Renewable Energy 68 (2014) 14e24

Contents lists available at ScienceDirect

Renewable Energy
journal homepage: [Link]/locate/renene

Unit-response function for ground heat exchanger with parallel, series

or mixed borehole arrangement
D. Marcotte*, P. Pasquier
Civil, Geological and Mining Department, Polytechnique Montral, C.P. 6079 Succ. Centre-ville, Montral H3C 3A7, Canada

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 22 July 2013
Accepted 20 January 2014
Available online 14 February 2014

A novel approach is presented that allows to predict uid temperatures entering a Ground Heat
Exchanger (GHE) for parallel, series and mixed arrangements of boreholes. The method determines at
each time step the heat transfer rates occurring at each borehole so as to reproduce the uid temperature
at the GHE inlet for a specic borehole arrangement. The analytical nite line source model is used to
compute the borehole wall temperatures, whereas the uid temperatures are assumed to vary linearly
along the pipes. The method requires to solve a linear system of equations at a small number of time
steps. The different systems of equations for each arrangement are determined. A comprehensive 3D
nite element numerical model shows good agreement with the computed uid temperatures. The
proposed approach is computationally very efcient. The uid temperature unit response function can be
convolved with any desired heat load to estimate uid temperatures at the GHE inlet for a wide variety of
scenarios.
2014 Elsevier Ltd. All rights reserved.

Keywords:
Finite line source
Parallel arrangement
Series arrangement
Ground heat exchanger

1. Introduction
Several analytical models are available in the literature [3,10] to
predict the temperature response resulting from operation of a
ground heat exchanger (GHE). Among the models, the nite line
source (FLS) is often used to compute the average temperature
along the wall of a borehole heat exchanger (BHE) [5,7,14,17,18]. Fast
computation of BHE wall temperature can be done efciently by
spectral methods using either Fast Fourier Transform (FFT) [16] or
Laplace transform [11,13]. The FFT approach is faster and easier to
apply than the Laplace approach when the heat transfer occurring
at each BHE is already known. However, in a GHE, the interactions
between the BHEs imply that the heat transfer at each BHE varies in
time according to its position in the network, hence, it has to be
determined at each time step.
Cimmino et al. [4] and Lazzarotto [15] both used the Laplace
transform approach of Lamarche [11] to determine sequentially the
heat transfer and the average wall temperature at the BHEs. They
used a linear system of equations solved at each time step. Their
approach requires the numerical evaluation, at each time step, of an
inverse Laplace transform. Pasquier and Marcotte [20,21] proposed
a quite different approach where they work simultaneously on all
time steps by perturbing iteratively an initial guess heat transfer

* Corresponding author. Tel.: 1 514 340 4711x4620; fax: 1 514 340 3970.
E-mail address: [Link]@[Link] (D. Marcotte).
0960-1481/$ e see front matter 2014 Elsevier Ltd. All rights reserved.
[Link]

distribution so as to meet imposed temperature signals on the BHE

wall temperatures. Their algorithm converges in a few tens of iterations in all examples tested, and is proven to be exact when the
number of iterations reaches the number of time steps.
Neither Cimmino et al. [4], Lazzarotto [15] and Pasquier and
Marcotte [20], compared their results to the temperatures obtained
with a full 3D numerical model, although Cimmino et al. [4]
compared their results to the numerical g-functions of Eskilson
[9] and nd noticeable differences at long times. Moreover, the
previous methods focused on the determination of BHE wall temperature which implicitly assume a parallel arrangement between
the BHEs. The examples in Lazzarotto [15] are all with parallel
arrangement although Lazzarotto [15] mentions, without providing
the pertaining equations, that his approach accommodates parallel
and series arrangements.
The objectives of this paper are three-folds: i. adapt the
sequential idea of Cimmino et al. [4] and Lazzarotto [15] to the
easier to apply FFT spectral approach, ii. develop the unit response
functions based on the uid temperature at the GHE inlet rather
than based on the BHE average wall temperature, hence allowing
simulation of temperatures for parallel, series and mixed arrangements, and iii. test the proposed approach with a full 3D nite
element numerical model.
The paper is structured as follows. The methodological section
presents the idea introduced by Lamarche and Beauchamp [13] of
splitting the response function in a historical and a contemporary
part. Using the FLS and the assumption of linear uid temperature

a)

q
q
; T t Tf t 
_ p out
_ p
2mC
2mC

(2)

out

Fig. 1. Different possible BHE arrangements: a) parallel; b) series; c) mixed.

two terms, the historical part hij(mDt) due to heat transfer from 0 to
(m  1)Dt and the present time step contribution qj(mDt)fDt(rij):

