Composite Structures 119 (2015) 477486

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Model order reduction applied to heat conduction in photovoltaic

modules
S.O. Ojo a, S. Grivet-Talocia b, M. Paggi c,
a
Department of Structural, Geotechnical and Building Engineering, Politecnico di Torino, C.so Duca degli Abruzzi 24, 10129 Torino, Italy
b
Department of Electronics and Telecommunications, Politecnico di Torino, C.so Duca degli Abruzzi 24, 10129 Torino, Italy
c
IMT Institute for Advanced Studies Lucca, Piazza San Francesco 19, 55100 Lucca, Italy

a r t i c l e i n f o a b s t r a c t

Article history: Modeling of physical systems may be a challenging task when it requires solving large sets of numerical
Available online 16 September 2014 equations. This is the case of photovoltaic (PV) systems which contain many PV modules, each module
containing several silicon cells. The determination of the temperature eld in the modules leads to large
Keywords: scale systems, which may be computationally expensive to solve. In this context, Model Order Reduction
Photovoltaics (MOR) techniques can be used to approximate the full system dynamics with a compact model, that is
Thermal system much faster to solve. Among the several available MOR approaches, in this work we consider the Discrete
Model order reduction
Empirical Interpolation Method (DEIM), which we apply with a suitably modied formulation that is spe-
Numerical methods
cically designed for handling the nonlinear terms that are present in the equations governing the ther-
mal behavior of PV modules. The results show that the proposed DEIM technique is able to reduce
signicantly the system size, by retaining a full control on the accuracy of the solution.
2014 Elsevier Ltd. All rights reserved.

1. Introduction So far, semi-analytical and numerical solutions [2,3] for the

A photovoltaic (PV) system usually consists of an array of PV
modules (e.g. 10), and each module contains several solar silicon
cells (e.g. 60 or 70 in commercial modules). Each module is a lay-
ered composite such that the silicon cells are sandwiched between
the different layers (see Fig. 1 for a schematic representation of a
module cross-section). The computation of thermal stresses and
thermo-elastic displacements is of paramount importance both
during the production process, and during the operating conditions
of the module. In the former case, compressive residual thermo-
elastic stresses may arise and are benecial in case of cracks within
silicon due to the activation of partial crack closure [1]. In the lat-
ter, cyclic thermoelastic stresses are responsible for crack growth
in silicon cells and a power-loss of the PV system in time. More-
over, thermoelastic deformation may induce failure of the busbars
connecting solar cells, due to an increase in the gap between cells,
as experimentally and numerically studied in [2,3]. In all of these
cases, it is important to accurately compute the temperature distri-
bution in the plane of the solar cells [4], but also the temperature
in the various layers [5] for the study of fully coupled thermome-
chanical problems.
Corresponding author. Tel.: +39 0583 4326 604; fax: +39 0583 4326 565.
E-mail address: marco.paggi@imtlucca.it (M. Paggi).
assessment of the change in the gap between solar cells have been
proposed by assuming a uniform temperature eld across the
module, which is an assumption holding for stationary conditions.
In reality, temperature contour plots obtained from nite element
thermal analysis [6] show that there is a temperature gradient
across each layer, with the regions near the frame being signi-
cantly cooler, while the temperature distribution across the cells
in the center of the module is found to be quite uniform. In addi-
tion, existence of cracks in the silicon cells may induce a localized
temperature increment (hot spots) in the region near the cracks
due to a localized electric resistance [7]. Moreover, transient
regimes, such as those taking place in accelerated environmental
tests within climate chambers, or under operating conditions, have
only marginally been investigated due to the inherent complexity
related to the very different thicknesses of the layers composing
a PV module. In such cases, accurate predictions require the solu-
tion of large systems of equations resulting from the nite element
or nite difference approximation of the (nonlinear) partial differ-
ential equation governing the problem of heat conduction. Suitable
techniques for reducing the computational requirements for such
simulations are therefore highly desirable.
The purpose of this study is to formulate a thermal model of a
PV system, and to identify an appropriate class of Model Order
Reduction (MOR) techniques, capable of approximating the full

http://dx.doi.org/10.1016/j.compstruct.2014.09.008
0263-8223/ 2014 Elsevier Ltd. All rights reserved.
478 S.O. Ojo et al. / Composite Structures 119 (2015) 477486

