Solar Energy 177 (2019) 241–254

Contents lists available at ScienceDirect

Solar Energy
journal homepage: www.elsevier.com/locate/solener

Apparent delay analysis for a flat-plate solar field model designed for control T
purposes
⁎
Gary Ampuñoa, Lidia Rocab, , Juan D. Gilc, Manuel Berenguelc, Julio E. Normey-Ricod
a
Universidad Politécnica Salesiana, Guayaquil, Ecuador
b
CIEMAT-Plataforma Solar de Almería, Centro Mixto CIESOL, Tabernas, Spain
c
Universidad de Almería, Centro Mixto CIESOL, ceiA3, Almería, Spain
d
Federal University of Santa Catarina, Department of Automation and Systems, Florianópolis, Brazil

A R T I C LE I N FO A B S T R A C T

Keywords: This study presents an analysis of the effect of the transport delay which occurs in solar flat-plate collector fields
Dynamic simulation and how to include its behavior in dynamic models suitable for control purposes. This investigation has been
Solar energy carried out using simplified models based on dynamic energy balances and models based on step response
Bilinear model methods and experimental tests. The solar flat-plate collector field encompasses parallel absorber pipes inter-
twined with pipelines which act as several first-order plus dead time systems in parallel. The effect is an apparent
delay observed at the outlet temperature that must be included in the dynamic model in order to reduce the error
between the real measurement and the model prediction. The main contribution of this paper is the procedure to
evaluate the apparent delay to obtain adequate dynamic models aimed to be used for control purposes.

1. Introduction of the most common models are the steady-state model following the
ASHRAE 93-77 standard and the quasi-dynamic model described in the
The major purpose of a distributed solar collector (DSC) field is to EN 12975 standard (Fischer et al., 2004). Both models describe an
collect the maximum solar energy available to produce as much thermal energy balance but while the ASHRAE 93-77 model considers only the
energy as possible. This thermal energy can be used for different pro- effect of the beam radiation and heat losses, the quasi-dynamic model
cesses such as electricity generation or seawater desalination (Camacho adds the effect of the diffuse radiation, wind effects, and a heat losses
et al., 1997; Ali et al., 2011). An improvement of the process operation term to describe the long-wave radiation. These models are mainly used
can be usually attained by controlling the outlet temperature of the for thermal performance characterization or for long-time evaluations.
solar field. In addition, the use of solar energy mitigates the impact of But, for transient studies, dynamic models are required (Kicsiny, 2017;
greenhouse emissions, especially the CO2 caused by different kinds of Aleksiejuk et al., 2018).
industrial processes (Ackermann et al., 2015). Regarding dynamic models based on energy balance equations,
Various types of solar thermal collectors and their applications are classifications in literature are based on the number of differential
presented in Camacho et al. (2012), including flat-plate, compound equations considered. The one-node lumped parameter model consists
parabolic, evacuated tube, parabolic-trough, Fresnel lens, parabolic on one differential equation for the fluid, it is easy to implement and the
dish and heliostat fields. As described in Kalogirou (2004), Duffie and results obtained are quite satisfactory. One example can be found in
Beckman (2013), flat-plates are non-tracking collectors with a ‘‘black’’ Pasamontes et al. (2009) in which a flat-plate solar collector field is
solar energy-absorbing surface designed to transfer the absorbed energy modeled to evaluate the water temperature that flows to an absorption
to a fluid, a transparent cover reduces convection and radiation heat heat pump or in Gil et al. (2018) in which a small flat-plate solar field
losses to the atmosphere and a back insulation reduces conduction heat provides thermal energy to a membrane distillation system. The two-
losses. In order to evaluate and improve the system performance, node lumped parameter models add a differential equation to consider
mathematical modeling is an important and established widely used the temperature of the absorber plate and a heat transfer term to in-
tool. clude the convection from the absorber plate to the heat transfer fluid.
A detailed review of different models for flat-plate solar collectors In (Deng et al., 2015) these two equations are combined to eliminate
can be found in Nayak and Amer (2000), Tagliafico et al. (2014). Two the dependence with the absorber plate temperature and to reach a

⁎
Corresponding author.
E-mail address: [email protected] (L. Roca).

https://doi.org/10.1016/j.solener.2018.11.014
Received 29 August 2018; Received in revised form 31 October 2018; Accepted 4 November 2018
0038-092X/ © 2018 Elsevier Ltd. All rights reserved.
G. Ampuño et al. Solar Energy 177 (2019) 241–254