 

2.1. Borehole interactions

In a network of n BHEs, the wall temperature of BHE i can be
obtained by spatial superposition. Summing the contribution from
each BHE, one has:

DTj/i t

(3)

j1

where DTj/i(t) is the temperature perturbation at BHE i caused by

heat emanating from BHE j.
The heat loads are assumed to be a step function with time step
Dt. The unit response function has to be calculated at times mDt,
m [Link]. It is convenient to split the temperature perturbations in

DTj/i t hij t qj tfDt rij

(4)

where fDt(rij) is the unit transfer function computed with the

analytical model for one time step Dt, and rij is the distance between boreholes i and j. For i j, one has rii rb. Note that the total
Pn
heat transferred by the n BHEs is qt
j 1 qj t. The historical
part hij(mDt) can be easily computed by a discrete convolution
product:

hij mDt


~ j *f mDt
q

(5)

where, at time t mDt, the step increment vector qej is the vector:

i
h
qej qj;1 ; qj;2  qj;1 ; .; qj;m1  qj;m2 ; qj;m1

Coord. y (m)

b)
7

3m

Coord. x (m)

(6)

and the transfer function f is the analytical model response for the
rst m time steps under a unit heat load at each BHE evaluated for a
distance rij; hij(mDt) is obtained as the mth element of the

a)

Coord. y (m)

From HP

To HP

(1)

_ is the uid mass ow rate and Cp is the specic heat of the

where m
uid.

T bi t T0

Tin

Tout

To HP

c)

Tin

From HP

b)

where q is the heat transferred by the BHE, H is the borehole length,

Rb is the BHE equivalent thermal resistance and T b is the average
temperature at the BHE wall. Assuming, for simplicity, a linear
variation of the uid temperature along the pipes, one has:

n
X

in

out

An analytical model (e.g. FLS [5,14]) enables to compute the

mean temperature at the borehole wall. Under a steady state hypothesis the mean uid temperature is obtained as:

Tin t Tf t

2. Methodology

Tf t q=HRb T b t

15

To HP

variation along the pipes, the remarkably simple linear systems of

equations for different BHE arrangements (parallel, series or mixed)
are then presented. The model is applied for a GHE with radial
symmetrical distribution of BHE, for parallel and mixed arrangements. The results are compared to those of a full 3D numerical
model developed in COMSOL. Computational aspects are discussed
to prove the practicality of the sequential FFT approach, even for
numerous BHE and long time series. Finally, a detailed numerical
example is provided in Appendix A.

From HP

D. Marcotte, P. Pasquier / Renewable Energy 68 (2014) 14e24

Coord. x (m)

Fig. 2. Location of boreholes and arrangements a) parallel, b) series.

16

D. Marcotte, P. Pasquier / Renewable Energy 68 (2014) 14e24

Table 1
Parameters.
Variable

Value, Units

ks
cs
Rb
T0
_
m
Cp
rb
H

2.5 W(m  C)1

2  106 J(m3  C)1
0.12 m  C W1
0 C
0.3 kg/s
4180 J kg1 K1
0.08 m
30 m

convolution vector qej *f. The convolution can be computed efciently using FFT after padding vectors qej and f with m  1 zeros to
account for the non-periodicity of functions as described in Marcotte and Pasquier [16], and Pasquier and Marcotte [20]. Note that

one convolution per BHE pair has to be computed. Moreover, the

historical contribution has to be computed at each time step.
However, Section 3.2 will show that the computations need to be
realized at only a very small subset of the time steps, therefore
making the approach practical.
Adopting a matrix notation and dropping the time reference,
one has simultaneously:

T h Gqt

(7)

where T is the vector with total temperature perturbation at each of

the n BHEs, h is the vector containing the n temperature perturbations due to the historical part, G is the n  n matrix whose
element (i,j) is the temperature perturbation caused, at BHE i, by
one unit of heat emanating from BHE j for the current time step, qt
is the n  1 vector of heat ux emanating from the BHEs for time

b)

a)

c)

Fig. 3. Short-term heat transfer ratios (in %) for a) parallel and b) series; c) T~ in for the two arrangements.

D. Marcotte, P. Pasquier / Renewable Energy 68 (2014) 14e24

17

a)

b)

c)

Fig. 4. Long-term heat transfer ratios (in %) for a) parallel and b) series; c) T~ in for the two arrangements.

mDt. Note that G is a function of the time step, not of the time,
therefore it needs to be computed only once.
In the next section, the inuence of the connections between
BHEs is examined. More specically, the pure parallel arrangement
and the pure series arrangement and a mix of parallel/series
arrangement are considered, as depicted in Fig. 1. As in Lazzarotto
[15], it is assumed that the uid coming from the heat pumps is
thorough-fully mixed when entering the GHE, so a unique Tin value
applies. Similarly, the uid leaving the GHE is mixed before
entering the heat pumps, so a unique Tout applies.

