Professional Documents
Culture Documents
Atam 2016
Atam 2016
art ic l e i nf o a b s t r a c t
In this paper, a compact overview of the state-of-the-art in modeling of ground-coupled heat pump (GCHP)
Keywords: systems and an in-depth review of their optimal control along with the associated research challenges are
Geothermal energy given. The main focus is on optimal control but since design of an optimal controller may require a model, a
Ground-coupled heat pumps relatively short literature review of modeling approaches is also discussed. Adopting the adage “a picture is
Hybrid ground-coupled heat pumps worth a thousand words”, we tried to include a minimal number of representative schematics and result
Borefield modeling figures for some of the reviewed studies for clarity and a better understanding of the presented material. In
Optimal control addition to the literature review, we included our comments, points of view, alternative solutions and some
potential future directions. This review paper is useful both for engineers and researchers involved in
modeling and optimal control of GCHP systems. The second part of the paper, “Ground-Coupled Heat
Pumps: Part 2 – Literature Review and Research Challenges in Optimal Design”, focuses on the literature
review on optimal design and the associated design challenges for GCHP systems.
& 2015 Elsevier Ltd. All rights reserved.
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Literature review on modeling of GCHPs and modeling challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1. Thermal response factor-based models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2. Numerical thermal models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3. Artificial neural network models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4. State-space models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.5. Challenges in modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5.1. Dynamic ground surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5.2. Moisture in the soil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5.3. Effect of groundwater . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.6. Comments on modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.7. Comparison of different GCHP modeling methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3. Literature review on control of (Hy)GCHPs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1. Rule-based control methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2. Model-based control methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.3. Other control methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.4. Control challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.4.1. Obtaining a control model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.4.2. State estimation issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.4.3. Robustness and practical implementation issues for controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.4.4. Defining a benchmark case study to compare different control methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.5. Comments on control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.6. Comparison of different GCHP control methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
n
Corresponding author.
E-mail address: Ercan.Atam@kuleuven.be (E. Atam).
http://dx.doi.org/10.1016/j.rser.2015.10.007
1364-0321/& 2015 Elsevier Ltd. All rights reserved.
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
2 E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3
alternative solutions, our points of view on the related subjects The rest of the paper is structured as follows. In Section 2,
and some future directions to follow based on this literature different modeling approaches for GCHPs or HyGCHPs together
review and our experience. with the associated modeling challenges are presented and the
advantages/disadvantages of each modeling approach from a
Table 2
Categories within modeling and optimal control of GCHP/HyGCHP systems based control-oriented point of view are discussed. Section 3 reviews the
on the current literature. relevant work on optimal control of (Hy)GCHPs, discusses advan-
tages/disadvantages of different control approaches, lists main
Modeling
challenges in control of (Hy)GCHPs and suggests some future work
Thermal response factor Numerical thermal Artificial State-space to be done based on our experience. Finally, the summary of this
models models neural net- models
review material and some concluding remarks can be found in the
work models
conclusions, Section 4.
Eskilson [11], Cimmino Rees and He[21], Esen et al. Verhelst and
and Bernier [12], Cui et al. [22], Li [29], Gang Helsen [33],
Yavuzturk and Spitler and Zheng [23], and Wang De Rideder
[13], Zeng et al. [14], Bauer et al. [24], [30], Yating et al. [34],
Lamarche and Beau- Diersch et al. [25], et al. [31], Atam et al. 2. Literature review on modeling of GCHPs and modeling
champ [15], Cui et al. Al-Khoury et al. Gang et al. [35], Atam challenges
[17], Marcotte and [27], Wooszyn and [32] et al. [36]
Pasquier [18], Goa [28] In this section, we present a survey of borehole/borefield
Lamarche [19], Fisher
and Simon [20] thermal modeling approaches to be discussed/analyzed in the
Control context of optimal control of (Hy)GCHPs. The application of an
optimal control method for a (Hy)GCHP system is very dependent
Rule-based control methods Mathematical-model based Other control
control methods methods
on the used borehole/borefield model. Some control approaches
can be designed based on specific borehole/borefield models only.
Cullin and Spitler [47], Hackel Verhelst [55], De Ridder Gang and Since the main focus of the paper is optimal control of (Hy)GCHP
and Pertzborn [48], Yang and et al. [34], Atam and Helsen Wang [30],
Wang [50], Sagia and Rako- [57], Atam et al. [58], Hu Yating et al.
systems, the review of modeling approaches is done from a
poulos [51], Yavuzturk and et al. [56], Antonov et al. [31], Madani control-oriented point of view, i.e, the usefulness, advantages/
Spitler [52], Man et al. [53], [59], Sundbrandt [8] et al. [7] disadvantages of modeling approaches in the context of design of
Arteconi et al. [54]
optimal controllers.