second-order differential equation for the heat transfer fluid. From a technology research, development and test center in Europe. This fa-
validation process described in Deng et al. (2017), authors emphasize cility has a flat-plate solar field which provides thermal energy to a
the importance of reducing the temperature measurement error to Multi-Effect Desalination (MED) system (Fernández et al., 2012;
make useful the second order term in the differential equation. Alarcón et al., 2005, 2007, 2010) and to a membrane distillation (MD)
Distributed parameter models based on energy balance in the ab- system, and was designed and built with the purpose of piloting studies
sorber plate and in the heat transfer fluid have also been used and of large-aperture static solar collector panels when joined with desali-
validated with good results for the case of flat-plate solar collectors nation processes. Fig. 1 displays the simplified schematic diagram of the
(Pasamontes et al., 2013). In this case the energy balance is described plant. Basically, it is made of two main circuits: the primary circuit and
by partial differential equations and the temperature can be evaluated the secondary circuit. Heat generation is performed in the primary
along the absorber tube dividing the model in different control vo- circuit. Thermal storage to be used for the MED unit takes place at the
lumes. When high accuracy is required for the evaluation of the tem- secondary circuit.
perature or velocity inside tubes, computational fluid dynamic (CFD) As shown in Ampuño et al. (2018) the primary circuit is constituted
software is a suitable tool. This kind of models include continuity, by a solar field, an air cooler, and a heat exchanger and connection
momentum and energy balances for the heat transfer fluids in three pipes. The static solar field has 60 collectors arranged in five loops
dimensions and they can be very useful for design optimization. In the linked in parallel. There are four loops with 14 flat-plate collectors each
study developed by (Facão, 2016) a CFD analysis is carried out with the (each loop has two rows or sections connected in series with 7 collectors
aim of designing the manifolds in a flat-plate collector which improves in parallel per row) and one loop with 4 flat-plate collectors connected
the flow distribution along the absorber tubes. in parallel. Each loop has its own pumping system so it can be operated
Nevertheless, for control purposes, simplified dynamic models are independently. As the main energy source, the solar irradiance, cannot
required to make the control algorithms more efficient from the com- be controlled (it acts as a disturbance from the control viewpoint), the
putational point of view. Although one-node lumped parameter models outlet temperature of the solar field is usually controlled by manip-
can be very useful as commented previously, transfer functions (Buzás ulating the fluid flow rate using the pumps (Roca et al., 2009).
and Kicsiny, 2014; Kicsiny, 2017; Aleksiejuk et al., 2018), and black- Depending on the operating requirements (solar field characteriza-
box identification (Brus and Zambrano, 2010) are other techniques tion, desalination, process heat, …) the hot fluid from the solar field
used to model flat-plate collectors. In the first case, authors obtain enters an air cooler or directly reaches a heat exchanger to heat the
different transfer functions (one for each input) based on one-node primary water circuit where thermal storage tanks are placed. Table 1
lumped parameter model. In the second case, the model is obtained describes the nomenclature used in this work.
using a recursive prediction error method.
One issue that it is not completely tackled in flat-plate solar col- 3. Bilinear model for a solar-field loop
lector models is the delay. The most common delay included in the
models is the one related directly with the transport delay in the inlet A solar collector field can be described by a concentrated parameter
temperature such as in Pasamontes et al. (2009). This delay varies model of the temperature. The model presented in Roca et al. (2008,
because of the transport time the heat transfer fluid takes to reach the 2009) has been used as reference (see Eq. (1)). This concentrated
sensor, it depends on the fluid flow rate and can be evaluated using the parameters model, which provides the evolution of the temperature at
approximation proposed by Normey-Rico et al. (1998). Brus and the outlet of loop j section w, Tj,w, out (t ), has been applied to each section
Zambrano (2010) added two delays affecting the outlet temperature of each loop independently. The model considers that the behavior of
caused by two inputs; the flow rate and irradiance. From experimental each loop-section can be characterized taking into account just one flat-
data, linear approximations are obtained to evaluate these delays and to plate absorber tube and that the loop water flow rate is equally divided
include them in the developed model. But, the source of these delays is in each tube. Therefore, the volumetric flow rate considered as input in
not explained. this model is Qj / cf where parameter cf includes the number of tubes in
This paper focuses on a flat-plate solar field model putting emphasis each loop-section and the L·min−1–L·s−1 conversion factor.
on the delay treatment; the source of the delay, how to include the
∂Tj, w, out (t ) H
delay in a lumped parameter model and approximations to reduce the ρcp Acs = βI (t ) − (Tm (t ) − Ta (t ))
∂t L
computational time to use the model for control purposes. This paper
ρ (Tj, w, out (t ) − Tj, w, in (t ))
follows a previous paper (Ampuño et al.,2018) where the flat-plate − cp Qj (t ) ,
collector dynamic model was presented. In that paper, the delays were cf L (1)
experimentally obtained and included in the model. This new paper Tj, w, out (t ) + Tj, w, in (t )
focuses in analyzing the causes of the observed delay, it demonstrates Tm (t ) = .
2 (2)
the existence of an apparent delay and the equations obtained are in-
cluded in the model to improve the results, especially under transient As explained in Ampuño et al. (2018), the outlet temperature of the
conditions. loop depends on several inputs, the manipulated one, Qj (t ) is the vo-
The paper is organized as follows: first, AQUASOL-II plant is shown lumetric flow [L·min−1], and the other ones act as disturbances: the
in Section 2, then the bilinear model, experimental campaign, calibra- inlet temperature, Tj,w, in (t ) , the ambient temperature, Ta (t ) , and the
tion and validation results are shown in Section 3, in Section 4 the delay solar irradiance, I (t ) . The remaining parameters are shown in Table 1.
treatment and the models considered are explained, in Section 5 a
simplified model to reduce the computational time and make the model 3.1. Experimental campaign
easier to be used for control purposes is proposed and in Section 6
experimental results are fully discussed. Some concluding remarks are In order to calibrate and validate the solar field model, dynamic
included at the end. data from AQUASOL II facility was collected. The experimental cam-
paign started on November 16th, 2016 and ended on December 21st,
2. System description 2016. The following data details the methodology carried on to perform
the tests:
The solar desalination facility AQUASOL-II (see Fig. 1) is placed at
Plataforma Solar de Almeria (PSA), a dependency of the Centro de In- • The secondary circuit of the AQUASOL II facility (see Fig. 1) is
vestigaciones Energéticas, Medioambientales y Tecnológicas (CIEMAT) stopped, therefore no water flows between the heat exchanger and
in the south of Spain, which is the largest concentrating solar the thermal storage system.

242
G. Ampuño et al. Solar Energy 177 (2019) 241–254

Fig. 1. Schematic diagram of the solar MED facility at PSA and photo of the solar field.

• The solar field inlet temperature is maintained at different values • Water flow rates in loops 3, 4 and 5 have been varied between
using the air cooler. A PID controller with anti-windup regulates the 18 L·min−1 and 36 L·min−1. Water flow rate in loop 1 has been
air-cooler fan speed to maintain the desired temperature at the varied between 7 L·min−1 and 18 L·min−1.
outlet of the air-cooler. Since the secondary circuit is stopped, the • Loop 2 was out of service during the experimental campaign.
air-cooler outlet temperature can be considered as the solar field • Tests were performed during sunny and cloudy days to analyze the
inlet temperature at steady-state conditions. The inlet temperature effect of real irradiance disturbances in the model.
range considered is from 50 °C to 85 °C. • The testing schedule ranges from 9:00am until 14:00 pm local time.

Table 1
Nomenclature.
Symbol Quantity Unit

A Cross-section area of the loop pipe m2

Acs Collector absorber cross-section area m2
β Parameter that modulates the solar irradiance m
cp Specific heat capacity J·kg−1·°C−1
cf Conversion factor to account for number of flat-plate tubes in each loop, and L·min−1 – L·s−1 conversion s·L·min−1·m−3
cf 2 Conversion factor to account for L·min−1 – m3·s−1 conversion s·L·min−1·m−3
dj, w, l, in Transport delay in the incoming water to each one of the absorber pipes that are connected in parallel of DSC loop j ( j = 1. .5) section w (w = 1..2) s
absorber pipes number l (l = 1..350)
dj, w, l, out Transport delay in the outlet water from each one of the absorber pipes that are connected in parallel of DSC loop j ( j = 1. .5) section w (w = 1..2) s
absorber pipes number l (l = 1..350)
dj, tout − tin Inlet - outlet water temperature related transport delay in loop j (j = 1. .5) s
dj, tout − Q Inlet water flow rate - outlet water temperature related apparent delay in loop j (j = 1. .5) s
dj, tout − I Inlet solar irradiance - outlet water temperature related apparent delay in loop j (j = 1. .5) s
dj, tout − tin, s Inlet-outlet water temperature related transport delay in loop j (j = 1. .5) for the simplified model s
dj, tout − Q, s Inlet water flow rate - outlet water temperature related apparent delay in loop j (j = 1. .5) for the simplified model s
dj, tout − I , s Inlet solar irradiance-outlet water temperature related apparent delay in loop j (j = 1. .5) for the simplified model s
H Thermal losses coefficient J·s−1·°C−1
I Solar irradiance W·m−2
L Absorber pipe length m
Lca Distance between two absorber pipes m
Lloop Length of the flat-plate collector loop m
n Number of absorber pipes in each loop-section (n = 200 for DSC loop 1 and n = 350 for DSC loops j = 2..5) –
N Number of samples –
Qj Volumetric flow of DSC loop j (j = 1. .5) L·min−1
Qca, j, w, l Water flow rate that circulates between absorber pipes l and l + 1 (l = 1..n) of DSC loop j (j = 1. .5) section w (w = 1..2) L·min−1
Ta Ambient temperature °C
Tj, w, in Inlet temperature of DSC loop j (j = 1. .5) section w (w = 1..2) °C
Tj, w, l, in Inlet temperature of absorber pipe l (l = 1..n), DSC loop j (j = 1. .5) , section w (w = 1..2) °C
Tj, w, out Outlet temperature of DSC loop j (j = 1. .5) section w (w = 1..2) °C
Tj, w, l, out Outlet temperature of absorber pipe l (l = 1..n), DSC loop j (j = 1. .5) section w (w = 1..2) °C
Ts, j, w, in Inlet temperature measured by the sensor of DSC loop j (j = 1. .5) , section w (w = 1..2) °C
Ts,j,w,out Outlet temperature measured by the sensor of DSC loop j (j = 1. .5) , section w (w = 1..2) °C
St Sampling time s
Tm Equivalent flat plate tube mean temperature °C
v Velocity rate m·s−1
ρ Water density kg·m−3
yi Plant outputs at sample i