2.2. Pure parallel arrangement

The n BHEs are fed simultaneously by a common uid source.
Therefore the Tin temperatures at the BHEs are all equal at each
time step. Moreover the sum of the heat ux over the n BHEs is set

to one unit of heat for the entire boreeld. The latter condition is
written in vector form as:

1T qt 1ct

(8)

where 1 is a n  1 vector of ones.

The n 1 unknowns at each time step are the n heat transfer
coefcients (one per BHE) and the Tin  T0 value. Solution of the
linear system of equations described by Equation (9) gives, for a
given time step, qt and Tin since T0 is known.

GD
1T

1
0



qt
Tin  T0

h
1


(9)

_ i Cp ; i [Link]n, where diag represents

with D diagRbi =Hi 1=2m
_ i are the
a diagonal matrix with the n terms on the diagonal. The m

18

D. Marcotte, P. Pasquier / Renewable Energy 68 (2014) 14e24

a)
8
Pr

of s
ym
m

etry

of
ile

se
ct
(F

Pla

ne

io
ig

.1

Coord. y (m)

1)

Exterior

Inflow
Outflow

Outflow

Profile (Fig. 14)

Inflow

Center

Fig. 5. Sensitivity of the T~ in to the time step; 9 boreholes as in Fig. 2 with parallel
arrangement; the four curves indicated in the legend are almost perfectly superposed,
therefore indistinguishable.

8
8

mass ow rate entering BHE i. They are assumed known at all time
steps.
2.3. Pure series arrangement
The n BHEs form an ordered sequence going from BHE 1 to BHE
n. The uid temperature at the outlet of a BHE is considered to be
the uid temperature at the inlet of the next BHE in the chain
(hence, uid heat losses occurring during the transfer between the
BHEs are neglected). Moreover, we have:

_ ci; i 1.n
_i m
m

(10)

and

Tout;i Tin;i1 ci; i 1.n  1

(11)

Fig. 6. T~ in computed at varying time steps from 1 h to 200 y; 9 boreholes as in Fig. 2

with parallel arrangement.

Coord. x (m)
Fig. 7. BHE locations, planes of symmetry, and proles used in Figs. 11 and 14. The two
series arrangements are illustrated: 1 e Exterior: Inow from the eight exterior
boreholes and outow from the eight center boreholes, and 2 e Center: Inow from
the eight center boreholes and outow from the eight external boreholes.

Tin;1 Tin Tout;n

Q
_ p
mC

(12)

The n 1 unknowns at each time step are the n heat transfer

coefcients of qt and Tin. All other uid and wall temperatures can
be obtained from these. They are obtained by solving the linear
system of equations:

GDL
1T

1
0



qt
Tin  T0

h
1

Fig. 8. Geometry and mesh in the horizontal plane.

(13)

D. Marcotte, P. Pasquier / Renewable Energy 68 (2014) 14e24

19

Table 2
Thermal properties and geometry of the reference model.
Description

Parameters, Units

Values

Fluid thermal conductivity

Pipe thermal conductivity
Grout thermal conductivity
Soil thermal conductivity
Fluid volumetric heat capacity
Pipe volumetric heat capacity
Grout volumetric heat capacity
Soil volumetric heat capacity
Borehole radius
Inner radius of the pipe
Outer radius of the pipe
Half pipe spacing
Depth of BHE summit
BHE length
Mass ow rate per BHE

kf, W/(m K)
kp, W/(m K)
kg, W/(m K)
ks, W/(m K)
rfcf, J/(m3 K)
rpcp, J/(m3 K)
rgcg, J/(m3 K)
rscs, J/(m3 K)
rb, m
ri, m
ro, m
D, m
D, m
H, m
_ kg/s
m,

0.55a
0.40
1.45
2.5
4 180
2 000
2 000
2 000
0.08
0.017
0.022
0.04
2.0
30.0
0.3

000
000
000
000

a
To represent convection in the pipes, a conductivity 500 times higher is used in
the xy plane.