system by a compact dynamical model characterized by a much capacity (J/m3 K), which is equal to an equivalent mass density
smaller number of degrees of freedom (state variables), and whose times the equivalent specic heat capacity (C = q  cp) taking into
numerical solution requires a signicantly reduced time with account the composite structure of the laminate, see next section.
respect to the full system. The function H(x, z, t) represents the heat losses by radiation and
Choosing a particular MOR technique that is suitable for a given convection taking place at any point within the PV module, and
problem depends on several factors, such as type of system (linear G(x, z, t) is the electrical energy generated by the cell layer. The coef-
or nonlinear), number and structure of equations to be solved for cients kx(x, z) and kz(x, z) are the thermal conductivities in the x and
and, in case of nonlinear systems, the degree of nonlinearity. Model z directions respectively. According to Fouriers law of heat conduc-
order reduction is a mature eld when applied to linear systems, tion, the heat ows in the x and z direction are related to these
and several excellent books on the subject are available [811]. thermal conductivities by
Many techniques are available for linear order reduction, including @T @T
Krylov subspace projection based on orthogonal Arnoldi [12,13] or qz kz ; qx kx 2
@z @x
biorthogonal Lanczos [14] processes, principal components analysis
and balanced truncation [15], Hankel norm approximation [16], and Substituting (2) into (1), we have:
singular value decomposition (SVD) based methods, which include @T @q @q
Proper Orthogonal Decomposition (POD) in its many variants [17]. C  x zHG 3
@t @x @z
Many extensions to nonlinear system are also available, see e.g.
[1822], which combine system projection or truncation with suit- Using now a nite difference (FD) discretization scheme dened by
able approximations of the nonlinear terms. grid spacing Dxi and Dzj in the x and z-direction, respectively, with
In this work, we propose a POD projection based approach for associated discretization indices i for 1 6 i 6 l and j for 1 6 j 6 s (see
the reduction of the model dynamic order, combined with a Dis- Fig. 2), we can rephrase (3) as
crete Empirical Interpolation Method (DEIM) [22] for handling dT i; j qi12; j  qi12; j qi; j12  qi; j12
the nonlinear terms. We show that this approach is ideally suited C i; j Hi; j  Gi; j 4
dt Dxi Dz j
to the inherent structure of the PV discretized heat transfer equa-
tions obtained through a nite difference scheme, allowing for where qi1; j and qi1; j are the heat ows through the left and right
2 2

massive order reduction at almost no loss of accuracy. A systematic boundaries of element (i, j), and qi; j1 and qi; j1 are the heat ows
2 2

validation study is conducted on a PV template system, showing through the upper and lower boundaries of the element in its plane.
the inuence of the various parameters governing the reduction Multiplying (4) by the area Ai,j = DxiDzj of each grid element in the
scheme. Numerical results conrm the excellent scalability of the FD discretization leads to
proposed technique. @T i; j
C i; j Ai; j Q i1; j  Q i1; j Q i; j1  Q i; j1 Ai; j Hi; j  Ai; j Gi; j 5
@t 2 2 2 2

2. Formulation of the differential problem and nite difference

where Q represents, consistently with energy conservation princi-
approximation
ples, the heat ow between adjacent cells, which can be further
expressed as Q = KDT [24], where DT is the temperature change
A 2D thermal model of a PV module is proposed here based on
between the two cells, and K is the corresponding thermal conduc-
the work by Jones [23]. We consider a PV module containing 12 sil-
tance. The latter is a function of the equivalent thermal conductiv-
icon cells embedded in a composite made of glass, EVA, Silicon,
ities of the two cells and the width and length of the elements, i.e.
EVA, backsheet and tedlar layers with the properties described in
Dzj and Dxi.
Table 1 [24]. Although solar cells are separated from each other
by a small amount of EVA in their plane, in this work we slightly
2.1. Thermal conductances and heat ows
simplify the structure and we consider all layers as uniform in
the x and y directions, see Fig. 1. Further, we consider the y direc-
The discretized thermal conductances Ki,j (W/m K) provide
tion as innite.
information on the thermal coupling between the elements in
Under the above assumptions, the following general 2D heat
the discretization of the PV module. Assuming perfect bonding at
equation for the composite PV panel holds:
the various interfaces between the layers, the thermal conductance
per unit length in the x-direction between cells (i  1, j) and (i, j) is
@T @2T @2T
C kx 2 kz 2 H  G 1 given by [25]
@t @x @z
Dz j
where T(x, z, t) represents the space and time dependent tempera- K i1; j 6
2 Dxi1 =2kx i1; j Dxi =2kx i; j Ri1; j
ture prole of the module. C(x, z) is an equivalent volumetric heat 2

Glass layer (h=4mm)

EVA (h=0.5mm)

Si cell
(h=0.166mm)

Backsheet
(h=0.1mm)
125 mm 2mm Tedlar (h=0.1mm)
x-axis
1580 mm
z-axis

Fig. 1. A sketch of a cross-section of a PV module, not in scale. For the actual value of the thicknesses, see the labels in the gure.
 S.O. Ojo et al. / Composite Structures 119 (2015) 477486 479

Table 1
Material properties of the layers of the PV module.

Layer Thickness (mm) Thermal conductivity (W/m K) Density (kg/m3) Specic heat capacity (J/kg K)
Glass 4 1.8 3000 500
EVA 0.5 0.35 960 2090
PV Cells 0.166 148 2330 677
EVA 0.5 0.35 960 2090
Back contact 0.1 237 2700 900
Tedlar 0.1 0.2 1200 1250

2.2. Boundary conditions

z-direction In this study, a constant (Dirichlet) temperature is applied to

the right and left boundary of the module. Thus, the heat ow at
the left and right boundary elements of the PV module (1, j) and
(l, j) are