yi Model outputs at sample i

243
G. Ampuño et al. Solar Energy 177 (2019) 241–254

• Three different kind of tests have been performed: (a) solar field One of the objectives of this study is to compare the solar field
temperature remains constant during the whole test and flow rate model under different kind of disturbances, such as variable solar ir-
changes are performed in each loop, (b) flow rates remain constant radiance due to passing clouds. On December 14th (shown in Fig. 4),
during the whole test and inlet temperature changes are performed cloudy conditions can be appreciated from local time 11:45am (11.75)
and (c) inlet temperature and flow rate changes are performed at the until 12:30 pm (125). For the case of loop 5 section 2 shown in Fig. 4,
same time. the water flow rate was maintained constant at 36 L·min−1 and the inlet
temperature varies between 55 °C and 75 °C. This figure shows that the
3.2. Calibration outlet temperature calculated by the model is similar to the real outlet
temperature of the solar field in a clear day (from 10:30 am to
Parameter calibration is usually a challenging task and its com- 11:45 am). When hard disturbances in irradiance occur, the model be-
plexity rises with the number of unknown variables. In this work, two havior also shows sudden changes, conversely to the real data, that
parameters must be calibrated; the irradiance parameter, β , and the presents a smoother profile. Section 4 deals with this sudden solar ra-
thermal losses coefficient, H (see Eq. (1)). With the obtained informa- diation change.
tion from the experimental campaign and the least squares method, This example also shows the delay effect between the two inputs,
these parameters were precisely and accurately calibrated. The least irradiance and inlet temperature, and the outlet temperature; when the
squares method is an optimization technique applied to linear in clouds appear and the solar irradiance falls, the outlet temperature has
parameters systems. The aim is to find the model output, the outlet a delay of around 216 s. Since the model does not include this delay, an
temperature in this case, that best approximates the real process output error of 7.85 °C is obtained.
according to the minimal sum of squared error function, so that, the Table 2 shows the RMSE and ME errors obtained in the validation
best combination of unknown parameters are obtained from this opti- campaign using the bilinear model for each solar-field loop section.
mization. An outline of numerous least square concepts can be found in Comparing the model results with real data, the ME errors are far from
Goodwin and Payne (1977). the expected results due to the fact that the delays observed in the real
Following this method, the obtained unknown parameters are those systems are not implemented in the model.
shown in Table 2, where errors are depicted in terms of maximum error,
ME, and the root mean square error, RMSE, estimated as: 4. The apparent and transport delays

1 N
RMSE =
N
∑i =1 (yi − yi )̂ 2 , (3)
Figs. 3 and 4 present a comparison between the temperature cal-
culated by the model and the temperature measured at the end of the
N being the number of samples from the experiment, and yi and yi ̂ return pipe. These figures intend to demonstrate that the dynamics of
are the plant and model outputs at sample i. It is important to mention the water temperature measured can be approximated using the pro-
that the sampling time considered, St, is 1 s. posed model described in Eq. (1) but it is mandatory to take into con-
The mean RMSE errors for each loop-section model are acceptable, sideration a variable dead-time to avoid huge errors during transients.
but the ME error is far from the desired results. This huge deviation has As will be explained in this section, the source of this delay that cause
been analyzed and it is caused by the delay between the outlet tem- fluctuations in the behavior of the collector water outlet temperature is
perature and the different inputs (inlet temperature, water flow rate a transport delay at the inlet and outlet of each absorber pipe that
and irradiance). Fig. 3 shows the results obtained with the model using produces an apparent delay at the outlet of the loop.
the data collected in November 10th, 2016 (see also solar irradiance
and ambient temperature depicted in Fig. 2). If the model does not take 4.1. The transport delay
into account the delay, huge errors can be obtained between the model
response and the real data during transients. Notice for example the Fig. 5 comprises three images showing the flat-plate collector solar
change performed at the inlet temperature of loop 1 section 1 between field and how the absorber tubes are connected to each other. Since the
13 h and 13.5 h; the real outlet temperature delays about 270 s to start inlet temperature sensor is located at the beginning of the loop (Ts, j, w, in ),
increasing, which causes a maximum error of 11.1 °C. Section 4 goes there is a transport delay in the incoming water to each one of the
into detail about the origin of this delay and how to model it. absorber pipes that are connected in parallel (dj, w, l, in ) where j is the loop
number ( j ∈ [1, 5]) , w the loop section (w ∈ [1, 2]) and l the absorber
3.3. Validation tube number (l ∈ [1, 350] ∀ j = {2⋯5}, l ∈ [1, 200] ∀ j = 1) . This trans-
port delay can be estimated at each sampling time as a flow-dependent
For the validation procedure, the performance of the model has delay as described by Normey-Rico et al. (1998).
been evaluated using a different data set. These tests present variability Tj, w, l, in (t ) ≈ Ts, j, w, in (t − dj, w, l, in ), (4)
in all the inputs: irradiance, inlet temperature, ambient temperature
and water flow rate. dj, w, l, in
St
i=m−1
Lca = ∫ v (t ) dt →
A
∑ Qca, j, w, l (k − i)·cf 2 = Lca ,
0 i=0 (5)
Table 2
Validation campaign results using the bilinear model. where k is the actual sampling time, St [s] is the sampling time, v (t ) is
Loop, section Parameters Global error the water velocity rate [m s−1], index m is equal to the delay (in
sampling times), Qca, j, w, l (k − i) [L·min−1] is the flow rate at sampling
H β RMSE ME time k − i that circulates between two absorber tubes connections, cf2 is
a conversion factor to account of the L·min−1 – m3·s−1 conversion, A is
[J s−1 °C−1] [m] [°C] [°C]
the cross-section area of the loop pipe [m2]; and the integral of v (t ) is
5,1 1.095 0.0959 2.410 11.790 approximated by a discrete time sum that accounts for different flow
5,2 1.106 0.0956 1.680 11.400 rates. The input, Tj, w, l, in (t ) , is estimated in every sample time using past
4,1 1.137 0.0949 1.640 7.900 measures of the inlet temperature Ts, j, w, in (t ) , and evaluating the delay
4,2 1.159 0.097 1.560 10.810 dj, w, l, in with the past flow rates.
3,1 1.101 0.0951 1.830 7.650
The procedure described in Eq. (5) must be repeated to estimate the
3,2 1.121 0.0946 1.690 10.890
1 1.107 0.094 1.790 7.550 inlet temperature for each absorber pipe inside the loop because the
volumetric flow rate is different for each connection. With the aim of