GDL
1T

1
0



qt
Tin  T0

h
1


(14)

where the D and G matrices are dened as before, and the matrix L
has the following structure:

L1
6 0
6
L 4
/
0

Fig. 9. Geometry and mesh for the whole model.

where L is a lower triangular matrix with zeros on the diagonal and

_ p ; cj < i; i [Link]n.
Lij 1=mC

0
L2
/
/

/
0
/
0

3
0
/7
7
/5
Lm

(15)

and where each Li corresponds to a head BHE. It is a lower triangular matrix of size ni  ni with zeros on the diagonal and value
_ i Cp for all other entries in the lower part of the matrix (m
_ i is the
1=m
uid mass ow entering head BHE i).
There is one linear system of Equations (9) and (13), or (14) per
time step to be solved. However, each system is of small size (n 1)
and moreover the left hand matrix is time invariant, so it needs to
be inverted only once. Appendix A presents a small numerical
example to illustrate the computations involved.

2.4. Mixed arrangements

2.5. GHE dimensionless unit response function
We consider a parallel arrangement for the m BHE that are the
heads of as many clusters of BHEs arranged in series (a cluster can
have a single BHE). The clusters are of size ni, i 1.m. Combining
the results of the two previous sections, one gets:

The Tin(mDt) function evaluated in Equations (9), (13) and (14),

represents the time temperature evolution (in  C) of the uid
entering the GHE due to the constant application of 1 W of heat

Fig. 10. Illustration of the boundary conditions used at the base of a BHE.

20

D. Marcotte, P. Pasquier / Renewable Energy 68 (2014) 14e24

load for the entire GHE. It is convenient to dene the dimensionless

unit response function T~ in as:

T~ in Tin  T0 2pks nH

(16)

This function is similar to the g-function of Eskilson [9], except

that it represents the uid temperature at the GHE inlet instead of
the average BHE wall temperature. Contrary to the g-function of
Eskilson, it already integrates the effects of BHE arrangement
(parallel, series or mixed) and can be used directly to predict uid
temperatures for any heat load scenarios.

3. Results
3.1. A simple example
The time varying heat transfer coefcients for the two arrangements described in Fig. 2 are computed with parameters
given in Table 1. The nine H 30 m long BHEs are located on a 3 m
regular grid. Fig. 3 shows the heat transfer coefcients for each BHE
and the T~ in obtained with the parallel and the series arrangements
in the short-term, i.e. within the rst 20 days (ts 2.537y, ln(t/
ts) 3.83, where ts is the Eskilsons characteristic time, ts H2ks/
(9Cs)). For the parallel arrangement, due to the symmetry, only 3
different heat transfer proportion curves appear. For the series
arrangement, each BHE shows a different heat transfer proportion
ordered according to the BHE sequence in the series. The T~ in is
signicantly lower for the parallel case than for the series case,
indicating a larger heat transfer to the ground for this arrangement
compared to the series arrangement. Note that the uid mass ow
per BHE is the same for both arrangements, implying that 9 times
more uid goes to the HP in the parallel case.
Fig. 4 shows the heat transfer ratios for each BHE and the T~ in
temperatures obtained with the parallel and the series arrangements for the long-term, i.e. up to 10 years. The parallel and series
curves have similar behavior except that the T~ in is lower for the
parallel case. After approximately 2.5 years (ln(t/ts) z 1), a quasisteady state is reached for the heat transfer coefcients in both
scenarios and for all BHEs. Note that in the series arrangement, the
BHE heat transfer depends of its location (corner, sides or center)
and its order in the sequence. Hence, corner BHEs transfer more
heat than side BHEs and side BHEs transfer more heat than the
central BHE. Among the corner BHEs, the heat transfer proportion
follows the BHE sequence and similarly for the side BHEs.

4. Comparison with a numerical model

The analytical model is compared to a full 3D numerical model
built in COMSOL Multiphysics [6]. The case of radially and symmetrically located BHEs depicted in Fig. 7 is used with parallel and
series feeding strategies. Two different strategies are considered for
the series case: feeding by the central BHE of each branch or
feeding by the external BHE of each branch.

4.1. Description of the numerical model in COMSOL

To provide a reference set of uid temperature, the model
developed by Marcotte and Pasquier [16] for a single BHE in the
Comsol Multiphysics [6] environment has been extended to include
12 complete BHEs. The 12 boreholes lie on the same side of any of
the symmetry planes found at azimuth 22.5 , 67.5 , 112.5 and
157.5 . The modeled geometry includes the heat carrier uid
circulating in the supply and return pipes, the pipe material itself
and the grout for each BHE, as well as the geological material
located below, above and around the BHEs. The model however
omits the horizontal return loop located at the base of every pair of
pipes, and the horizontal pipe segment transporting the heat carrier uid between the BHEs. The geometry around a single BHE is
shown in Fig. 8 while the complete geometry is shown in Fig. 9.
The state equation behind the model is given by

rCp

vT
rCp u,VT V,kVT
vt

(17)