Q 1; j K j T b1  T 1; j 15
2

x-direction Q l1; j K j T l; j  T b2 16
2

Fig. 2. Finite difference discretization of the PV module in its plane.
where R is the thermal resistance at the interface between the ele-
ments. Since in the present approximation, the PV layers are uni-
form in the x-direction, the thermal conductivity kx does not
change in the x direction and we have Ri1; j 0 and we can simplify
2
notation as kx i1, j = kx i, j = kj. This assumption is still reasonable for
where Tb1 and Tb2 are the xed temperatures imposed at the right
and left of the module. In all subsequent simulations, we will set
them equal to 343 K and 313 K in order to simulate a distinct differ-
ential temperature prole from one end of the PV module to the
other.
The heat ow at the top and bottom boundary elements of the
PV module (i, 1) and (i, s) are instead
Q i; 1 K 1 T sky  T i;1 17
2 2

the silicon cell layer, since the cells are separated from each other Q i; s1 K s1 T i; s  T sky 18
2 2
by a small amount of EVA (2 mm), much smaller than the lateral
size of each silicon cell (125 mm). Grid spacing Dx is also consid- where Tsky is the temperature of the sky.
ered to be constant in the x direction. Therefore (6) becomes
2.3. Heat loss
Dz j Dz j
K i1;j kj 7
2 Dx=2kj Dx=2kj Dx
The heat loss, which varies through the layer thickness of the
and K i1; j K i1; j K j due to material homogeneity in the x- module, is given by the sum of the following contributions [23]
2 2
direction.
Hx; z; t qlw x; z; t qsw x; z; t qconv x; z; t 19
In the z-direction, we have:
where the long wave, short wave and convection heat transfers are
Dx
K j1 8 denoted by qlw, qsw and qconv respectively.
2 Dzj1 =2kz j1 Dzj =2kz j Rj1
2 The short wave radiation heat transfer of a body of area A is
Dx given by
K j1
2 Dzj1 =2kz j1 Dzj =2kz j Rj1
2 qsw AaU 20
At the top and bottom boundary elements of the PV module, (i, 1) where a and U are the absorptivity of the material and the total
and (i, s), we have incident irradiance input to the module surface, respectively. The
Dx long wave radiation heat transfer is given by the Stefan Boltzmann
K1 9 law
2 Dz1 =2kz 1 R1
2

Dx qlw reT 4 21
K s1 10
2 Dzs =2kz s Rs1
2 where r is the Boltzmanns constant (5.607  10 J s m K4) 8 1 2

where R1 and Rs1 are the thermal resistances between the top and and e is the emissivity of the body. It is assumed that the net long
2 2
bottom elements and the free surfaces. wave exchange is negligible for the rear of the module. Thus, it is
From (5), the heat ows through the left and the right bound- only necessary to calculate the long wave exchange from the sur-
aries of the element (i, j) are thus dened as face of the module. The net long wave radiation exchange between
two surfaces x and y is given by [23]
Q i1; j K j T i1;j  T i; j 11
2
qlwxy Ax F xy Lx  Ly Ay F xy Ly  Lx 22
Q i1; j K j T i;j  T i1; j 12
2
Here Lx and Ly are long wave irradiance per unit area for surface x
whereas the heat ows through the lower and upper boundary of and y respectively which are given by
the element (i, j) are
Lx rex T 4x and Ly rey T 4y
Q i; j1 K j1 T i; j1  T i; j 13
2 2
where Fxy is the view factor, a fraction of the radiation leaving sur-
Q i; j1 K j1 T i; j  T i; j1 14
2 2 face x that reaches surface y.
480 S.O. Ojo et al. / Composite Structures 119 (2015) 477486

A tilted module surface not overlooked by adjacent buildings at 800

an angle h from the horizontal has a view factor of 1cos2
h
for the Experimental data
Simulated data
sky and 1cos
2
h
for the horizontal ground [23]. Thus, inserting the 700

view factor coefcient for sky and ground into Lx we have

600
1 cos h 1  cos h

Irradiance (W/sq.m)
Lx r esky T 4sky r eground T 4ground 23
2 2 500
4
Ly remod T 24
400
where emod is the module emissivity.
Substituting (23) and (24) into (22), we have 300
 