244
G. Ampuño et al. Solar Energy 177 (2019) 241–254

Fig. 2. Experimental disturbances, November 10th, 2016.

reducing the computational cost, it is possible to estimate the delay dj, w , n − 1

dj, w, l, in = ∀ l ∈ [1, n − 1],
between the sensor located at the beginning of the loop and the inlet of n−l (7)
each absorber pipe by only evaluating the delay between the sensor and
Notice that these delays are between the connections of two ab-
the inlet of the last absorber pipe.
sorber pipes. The inlet temperature to each absorber pipe will be:
Since the distance between two absorber pipes is the same
(Lca =Lloop / n) and the volumetric flow rate inside each absorber pipe is ⎛
i=l

the same (Qj / n) (see Fig. 6), it is easy to follow that the volumetric flow Tj, w, l, in = Ts, j, w, in ⎜t − ∑ dj,w,i⎞⎟,
rate in the connection between absorber pipe l and l + 1 can be esti- ⎝ i=1 ⎠ (8)
mated as: The other transport delay observed in the flat plate collectors is
present at the outlet of the distributer pipe which affects to the loop
l
Qca, j, w, l, in = Qj ⎛1 − ⎞ ∀ l ∈ [1, n − 1]. outlet temperature Tj, w, out . Due to the connections between the absorber
⎝ n⎠ (6) pipes in the upper part (see Fig. 7), the volumetric flow rate is different
in each connection, giving rise to different transport delays. Using Eq.
Then, to estimate the delay dj, w,(n − 1), in between absorber pipe n − 1
(5), it is possible to calculate the delay in each connection and to assess
and n , we have to consider the volumetric flow rate Qj / n and the length
the delay between the outlet of each absorber tube and the sensor lo-
Lloop / n in Eq. (5).
cated at the outlet of the collector loop.
To calculate the delay between absorber pipe n − 2 and n − 1with
For this, the volumetric flow rate in the connection between ab-
Eq. (5), it is necessary to consider the volumetric flow rate 2·Qj / n and
the lengthLloop / n , but it is easy to deduce that this delay will exactly be sorber pipe l and l + 1 can be estimated as:
the half of the one calculated previously because it is the time required l·Qj
to go across a pipe with the same cross section area and length but with Qca, j, w, l, out = ∀ l ∈ [1, n − 1],
n (9)
the double volumetric flow rate. So, it can be concluded that:
Then, to estimate the delay dj, w,1, out between the absorber pipe 1 and
2 the volumetric flow rate Qj / n and the length Lloop / n has to be

Fig. 3. Solar field model validation experiment performed in loop 1, section 1, November 10th, 2016.

245
G. Ampuño et al. Solar Energy 177 (2019) 241–254

Fig. 4. Experimental disturbances and simulation results using the solar field model, loop 5, section 2, December 14th, 2016.

considered in Eq. (5). dj, w,1

dj, w, l, out = ∀ l ∈ [1, n].
To calculate the delay between absorber tube 2 and 3 with Eq. (5), nl (10)
the volumetric flow rate 2·Qj / n and the lengthLloop / n have to be con-
sidered, but the number of equations can be simplified again following Notice that these delays are between the connections of two ab-
the same reasoning used previously: sorber tubes. The outlet temperature of the loop j section w will be:

Fig. 5. AQUASOL II solar field and absorber tubes connections.

246
G. Ampuño et al. Solar Energy 177 (2019) 241–254

rate, inlet temperature and irradiance. From these steps, first-order

models are obtained, the results are compared with real data and ex-
pressions are obtained to evaluate those apparent delays as a function
of the flow rate.
The linear approximation is usually found by the Taylor’s series
expansions of the non-linear vector field in the vicinity of a steady state
(equilibrium). The approximation depends on the system parameters
and the specific steady state (an explanation of nonlinear models line-
arization can be found in Ogata (1998)).
As a first stage, considering the first section of a collector loop, first-
order models relating the water outlet temperature to each one of the
inputs have to be obtained. These first-order models can be obtained
Fig. 6. Inlet temperature delay within the inlet distributer pipe. from the bilinear model described in Eq. (1) for a steady state
(ρc Ap cs
∂Tj,w, out (t )
∂t
=0 .)
cp ρ H T¯j, w, out + T¯j, w, in
0 = βI¯ − Q¯ j (T¯j, w, out − T¯j, w, in ) − ⎛⎜ − T¯a⎞⎟,
cf L L⎝ 2 ⎠ (12)

where the linearization is carried out around the operating points de-
fined by (Q¯ j , T¯j, w, out , T¯j, w, in, T¯a, I¯ ). System variables
Qj (t ), Tj, w, out (t ), Tj, w, in (t ), Ta (t ) and I (t ) are represented as the sum of
the steady-state value that defines the operating point with the value of
the deviation variable around that equilibrium state
∼ ∼ ∼ ∼ ∼
(Qj (t ), T j, w, out (t ), Tj, w, in (t ), Ta (t ), I (t )) :
∼
Qj (t ) = Q¯ j + Qj (t ),
∼
Tj, w, out (t ) = T¯j, w, out + Tj, w, out (t ),
∼
Fig. 7. Outlet temperature delay within the outlet distributer pipe. Tj, w, in (t ) = T¯j, w, in + Tj, w, in (t ),
∼
Ta (t ) = T¯a + Ta (t ),
∼
i=n ii = n
∑i = 1 Tj, w, i, out (t − ∑ii = 1 dj, w, ii ) I (t ) = I¯ + I (t ), (13)
Tj, w, out (t ) =
n (11) The deviation variables represent the linear dynamic system when
Therefore, to evaluate the water temperature at the outlet of each variations occur in the dynamics of the system around the operating
loop section, the temperature at the outlet of each absorber pipe has to point defined by (Q¯ j , T¯j, w, out , T¯j, w, in, T¯a, I¯ ).
be estimated, which requires solving 3000 bilinear equations described Using the Taylor’s series approximation in the non-linear part of the
by Eq. (1). In order to reduce the required computational time, the model, the following linearization is obtained:
concept of apparent delay is introduced in the following section to ∼
Qj (t )
∂Qj (t ) Tj, w, out (t ) ⏞
explain the approximation considered to evaluate the water tempera- Qj (t ) Tj, w, out (t ) ≈ Q¯ j T¯j, w, out + (Qj (t ) − Q¯ j )
ture at the outlet of each loop-section. ∂Qj (t ) Q¯ j, T¯ j, w, out
∼
Tj, w, out (t )
∂Qj (t ) Tj, w, out (t ) ⏞
4.2. The apparent delay + (Tj, w, out (t ) − T¯j, w, out )
∂Tj, w, out (t ) Q¯ j, T¯ j, w, out