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
4 E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
6 E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎
reduced-order model for the system through application of proper 2.6. Comments on modeling
orthogonal decomposition method which allows to integrate the
effect of multiple time scales by appropriately constructing a The different categories of (Hy)GCHP system modeling approaches
matrix (known as snapshot matrix) used in the reduced-order have pros and cons relative to each other. The model of a (Hy)GCHP
modeling. An approach for obtaining parameterized state-space should have at least one of the following three properties, preferably
models as a function of physical variables such as borehole length, having all of them, (a) to predict the response of the underlying sys-
radius, and mass flow rate was derived by Atam et al. [35]. The tem with an acceptable accuracy, (b) to be used in optimal control,
main advantage of such a parameterized borehole model is the (c) to be used in optimal design. The first property, in fact, is a must.
fact that it can be used in a co-design (simultaneous optimal Now let us analyze the presented different modeling approaches in
design and optimal control, which will be discussed in the second the context of the first two properties (the discussion in the context of
part of the paper) framework. All of these state-space based third property will be done in the second part of the paper).
modeling approaches are either valid for a single borehole only or
in case of borefields their derivation relies on some idealized ANN models can accurately model an existing complex (Hy)
assumptions which are hard to meet in a real-time applications. GCHP system but, as mentioned in Section 2.3, they are not
useful for optimal control in long-term application practice
since their training requires long-term data. ANN is not the
2.5. Challenges in modeling
preferred approach to model BHEs, but may be a good approach
for modeling other components within the global system
2.5.1. Dynamic ground surface
having smaller time constants.
The ground surface involves many complex processes and inter-
The thermal response factor models are not so useful for con-
actions happening and an accurate modeling should take into account
trol. It is extremely difficult to use such complex models in a
solar radiation, cloud cover, surface albedo, ambient temperature,
model based control design framework. However, these models
relative humidity, rainfall, snow cover, wind speed, etc [38]. It is
can be used indirectly for control purpose. For example, system
almost impossible to model analytically the above complex processes
identification can be applied to obtain reduced-order models by
and it is very hard also to consider their effects in numerical models.
using the input–output data from the simulation of thermal
As a result, many researchers simplified the dynamic ground surface
response factor models and a controller can be designed using a
by assuming the ground surface temperature variation in the form of a
reduced-order model.
sine wave or Fourier series [39,40]. Some other researchers even made
Numerical thermal models are, in general, very accurate, and
a more simplified assumption: they assumed the ground surface to
hence may be considered ideal for control purposes. However,
have a constant temperature value or they assumed an adiabatic
they include too many states (in the order of millions) and
boundary condition on the ground surface.
hence they cannot be used directly for control purposes. To that
end, first a model-order reduction is necessary to obtain a
2.5.2. Moisture in the soil
reduced-order, controllable/observable model, based on which a
Heat transfer in the ground with the presence of soil moisture
controller can be designed. Even in this case, the controlled
requires system modeling with coupled heat and mass transfer
system will be of limited capacity since thermal numerical
balances which make the modeling problem more difficult and
approaches can only model a borefield system with a few BHEs.
complex. However, taking into account the moisture effect is very State-space models (especially, linear, time-invariant state-
crucial for the optimal design and performance prediction of a space models (LTI-SS)) are desired forms for optimal control.
GCHP system since moisture content directly affects the soil The main reason for this is the fact that the related theory for
thermal conductivity. In Fayegh and Rosen [38], it was specified optimal control of LTI-SS systems is very well-developed.
that the effect of moisture is more pronounced in horizontal BHEs However, as we observed from the literature review on model-
than vertical BHEs because of the effect of temperature and ing, the most general models are thermal response factor
moisture variation near the ground surface. Some literature stu- models and these models are not in state-space form. However
dies [41–43] reported that the moisture content of soil affects the for optimal control purposes, as we mentioned before, it is
BHE design length and GCHP performance in the long term. worth trying the following: once a borefield is modeled using
thermal response factor approaches, system identification can
2.5.3. Effect of groundwater be used to obtain a state-space model approximating it. As
Another factor affecting the thermal response of a borefield input, for example, we may consider the inlet temperature of
system is the presence of groundwater flow. Groundwater flow the circulating fluid exiting the heat pump and as output we can
requires the heat conduction equation to be coupled to the heat take the temperature of the fluid entering the heat pump.
advection equation. The direction of groundwater flow is a key
parameter and in many methods dealing with the effect of 2.7. Comparison of different GCHP modeling methods
groundwater flow on the heat transfer response of a borefield
system, it was assumed that the direction of flow was parallel to a A qualitative comparison of presented different (Hy)GCHP mod-
coordinate axis. This assumption may be violated in complex eling methods with respect to their capabilities of handling the
borefield configurations and, in fact, based on flow direction the mentioned research challenges and with respect to model character-
borefield orientation may be decided for a better performance. In istics (accuracy, complexity and speed) is given in Table 3.
Lee and Lam [44], the effect of groundwater flow direction was
scrutinized in detail through a 3D numerical modeling study.