1 cos h 1  cos h
qlw Ar esky T 4sky eground T 4ground  emod T 4 200
2 2
25 100

Further, Tsky = Tambient  dT for clear sky condition in which dT = 20 K

0
and Tsky = Tambient for overcast condition. 0 5 10 15 20 25 30
The convection heat transfer is related to the temperature gap Time (mins)
between the upper part of the solar panel and the ambient [23]
qconv Ahc;forced hc;free T  T ambient 26 Fig. 3. Experimental and simulated irradiance input (from 09:5210:22, 11/01/
2011).
where hc,forced and hc,free in W/m2 K are the forced and free convec-
tion heat transfer coefcients which depend on the wind speed.
Collecting all the heat loss contributions together, we nally
obtain At the left and right boundaries of the PV module, Q i1; j and Q i1; j
2 2
   from (5) are replaced by Q 1; j and Q l1; j respectively from (15) and
1 cosh 1  cosh 2 2
Hi; j Ai; j ai; j U r esky T 4sky eground T 4ground  ei; j T 4i; j (16), while at the top and bottom boundaries of the PV module,
2 2
   Q i; j1 and Q i; j1 are replaced by Q i; 1 and Q i; s1 respectively from
2 2 2 2
 hc; forced hc; free T i; j  T ambient 27 (17) and (18).
The discretized thermal Eq. (29) can nally be written in a com-
where ai, j and ei, j denote the absorptivity and emissivity coef- pact matrix form
cients of the discretized elements in the different layers. dTt
CA KTt AITt; Et 30
dt
2.4. Power generated by the PV Cell
or, in explicit form, as
The power generated by the PV cell at location (i, j) can be esti- dTt
mated as [23] KTt FTt; Et 31
dt |{z} |{z}
E lncE linear term Non-linear term
Gi; j C FF 28
T i; j where K CA1  K and F = C1I. The independent variable
where CFF is the ll factor model constant (1.22 K m ) and E(t) in W/ 2 t 2 [0, h] denotes time, and Tt T 1 t; . . . ; T n tT 2 Rn is the
m2 is the incident irradiance input through the thickness of the PV unknown temperature vector for all the elements in the FD discret-
module. The constant c is equal to 106 m2/W. It should be noted ization, where we use a single subscript with n denoting the total
that the power generated by the discretized PV cells in (28) is number of nodes. E(t) is the time-varying irradiance input to the
non-zero only for the silicon cell layer. system, the matrix K 2 Rnn contains constants and F(T(t), E(t)) is
The incident irradiance input into the system is obtained from a nonlinear function evaluated at T(t) component-wise i.e.
experimental data [23]. To obtain a validated result of the reduced
F F 1 ; . . . ; F n T 32
order model to be derived in Section 3, a minute by minute irradi-
ance input obtained from the solar resource and meteorological
assessment project website (http://www.nrel.gov/midc/kalael- 2.6. Reference solution for the complete system
oa_oahu/) will be used. The plots for irradiance for a period of
30 min are shown below in Fig. 3. Before applying our proposed model order reduction approach,
we derive a reference solution for (31) by direct time discretiza-
2.5. System of nonlinear ODEs for the PV module tion. A backward nite difference scheme is selected to solve the
thermal problem to avoid any convergence issues associated with
Considering all the relations established in Section 2, the dis- explicit methods in terms of choice of time step. The numerical
cretized thermal Eq. (5) can be rewritten after substituting the cor- method is implemented in Matlab. A uniform discretization of
responding expressions for Q, H and G as the module in the x-direction is adopted with l = 361 grid points,
  while there are s = 6 strips in the z-direction with different thick-
@T i; j
C i; j Ai; j  K j T i1; j  K j T i1; j T i; j 2K j K i; j1 K i; j1 nesses so that n = s  l = 2166. The solution of this problem is per-
@t  
2 2
formed for ns = 186 time steps, each step representing 10 s of
1 cos h
 K j1 T i; j1  K j1 T i; j1  Ai; j ai; j U r esky T 4sky physical time. Fig. 4(a) shows the temperature prole for node
2 2 2
 741 in the silicon layer for all the 186 time steps. The temperature
1  cos h along the silicon layer vs. position at the last time step is shown in
eground T 4ground  ei; j T 4i; j
2 Fig. 4(b). As it can be seen, the transient regime is quite evident and

 E lncE the temperature in the silicon cell layers is signicantly different
 hc; forced hc; free T i; j  T ambient Ai; j C FF 29
T i; j from cell to cell.
 S.O. Ojo et al. / Composite Structures 119 (2015) 477486 481

Temperature for node 741

340 last time step
345

330 340

Cell Temperature(K)
335
Temperature (K)

320 330

325
310
320

300 315

310
290 305
0 50 100 150 200 0 0.5 1 1.5
Time step Length of the module(m)
(a) (b)
Fig. 4. Temperature prole of the full system for (a) a node within the silicon layer vs. time step and (b) along the silicon layer at the last time step of the simulation.

3. Model order reduction In particular, we collect the ns snapshots T(th) obtained from the full
solution of the system at discrete time steps of size h in the follow-
The direct numerical simulation of (31) may be quite demand- ing snapshot matrix
ing in terms of computing resources, especially in view of its exten- S Tt 1 ; . . . ; Ttns  35
sion to a full 3D geometry. For this reason, we investigate in this
work a MOR technique, with the objective of approximating the and we apply the POD algorithm below
large-scale system (31) with a lower order compact dynamical
INPUT: S Tt1 ; . . . ; Ttns   Rnns
model, that is able however to preserve accuracy in its input/out-
OUTPUT: V k v 1 ; . . . ; v k  2 Rnk
put transient response. The two key aspects of proposed MOR
1. Form the snapshot matrix S = {T(t1), . . . , T(tns)}
approach are: (i) a massive reduction in the degrees of freedom
2. Perform the singular value decomposition T = VRWT to
(states), and (ii) an accurate representation of the nonlinear terms
produce orthogonal matrices V v 1 ; . . . ; v r  2 Rnr and
that inuence the heat exchange of the structure. These two
W w1 ; . . . ; wr  2 Rns r and diagonal matrix
aspects are analysed in detail.
R diagr1 ; . . . ; rr 2 Rrr where r is the rank of S.
3. Set a threshold to pick the k highest modes from the diag-
3.1. System projection
onal matrix R
4. Pick the columns in matrix V which correspond to the
The reduction in the degrees of freedom is here performed
modes selected in 3 to generate the POD basis
through a standard projection approach. The vector T collecting
V k v 1 ; . . . ; v k  2 Rnk
all n cell temperatures is approximated as a linear superposition
of a small number k of basis vectors, which span a reduced order
subspace. More precisely, we consider the representation T  V k T e,
Note that in the present case we choose ns = 186, as the total
where T e 2 Rk is a reduced temperature vector collecting the coef-
number of time steps in the full solution. The choice of ns should
cients of T into a reduced basis, dened by the columns of matrix be carefully considered since it can strongly inuence the accuracy
V k 2 Rnk . We consider an orthonormal basis, so that of the approximation and the computational cost, as shown later.
V Tk V k I 2 Rkk k  n with I an identity matrix. Introducing the
above reduced expression for T into (31), we have 3.2. Discrete Empirical Interpolation Method (DEIM)
e t
dT
Vk e t FV k T
 KV k T e t; Et 33 System (34) is a reduced order model, but the evaluation of the
dt
nonlinear term still requires the mapping Vk to the full-size space.
Projecting now these equations along the subspace generated by Vk The DEIM approach is used here to further approximate the nonlin-
leads to ear terms, thus reducing the computational cost associated with
the simulation of the reduced model. According to [10], we write
d Te t e t V T FV k T
e t; Et
 V Tk KV k T k
the nonlinear term of (34) as
dt |{z}
e and where K
e 2 Rkk e V T FV k T
N T e t; Et 36
with V Tk KV k K k
|{z} |{z}
The reduced form of the thermal Eq. (31) reads kn n1