As described by Normey-Rico et al. (2007), when there are several ∼

Qj (t )
systems connected in parallel, as happens in the absorber pipes, and ∂Qj (t ) Tj, w, in (t ) ⏞
Qj (t ) Tj, w, in (t ) ≈ Q¯ j T¯j, w, in + (Qj (t ) − Q¯ j )
each of them has a small delay, the accumulation of every one of these ∂Qj (t ) Q¯ j, T¯ j, w, in
delays produces an apparent delay. In this case the apparent delay ∼
Tj, w, in (t )
depends on the number of absorber pipes of the loop, n . ∂Qj (t ) Tj, w, in (t ) ⏞
The idea behind this section is to evaluate the apparent delay ob- + (Tj, w, in (t ) − T¯j, w, in )
∂Tj, w, in (t ) Q¯ j, T¯ j, w, in
served when step changes are performed at the following inputs: flow

a) Schematic diagram when the input u is the inlet b) Schematic diagram when the input u is the
temperature water flow rate, the irradiance or the ambient
temperature.

Fig. 8. Outlet temperature model combining first-order models.

247
G. Ampuño et al. Solar Energy 177 (2019) 241–254

Fig. 9. Experimental disturbances, outlet temperature and simulation results when a flow rate step input is applied, loop 5, section 1, December 21st 2016.

Fig. 10. Experimental disturbances and simulation results using solar irradiance step input, loop 5, section 1, December 14th2016.

∼ ∼ ∼
Qj (t ) Tj, w, out (t ) ≈ Q¯ j T¯j, w, out + T¯j, w, out Qj (t ) + Q¯ j Tj, w, out (t ) (14) ∂Tj, w, out (t )
ρcp Acs
∂t
∼ ∼ ⏞
0
Qj (t ) Tj, w, in (t ) ≈ Q¯ j T¯j, w, in + T¯j, w, in Qj (t ) + Q¯ j Tj, w, in (t ) cp ρ cp ρ H T¯j, w, out + T¯j, w, in
= βI¯ − Q¯ T¯j, w, out + Q¯ j T¯j, w, in − ⎜⎛ − T¯a ⎟⎞
∂Tj, w, out (t )
∼
∂Tj, w, out (t )
cf L j
cf L L⎝ 2 ⎠
Since ρcp Acs = ρcp Acs , the bilinear model can be cp ρ cp ρ ∼ cp ρ
∂t ∂t ∼ ∼ ¯ ∼
approximated as: + βI (t ) − T¯j, w, out Qj (t ) − Qj Tj, w, out (t ) + T¯j, w, in Qj (t )
cf L cf L cf L
∼ ∼ ∼
∂Tj, w, out (t ) ∼ cp ρ ∼ H ⎛ Tj, w, out (t ) + Tj, w, in (t ) ∼ ⎞
ρcp Acs = β (I¯ + I (t )) + Q¯ j Tj, w, in (t ) − ⎜ − Ta (t )⎟,
∂t cf L L 2
⎝ ⎠
cp ρ ∼ ∼
− ((Q¯ j T¯j, w, out + T¯j, w, out Qj (t ) + Q¯ j Tj, w, out (t ))
cf L
∼
∼ ∼ ∂Tj, w, out (t ) ∼ cp ρ ∼
− (Q¯ j T¯j, w, in + T¯j, w, in Qj (t ) + Q¯ j Tj, w, in (t ))) ρcp Acs = βI (t ) − ((T¯j, w, out − T¯j, w, in ) Qj (t )
∼ ∼ ∂t cf L
H ⎛ T¯j, w, out + Tj, w, out (t ) + T¯j, w, in + Tj, w, in (t ) ∼ ∼
− ⎜ + Q¯ j (Tj, w, out (t ) − Tj, w, in (t )))
L 2
⎝ ∼ ∼
H ⎛ Tj, w, out (t ) + Tj, w, in (t ) ∼ ⎞
∼ ⎞ − ⎜ − Ta (t ) ⎟.
− (T¯a + Ta (t )) ⎟. L
⎝
2
⎠ (15)
⎠
∼ ∼ ∼ ∼
Considering that small variations of Qj (t ), Tj, w, in (t ), Ta (t ) and I (t )
Taking into account steady state relationships: are produced around the equilibrium point
∼ ∼ ∼ ∼
Qj (t) = ΔQj , Tj, w, in (t) = ΔTj, w, in, Ta (t) = ΔTa, I (t) = ΔI , a temporal evo-
∼
lution of Tj, w, out (t ) from T̄j, w, out to a new state value T¯j, w, out + ΔTj, w, out is
produced and the linearized system can be represented by a first order
(
∂y (t )
model τ ∂t + y (t ) = ku (t ) : )

248
G. Ampuño et al. Solar Energy 177 (2019) 241–254

Fig. 11. Outlet temperature time delay curve. Black markers are the experimental data.

⎡ − cp ρ (T¯j, w, out − T¯j, w, in ) ⎤ first-order model because the variation of this input is negligible inside
∼
ρcp Acs ∂Tj, w, out (t ) ∼ ⎢ cf L ⎥∼ a temporal window equivalent to the observed delay.
+ Tj, w, out (t ) = ⎢ ⎥Qj (t )
cp ρQ¯ j H ∂t cp ρQ¯ j H Once the models have been implemented, the results have been
⎡ + ⎤ ⎢ ⎡ + ⎤ ⎥
cf L 2L
⎣ ⎦ ⎣ ⎣ cf L 2L
⎦ ⎦ compared to real data. Since the goal is to evaluate the apparent delay
⏟
τ ⏟kQ obtained using first-order models, the other inputs must be maintained
β ∼ ∼ as constant as possible. Therefore, to compare the experimental and real
+ I (t )− 1 ·Tj, w, in (t )
cp ρQ¯ j H ⏟ delay related with the flow rate, irradiance and inlet temperature must
⎡ + ⎤ kTin
⎣ cf L 2L
⎦ be maintained constant.
⏟
kI Fig. 9 represents the recording data of December 21st, 2016. The
H test consists in applying a step signal (from 35 L·min−1 to 30 L·min−1)
L ∼
+
cp ρQ¯ j
Ta (t ). at the inlet flow of the loop 5, Q5 . The linearized model is obtained
H
⎡ + ⎤ around an operational point set up by the outlet temperature around
cf L 2L
⎣ ⎦
k⏟ (16) 73 °C, the irradiance level is about 860 W m−2 and the inlet tempera-
Ta
ture was maintained constant in 55 °C. Notice that the variation of the
For each absorber pipe and for each input, the steady-state gain and ambient temperature in that time interval is less than 0.5 °C and the
the time constant are almost the same (the properties of the water can contribution of this input in the linearized model is neglected.
be slightly different) but the output delay will be different according to Fig. 9 shows that the lumped parameter model gives a result
the transport delay explained in Section 4.1. Considering the first order without including any type of delay. On the other hand, the set of n first
models obtained for each absorber pipe, the outlet temperature of one order models obtained delivers the total temperature of the loop with
loop section as a function of each one of the inputs u (Qj , I , Tj, w, in, Ta) an apparent delay when there are changes in the inlet flow rate, which
can be evaluated by combining n first-order models and considering Eq. is the hypothesis set out at the beginning. As can be observed the ob-
(8) and Eq. (11) such as Fig. 8 shows. tained delay is quite similar to the real one.
Notice that the inlet delay is only considered when the input of the As commented previously, changes of the solar radiation over the
first-order models is the inlet temperature. Irradiance, ambient tem- solar field also generate an apparent delay at the outlet temperature
perature and water flow rate changes affect to the absorber pipes at the (see Fig. 10).
same time, the only delay in these cases is the transport delay of the
water to go across the outlet connection pipe. Nevertheless, the ambient
temperature is assumed constant and its effect is not evaluated with the

