quality (EA4EPQ) Granada (Spain), 23th to 25th March, 2010

Identification of Photovoltaic Array Model Parameters.

Modelling and Experimental Verification

I. Houssamo1, M. Sechilariu1, F. Locment1 and G. Friedrich2

University of Technology of Compiegne

1
AVENUES Team Research
BP 60319 - 60203 Compiegne (France)
Phone/Fax number: +33 344234964/+33 344235262
e-mail: issam.houssamo@utc.fr, manuela.sechilariu@utc.fr, fabrice.locment@utc.fr
2
LEC, EA 1006
BP 20529 – 60205 Compiegne Cedex (France)
Phone/Fax number: +33 344234515/+33 344237937
e-mail: guy.friedrich@utc.fr

published literatures, a model based on the electrical

Abstract. This paper aims to identify the model parameters equivalent circuit of the cell is proposed in this paper. In
of a photovoltaic array installed in our experimental platform. [1], a two-parameter PV model was characterised by
Two methods used for parameters identification in order to adjusting the current-voltage (I-V) characteristic curve at
characterize the photovoltaic array are compared. Based on the the three points given by the manufacturer: open circuit,
electrical equivalent circuit of the photovoltaic cell, the maximum power point and short-circuit. According to the
mathematical model is deduced. From the data-sheet values, cell equivalent circuit, an analytical four and five-
given by the manufacturer, we were able to determine the parameter PV panel models were deduced and
parameters of this model. On the other hand, the second method experimentally verified [2]. A simplified graphical
was applied by using one of the least squares fitting approaches. method was used in [3]. Other publications use the robust
The measurement of the outdoor solar irradiance, cell
temperature and current taken by dSPACE controller board and
linear regression methods [4] or the artificial neural
for a constant terminal voltage level includes the necessary data network [5].
to be fitted with the model. The implementation of two methods
in MATLAB provides the model parameters which have to In this paper, two methods of PVA parameters
minimize as soon as possible the error involved between the identification are analyzed and compared in order to
calculated and measured output current. For the minimum choose the most appropriate one to use in modelling of
obtained error, the corresponding method is the best one for the PVA installed in our experimental platform. Based on a
photovoltaic array characterization. single-diode PV cell circuit, a mathematical model is
deduced. At the three operating points mentioned above,
Key words the first method is applied and allows finding its
parameters. Concerning the other one, the error occurred
Photovoltaic, Modelling, Least squares fitting, between the calculated and measured output PVA current
Simulation. was minimized by using one of the least squares fitting
approaches. Both methods are programmed in MATLAB.
1. Introduction First, the PVA is described. Then, the mathematical
model is presented. Subsequently, the measurement
For self-feeding building with renewable electricity, the bench is given. After that, each algorithm is explained.
semi-isolated and safety network is proposed to Finally, for both cases, the output PVA currents are
overcome the technical constraints related to connecting compared with the measured ones in order to find the
distributed sources with the utility grid. For this purpose best PVA parameter identification method.
we are interested to modelling a photovoltaic array
(PVA). The knowledge of the model parameters is 2. PVA description
essential to evaluate the PVA behaviour under all
operating conditions. However, the manufacturers give The experimental platform, shown in Fig. 1, has two
just the electrical features of photovoltaic (PV) panel PVA installed on the roof of centre Pierre Guillaumat 2 at
under the standard test conditions (STC). Thus, many the University of Technology of Compiegne in France.
works focused on the identification of the unknown
parameters of the PV or PVA model. As for many
 current is noted iD. It is function of T, as given in (3). The
dark saturation current, isat, is expressed in (4).
⎛ G ⎞
(
iL = I sc∗ ⎜ ∗ ⎟ 1 + K i T − T ∗
⎝G ⎠
( )) (2)