Some other groundwater effect studies are presented in [45,46]. 3. Literature review on control of (Hy)GCHPs
The groundwater level has also a direct effect on the performance
of a GCHP system and hence on the model to predict its COP. As the 3.1. Rule-based control methods
water level increases, the thermal interaction of groundwater with
borehole through advection and natural convection increases. This A rule-based controller that uses the most efficient machine at
increased interaction decreases the borehole thermal resistance, each time step might be the simplest algorithm to use. This kind of
which in turn increases COP of the GCHP system. controller is, however, not able to make full use of the temperature
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
8 E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎
are model-based control approaches, they are looking at different actions for buildings may require sampling times in order of
problems using different approaches. As a result, each one is minutes or hours. Moreover, the realization of the designed con-
summarized separately. troller requires the measurement of the underground borefield
De Ridder et al. [34] used dynamic programming. Dynamic temperature, for which measurement may be either difficult or
programming is a powerful method since it is a closed-loop, global non-accurate. As a result, the approach of [34] involves both some
optimal control algorithm. However, the model used in [34] for modeling simplifications and a hard-to-realize implementation.
dynamic programming is a very simple first-order model for the Verhelst [55] applied a linear optimal control method for a
ground mean temperature. The chosen sampling period for the HyGCHP system to satisfy heating/cooling demands of a typical
system is one week, which is very long since typical control office zone building. The COP for the heat pump and chiller were
Fig. 3. (a) Schematic diagram of the solar-assisted GCHP system. (b) Rule-based control flow chart [50].
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 9
taken to be constant, in contrast to being functions of source and Atam and Helsen [57] presented a convex approach through
sink temperatures. COP values were taken to be constant because convexification for non-convex optimal control-based and non-
otherwise the optimization problem is a non-convex problem, convex model predictive control-based energy-use minimization
which cannot be solved over a couple of years especially when of buildings equipped with hybrid ground-coupled heat pump
short control time steps are considered. Although a mathematical systems. The original non-convex problems were convexified
model-based optimal control was considered, the simplifications using a convex envelope approach. The results of the convexified
of taking the mentioned COPs as constant values without a formal optimal control problem were compared with dynamic
justification restricts the work of [55]. Moreover, the model used programming-based optimal control, which is a global and closed-
for control and emulator are the same, which eliminates the loop control approach. The comparison results showed that the
impact of model mismatch and therefore limits the generality of convex approximation of the optimal control problem gave almost
the approach followed in [55]. global optimal results in terms of responses and cost criteria. The
Hu et al. [56] presented an extremum seeking control (ESC) suggested method is especially useful for optimal/model pre-
strategy for a HyGCHP system where a cooling tower is included as dictive control of HVAC systems integrated in buildings, where the
the supplementary device. The system and control configurations source and sink temperature-dependent coefficients of perfor-
are shown in Fig. 4. The ESC method is seen, most of the time, as a mance of some components introduce non-convexity into the
“model-free” control method [60], which means that it does not system.
require a state-space model of the system to be controlled but Atam et al. [58] presented a prediction-based dynamic pro-
requires a relation between control inputs and controlled outputs. gramming control approach, a non-linear model predictive control
approach and a linear optimal control approach to analyze the
Hence, it is still convenient to consider such an approach as a
minimization of the total energy-use of a HyGCHP system (incor-
model-based control method based on the mathematical relation
porating a ground-coupled heat pump, a gas boiler, a passive
between control inputs and controlled outputs. The optimal con-
cooler and an active chiller) under operational constraints. A large-
trol is determined online from minimization of an objective
scale emulator model (based on finite-volume method and the
function f ðt; uÞ:
equivalent diameter approach) was used for the borehole system
uopt ¼ arg min f ðt; uÞ: ð5Þ and for assessment of different control algorithms. A non-linear
autoregressive exogenous model (NARX) was identified from the
The ESC scheme used takes as feedback the total power for the input–output data generated by the emulator model to be used in
heat pump compressor, the tower fan and the water pump. The the dynamic programming-based controller. Since dynamic pro-
control inputs are the tower fan speed and the condensing water gramming is a global optimal control method, it was used as a
flow rate. The performance of the ESC controller was tested for reference for the performance assessment. Next, a state-space,
different constant part-load cooling scenarios (part-loads are reduced-order, control-oriented model with a larger sampling
defined based on the evaporator entering water temperature) and time was obtained from the emulator model using the so-called
dynamic cooling load scenarios. The control results for partial-load proper orthogonal decomposition (POD) model reduction techni-
cases are shown in Fig. 5. In Fig. 5, the ESC controller is tuned at que. This model was used in a non-linear model predictive control
t ¼ 50; 000 s and the ESC controller achieves the estimated optima algorithm with the purpose of energy-use minimization over
in the static maps, with steady-state errors less than 3% in both sequences of shorter time intervals (prediction interval) to see
cases. Although the results are not shown here, ESC control for how far it is suboptimal with respect to dynamic programming.
dynamic loading was also successful, where they incorporated an Finally, a series of linear optimal controllers based on constant
anti-windup scheme to alleviate the actuator saturation effects on heat pump coefficients of performance were tested to see how
control results. The authors concluded that ESC-based control much the system performance deteriorates. The control algo-
resulted in power savings in the range of 24–30% relative to rithms were used for satisfaction of heating–cooling demands of
operating the system with a relative flow rate of 0.9. three types of buildings: heating-dominated, cooling-dominated
Fig. 4. (a) Schematic diagram of the hybrid GCHP system. (b) Control system diagram [56].