3.2. Sensitivity to the time step

Fig. 5 shows the T~ in obtained for the parallel arrangement of
Fig. 2 with different choice of time steps Dt. Whatever, the time step
adopted, the results are almost identical at the same time, the
maximal relative difference between any pair of curves being only
0.11%. The drawback to using a large time step is to provide the rst
value at a larger time. This suggests, as in Cimmino et al. [4] to
diminish the total number of time steps by adopting increasing
time steps and merging the results. As an example, the solution can
be computed at every hour for one day, then at every day for one
month, then at every month for one year, then at every year for 10
years, then every 10 years for a few centuries (if required). This
strategy ensures the number of time steps remains always
tractable.
Fig. 6 shows the result of such strategy for the previous case
with 9 boreholes. It was computed in 1.5 s on a standard laptop
(Intel coreI3 at 2.66 GHz, Windows 7; average time over 100
simulations).

Fig. 11. Temperature distribution in a vertical section including six BHEs, after 180 days
of 1 W heat injection for the whole GHE.

D. Marcotte, P. Pasquier / Renewable Energy 68 (2014) 14e24

where r is the material density, Cp is the specic heat capacity, k is

the thermal conductivity and where u is a velocity vector. The latter
is used only to represent the vertical uid advection within the
pipes and is equal to zero everywhere else in the domain. In every
_ rf pri2 while a value of m=
_ rf pri2 is used in
supply pipe, u worths m=
every return pipe.
The boundary conditions used include a zero constant temperature condition along the surface (z 0 m) and the base (z 67 m)
of the domain. To represent a constant temperature boundary
located at innity, innite boundary conditions constrained at T 0
are used for the lateral boundary located at r 20 m. To reduce the
models size, a symmetry boundary condition is used to split in two
equal parts the problem. To create small inner boundaries corresponding to the inlet and outlet uid at the extremities of each
pipe, small voids are created in the model as depicted in Fig. 10. This
feature allows to assign an outow boundary condition along the
outlet boundary of each pipe. To model the U-loop at the base of a
BHE, a numerical integration is performed at the outlet of each
supply pipe. The corresponding mean temperature is then used as a
time-dependent prescribed boundary condition along the inlet
boundary of the return pipe, leading to a coupled nonlinear problem. The same technique is used to emulate BHE in series and
ensure the same uid temperature between the outlet of each

21

return pipe and the inlet of the supply pipe located downstream.
Finally, for BHE connected in series, a prescribed temperature
_ is used
boundary condition given by Tin t Tout t qt=Cp m
to link the temperature leaving the last BHE (Tout) to the inlet
temperature of the rst BHE (Tin). If the BHEs are connected in
parallel, the same condition is used to link the supply and return
pipes of each BHE.
The sizes and the thermal properties of the materials present in
the model are summarized in Table 2. Care has been taken to ensure
that the equivalent borehole resistance of the numerical model
corresponds to the Rb value integrated in matrix D. In fact, when
used in conjunction with the Multipole Method [2], the parameters
presented in Table 2 correspond to a Rb value of 0.121 mK/W. Note
that turbulent ow conditions within the pipes are represented by
assigning to the uid a much higher thermal conductivity in the
horizontal direction than in the vertical direction.
The volume where the BHEs are embedded is divided into 60
layers of triangular prisms while an unstructured mesh is used
above and below the BHEs. Ultimately, the domain is discretized
in 2 334 674 linear nite elements, which corresponds to a
problem of 1 006 175 degrees of freedom. This meshing was then
used to compute a solution at each time step from an initial
ground temperature of T(0) 0. A cross-section showing a

a)

b)

c)

Fig. 12. Heat transfer ratios (%) obtained by the analytical and numerical model for parallel (a) and series arrangements fed by (b) BHE1 (center) or (c) BHE3 (external). The sum of %
equals 100/8 as there are 8 branches in the GHE.

22

D. Marcotte, P. Pasquier / Renewable Energy 68 (2014) 14e24

temperature solution across two of the eight branches is illustrated in Fig. 11.