⎛ ⎛ v + irs ⎞ ⎞
iD = isat ⎜ exp ⎜ ⎟ − 1⎟⎟ (3)
⎜
⎝ ⎝ nVT ⎠ ⎠
Fig. 1. Experimental platform
⎛3⎞
The PVA consists of 8 PV panels Solar-Fabrik SF-130/2- ∗⎛T ⎞⎜⎝ n ⎟⎠ ⎛ −qEg ⎛ 1 1 ⎞ ⎞
125 electrically coupled on four parallel branches formed
I sc ⎜ ∗ ⎟ exp ⎜ nk ⎜ T − ∗ ⎟ ⎟
⎝T ⎠ ⎝ ⎝ T ⎠⎠
by two series PV panels (Fig. 2). isat = (4)
⎛ ⎛ qV ⎞ ⎞
∗
1 2 3 I PVA = I n pp ⎜ exp ⎜ oc∗ ⎟ − 1⎟
4
⎜ ⎜ nkT ⎟ ⎟
I =i ⎝ ⎝ ⎠ ⎠

The STC are assumed to be the reference solar irradiance

V G*=1000W/m2 and the reference cell temperature
T*=298K at air masse equal to 1.5. Under these
PVA VPVA = V n ps
conditions, the manufacturer specifies some electrical
features of the PV panel: short circuit current Isc*, open
circuit voltage Voc*, temperature coefficient for current Ki
expressed in mA/K, maximum output current Impp and
maximum output voltage Vmpp. The quantities k, q are
Fig. 2. Photovoltaic array Boltzmann’s constant and the electron charge. The
thermal voltage is VT=kT/q. Eg, is the band gap energy of
In order to protect the PV panels from the reverse semiconductor (it equals to 1.12eV for the silicon). The
current, one diode is placed in the head of each branch. unknown parameter, n, is the diode ideality factor which
Several poly-crystalline silicon cells are interconnected in takes a value between 1 and 2. Other unknown
series and packaged in each PV panel. parameters are the series resistance, rs, and the parallel
resistance, rp. The series resistance represents the
semiconductor material resistance and those for contact
3. PVA modelling interfaces. The parallel resistance represents the leakage
current across the p-n junction.
Many models of varying complexity describing the
behaviour of a PV cell are available. So, the number of Let ns be the number of cells in series in one PV panel.
parameters that need to be identified is different. In this The output panel voltage is V = v ns , and the
paper, a single-diode PV cell model is proposed. It is the characteristic (I-V) of PV panel is given by (5).
most classical model available in the literature [1]-[5]-
[6]-[7]-[8] and has the electrical scheme shown in Fig. 3. ⎛ ⎛ V + IRs ⎞ ⎞ V + IRs
I = iL − isat ⎜ exp ⎜ ⎟ − 1⎟⎟ − (5)
It consists of a current generator for modelling the ⎜
⎝ ⎝ nnsVT ⎠ ⎠ Rp
incident solar irradiance, a diode for the polarization
phenomena, a series resistance and a parallel resistance
for representing the power losses. where Rs = ns rs and R p = ns rp are the panel series and
rs i parallel resistances. Considering that the numbers of the
ip PV panel in series and respectively the parallel branches
G iL iD
in the PVA are known as nps and respectively as npp, the
vD rp v general characteristic (IPVA-VPVA) of the PVA is:

T ⎡ ⎛ ⎛ VPVA n pp + I PVA Rser ⎞ ⎞ ⎤

I PVA = n pp ⎢iL − isat ⎜ exp ⎜ ⎟ − 1⎟⎟ ⎥
⎢⎣ ⎜ nNVT
⎝ ⎝ ⎠ ⎠ ⎥⎦
Fig. 3. Three-parameter PV cell electrical equivalent model (6)
VPVA n pp + I PVA Rser
Using Kirchhoff’s first law, the terminal current of the −n pp
cell is: R par
The PVA series and parallel resistances are
v + irs Rser = ns n ps rs and R par = N rp , where N = ns n ps n pp .
i = iL − iD − (1)
rp So, the (IPVA-VPVA) characteristic represents an implicit
where iL, given in (2), is the light-generated current. It is and non linear equation with two variables, G and T, and
directly proportional of the incident solar irradiance (G), three unknown parameters: n and Rser and Rpar. This
linearly related to cell temperature (T), and depends on function can be expressed as follows:
the materials used and fabrication processes. The diode
 I PVA = f (G, T , I PVA , VPVA , P ) (7)
5. Identification process
where P is the unknown parameters vector:
A. Determination of the PV panel model parameters from
P = [n, Rser , R par ] (8) data-sheet values