249
G. Ampuño et al. Solar Energy 177 (2019) 241–254

Fig. 12. Experimental disturbances and simulation results using the solar field model including delays, loop 5, section 1, December 14th, 2016.

4.3. The bilinear model with delays dj, tout − Q = 0.13·Qj 2 − 9.9·Qj + 210.17, (20)

From the results obtained in the above section, it is demonstrated dj, tout − I = 0.124·Qj 2 − 9.43·Qj + 201.38, (21)
the existence of the transport delay that can be treated as different input
apparent delays. With the aim of improving the results of the bilinear d1, tout − tin = 2.6·Q12 − 86.3·Q1 + 876.2, (22)
model, it is necessary to introduce these delays in the inputs, as it
d1, tout − Q = 0.11·Q12 − 3.87·Q1 + 64, (23)
shown in Eqs. (17) and (18).
∂Tj, w, out (t ) H d1, tout − I = 0.067·Q12 − 3.12·Q1 + 61.1, (24)
ρcp Acs = βI (t − dj, tout − I ) − (Tm (t ) − Ta (t ))
∂t L where j is the loop number ( j ∈ [2, 5]) and loop 1 is depicted by its own
ρ
− cp Qj (t −dj, tout − Q ) equation due to the difference in the number of absorber tubes.
cf
Although these equations have been obtained from steady state
(Tj, w, out (t ) − Tj, w, in (t − dj, tout − tin )) conditions, the delays are also a good approach for transient data as will
,
L (17) be shown in the following results.
If the delays in the form of quadratic functions are included in the
Tj, w, out (t ) + Tj, w, in (t − dj, tout − tin )
Tm (t ) = . bilinear model, the EM and RMSE errors are considerably reduced. As
2 (18)
an example, Fig. 12 compares the real data with the results obtained
Experiments in the real plant and in the linearized model (see with the bilinear model with and without delays aiming to demonstrate
Fig. 11) show that the relationship between flow rate and the delay can the improvements in the transient performance.
be approximated as quadratic functions. For the solar irradiance only For the case of loop 5 shown in Fig. 12, the water flow rate was
simulations with the linearized models have been used. maintained constant to 35 L·min−1. The inlet temperature was con-
Eqs. (19)–(24) give the approach obtained for the inlet- outlet time trolled with the air-cooler to 55 °C, but due to irradiance disturbances,
delay(dj, tout − tin ), flow rate-outlet time delay (dj, tout − Q ) and solar irra- it varies between 49 °C and 61 °C. This figure shows that under hard
diance time delay (dj, tout − I ) .: solar radiation disturbances, the model with delays improves the results
obtained in the previous section.
dj, tout − tin = 0.24·Qj 2 − 18.42·Qj + 480.7, (19) Another example is shown in Fig. 13 for the case of water flow rate

250
G. Ampuño et al. Solar Energy 177 (2019) 241–254

Fig. 13. Experimental disturbances and simulation results using the solar field model, loop 4, section 2, November 17th, 2016.

Table 3 and inlet temperature steps performed on November 17th, 2016. The
Campaign results using the bilinear model with delays. characteristics of this day are the following ones:

• In the test day, solar irradiance does not show cloudy conditions or
Loop-section Parameters Global error

H β RMSE ME transitions.
[J s−1 °C−1] [m] [°C] [°C]
• Ambient temperature varies between 18 °C and 21 °C.
5–1 1.095 0.0959 2.170 4.610 • Inlet temperature for Loop 4 varies between 45 °C and 60 °C in
5–2 1.106 0.0956 0.830 3.240 Section 1 and between 62 °C and 83 °C in Section 2. The inlet flow
4–1 1.137 0.0949 0.860 2.640 rate varies between 25 L·min−1 and 36 L·min−1.
4–2
3–1
1.159
1.101
0.097
0.0951
0.833
0.930
3.820
3.340
• The results obtained with the model including delays are quite si-
milar to the real data, being the biggest difference 3.1 °C at 13:15am
3–2 1.121 0.0946 0.960 4.280
1 1.107 0.094 0.990 3.670 (13.25) and 13:50 pm (13.5) that occurs by a hard disturbance in the
inlet temperature.

Table 4 The RMSE and ME errors obtained using the bilinear model with
Test campaign results with the simplified model. delays are included in Table 3. The results demonstrate a notorious
Loop Parameters Global error
reduction of the EM and RMSE errors in comparison with the errors
described in Table 2.
H β RMSE ME
[J s−1 °C−1] [m] [°C] [°C]
5. A simplified model for control purposes
5 1.145 0.0972 2.120 4.560
4 1.137 0.0948 1.660 4.030 As indicated in the preceding section, the solar field model study has
3 1.107 0.0926 1.720 4.610 been divided into two parts aimed to demonstrate the validity of solar
1 1.107 0.0936 1.790 5.910
field model; the bilinear model and the input delays. Parametric
equations to evaluate the delay for each input have been identified to
reduce the error of the model obtained mainly during transients.

251
G. Ampuño et al. Solar Energy 177 (2019) 241–254

Fig. 14. Experimental disturbances and simulation results using the solar field model, loop 3, section 2, November 16th, 2016.