e t
dT and we dene
eT
K e t V T FV k T
e t; Et 34
k
dt e t; Et
f t FV k T 37
The above system represents a reduced order model, since its main
variables are the coefcients of a reduced basis. In order to deter- The basic idea is to approximate f(t) by projecting it onto the sub-
mine Vk, we use a Proper Orthogonal Decomposition (POD), which space spanned by a suitable set of m  n basis vectors u1, . . . , um via
extracts the basis vectors from the actual transient solution of the
full system by means of a truncated singular value decomposition. f t  Uct 38
482 S.O. Ojo et al. / Composite Structures 119 (2015) 477486

where U u1 ; . . . ; um  2 Rnm . The corresponding coefcient vector term in the thermal system formulation has most impact on the
c(t) is determined by selecting m signicant rows from the overde- surface of the PV module. On the other hand, the power output is
termined system (38). This can be achieved by considering the map- generated only by the silicon cell (third layer). On this note, the
ping matrix DEIM operation is here performed separately for the two nonlinear
terms using two different sets of snapshots. Accordingly, we have
P e.1 ; . . . ; e.m  2 Rnm 39
Eq. (43) respectively expressed for the two nonlinear terms in
T n
where e.i 0; . . . ; 0; 1; 0; . . . ; 0 2 R is the .ith column of the iden- the reduced system as
tity matrix I 2 Rnn for i = 1, . . . , m. The coefcient c(t) can thus be  1  
e t; Et  U1 PT U1
F 1 V k T Fe 1 PT1 V k T
e t; Et 45
determined by inverting system 1
 1  
PT f t PT Uct 40 e t; Et  U2 PT U2
F 2 V k T Fe 2 PT2 V k T
e t; Et 46
2
T
Provided P U is non-singular, the nal approximation of (38) is Finally, the nonlinear term in (36) can now be represented as
T 1 T  1  
f t  Uct UP U P f t 41 e V T U1 PT U1
N T k 1 Fe 1 PT1 V k T e tEt
e t; Et, (41) can thus be written as
Since f t FV k T |{z} |{z}
km1 m1 1
e t; Et  UPT U PT FV k T 1
e t; Et  1  
FV k T 42 V Tk U2 PT2 U2 e 2 PT V k T
F e t; Et 47
2
Eq. (42) ensures that the nonlinear function F is evaluated for the |{z} |{z}
km2 m2 1
full system and then interpolated by matrix P, an operation which
still shows the dependence of the reduced system on the complete With this modication, the interpolation of the nonlinear terms can
system size. To avoid this dependence, DEIM interpolates the input be handled independently, enabling a ner control on reduced sys-
vector of the nonlinear function F and then evaluates F component- tem complexity and efciency.
wise at its interpolated input vector. Based on this, (42) can be writ-
ten as 4. Numerical results