4. Test bench
For this first method, the values of the model parameters
The parameters identification can be carried out by are calculated by using the specifications described in
means of least squares fitting and by using the data-sheet. Accordingly, the operation conditions are the
experimental measurements recorded at our test bench STC. So, the thermal voltage becomes:
shown in Fig. 4. These measurements are Im, Vm, Gm and
kT ∗
T m. VT ∗ = (10)
q
Writing the equation (5) for short-circuit point, maximum
power point and open circuit point gives these three
equations:

⎛ ⎛ I ∗R ⎞ ⎞ I ∗R
I sc∗ = iL − isat ⎜ exp ⎜ sc s∗ ⎟ − 1⎟ − sc s (11)
⎜ ⎜ nn V ⎟ ⎟ Rp
⎝ ⎝ s T ⎠ ⎠

⎛ ⎛ Vmmp + I mmp Rs ⎞ ⎞
I mmp = iL − isat ⎜ exp ⎜ ⎟⎟ − 1⎟
⎜ ⎜ nnsVT ∗ ⎟
Fig. 4. Test Bench ⎝ ⎝ ⎠ ⎠
(12)
Vmmp + I mmp Rs
All the experimental measurements are controlled in real −
time by dSPACE controller board. The measurements Rp
relating to the irradiance Gm and cell temperature Tm are
carried out by a pyranometer and respectively by a ⎛ ⎛ V ∗ ⎞ ⎞ V ∗
0 = iL − isat ⎜ exp ⎜ oc ∗ ⎟ − 1⎟ − oc (13)
PT100 sensor. The electrical equivalent circuit test is ⎜ ⎜ nn V ⎟ ⎟ R
shown in Fig. 5. ⎝ ⎝ s T ⎠ ⎠ p

From (13) the current iL and the current isat can be

I PVA Rl expressed:
i
vd
Im ⎛ ⎛ V ∗ ⎞ ⎞ V ∗
iL = isat ⎜ exp ⎜ oc ∗ ⎟ − 1⎟ + oc (14)
PDCEL ⎜ ⎜ nn V ⎟ ⎟ R
⎝ ⎝ s T ⎠ ⎠ p
Vm CVS
By insertion (14) into (11), we obtain this equation:
VPVA
⎛ ⎛ V ∗ ⎞ ⎛ I ∗R ⎞ ⎞
I sc∗ = isat ⎜ exp ⎜ oc ∗ ⎟ − exp ⎜ sc s∗ ⎟ ⎟
⎜ ⎜ nn V ⎟ ⎜ nn V ⎟ ⎟
PVA ⎝ ⎝ s T ⎠ ⎝ s T ⎠⎠
(15)
Fig.5. Measurement circuit ∗ ∗
Voc − I sc Rs
+
Rp
Using a controlled voltage source (CVS) and a
programmable DC electronic load (PDCEL), for a From the above equation, the current isat results in:
constant voltage level (Vm), these measurements are
made. In this test the CVS is just used for setting the load isat =
(
I sc∗ Rs + R p − Voc∗ )
voltage at a constant level and it doesn’t supply any ⎛ ⎛ V ⎞ ∗ ⎛ I ∗R ⎞ ⎞ (16)
current. All the PVA power is absorbed and dispersed by R p ⎜ exp ⎜ oc ∗ ⎟ − exp ⎜ sc s∗ ⎟ ⎟
⎜ ⎜ nn V ⎟ ⎜ nn V ⎟ ⎟
the PDCEL, thus I PVA = I m . The fact that the PVA is ⎝ ⎝ s T ⎠ ⎝ s T ⎠⎠
installed on the roof compels us to measure the voltage
after the protection diode and the connecting cable. So,
the voltage drop across the diode (vd) and the connecting
cable resistance (Rl), are accounted for as follows:
VPVA = Vm + I m Rl + vd (9)
Insertion (14) and (16) into (12), the maximum output Start