Nevertheless, in order to obtain the water temperature at the outlet of These equations include the delays obtained in Eqs. (19)–(21), but
each loop, two bilinear models are required for each loop. With the aim multiplied by 2, because the simplified model incorporates the two
of reducing the computational time and making the model of each loop sections of each loop in only one model, this means, that number l of
easier to be used for control purposes, a simplified model is proposed in absorber pipes pipe (l ∈ [1, 700]) and the total distance from the be-
this section. The idea is to use just one bilinear equation to evaluate the ginning of the loop (Ts, j − k, in [°C]) to the last absorber (l = 700) has been
outlet temperature for each loop and include parametric equations to increased. Furthermore, the transport delay in the pipe between the end
take into account the delays at the input: of Section 1 and the beginning of Section 2 has been added to Eqs.
(26)–(28). This transport delay has been estimated using Eq. (5) as
∂Tj, out (t ) H ⎛ Tj, out (t ) + Tj, in (t − dj, tout − tin, s )
ρcp Acs = βI (t − dj, tout − I , s ) − ⎜ described by (Normey-Rico et al., 1998) as well as in Section 4.
∂t L⎝ 2 Once the delays are included, parameters H and β have been cali-
brated again (see Table 4) with data from the experimental campaign
− Ta (t ) ⎞⎟

⎠ carried on for the calibration process.

ρ Fig. 14 shows the model validation with data from the test carried
− cp Qj (t −dj, tout − Q, s ) on in November 16th, 2016 in which several changes were performed
cf
in the inlet flow rate. As can be seen, the results of the simplified model
(Tj, out (t ) − Tj, in (t − dj, tout − tin, s ))
, including the delays are very similar to real data. The characteristics of
L (25)
this day are the following ones:
where j is the loop number ( j ∈ [1, 5]) and the delays for the simplified
model are the following ones: • In the test day, solar irradiance does not show cloudy conditions or
2 transitions.
• Ambient temperature varies between 14 °C and 20 °C.
dj, tout − Q, s = 2·dj, tout − Q + 0.037·Qj − 3.37·Qj + 92.63 (26)

dj, tout − tin, s = 2·dj, tout − tin + 0.037·Qj 2 − 3.37·Qj + 92.63 (27) • Inlet temperature for the loop 3 remains almost constant at 50 °C.
−1 −1
The inlet flow rate varies between 18 L·min and 36 L·min .
2
dj, tout − I , s = 2·dj, tout − I + 0.037·Qj − 3.37·Qj + 92.63 (28) • The comparison between the simulated and the real loop outlet
temperature shows that the biggest difference is 3.53 °C at 09:52 am

252
G. Ampuño et al. Solar Energy 177 (2019) 241–254

Fig. 15. Experimental disturbances and simulation results using the simplified solar field model, loop 3, section 2, December 20th, 2016.

(9.88) and 3.1 °C at 11:16 am (11.28). simplified model results with the real ones are noticed, being the
biggest difference 3.59 °C at 12:00 pm.
Notice that using the simplified model the main source of error
comes from the fact that the inlet temperature of section two is not As shown in Table 4, a comparison between the results of the sim-
considered as model input. Although the outlet temperature of section 1 plified model with the real data outlet temperature is performed. For
should be the same than the inlet temperature of section 2 solar irradiance disturbance the simplified model reacts according to
(Tj, l, out ≈ Tj,2, in ) , in some occasions a non-desirable behavior at the inlet the outlet temperature dynamics but for the inlet flow rate changes the
temperature of section 2 has been detected; when flow rate steps are behavior can be a little bit different to the real data, as a result to
performed, inlet temperature in section 2 decreases immediately such the non-considered dynamic observed in some occasions in the second
as shown in Fig. 14. This dynamic, which is not included in the model, section inlet temperature. This makes that the results of the simplified
is the cause of the error obtained. In order to check the simplified model model not as good as expected.
behavior against solar irradiance disturbances, several tests under Despite this, the ME and RMSE errors are low and it is possible to
cloudy conditions were carried out. The one performed in December infer that the simplified model can be used for future investigations of
20th, 2016 is shown in Fig. 15. The characteristics of this day are the solar field controllers and MED operating strategies optimization.
following ones: The simplified model allows estimating the water temperature at
the outlet of each loop while reducing the computational time about
• In the test day, solar irradiance shows cloudy conditions or transi- 16% (computational time is reduced from 38 to 32 s when a simulation
tions. The lowest value is 400 W m−2 at 13:00am. As shown in of 10000 s is performed), because it resolves one differential equation
Fig. 15, the most significant disturbances occurred approximately per loop while the complete model includes two differential equations
between 11:48 am and 13:15 pm. per loop. Nevertheless, the simplified model shows more error than the
• Ambient temperature varies between 12 °C and 16 °C. Inlet flow rate one expected due to a non-desirable behavior that occurs in some ex-
and inlet temperature remain constant. periments at the inlet temperature of the second section when step flow
• The comparison between the simulated and the real solar field outlet rates are performed. This fact is the cause of an error increment of about
temperature can be observed in Fig. 15, where the similarity of the 2 °C in comparison to the complete model. On the order hand, the