e tEt  UPT U1 Fe PT V k T

FV k T e t; Et 43 In this section, we assess the accuracy of the proposed reduced
e denotes the selected components of F. This approximation modeling scheme by comparing the responses of the compact
where F
model and the original system. In particular, we investigate the
is particularly effective when the full nonlinear function F is evalu-
convergence of the reduced system as a function of the three
ated independently for each component of its vector argument, as
parameters that measure its complexity, namely the size k of the
in present FD formulation. The nonlinear term in (36) can now be
reduced basis used in the state-space (linear) projection, and the
represented as
two orders m1, m2 of the nonlinear interpolation.
e t V T UPT U1 Fe PT V k T e t; Et As shown in Fig. 5, the cell temperatures at the last time step of
N T k 44
|{z} |{z} the iterative solution of the full system and of the reduced system
km m1
are in fair good agreement by increasing the order k of the reduced
Now, to evaluate N T e in (44), we must specify the projection system. As the order of the reduced system increases from k = 1 to
basis [u1, . . . , um] and the interpolation indices [.1, . . . , .m]. We k = 7, the approximation of the reduced system approaches the
can obtain the basis [u1, . . . , um] by applying the above described exact value of the complete system. At k = 7, the reduced system
POD scheme to the matrix collecting the nonlinear snapshots approximation ts well the complete system such that further
F = {F(T(t1)), . . . , F(T(tns))} resulting from a direct evaluation of the increasing the order of the reduced system does not change the
nonlinear function of the full system at different time steps, and result signicantly.
then using the DEIM algorithm described in [22]. The following It is noticed that the number of snapshots ns used to construct
implementation is used to iteratively construct the basis vectors the compact model also affects its convergence to the full solution.
and the set of interpolation indices. Convergence is achieved more efciently using a high number of
samples in the snapshots matrix S than using a small number of
INPUT: fui gm n
i1  R linearly independent samples. The plots in Fig. 6 illustrate the convergence of the
!
OUTPUT: . .1 ; . . . ; .m  2 Rm reduced solution using 185, 100, 70 and 40 snapshots while xing
1. [jqj,.1] = max{ju1j} the order of k = 7 and interpolation order m1 = m2 = 5.
!
2. U ju1 j; P e.1 ; . .1 
3. for i = 2 to m do 4.1. Validation of the reduced model using simulated irradiance data
4. Solve (PTU)c = PTui for c
5. r = ui  Uc In this section, we investigate the sensitivity of the reduced
6. [jqj,.i] = max{jrj} " # model to the input irradiance signal. To this end, the reduced
!
U ui ; P P e.i ; . .
!
7. U model is rst constructed based on snapshots derived from exper-
end for
.i imental irradiance data, see Fig. 3. Then, simulated irradiance data
is used to excite the model, and the corresponding response is
compared to the full system response computed by direct time dis-
3.3. Modication of DEIM formulation cretization. The irradiance data for this validation was carefully
selected to have different environmental characteristics with the
To control the accuracy of the reduced system more efciently, identication experimental data. On this basis, a day is chosen in
we notice that: (i) there are two nonlinear terms with different autumn of November, 2011 with average air temperature of
characteristics in the thermal formulation of the PV module, and 22 C and average wind speed of 3 m/s. It is veried that the
that (ii) these two terms inuence different layers of the PV mod- reduced model approximates the full system to a reasonable
ule. In fact, since we assumed that the net long wave exchange for degree of accuracy also under this different excitation, as shown
the rear of the module is negligible (see Section 2.3), the heat loss in Fig. 7.
 S.O. Ojo et al. / Composite Structures 119 (2015) 477486 483

last time step

Temperature for node number 741 360
350 Complete
350 k1=1
340 k1=3

Cell Temperature(k)
k1=5
340
k1=7
Temperature (K)

330
Complete
k=1 330
320 k=3
k=5
320
310 k=7

300 310

290 300
0 50 100 150 200 0 0.5 1 1.5
Time step Length of the module(m)
(a) (b)
Fig. 5. Convergence of reduced solution to the full solution with increasing order of k and xed interpolation points m1 = m2 = 3 for (a) a specic point in the silicon layer, and
(b) for the entire silicon layer.

Temperature for node 250 noted that the error obtained by using only one interpolation point
302 (i.e. an equivalent of a linear system) is in the range of 105 to 106,
Complete
185 Snapshots an indication that the nonlinearity in the system is not at all strong.
300 100 Snapshots Increasing the number of interpolation points further shifts the
70 Snapshots error bound from 105 to less than 108 which conrms the excel-
Temperature (K)

40 Snapshots
298
296
294
292
0 50 100 150
Time step
Fig. 6. Convergence of the reduced solution by increasing the number of snapshots
for order of basis k = 7 and interpolation points with m1 = m2 = 5.
4.2. Error analysis
To demonstrate the efciency and accuracy of the nonlinear
order reduction, an error plot is deemed necessary to observe the
convergence of the responses by varying the interpolation points
m1,m2 and the dynamical order k. To do this, we compute a
normalized error e(k, m1, m2) as
ek
kT  T 2
e 48
kTk2
where the norm is dened either in time domain by xing the cell
location, or in the space domain by xing time step.
In order to observe the rate of convergence of the reduced sys-
tem as we increase its dynamical order k with a xed number of
interpolation points m1 = m2 = m, we compute the error at specic
time steps representative of the beginning, middle and end of sim-
ulation. A time variation error plot is also obtained for a selected
element in the PV module for all the time steps. The results are
shown in Fig. 8.
A clear lower error bound can be observed from the error plots,
which is an indication that the reduced system response converges,
as the order of the system is increased, only to the extent allowed
by the representation of the nonlinear terms. It should be generally
lent suitability of the DEIM algorithm for this thermal modeling
task.
In order to verify the improved efciency that can be achieved
by using different interpolation points for the two nonlinear terms,
as against using the same interpolation order for the two nonlinear
terms, we perform various experiments by independently varying
m1 and m2. The results are shown in Fig. 9.
200
The plots shown in Fig. 9 conrm that by independently varying
m1 and m2, we can achieve better control of the accuracy of the
reduced model. In Fig. 9(a) where the same interpolation order is
used for the two nonlinear terms, the reduced model becomes ef-
cient as order k is increased above 50 when the lower bound error
becomes more stable. By varying m1 and m2 independently, the
stability of the lower error bound is attained with a smaller order
k. It can be clearly observed in Fig. 9(b)(d) that there is a reduction
in the lower bound error from less than 106 to less than 108 as
m1 is increased from 1 to 7 while m2 is varied for each xed m1.
Furthermore with independent variation of m1 and m2, a lower
order k (<50) of the linear subspace projection is required to attain
a stable error bound. This observation proves that, a better approx-
imation of the full system can be achieved by independent varia-
tion of the interpolation of the nonlinear terms.
4.3. Runtime
Finally, we emphasize the advantages of proposed MOR tech-
nique by reporting the runtime required for the various simula-
tions on the same commodity laptop. A transient analysis of the
full system requires 61.40 s. Based on available snapshots, the con-
struction of the reduced model via the proposed POD/DEIM
requires as few as 0.86 s, whereas the transient simulation of the
reduced model (k = 7 and m1 = m2 = 3) takes only 1.40 s. Excluding
model setup and construction, the overall speedup is almost 44X.
Considering that the proposed model is a quite simplied and 2D
structure, more dramatic speedup is expected when applying this
process to a full 3D geometry.
484 S.O. Ojo et al. / Composite Structures 119 (2015) 477486