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
10 E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Fig. 5. ESC control for partial cooling load cases. (a) Static map under 80% cooling load. (b) ESC control results for 80% cooling load. (c) Static map under 50% cooling load.
(d) ESC control results for 50% cooling load [56].
and thermally-balanced. The effects of constraining the thermal which MPC was successful. In this study, although a more realistic
build-up/depletion of the ground and variable electricity profiles and detailed modeling approach was followed, the ground
on the performance of the applied different controllers were also dynamics was not taken into account and it was simply assumed
separately analyzed. The results showed that the performance of that the GCHP system was able to provide the desirable heating/
MPC and linear optimal controllers were close to dynamic cooling.
programming-based control results: less than 10% performance Antonov et al. [59] analyzed the annually optimal operation of a
deterioration. HyGCHP system. The ground heat exchanger model was derived
Sundbrandt [8] controlled a GCHP system through model pre- by Verhelst and Helsen [33] from a resistor-capacitor (RC) net-
dictive control. The control problem was defined as to keep the work, which was composed following the guidelines of Eskilson
inside temperature of an office building and the domestic hot [11]. The model was validated by simulation data from a detailed
water temperature within certain bounds compatible with ther- emulator model of a duct storage system. A linear state-space
mal comfort while minimizing the energy used. The GCHP system model was used, which integrated the models of an office building
was modeled as close as possible to reality: the pulse width (4-th order), a ground heat exchanger (11-th order) and algebraic
modulation and on/off nature of the heat pump, and minimization models of primary and backup heating/cooling devices. It was
of maximum number of heat pump start-ups and shut-downs per shown that the best performance of the system investigated was
hour to get a good life time expectancy were taken into account. achieved by MPC with a time horizon of longest one week.
This resulted in a hybrid model for the overall system which Moreover, it was noted that there was no potential for further
includes logic, discrete and continuous dynamics, and the mixed decreasing the system operation cost by forcing seasonal under-
logical dynamical (MLD) system modeling approach [61] was used ground thermal energy storage. This was concluded by solving an
to mathematically model the system in a suitable form for the off-line optimal control problem over a time horizon of one year
MPC framework. This hybrid model required the mixed integer with a control time step of one hour. For analysis of the seasonal
quadratic programming (MIQP) for MPC control, as opposed to the underground thermal energy storage, the solution of this optimal
use of QP in classical MPC. The heat pump was modeled by a state- control problem was compared to the MPC controlled system with
space model. Different scenarios were considered depending on a prediction horizon of one week (control time step also one hour),
different outside conditions in different times of the year and the which recedes with one day (24 time steps) on each iteration.
control results were compared to the results of a conventional
controller. Without night setback, the energy saving performance 3.3. Other control methods
of MPC was 1–3% better compared to the conventional controller,
which increased to 7.5% when night setback was considered. From The control approaches used in this category are approaches
the extensive simulations it was observed that a conventional which cannot be categorized as solely rule-based or model-based.
controller cannot achieve an acceptable reference tracking for However, some of them can be seen as a hybrid combination of
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 11
them. The first example of such a study is by Gang and Wang [30] this case. After detailed simulations with the two different control
which considered control of a HyGCHP system for a cooling- strategies (fixed capacity, on–off controlled GCHPs versus the
dominated office building where a supplementary cooling tower variable capacity GCHP) under different operating and boundary
was used in parallel to a GHE as shown in Fig. 6. They used a model conditions (like ambient temperature, solar radiation and internal
with a rule based controller. They used ANNs for modeling both gains), the authors observed the following:
the CT and BHE. They indicated that during the operation of such a
hybrid system only one of the exit temperatures of GHE or CT was Comparing the performances of fixed capacity and variable
measurable since only one of them was working at a given time. As capacity GCHP systems, the capacity of the fixed GCHP system is
a result, they claimed that there was a need to predict the exit very crucial: if the fixed capacity GCHP is designed to cover only
temperature of GHE or CT and for that purpose they used ANN- 55% of peak building load, then the SPF of the fixed capacity
based prediction models. For ANN-based GHE modeling they GCHP system is lower than that of the variable capacity GCHP
concluded that a dynamic ANN model with the Levenberg–Mar- system. On the other hand, if the fixed capacity GCHP is
guardt (LM) learning algorithm was the best option and such a designed to cover more than 65% of peak building load, then
model had a prediction error performance of less than 0.2 °C for SPFs are close to each other.
the GHE exit fluid temperature. Their control strategy was as fol- The fixed capacity GCHP systems have a higher SPF when the
lows. They compared the exit fluid temperatures of GHE and CT auxiliary heater is designed to cover less than 5% of the annual
and chose the lower one. They compared the performance of their heat demand.