13

Center
Exterior
Parallel

12

4.2. Results of comparison test

The heat transfer ratios and the unit response functions obtained by the analytical and the numerical models for the parallel
and the two series arrangements are presented in Figs. 12 and 13.
The numerical model shows similar results to the analytical
model regarding the values and the time evolution of the heat
transfer occurring at the different BHEs. The numerical values obtained for T~ in are slightly smaller than the analytical model values.
The steady 3.5% difference for T~ in is deemed acceptable and can be
attributed to the difference of assumptions inherent to the two
approaches. Hence, the analytical model assumes that the heat
transfer is distributed evenly along the BHE length and that the
uid temperature varies linearly along the pipes. These assumptions are not made in the numerical model. Moreover, the analytical
model assumes instantaneous effects of heat transferred to the
uid, whereas the numerical model incorporates the time delay
due to advection. Nevertheless, the analytical model is much easier
to manipulate and constitutes a good and exible alternative
approximation to the heavy numerical model.
One notes that the parallel T~ in presents only slightly inferior
values compared to the mixed T~ in . This reects that the series is
short along each branch with only 3 BHEs. More important, and to
our knowledge this was not previously noticed in the literature, the
two series arrangements, i.e. fed by the central BHE or by the
external BHE, provide identical T~ in , despite the fact that the heat
distribution in the ground is evidently different in both cases, as
veried in Fig. 14. This observation is valid for both the analytical
and the numerical model.
5. Discussion
The analytical model described enables to compute, with
simple algebraic manipulations, the heat transfer rate and the
Tin temperature at the end of each time step for any arrangement of the BHEs, parallel, series or mixed, and any locations of
the BHEs. The model relies on the FLS, and the assumption of
linear variation of uid temperature along the pipes. In

Fig. 13. T~ in obtained by the analytical and numerical models for parallel and series
arrangements.

o
Ground Temp. ( C)

11

10

5
0

B
1

Position along profile (m)

Fig. 14. Ground temperature perturbation (T(x)  T0) along prole AeB of Fig. 7, for a
constant heat injection of 15 W/m, at day 180. The ground temperature is the average
temperature integrated along a vertical line parallel to the BHE.

principle, any model allowing computation of the transfer

function f in Equation (4) can be used (e.g. the inclined FLS Cui
et al. [7], Marcotte and Pasquier [17], Lamarche [12]). The algorithm involves only convolutions and the computed responses
are exact in the sense that the FLS does reproduce exactly the Tin
temperature with the time varying heat transfer qt computed at
each BHE under the same assumption for the linear variation
along the pipes of uid temperature. As for the CPU aspect, the
discrete convolutions can be computed efciently by FFT as in
Marcotte and Pasquier [16]. More importantly, only a small
subset of the time steps need to be evaluated to estimate the
whole T~ in function as shown in Fig. 6, therefore reducing
considerably the CPU effort required. The T~ in function is then
interpolated to the desired time step for the convolution with
the heat load function, allowing to predict the uid temperature
for any scenario of heat load.
One of the interesting ndings is that, for the mixed arrangement in the radial case, the T~ in is insensitive to the location of the
start of the series, center or exterior. This somewhat surprising
result was conrmed by the 3D numerical model and can have
important consequences for heat storage applications.
The FLS model is of course a simplication of the heat transfer
process occurring in a GHE. It was shown to lie within 3.5% of the
3D numerical solution obtained with a comprehensive nite
element GHE model developed in COMSOL. However, for very short
times (less than 1 h) the FLS cannot represent the complex heat
transfer occurring within the BHE radius. But at short times the
interactions between BHEs are non-perceptible (as a consequence,
matrix G becomes diagonal in Equations (9), (13) and (14)). This
suggests to use alternative models to the FLS for the short times.
One approach is to use the thermal resistance-capacitance models
(TRCM) [1,8,19]. With these models, Tin can be computed directly at
short times and merged with the FLS Tin obtained at longer times. A
single TRCM model would be required for the parallel arrangement
whereas the TRCM can easily be coupled in series for the series case
due to the lack of interactions occurring between BHEs at these
times. This opens the possibility to generate realistic uid temperature unit response function covering time intervals from minutes to centuries.
One limitation of the proposed approach is the assumption,
intrinsic to the FLS model, of spatially constant ground thermal