current will take the form: Input tol

ε = 2 × tol
⎛ ⎛ V ∗ ⎞ ⎛ Vmmp + I mmp Rs ⎞ ⎞ Initialize n
⎜ exp ⎜ oc ∗ ⎟ − exp ⎜ ⎟⎟ ⎟
⎜ ⎜ ⎟ ⎜ nnsVT ∗ ⎟
⎝ ⎝ nnsVT ⎠ ⎝ ⎠⎠
I mmp = while n <= 2
No
⎛ ⎛ V ∗ ⎞ ⎛ I ∗R ⎞ ⎞
R p ⎜ exp ⎜ oc ∗ ⎟ − exp ⎜ sc s∗ ⎟ ⎟ Yes
⎜ ⎜ nn V ⎟ ⎜ nn V ⎟ ⎟ (17) Search the minimum MAE
⎝ ⎝ s T ⎠ ⎝ s T ⎠⎠ G = G ∗ and T = T ∗
Output the corresponding
values of Rs , R p and n

Voc∗ − Vmmp − I mmp Rs

( ( )
Initialize Rs

)
× I sc∗ Rs + R p − Voc∗ +
Rp Input Gm , Tm , I m and Vm
Calculate R p from (18) End

Calculate VPVA from ( 9 )

From this equation we can express the parallel resistance: Calculate I PVA from ( 6 )
No
while ε > tol
Calcul the error ( MAE )
A Yes
Rp = (18) between I PVA and I m
B Calculate I from ( 5)

Where, A and B are given in (19) and (20). Increment n for V ∈ [0,Voc∗ ]

⎛ V ∗ ⎞
(
A = Rs I sc∗ + I mmp − Vmmp exp ⎜ oc ∗ ⎟
⎜ nn V ⎟ ) Calculate the output power P = I × V
find the maximum power Pmax
⎝ s T ⎠ Calculate the error ε = I mmp × Vmmp − Pmax

⎛ I ∗R ⎞
( ( )
− Rs I mmp − Vmmp + Voc∗ exp ⎜ sc s∗ ⎟
⎜ nn V ⎟
⎝ s T ⎠
) (19)
Increment Rs

⎛ Vmmp + I mmp Rs ⎞ Fig. 6. Algorithm for determination of the PV panel parameters

( )
+ Voc∗ − I sc∗ Rs exp ⎜
⎜ nnsVT ∗
⎟⎟
⎝ ⎠ The best values of the parameters Rs, Rp and n are those
for which the model involves the most minimal error. In
⎛ ⎛ V ∗ ⎞ ⎛ I ∗R ⎞ ⎞ the beginning of the algorithm, the error (ε) between the
B = I mmp ⎜ exp ⎜ oc ∗ ⎟ − exp ⎜ sc s∗ ⎟ ⎟
⎜ ⎜ nn V ⎟ ⎜ nn V ⎟ ⎟ calculated and maximum power given by the
⎝ ⎝ s T ⎠ ⎝ s T ⎠⎠ manufacturer, takes an initial value higher than the
(20) tolerance criterion (tol) in order to assure the programme
⎛ ⎛ Vmmp + I mmp Rs ⎞ ⎛ V ∗ ⎞⎞ running once at least.
+ I sc∗ ⎜ exp ⎜ ⎟⎟ − exp ⎜⎜ oc ∗ ⎟⎟ ⎟
⎜ ⎜ nnsVT ∗ ⎟
⎝ ⎝ ⎠ ⎝ nnsVT ⎠ ⎠
B. Determination of the PVA model parameters based on
the measurements
In [1], an algorithm is used in order to identify two-
parameter model under the STC. Those parameters were The output PVA current carried out by (7) enables us to
Rs and Rp. However, the diode ideality factor is taken as compute the error committed by the model as follows:
known parameter. As we know that n ∈ [1, 2] . Using this
z