control strategy using the developed models with the performance
of a time-based scheduled control over two days. Their control 3.4. Control challenges
strategy proved minor advantages over the time-based scheduled
control. Authors concluded that though ANN modeling provided 3.4.1. Obtaining a control model
good prediction accuracy, still there was need for improvement of By control model of a (Hy)GCHP system we mean a model to be
the proposed control strategy and a comparison over the life cycle used in a model-based control method. This control model is, in
of the system. general, different from the simulation model (or the real system),
A similar work was done by Yating et al. [31] where they used which is sometimes called the emulator model. Emulator models
radial basis function neural networks (RBFNNs) to predict supply are complex compared to control models for complex systems and
water temperature of the building and COP of the GCHP system obtaining a control model from an emulator model is a very hard
unit. They used adaptive particle swarm optimization (APSO) for task. For example, the control model used by Atam et al. [58] is a
determination of control inputs of the predictive control algorithm state-space model obtained from a large-scale, finite volume
which minimizes the total consumed energy by the system under numerical model of the borehole heat exchanger by applying a
operational constraints. The authors indicated a 5% energy saving model order reduction scheme. Similarly, the control model used
potential of the RBFNN-APSO method. by Ridder et al. [34] is a simple state-space model but the emulator
Next, we consider an interesting and a different control model is a complex model in TRNSYS.
approach called “capacity-based control” (meaning the use of Emulator models are needed for testing the control algorithm
different components with different capacities) by Madani et al. before application to the real system to test a priori the expected
[7]. They developed a simulation model for a building equipped performance of the controlled system and redesign the controller
with a GCHP system by developing submodels for the heat pump in case of poor performance. System identification and model
unit, the building, the heat source, the thermal storage tank, and order reduction techniques are among the methods applied to
the climate, in order to compare some of the strategies used for obtain control models. In system identification, measured input–
capacity control of a heat pump system, such as using a variable output data are used to obtain black-box or grey-box models of the
speed compressor and/or variable speed pumps in the HP system. system and hence no or little physical knowledge about the system
TRNSYS and EES [62] were used as the simulation tools in so-called is required. System identification may fail if the underlying system
co-solving manner. The seasonal performance factors (SPFs) of is highly non-linear or if it has many input and outputs. Another
four constant speed, on–off controlled GCHPs were compared with disadvantage of system identification for (Hy)GCHP systems is that
SPF of an inverter-driven, variable speed ( ¼variable capacity) it requires long-term data (over 15–20 years), which is not feasible
GCHP system. The borehole used for ground energy source was to collect from field measurements over such a long period.
taken to be 200 m deep. They used an electrical auxiliary heater Moreover, the data to be used should have a rich-frequency con-
for on–off controlled systems to cover the building load demand tent to be able to identify the relevant modes in the system
not covered by the GCHP. In the variable speed GCHP system, it dynamics [63]. An example of obtaining a control-oriented, grey-
was expected that this system alone would cover the whole box model of a borehole using TRNSYS to represent the real system
building load and hence no electrical auxiliary heater was used in to avoid real data is given by Clara and Verhelst [64].
In model order reduction techniques, a simplified model of the
system is obtained from the emulator model to be used as a
control model by projection of the large-scale model dynamics
onto a lower subspace [65]. Model reduction techniques require
mathematical equations of the emulator model, which may not be
available or may be very complex. Obtaining a control model is the
core and the most difficult stage of designing a model-based
controller for GCHP or HyGCHP systems.
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
12 E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎
observable, one may try the following: (a) apply a model reduction As a result, to asses the nominal or robust performance of these
technique to the control model to eliminate non-observable different approaches on the same system, a benchmark case study
modes, (b) try to measure more outputs. It is very crucial that is required.
the outputs of interest are accurately measurable in practice. In
this sense, for example, for a GHE in a GCHP system, one may try 3.5. Comments on control
to obtain a control model where the temperature of the exit fluid
from the heat pump is the input to the model and the temperature From the reviewed material on (Hy)GCHP control, we have the
of the inlet fluid to the heat pump from ground side is the output, following observations:
both of which can be measured. If the output of interest is ground
mean temperature as in Ridder et al. [34], then the estimator may Rule-based control methods are suboptimal compared to
not be realized or it may be non-accurate in a practical application, model-based control methods or model-free control methods
since the measurement of the mean ground temperature at some like ESC. However, they have the advantage that a control model
depth may not be easy, accurate or practical. The determination of of the underlying system is not needed, which is a big plus
outputs to measure and the sensor location are key factors for the especially for the hard-to-model borefield dynamics.
success of an estimator. Model-free approaches like ESC have two advantages (a) in
general, they give results which are, in terms of performance,
3.4.3. Robustness and practical implementation issues for controllers between rule-based and model-based control results, (b) a
Robustness of the control algorithm is another important control model is not needed.
aspect. If in the control design framework used for the (Hy)GCHP Model-based control methods are, by far, better than all other
systems it is also possible to design a robust controller, then a approaches and they can provide significant cost savings.
robust controller should be designed with a certain robustness However, they require an accurate mathematical control model,
degree against parametric or dynamic uncertainties of the con- which may not be easy for complex borefields. Among the
trol model. If it is difficult to design a robust controller or if there model-based control approaches, MPC seems the best one
does not exist a robust control design framework for the used thanks to its ability of handling constraints and working for
control approach, then the performance of the controller should MIMO systems.