D. Marcotte, P. Pasquier / Renewable Energy 68 (2014) 14e24

parameters. Situations where the ground thermal parameters are

deemed to vary in space would have to be dealt with by other
approaches such as TRCM or numerical models.
6. Conclusion
The different BHE arrangements, parallel, in series or mixed, can
be represented in a unied way by a linear system of equations that
can be solved at each time step. This enables to compute the heat
transfer specic to each BHE and the temperature unit response
functions T~ in proper to the arrangement and the BHE network.
Only a few time steps need to be considered, which allows efcient
computation of the uid unit response function T~ in . The latter can
then be convolved in the spectral domain with any desired heat
load to provide uid or ground temperature solutions.
Acknowledgments
This research was nanced by the National Science and Engineering Research Council of Canada. Helpful comments by two
anonymous reviewers are acknowledged.
Nomenclature
cs
ground volumetric heat capacity (J/(m3 K))
Cp
uid specic heat capacity (J kg1 K1)
D
diagonal matrix for effects due to the BHE resistance and
the uid temperature linear variation along the pipes (K/
W)
fDT
unit temperature response at one time step
G
matrix n  n of temperature perturbation caused, at BHE i
by one unit of heat transfer at BHE j (K/W)
hij
historical part of the temperature perturbation caused by
BHE j at BHE i (K)
h
vector of temperature perturbations at the n BHEs due to
the historical part (K)
H
borehole length (m)
k
thermal conductivity (W/(m K))
ks
ground thermal conductivity (W/(m K))
L
lower triangular matrix for effects due to series
arrangement (K/W)
_
m
mass ow rate (kg/s)
n
number of BHEs
r
ground density (g/cm3)
q(t)
heat transfer for the whole boreeld at time t
Pn
qt
j 1 qj t (W)
qj(t)
heat transfer by BHE j at time t (W)
qej
vector of m heat transfer increments at BHE j (W)
qt
vector of the n BHE heat transfer coefcients at time t (W)
rb
borehole radius (m)
rij
distance between BHEs i and j (m)
Rb
BHE effective thermal resistance (m K/W)
t
time
ts
Eskilsons characteristic time, ts H 2 Cs =9ks (s)
Tf
uid temperature (K)
Tin
uid temperature at the entrance of the boreeld (K)
T~ in
dimensionless uid temperature response ()
Tout
uid temperature at the outlet of the boreeld (K)
T0
undisturbed ground temperature (K)
T
vector of temperatures at the n BHEs (K)
u
uid velocity vector (m/s)
x, y
spatial coordinates (m)
1
vector of size n of ones
Subscripts
f
Fluid

i, j
in, out
m
s
t
Dt

23

Borehole index
inlet, outlet of the boreeld
time step index
ground
time (s)
time step (s)

Appendix A
Three 30 m long BHEs are located along a line at x 0 m,
x 2 m and x 5 m. The parameters used for the computations
are given in Table 1. Computations are illustrated for a 1000 h time
step at times 1000 h and 2000 h, for the three different arrangements: parallel, series and mixed. In the mixed cases, BHE1 and
BHE2 form one branch, BHE3 is connected in parallel with the
BHE1e2 branch.
The G matrix is time invariant and invariant for the different
arrangements. It is computed using the FLS model evaluated at the
rst time step (here 1000 h) for a 1 W heat injection per BHE (i.e. 1/
30 W/m as H 30 m), for the distances:

r11
4 r21
r31

r12
r22
r32

3
2
0:08
r13
5
r23 4 2
r33
5

2
0:08
3

3
5
3 5m
0:08

(18)

One nds:

7:6489
G 4 1:1413
0:11671

1:1413
7:6489
0:55767

3
0:11671
0:55767 5  103 K=W
7:6489

(19)

D is diagonal with the constant value 0.12/30 1/(2  0.3 

4180) 4.3987  103 (K/W) as each BHE has the same resistance
and the same mass ow rate for the 3 arrangements. The time
invariant L matrices are for the series arrangement:

Lseries

0
4 0:79745
0:79745

0
0
0:79745

3
0
0 5  103 K=W
0

(20)

and for the mixed arrangement:

Lmixed

0
4 0:79745
0

0
0
0

3
0
0 5  103 K=W
0

(21)

A1. Solution for the rst time step (t 1000 h)

At the rst time step, there is no history to account for. Therefore
h 0. Applying Equations (9), (13) and (14) with the above matrix
denitions, one nds the solutions:

q1 1000
q2 1000
q3 1000
Tin  T0

Parallel
0:33283
0:31874
0:34842
0:0044143

Series
0:35625
0:31699
0:32676
0:0046919

Mixed
0:34153
0:30215
0:35631
0:0045011

and T~ in 1000 6:2405; 6:6330; 6:3633.

A2. Second time step (t 2000 h)
The response at t 2000 h due to the heat injected continuously at each BHE during the rst 1000 h has to be computed
using the heat transfer rate obtained in the previous solution. As
an example, for the parallel case, the historical contribution of

24

D. Marcotte, P. Pasquier / Renewable Energy 68 (2014) 14e24

BHE2 to the BHE1 wall temperature is obtained by evaluating

~ j f 0:31874; 0:318740:0011413; 0:001703 and retainq
ing the mth element, here: 0.00017905. The process is repeated
for each pair of BHEs. One nds the interactions matrix:

Injector j

Receiver i

1
2
3

0.0002194
0.00018697
8.1957e-005

0.00017905
0.00021011
0.00014677

8.5797e-005
0.00016043
0.00022968

Finally, vector h is computed by summing along a given row the

contributions found in the different columns. Hence,
h [0.00048425, 0.00055751, 0.0004584]T. The vectors at time
2000 h for the different arrangements are computed similarly:

Parallel
0:00048425
0:00055751
0:0004584

Series
0:00049336
0:00055953
0:00044908

Mixed
0:00048261
0:00055509
0:0004581

Finally the different solutions are obtained as:

q1 2000
q2 2000
q3 2000
Tin  T0

Parallel
0:33448
0:31353
0:35198
0:0049129

Series
0:35711
0:31153
0:33136
0:0051898

Mixed
0:34324
0:29695
0:35982
0:0049987

from which T~ in 2000 6:9454; 7:337; 7:0667 is computed. The

process is repeated sequentially up to the last time. Once
completed, the T~ in signal can be convolved with any heat load
signal so as to compute the uid temperature over time. Recall that
T~ in is a normalized (dimensionless) uid temperature corresponding to the constant injection of 1 W/m. Therefore, the incremental heat load signal to convolve with T~ in must be expressed
in W/m.

References
[1] Bauer D, Heidemann W, Mller-Steinhagen H, Diersch HJG. Thermal resistance
and capacity models for borehole heat exchangers. Int J Energy Res 2011;35:
312e20.
[2] Bennet J, Claesson J. Multipole method to compute the conductive heat ows
to and between pipes in a composite cylinder; 1987. Technical Report. Lund,
Sweden.
[3] Carslaw H, Jaeger J. Conduction of heat in solids. 2nd ed. Oxford: Clarendon
Press; 1959.
[4] Cimmino M, Bernier M, Adams F. A contribution towards the determination of
g-functions using the nite line source. Appl Therm Eng 2013;51:401e12.
[5] Claesson J, Javed S. An analytical method to calculate borehole uid temperatures for timescales from minutes to decades. ASHRAE Trans 2011;117(2):
279e88.
[6] Comsol AB. Comsol Multiphysics, version 4.3. Comsol AB; 2012.
[7] Cui P, Yang H, Fang Z. Heat transfer analysis of ground heat exchangers with
inclined boreholes. Appl Therm Eng 2006;26:1169e75.
[8] De Carli M, Tonon M, Zarrella A, Zecchin R. A computational capacity resistance model (CARM) for vertical ground-coupled heat exchangers. Renew
Energy 2010;35:1537e50.
[9] Eskilson P. Thermal analysis of heat extraction boreholes. Ph.D. thesis. Sweden: University of Lund; 1987.
[10] Ingersoll L, Zobel O, Ingersoll A. Heat conduction with engineering, geological
and other applications. New York: McGraw-Hill; 1954.
[11] Lamarche L. A fast algorithm for the hourly simulations of ground-source heat
pumps using arbitrary response factors. Renew Energy 2009;10:2252e8.
[12] Lamarche L. Analytical g-function for inclined boreholes in ground-source
heat pump systems. Geothermics 2011;40:241e9.
[13] Lamarche L, Beauchamp B. A fast algorithm for the simulation of GCHP systems. ASHRAE Trans 2007;113(1). DA-07-050.
[14] Lamarche L, Beauchamp B. A new contribution to the nite line-source model
for geothermal boreholes. Energy Build 2007;37:188e98.
[15] Lazzarotto A. A network-based methodology for the simulation of borehole
heat storage systems. Renew Energy 2014;62:265e75.
[16] Marcotte D, Pasquier P. Fast uid and ground temperature computation for
geothermal ground-loop heat exchanger systems. Geothermics 2008;37:
651e65.
[17] Marcotte D, Pasquier P. The effect of borehole inclination on uid and ground
temperature for GLHE systems. Geothermics 2009;38:392e8.
[18] Marcotte D, Pasquier P, Sheriff F, Bernier M. The importance of axial effects for
borehole design of geothermal heat-pump systems. Renew Energy 2008;35:
763e70.
[19] Pasquier P, Marcotte D. Short-term simulation of ground heat exchanger with
an improved TRCM. Renew Energy 2012;46:92e9.
[20] Pasquier P, Marcotte D. Efcient computation of heat ux signals to ensure the
reproduction of prescribed temperatures at several interacting heat sources.
Appl Therm Eng 2013;59:515e26.
[21] Pasquier P, Marcotte D, Kummert M, Bernier M. Simulation of ground-coupled
heat pump systems using a spectral approach. In: Proceedings of BS2013:
13th conference of international building performance simulation association,
Chambry, France; 2013. pp. 2691e8.