253
G. Ampuño et al. Solar Energy 177 (2019) 241–254

results of the simplified model with the delays included are quite sa- Alarcón, D., García, L., Blanco, J., 2010. Design recommendations for a multi-effect
tisfactory to be used for control purposes. distillation plant connected to a double-effect absorption heat pump: A solar desali-
nation case study. Desalination 262 (1–3), 11–14. https://doi.org/10.1016/j.desal.
2010.04.064.
6. Conclusions Aleksiejuk, J., Chochowski, A., Reshetiuk, V., 2018. Analog model of dynamics of flat-
plate solar collector. Sol. Energy 160, 103–116. https://doi.org/10.1016/j.solener.
2017.11.079.
The simplified model proposed in this paper implements a con- Ali, A., Al-karaghouli, Kazmerski, L.L., 2011. Renewable Energy Opportunities in Water
centrated parameter model of flat plate collectors augmented with Desalination. Desalination, trens and technologies. InTech, India.
quadratic functions relating flow rate and delay, providing the evolu- Ampuño, G., Roca, L., Berenguel, M., Gil, J.D., Pérez, M., Normey-Rico, J.E., 2018.
Modeling and simulation of a solar field based on flat-plate collectors. Sol. Energy
tion of outlet temperature. This model shows the following advantages: 170, 369–378. https://doi.org/10.1016/j.solener.2018.05.076.
(i) its implementation is straightforward; (ii) fast computational time Brus, L., Zambrano, D., 2010. Black-box identification of solar collector dynamics with
simulation; (iii) adequate performance under different conditions of variant time delay. Control Eng. Pract. 18 (10), 1133–1146. https://doi.org/10.
1016/j.conengprac.2010.06.006.
solar irradiance (sunny and cloudy days), and (iv) it describes variable
Buzás, J., Kicsiny, R., 2014. Transfer functions of solar collectors for dynamical analysis
delays from flow, input temperature and irradiance changes. Simulation and control design. Renew. Energy 68, 146–155. https://doi.org/10.1016/j.renene.
results compared with real data have shown that modelling errors when 2014.01.037.
changing values of flow rate, inlet loop temperature and solar irra- Camacho, E.F., Berenguel, M., Rubio, F.R., Martínez, D., 2012. Control of Solar Energy
Systems. Springer Science & Business Media, London.
diance are quite low, thus demonstrating that the simplified model is Camacho, E.F., Rubio, F.R., Berenguel, M., 1997. Advanced Control of Solar Plants.
reliable. Real experiments on a solar field have been included to show Springer, London.
the obtained promising results. Deng, J., Yang, X., Wang, P., 2015. Study on the second-order transfer function models for
dynamictests of flat-plate solar collectors Part I: A proposed new model and a fitting
As demonstrated in Ampuño et al. (2018), a good approximation to methodology. Sol. Energy 114, 418–426. https://doi.org/10.1016/j.solener.2015.01.
model the outlet temperature of the solar field is considering input- 046.
output related transport delays experimentally obtained. Nevertheless, Deng, J., Yang, X., Wang, P., 2017. Study on the second-order transfer function models for
dynamic tests of flat-plate solar collectors Part II: Experimental validation. Sol.
that solution requires an experimental campaign in which all the inputs Energy 141, 334–346. https://doi.org/10.1016/j.solener.2015.01.046.
must be manipulated to cover the operating range, which is impossible Duffie, J.A., Beckman, W.A., 2013. Solar Engineering of Thermal Processes, fourth ed.
for the irradiance variable, as it is not manipulable (it constitutes a John Wiley & Sons Inc, New Jersey, Hoboken.
Facão, J., 2016. Optimization of flow distribution in flat plate solar thermal collectors
disturbance from the control point of view). For small solar fields, a with riser and header arrangements. Sol. Energy 120, 104–112. https://doi.org/10.
mean value for the irradiance-output transport delay can be assumed, 1016/j.solener.2015.07.034.
but for large solar fields it is mandatory to define an irradiance-output Fernández, P., García, L., Alarcón, D., Palenzuela, P., Martín, I., 2012. Experimental
analysis of a multi-effect distillation unit operated out of nominal conditions.
transport delay as a function of the flow rate. One of the main objectives
Desalination 284, 233–237. https://doi.org/10.1016/j.desal.2011.09.004.
of the work is to demonstrate the existence of an apparent delay at the Fischer, S., Heidemann, W., Müller-Steinhagen, H., Perers, B., Bergquist, P., Hellström, B.,
output of the flat plate collector solar field due to the parallel connec- 2004. Collector test method under quasi-dynamic conditions according to the
tion of the absorber tubes. The raised hypothesis has been demonstrated European Standard EN 12975–2. Sol. Energy 76 (1–3), 117–123. https://doi.org/10.
1016/j.solener.2003.07.021.
using linearized process models arranged in parallel, following the Gil, J.D., Roca, L., Zaragoza, G., Berenguel, M., 2018. A feedback control system with
layout of the real solar field. In addition, the methodology explained in reference governor for a solar membrane distillation pilot facility. Renew. Energy
this paper permits to obtain the apparent delays from the model de- 120, 536–549. https://doi.org/10.1016/j.renene.2017.12.107.
Goodwin, G.C., Payne, R.L., 1977. Dynamic System Identification. Experiment Design and
scribed in Eq. (1) without additional experimental campaigns. Data Analysis, Mathematics in Science and Engineering. Academic Press, London, pp.
Finally, it is important to mention that obtaining a solar field outlet 136.
temperature model is the first step in the goal of designing control Kalogirou, S.A., 2004. Solar thermal collectors and applications. Prog. Energy Combust.
Sci. 30, 231–295. https://doi.org/10.1016/j.pecs.2004.02.001.
systems that: (i) act against disturbances of solar irradiance, (ii) prevent Kicsiny, R., 2017. Transfer functions of solar heating systems with pipes for dynamic
damages in collectors due to high temperatures, (iii) improve the per- analysis and control design. Sol. Energy 150, 596–607. https://doi.org/10.1016/
formance when working at low temperatures, (iv) minimize energy solener.2017.05.006.
Nayak, J.K., Amer, E.H., 2000. Experimental and theoretical evaluation of dynamic test
consumption of the plant and (v) maximize thermal performance.
procedures for solar flat-plate collectors. Sol. Energy 69 (5), 377–401. https://doi.
org/10.1016/S0038-092X(00)00108-0.
Acknowledgements Normey-Rico, J.E., Bordons, C., Berenguel, M., Camacho, E.F., 1998. A robust adaptive
dead-time compensator with application to a solar collector field. IFAC Proc. Volumes
31 (19), 93–98. https://doi.org/10.1016/S1474-6670(17)41134-7.
This work has been funded by the National Plan Projects DPI2014- Normey-Rico, J.E., Camacho, E.F., 2007. Control of Dead-time Processes. Springer,
56364-C2-1/2-R and DPI2017-85007-R of the Spanish Ministry of London.
Economy, Industry and Competitiveness and ERDF funds. Julio E. Ogata, K., 1998. Ingeniería de Control Moderna. Prentice Hall, México.
Pasamontes, M., Álvarez, J.D., Guzmán, J.L., Berenguel, M., 2009. Hybrid modeling of a
Normey-Rico thanks CNPq under project 305785/2015-0 for financial solar cooling system. IFAC Proc. Volumes 42 (17), 26–31. https://doi.org/10.3182/
support. 20090916-3-ES-3003.00006.
Pasamontes, M., Álvarez, J.D., Guzmán, J.L., Berenguel, M., Camacho, E.F., 2013. Hybrid
modeling of a solar-thermal heating facility. Sol. Energy 97, 577–590. https://doi.
References org/10.1016/j.solener.2013.09.024.
Roca, L., Berenguel, M., Yebra, L., Alarcón-Padilla, D., 2008. Solar field control for de-
Ackermann, T., Carlini, E., Bernhard, E., Groome, F., Orths, A., O’Sullivan, J., de la Torre, salination plants. Sol. Energy 82 (9), 772–786. https://doi.org/10.1016/j.solener.
M., Silva, V., 2015. Integrating variable renewables in Europe. IEEE Power Energ. 2008.03.002.
Mag. 13 (6), 67–77. https://doi.org/10.1109/MPE.2015.2461333. Roca, L., Guzmán, J.L., Normey-Rico, J.E., Berenguel, M., Yebra, L., 2009. Robust con-
Alarcón, D., Blanco, J., Malato, S., Maldonado, M.I., Fernandez, P., 2005. Design and strained predictive feedback linearization controller in a solar desalination plant
setup of a hybrid solar seawater desalination system : The AQUASOL project. In: ISES collector field. Control Eng. Pract. 17 (9), 1076–1088. https://doi.org/10.1016/j.
2005 Solar World Congress. Orlando, FL, USA. conengprac.2009.04.008.
Alarcón, D., García, L., Blanco, J., 2007. Assessment of an absorption heat pump coupled Tagliafico, L.A., Scarpa, F., De Rosa, M., 2014. Dynamic thermal models and CFD analysis
to a multi-effect distillation unit within AQUASOL project. Desalination 212, for flat-plate thermal solar collectors - A review. Renew. Sustain. Energy Rev. 30,
303–310. https://doi.org/10.1016/j.desal.2006.10.015. 526–537. https://doi.org/10.1016/j.rser.2013.10.023.

254