be tested on an emulator model through extensive worst-case Control strategies based on ANNs did not provide satisfactory
simulation scenarios that may be encountered in a real imple- performance, despite their detailed model structure. The main
mentation. Nominal controller design (controller design without cause for this is the lack of a mathematical control theory
robustness considerations) for a large-scale system in which working optimally for ANN models.
renewable energy is provided by a GCHP system is itself already The successful control of an integrated (Hy)GCHP system is also
difficult and an additional request for robustness may make the dependent on the building or structure it is integrated into. As a
problem extremely complex, if not impossible. Even if it is not result, the overall modeling and control of an energy-efficient
possible to design a robust controller, it is very important to building with (Hy)GCHP systems also involve modeling and
check performance degradation or constraint violations of the control of the building part separately, if a decentralized control
nominal controller under worst-case scenarios. Hence, in any approach is used. In case of a centralized control approach,
case, control performance test on an emulator model against building and (Hy)GCHP models should be coupled to design a
robustness is strongly recommended. If it is possible to use an single centralized controller for the overall system. It should be
adaptive control framework, then that may be optimal in the noted that modeling and control of multi-zone buildings is a
sense of robustness because by their design paradigm adaptive challenge, not less than modeling and controlling (Hy)GCHP
control algorithms adapt to changing environments and hence systems. So far, in the literature successful modeling and real-
may handle uncertainty issues better compared to other robust time control of maximum 3–4 zones have been achieved [66].
control methods. As a result, the combined system of building and (Hy)GCHP
system is very challenging to control.
3.4.4. Defining a benchmark case study to compare different control
methods 3.6. Comparison of different GCHP control methods
From the reviewed literature on control of (Hy)GCHP systems,
it was observed that different control approaches had different A qualitative comparison of presented different (Hy)GCHP
energy saving performances and the results were case dependent. control methods with respect to their capabilities of handling the
Table 4
A qualitative comparison of different (Hy)GCHP control methods in terms of research challenges and controller characteristics using a scale of 0–5 (0: lowest, 5: highest).
Optimality 1 5 3
Complexity 1 3 3
Speed 5 3 2
State estimation difficulty 0 3 3
Easiness to obtain a model 5 2 3
for the associated
control approach
Robustness level 1 4 2
Implementation easiness 5 3 3
Easiness to design 5 3 3
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 13
mentioned research challenges and with respect to controller potential time-varying parameters and which probably excludes
characteristics (optimality, complexity and speed) is given in some unmodeled dynamics. Therefore, an adaptive version of
Table 4. MPC which updates online the control model and controller
parameters may be very useful to consider.
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
14 E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Acknowledgments [31] Yating Z, Guiyang W, Guang H. Gchp system optimal predictive control based
on rbfnn and apso algorithm. In: Proceedings of the 32nd Chinese control
conference, July 25–28, Xi'an China; 2013.
The authors want to acknowledge the Agency for Innovation by [32] Gang W, Wang J, Wang S. Performance analysis of hybrid ground source heat
Science and Technology in Flanders (IWT) and the Scientific and pump systems based on ann predictive control. Appl Energy 2014;136:1138–44.
Technical Center for the Building Industry (WTCB) for supporting [33] Verhelst C, Helsen L. Low-order state-space models for borehole heat
exchangers. HVAC & R Res 2011;17(6):928–47.
the post-doctoral research work of Ercan Atam within the IWT [34] De Ridder F, Diehl M, Mulder G, Desmedt J, Van Bael J. An optimal control
project VIS SMART GEOTHERM. algorithm for borehole thermal energy storage systems. Energy Build 2011;
43(10):2918–25.
[35] Atam E, Verhelst C, Helsen L. Borehole dynamics parameterization for thermal
performance optimization under operational constraints. In: European geo-
References thermal congress, Pisa-Italy, June 3–7; 2013(a).
[36] Atam E, Verhelst C, Helsen L. Development of a control-oriented model for
[1] EPBD energy performance of buildings directive, directive 2010/31/ue; 2010. borehole dynamics for buildings equipped with ground-coupled heat pumps,
[2] Stuebi RT. Geothermal heat pumps: green for your wallet, green for our planet. In: BS 2013 Conference, Chambery, France, 2013(b).
In: Renewable energies conference, November 29, Denver, USA; 2000. [37] A transient systems simulation program, Available from: 〈http://sel.me.wisc.
[3] Geothermal heat pumps – DOE (USA), Available from: 〈http://www.eren.doe. edu/trnsys/〉; 2015.
gov〉; 2014. [38] Fayegh SK, Rosen MA. A review of the modeling of thermally interacting
[4] Wu R. Energy efficiency technologies – air source heat pump vs. ground source multiple boreholes. Sustainability 2013;5(6):2519–36.
heat pump. J Sustain Dev 2009;2(2):14–23. [39] Jacovides CP, Mihalakakou G, Santamouris M, Lewis JO. On the ground tem-
[5] Corberan JM, Fin DP, Montagud CM, Murphy FT, Edwards KC. A quasi steady- perature profile for passive cooling applications in buildings. Sol Energy
state mathematical model of an integrated ground source heat pump for 2002;57(3):167–75.