Temperature for node number 741 last time step

350 360
Complete
350 k1=1
340
k1=3

Cell Temperature(k)
k1=5
340
Temperature (K)

330 Complete k1=7

k=1
320 k=3 330
k=5
310 k=7 320

300 310

290 300
0 50 100 150 200 0 0.5 1 1.5
Time step Length of the module(m)

(a) (b)
Fig. 7. Convergence of reduced solution to the full solution with increasing order of k and xed interpolation points m1 = m2 = 3 for (a) a specic point in the silicon layer, and
(b) for the entire silicon layer.

Error plot for 50th time step Error plot for 150th time step
0 0
10 10
m=1 m=1

m=2 m=2
-2 -2
10 10 m=4
m=4
m=5 m=5
-4
m=7
-4 m=7
Error

10 10
Error

-6 -6
10 10

-8 -8
10 10

-10 -10
10 10
0 50 100 150 200
0 50 100 150 200
k k
(a) (b)

Error plot for 186th time step Error plot for 150th node
0 0
10 10
m=1 m=1
m=2 m=2
-2 -2
10 m=4 10 m=4
m=5 m=5
-4 m=7 -4 m=7
10 10
Error

Error

-6 -6
10 10

-8 -8
10 10

-10 -10
10 10
0 50 100 150 200 0 50 100 150 200
k k

(c) (d)
Fig. 8. Error plot for the topmost layer for xed m1 = m2 = m at (a) 50th time step (b) 150th time step (c) 186th time step, and for (d) node number 150 in the discretized
module for all 186 time steps.
 S.O. Ojo et al. / Composite Structures 119 (2015) 477486 485

Error plot for 186th time step 0

Error plot for 186th time step
0
10 10
m=1 m2=1
-2 m=2 m2=2
10 m=4 -2
m2=4
10
m=5 m2=5
10
-4 m=7 m2=7

Error
Error

-4
10
-6
10
-6
-8 10
10

-10 -8
10 10
0 50 100 150 200 0 50 100 150 200
k k
(a) (b)

Error plot for 186th time step 0

Error plot for 186th time step
0 10
10
m2=1
m2=1
m2=2
-2 m2=2 10
-2
10 m2=4
m2=4
m2=5 m2=5
-4 m2=7 10
-4 m2=7
10
Error
Error

-6 -6
10 10

-8 -8
10 10

-10 -10
10 10
0 50 100 150 200 0 50 100 150 200
k k
(c) (d)
Fig. 9. Error plot for the third layer at the 186th time step for (a) m1 = m2 = m (b) m1 = 1 (c) m1 = 5 (d) m1 = 7.