∑( I )
algorithm, when this parameter takes all the values 2
y
r= m − I PVA y (22)
included in this interval, the three unknown parameters
y =1
are calculated by MATLAB implementation as shown in
Fig. 6. This error represents the objective function, which must
Firstly, for each value of n the resistances Rs and Rp are be minimised. Since the function (7) is implicit, the
determined under the STC (all the values Isc*, Voc*, Impp calculation of the current IPVA requires an iterative
and Vmpp are used) in order to obtain a maximum output method which imposes a substantial calculating time.
power equal to that one available in the data-sheet. Furthermore, this calculation would be repeated for each
iteration of the minimisation method. For these reasons
Then, the error of current, whose the model causes is and taking into account that I PVA = I m , the function that
evaluated. So, the current IPVA is calculated according to gives the PVA current is:
(6) by using the Newton-Raphson method. For a package
I PVA = f (Gm , Tm , I m , Vm , P) (23)
of measurements, the mean absolute error (MAE) is
calculated as in (21) and stocked. Thus, the objective function to be minimised becomes:
z

∑( I )
2
z I my − I PVA y r= y
− f (Gmy , Tmy , I my , Vmy , P)
MAE = ∑
y =1
z
(21)
y =1
m (24)

Using the MATLAB “lsqcurvefit” function the retained

where z is the measured points number, it depends on the
parameters are those which give the better fit of this
measurement period and the sampling step of the
function.
dSPACE controller board.
6. Experimental verification and discussion 20
Measured current
Data-sheet current
Fig. 7 and Fig. 8 show the operating conditions (Gm, Tm) Fitting method current
during 8 hours on 20th of November and on 15th of 15

December, 2009, in Compiegne, France. During these

Current (A)
measurements, the voltage Vm was settled at 30V. The 10
obtained parameters by both methods are presented in
Tables I-II. The current calculated using these parameters
is compared with measured one (Fig. 9 and Fig. 10). 5

The Tables I and II show that the error related to the first 0
method for the two measurements is slightly smaller than 9:00 10:00 11:00 12:00 13:00
Time of day
14:00 15:00 16:00 17:00

those committed with the second one. Furthermore, the Fig. 10. Current of PVA calculated from data-sheet values and
parameters values belonging to the same method change by fitting method versus that measured on 15th of December
according to the operating conditions.
Table I. – Parameter Values of PV Panel Derived from the
Table III shows the error MAE obtained when the current Data-Sheet Values and Related Error for Each Recorded
is calculated under operating conditions (measurements Measurement
on 15th of December) to which the parameters belonged
to 20th of November measurements; similarly in the Table PARAMETER MEASURES OF MEASURES OF
IV, made with the measurements recorded on 20th of AND ERROR NOVEMBER DECEMBER
November. MAE THE 20th THE 15th
n 1.8 1.65
750 30 Rs 1.54.10-3Ω 52.10-3Ω
T
m Rp 4.001.103Ω 0.121.103Ω
Cell temperature (C)
Irradiance (W/m²)

G
500
m
20
MAE 0.8A 1.407A

Table II. – Parameter Values of PV Panel Derived With the

250 10 Fitting Approach Method and Related Error for Each Recorded
Measurement

0
9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 PARAMETER MEASURES OF MEASURES OF
Time of day AND ERROR NOVEMBER DECEMBER
MAE THE 20th THE 15th
Fig. 7. Operating conditions recorded on 20th of November
n 1.802 1.627
Rs 0.506.10-3Ω 110.10-3Ω
600 15
T
m
Rp 156.103Ω 397.103Ω
Cell temperature (C)
Iirradiance (W/m²)

G
m
MAE 0.84A 1.413A
400 10
Table III. – Mean Absolute Error Committed with the
Parameters Determined on November 20 under Operating
200 5 Conditions Taken on December 15