building space control. Energy Build 2011;43:82–92. [40] Mihalakakou G. On estimating ground surface temperature profiles. Energy
[6] Paris B, Eynard J, Grieu S, Talbert T, Polit M. Heating control schemes for Build 2002;34(3):251–9.
energy management in buildings. Energy Build 2010;42:1908–17. [41] Piechowski M. Heat and mass transfer model of a ground heat exchanger:
[7] Madani H, Claesson J, Lundqvist P. Capacity control in ground source heat theoretical development. Int J Energy Res 1999;23(7):571–88.
pump systems part ii: comparative analysis between on/off controlled and [42] Leong WH, Tarnawski VR. Effects of simultaneous heat and moisture transfer
variable capacity systems. Int J Refrig 2011;34(8):1934–42. in soils on the perforamnce of a ground source heat pump system. In: ASME-
[8] Sundbrandt M. Control of a ground source heat pump using hybrid model ATI-UIT conference on thermal and environmental issues in energy systems,
predictive control [M.Sc. thesis]. Linkopings University, Sweden; 2011. Sorrento, Italy; 2012.
[9] Ingersoll L, Adler F, Plass H, Ingersoll A. Theory of earth heat exchangers for [43] Erdogan B. The effect of mositure content in the soil on sizng of vertical single
the heat pump. Heat Pip Air Cond 1950:113–22. u-tube ground heat exchanger for ground source heat pump. In: Unitech10:
[10] Carslaw HS, Jaeger JC. Conduction of heat in solids. UK: University Press; 1959. international scientific conference, 19–20 November Gabrovo, Bulgaria; 2010.
[11] Eskilson P. Thermal analysis of heat extraction boreholes [Ph.D. thesis]. Uni- [44] Lee CK, Lam HN. Effects of groundwater flow direction on performance of ground
versity of Lund, Sweden; 1987. heat exchanger borefield in geothermal heat pump systems using a 3-d finite
[12] Cimmino M, Bernier M. A semi-analytical method to generate g-functions for difference method. In: Building simulation conference, Beijing, China; 2007.
thermal borefields. Int J Heat Mass Transf 2014;70:641–50. [45] Dehkordi SE, Schincariol RA. Impact of groundwater flow on thermal energy
[13] Yavuzturk C, Spitler J. A short-time step response factor model for vertical storage and borehole thermal interference, In: European Geosciences Union
ground loop heat exchangers. ASHRAE Trans 1999;105(2):475–85. General Assembly, Vienna, Austria, 2013.
[14] Zeng HY, Diao NR, Fang ZH. A finite line-source model for boreholes in geo- [46] Diao NR, Li Q. Heat transfer in ground heat exchangers with groundwater
thermal heat exchangers. Heat Transf – Asian Res 2002;31(7):558–67. advection. Int J Therm Sci 2004;43(12):1203–11.
[15] Lamarche L, Beauchamp B. A new contribution to the finite line-source model [47] Cullin J, Spitler J. Comparison of simulation-based design procedures for
for geothermal boreholes. Energy Build 2007;39(2):188–98. hybrid ground source heat pump systems. In: Proceedings of the 8th inter-
[16] Claesson J, Javed S. An analytical method to calculate borehole fluid tem- national conference on system simulation in buildings, Liege, Belgium; 2010.
peratures for time-scales from minutes to decades. ASHRAE Trans 2011;117 [48] Hackel S, Pertzborn A. Effective design and operation of hybrid ground source
(Part 2):279–88. heat pumps: three case studies. Energy Build 2011;43(12):3497–504.
[17] Cui P, Yang H, Fang Z. Heat transfer analysis of ground heat exchangers with [49] Wetter M. Genopt, generic optimization program. In: Seventh international
inclined boreholes. Appl Therm Eng 2006;26(11-12):1169–75. IBPSA conference, Rio de Janeiro, Brazil, August 13–15; 2001.
[18] Marcotte D, Pasquier P. The effect of borehole inclination on fluid and ground [50] Yang R, Wang L. Efficient control of a solar assisted ground-source heat pump
temperature for glhe systems. Geothermics 2009;38(4):392–8. system based on evaluation of building thermal load demand. In: North
[19] Lamarche L. Analytical g-function for inclined boreholes in ground-source heat American power symposium (NAPS), 2012, September 9–11; 2012.
pumps. Geothermics 2011;40(4):241–9. [51] Sagia Z, Rakopoulos C. New control strategy for a hybrid ground source heat
[20] Fisher DE, Simon JR. Implementation and validation of ground-source heat pump system coupled to a closed circuit cooling tower. J Appl Mech Eng
pump system models in an integrated building and system simulation
2012;1(2):1–8.
environment. HVAC & R Res Spec Issue 2006;12(3a) 693–619.
[52] Yavuzturk C, Spitler JD. Comparative study to investigate operating and con-
[21] Rees SJ, He M. A three-dimensional numerical model of borehole heat
trol strategies for hybrid ground source heat pump systems using a short
exchanger heat transfer and fluid flow. Geothermics 2013;46:1–13.
time-step simulation model. ASHRAE Trans 2000;106(2):192–209.