5. Conclusion to model temperature-induced debonding of the back-sheet, should

The partial differential equation governing the heat conduction
within a 2D photovoltaic system has been derived and a numerical
solution based on a nite difference scheme has been proposed. In
order to limit the computational burden, we have identied the
DEIM as a suitable technique to reduce the system through a combi-
nation of linear subspace projection and interpolation of nonlinear
terms. In the numerical examples, we show that the heat conduction
of a PV system discretized into 2166 nodes along the module span is
successfully reduced to a compact model with dynamical order
k = 7, based on interpolation with only m1 = m2 = 3 points. For the
specic system under investigation, we see that further increment
in k does not result into a signicant accuracy, and that the efciency
in the numerical solution of the reduced model, which depends on
the number of interpolation points, can be ne- tuned by carefully
selecting a different interpolation order for individual nonlinear
terms. From the validation results, it is also concluded that the
reduced solution is not very sensitive to the input function as it
approximates well the simulated irradiance data much the same
way as the experimental irradiance data used in the construction
of the compact model. As a future development of the present
research, the generalization of the present formulation for coupled
thermo-mechanical problems is envisaged. In this case, the pres-
ence of cohesive crack bi-material interfaces between the layers,
be exploited in order to assess the durability of PV modules. Also,
extension to a full 3D geometry is under way.
Acknowledgments
MP would like to thank the European Research Council for the
ERC Starting Grant Multi-eld and multi-scale Computational
Approach to Design and Durability of PhotoVoltaic Modules
(ERC Grant Agreement n. 306622 CA2PVM) granted under the
European Unions Seventh Framework Programme (FP/2007-
2013). SOO would like to acknowledge the support of the Italian
Ministry of Education, University and Research to the project FIRB
2010 Future in Research Structural mechanics models for renew-
able energy applications (RBFR107AKG).
References
[1] Paggi M, Berardone I, Infuso A, Corrado M. Fatigue degradation and electric
recovery in Silicon solar cells embedded in photovoltaic modules. Sci Rep
2014;4:4506. http://dx.doi.org/10.1038/srep04506.
[2] Eitner U, Kntges M, Brendel R. Use of digital image correlation technique to
determine thermomechanical deformations in photovoltaic laminates:
measurements and accuracy. Sol Energy Mater Sol Cells 2010;94:134651.
[3] Kntges M, Kunze I, Kajari-Schrder S, Breitenmoser X, Bjrneklett B. The risk of
power loss in crystalline silicon based photovoltaic modules due to
microcracks. Sol Energy Mater Sol Cells 2011;95:11317.
486 S.O. Ojo et al. / Composite Structures 119 (2015) 477486
[4] Sapora A, Paggi M. A coupled cohesive zone model for transient analysis of
thermoelastic interface debonding. Comput Mec 2014;53:84557.
[5] Paggi M, Corrado M, Rodriguez MA. A multi-physics and multi-scale numerical
approach to microcracking and power-loss in photovoltaic modules. Compos
Struct 2013;95:6308.
[6] Leea Y, Andrew A, Taya B. Finite element thermal analysis of a solar
photovoltaic module. Energy Procedia 2012;15:41320.
[7] Paggi M, Sapora A. Numerical modelling of microcracking in PV modules
induced by thermo-mechanical loads. Energy Procedia 2013;38:50615.
[8] Antoulas AC. Approximation of large-scale dynamical systems. Philadelphia,
PA, USA: Society for Industrial and Applied Mathematics; 2005.
[9] Benner Peter, Hinze Michael, Ter Maten E Jan W. Model reduction for circuit
simulation. Lecture notes in electrical engineering, vol. 74. Springer; 2010.
[10] Tan SXD, He L. Advanced model order reduction techniques in VLSI
design. New York: Cambridge University Press; 2007.
[11] Schilders WHA, Van Der Vorst HA, Rommes J. Model order reduction: theory,
research aspects and applications. Springer Verlag; 2008.
[12] Odabasioglu A, Celik M, Pileggi LT. PRIMA: passive reduced-order interconnect
macromodeling algorithm. IEEE Trans Comput Aided Des Integr Circuits Syst
1998;17(8):64554.
[13] Arnoldi WE. The principle of minimized iterations in the solution of the matrix
eigenvalue problem. Q Appl Math 1951;9(1):1729.
[14] Lanczos C. An iteration method for the solution of the eigenvalue problem of
linear differential and integral operators1. J Res Natl Bureau Stand 1950;45(4).
[15] Moore B. Principal component analysis in linear systems: controllability,
observability, and model reduction. IEEE Trans Autom Control
1981;26(1):1732.
[16] Glover K. All optimal Hankel-norm approximations of linear multivariable
systems and their l inf error bounds. Int J Control 1984;39(6):111593.
[17] Willcox K, Peraire J. Balanced model reduction via the proper orthogonal
decomposition. AIAA J 2002;40(11):232330.
[18] Bond B, Mahmood Z, Li Y, Sredojevic R, Megretski A, Stojanovic V, et al.
Compact modeling of nonlinear analog circuits using system identication via
semidenite programming and incremental stability certication. IEEE Trans
CAD Integr Circuits Syst 2010;29(8):114962.
[19] Phillips JR. Projection-based approaches for model reduction of weakly
nonlinear, time-varying systems. IEEE Trans Comput Aided Des Integr
Circuits Syst 2003;22(2):17187.
[20] Rewienski M, White J. A trajectory piecewise-linear approach to model order
reduction and fast simulation of nonlinear circuits and micro-machined
devices. IEEE Trans Comput Aided Des Integr Circuits Syst 2003;22(2):15570.
[21] Astol A. Model reduction by moment matching for linear and nonlinear
systems. IEEE Trans Autom Control 2010;55(10):232136.
[22] Chaturantabut S, Sorensen D. Nonlinear model reduction via discrete empirical
interpolation. SIAM J Sci Comput 2010;32(5):273764.
[23] Jones A, Underwood C. A thermal model for photovoltaic systems. Sol Energy
2001;70(4):34959.
[24] Armstrong S, Hurley W. A thermal model for photovoltaic panels under
varying atmospheric conditions. Appl Therm Eng 2010;30:148895.
[25] Blomberg T. Heat conduction in two and three dimensions: computer
modelling of building physics applications. Report TVBH-1008, Department
of Building Physics, Lund University, Sweden; 1996. p. 513.