0 0
PARAMETERS OF 20/11/2009 MAE
9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 n Rs (Ω) Rp (Ω) (A)
Time of day
th
1.8 1.54.10-3 4.001.103 1.536
Fig. 8. Operating conditions recorded on 15 of December 1.802 0.506.10-3 156.103 1.533

25 Table IV. – Mean Absolute Error Committed with the

Measured current
Data-sheet current
Parameters Determined on December under Operating
20 Fitting method current Conditions Taken on November 20

PARAMETERS OF 15/12/2009 MAE

Current (A)

15
n Rs (Ω) Rp (Ω) (A)
10
1.65 52.10-3 0.121.103 0.92
1.627 110.10-3 397.103 0.81
5
Even that the parameters related to the fitting method
0
give a little smaller error than those resulted using the
9:00 10:00 11:00 12:00 13:00
Time of day
14:00 15:00 16:00 17:00 data-sheet parameters values, and the calculation time is
shorter than the time elapsed while the data-sheet
Fig. 9. Current of PVA calculated from data-sheet values and algorithm is running, the quality of the minimisation by
by fitting method versus that measured on 20th of November
the MATLAB function “lsqcurvefit” depends on the
initial values, lower and upper limits.
Thus, the model is not robust and could lead to some [8] D. Sera, R. Teodorescu and P. Rodriguez, “PV panel model
wrong predictions of PVA behaviour. A bad conclusion based on data-sheet values”, in Proc. International Electronics,
may be drawn if the model built on the parameters ISIE, Vigo, Spain, 2007.
resulted for a measurement, is tested under different
conditions.

As in our research work, we don’t need to use the PVA

model in real time, the model identified using the data-
sheet algorithm is simple and can be suitable to
modelling the behaviour of PVA under all the operating
conditions.

7. Conclusion
The study of the semi-isolated and safety network for the
self-feeding building with renewable electricity,
especially generated by a PVA, requires a model that
allows knowledge of the PVA behaviour under various
meteorological conditions.

This paper aims to identify numerical and experimental

parameters of the one diode model PV cell extended for a
PVA. Using the data-sheet values and one of least
squares fitting methods, the error between the calculated
and measured current of the output PV array, is
minimised. Under various meteorological conditions,
based on the carried out parameters, the two methods are
compared in regarding to error between the calculated
and measured output PVA current, calculation time and
easiness of implementation. The method based on the
data-sheet values is more appropriate to identify the PVA
model parameters

References

[1] M.G. Villalva, J.R. Gazoli and E.R. Filho, “Comprehensive

Approach to Modeling and Simulation of Photovoltaic Array”,
IEEE Trans on Power Electronics, Vol. 24, n°5, pp. 1198-1208,
May 2009
[2] A.N. Celik and N. Acikgoz, “Modelling and experimental
verification of the operating current of mono-crystalline
photovoltaic modules using four- and five-parameter models”,
Applied Energy, Vol. 84, n°1, 2007, pp. 1-15.
[3] V. Autier, V. Molcrette and H. Roisse, “Contribution to the
determination of the elements of photovoltaic panel (PV)
equivalent electric model. Application in the maximal power
point (MPP) available analytical expression”, in Proc.
ICREPQ2003.
[4] M.C. Di Piazza, A. Ragusa and G. Vitale, “Identification of
Photovoltaic Array Model Parameters by Robust Linear
Regression Methods”, in Proc. ICREPQ2009.
[5] E. Karatepe, M. Boztepe and M. Colak, “Development of a
suitable model for characterizing photovoltaic arrays with
shaded solar cells”, Solar Energy, Vol. 81, n°8, pp. 977-992,
2007.
[6] O. Gergaud, B. Multon and H. Ben Ahmed, “Analysis and
Experimental validation of various photovoltaic system
models”, in Proc. 7th International ELECTRIMACS, Montréal,
August, 2002.
[7] T. Tafticht, K. Agbossou, M.L. Doumbia and A. Chériti,
“An improved maximum power tracking method for
photovoltaic systems”, Renewable Energy, Vol. 33, pp. 1508-
1516, 2008.