[22] Cui P, Yang H, Fang Z. Numerical analysis and experimental validation of heat
[53] Man Y, Yang H, Wang J. Study on hybrid ground-coupled heat pump system
transfer in ground heat exchangers in alternative operation modes. Energy
for air-conditioning in hot-weather areas like Hong Kong. Appl Energy
Build 2008;40(6):1060–6.
2010;87:2826–33.
[23] Li Z, Zheng M. Development of a numerical model for the simulation of ver-
[54] Arteconi A, Brandoni C, Rossi G, Polonara F. Experimental evaluation and
tical u-tube ground heat exchangers. Appl Therm Eng 2009;29(56):920–4.
dynamic simulation of a ground-coupled heat pump for a commercial build-
[24] Bauer D, Heidemann W, Müller-Steinhagen H, Diersch HJ. Modeling and
simulation of groundwater influence on borehole thermal energy stores. In: ing. Int J Energy Res 2013;37(15):1970–80.
Proceedings effstock-11th international conference on energy storage, Stock- [55] Verhelst C, Model predictive control of hybrid ground-coupled heat pump
holm; 2009. systems in office building [Ph.D. thesis]. Katholieke Universiteit Leuven, Bel-
[25] Diersch HJG, Bauer D, Heidemann W, Ruhaak W, Schatzl P. Finite element gium; 2012.
modeling of borehole heat exchanger systems part 1. fundamentals. Comput [56] Hu B, Li Y, Mu B, Wang S, Seem JE, Cao F. Extremum seeking control of hybrid
Geosci 2011;37(8):1122–35. ground source heat pump system. In: 15th international refrigeration and air
[26] Diersch HJG, Bauer D, Heidemann W, Ruhaak W, Schatzl P. Finite element conditioning conference at Prdue, July 14–17; 2014.
modeling of borehole heat exchanger systems part 2. Numerical simulation. [57] Atam E, Helsen L. A convex approach to energy use minimization of buildings
Comput Geosci 2011;37(8):1136–47. equipped with hybrid ground-coupled heat pump systems. Energy Build
[27] Al-Khoury R, Kolbel T, Schramedei R. Efficient numerical modeling of borehole 2015;93:269–81.
heat exchangers. Comput Geosci 2010;36(10):1301–15. [58] Atam E, Patteeuw D, Antonov S, Helsen L. Optimal control approaches for
[28] Wooszyn J, Goa A. Modeling of a borehole heat exchanger using a finite ele- analysis of energy use minimization of hybrid ground-coupled heat pump
ment with multiple degrees of freedom. Geothermics 2013;47:13–26. systems. IEEE Trans Control Syst Technol; 2015. http://dx.doi.org/10.1109/
[29] Esen H, Inalli M, Sengur A, Esen M. Modeling a ground-coupled heat pump TCST.2015.2445851, in press.
system using adaptive neuro-fuzzy inference systems. Int J Refrig 2008; [59] Antonov S, Verhelst C, Helsen L. Should the optimization horizon in optimal
31(1):65–74. control of ground-coupled heat pump systems cover the inter-seasonal time
[30] Gang W, Wang J. Predictive ann models of ground heat exchanger for the scale? ASHRAE Trans 2014;120(2):346–56.
control of hybrid ground source heat pump systems. Appl Energy [60] Ariyur KB, Kristic M. Real-time optimization by extremum-seeking control.
2013;112:1146–53. New York: John Wiley & Sons, Inc; 2003.
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i
E. Atam, L. Helsen / Renewable and Sustainable Energy Reviews ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 15
[61] Bemproad B, Morari M. Control of systems integrating logic, dynamics, and [66] Zacekova E, Vana Z, Cigler J. Towards the real-life implementation of mpc for
constraints. Automatica 1999;35(3):407–27. an office building: identification issues. Appl Energy 2014;135(15):251–9.
[62] Klein SA. EES: Engineering equation solver, professional version. F-chart [67] Hou Z, Jin S. Model free adaptive control: theory and applications. US: CRC
software, 〈http://www.fchart.com〉; 2003. Press, Taylor and Francis Group; 2013.
[63] Ljung L. System identification: theory for the user. PTR Prentice Hall Infor- [68] Domahidi A, Ullmann F, Morari M, Jones JN. Learning near-optimal decision
mation and System Science Series 1999. rules for energy-efficient building control. In: IEEE Conf Decis Control, Maui,
[64] Verhelst C, Helsen L. A linear dynamic borehole model for use in model based HI, USA; 2012.
predictive control. In: Effstock 2009, Stockholm, Sweden, June 14–17; 2009. [69] Domahidi A, Ullmann F, Morari M, Jones JN. Learning decision rules for energy
[65] Alfio Q, Gianluigi R. Reduced order methods for modeling and computational efficient building control. J Process Control 2014;24(6):763–72.
reduction. Switzerland: Springer International Publishing; 2014.
Please cite this article as: Atam E, Helsen L. Ground-coupled heat pumps: Part 1 – Literature review and research challenges in
modeling and optimal control. Renewable and Sustainable Energy Reviews (2015), http://dx.doi.org/10.1016/j.rser.2015.10.007i