You are on page 1of 9

G Model

JJPC-1712; No. of Pages 9 ARTICLE IN PRESS


Journal of Process Control xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Journal of Process Control


journal homepage: www.elsevier.com/locate/jprocont

Energy efficient control of HVAC systems with ice cold


thermal energy storage
Alessandro Beghi a , Luca Cecchinato b , Mirco Rampazzo a,∗ , Francesco Simmini a
a
Dipartimento di Ingegneria dell’Informazione, Università di Padova, via Gradenigo 6/B, I-35131 Padova, Italy
b
Everest s.r.l., via Roveredo 20, I-33170 Pordenone, Italy

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

Article history: In heating, ventilation and air conditioning (HVAC) systems of medium/high cooling capacity, energy
Received 1 April 2013 demands can be matched with the help of thermal energy storage (TES) systems. If properly designed,
Received in revised form 14 January 2014 TES systems can reduce energy costs and consumption, equipment size and pollutant emissions. In order
Accepted 14 January 2014
to design efficient control strategies for TES systems, we present a model-based approach with the aim of
Available online xxx
increasing the performance of HVAC systems with ice cold thermal energy storage (CTES). A simulation
environment based on Matlab/Simulink® is developed, where thermal behaviour of the plant is analysed
Keywords:
by a lumped formulation of the conservation equations. In particular, the ice CTES is modelled as a
HVAC
Thermal energy storage hybrid system, where the water phase transitions (solid–melting–liquid and liquid–freezing–solid) are
Modeling described by combining continuous and discrete dynamics, thus considering both latent and sensible
Control heat. Standard control strategies are compared with a non-linear model predictive control (NLMPC)
approach. In the simulation examples model predictive control proves to be the best control solution for
the efficient management of ice CTES systems.
© 2014 Elsevier Ltd. All rights reserved.

1. Introduction building requires cooling, heat is removed from the building and
added to the ice. The melted ice is reused during the next charg-
Heating, ventilation, and air-conditioning (HVAC) systems in ing period. The advantage of this cooling scheme is that the main
residential, commercial, and industrial sectors are responsible for electrically driven device in cooling systems, namely, the compres-
a major portion of energy utilisation. Energy demands vary with sor, is operated during low-electrical cost periods, i.e. at night [3].
daily, weekly, and seasonal trends. These demands can be matched However, experience with operating TES systems demonstrates
with the help of thermal energy storage (TES) systems that, if that poor design and operation of the control systems can lead to
properly designed, installed, operated, and maintained, can be bad energy efficiency [4]. It is worthwhile noting that a cooling
effectively used to shift peak loads to off-peak periods. The use of plant with TES is a complex system. Highly non-linear behaviour
TES systems often results in significant benefits, such as: reduced and strong cross coupling of inputs and outputs make its model-
energy costs and energy consumption, increased flexibility of oper- ing and control a non-trivial task. Classical control design methods
ation, decreased initial and maintenance costs, reduced equipment seem not to be able to achieve the right trade-off between energy
size, more efficient and effective utilization of equipment, conser- efficiency and demand satisfaction: they cannot provide reason-
vation of fossil fuels (by facilitating more efficient energy use) and able comfort at minimum energy use and financial costs. Different
reduced pollutant emissions (e.g. CO2 ) [1,2]. In particular, in HVAC approaches have been proposed in the literature in order to find
cooling systems, a popular thermal storage medium is ice: the cool- suitable controls for TES systems. In [5] a simulation environment
ing capacity of an ice Cold TES (CTES) system under total freezing is for the analysis of ice storage controls is presented. In [6] Henze
18 times as high as that of a water CTES system operating between et al. provide some guidelines to obtain an improvement of TES
12 ◦ C and 7 ◦ C [3]. An ice CTES has operating phases, namely, a system energy performances; these guidelines are derived from
charging phase where (typically at night) heat is removed from the analysis of optimal control and its comparison to TES sys-
water to produce ice, and a discharging phase, where, when the tem standard control strategies. In [7] predictive control design
for a three-story office building equipped with two chillers with
constant coefficient of performance and a thermal energy storage
∗ Corresponding author. Tel.: +39 0498277746.
system is illustrated.
E-mail addresses: beghi@dei.unipd.it (A. Beghi), lcecchinato@energyeverest.com
In this paper, a model-based approach is developed to design
(L. Cecchinato), mirco.rampazzo@dei.unipd.it (M. Rampazzo), efficient control strategies for HVAC systems equipped with ice
simminif@dei.unipd.it (F. Simmini). CTES. The thermal behaviour of the HVAC plant is analysed by

0959-1524/$ – see front matter © 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jprocont.2014.01.008

Please cite this article in press as: A. Beghi, et al., Energy efficient control of HVAC systems with ice cold thermal energy storage, J.
Process Control (2014), http://dx.doi.org/10.1016/j.jprocont.2014.01.008
G Model
JJPC-1712; No. of Pages 9 ARTICLE IN PRESS
2 A. Beghi et al. / Journal of Process Control xxx (2014) xxx–xxx

on model predictive control (MPC), which has been successfully


Nomenclature used in building cooling systems with water TES [8,9]. MPC refers
to a class of algorithms that compute a sequence of manipulated
C price of electricity [D W−1 min−1 ] variable adjustment in order to optimise the future behaviour of a
ChN nighttime-chiller controlled process. In this study, the future outputs of the plant are
ChD daytime-chiller predicted by using a white-box model fed by future control vari-
cpl water specific heat [J kg−1 K−1 ] ables together with disturbances forecasting. The optimisation step
cps ice specific heat [J kg−1 K−1 ] is developed by using a stochastic technique, the particle swarm
h tank well-mixed section fraction [–] optimisation (PSO) algorithm, that has already been proved to be
HS CTES hybrid system a practical solution in energy-related industrial applications [10].
ṁ massflow rate [kg s−1 ] In the provided examples two typical commercial building cooling
N moving window size [–] load profiles are taken into account. Furthermore, the case where
PLR part load ratio [–] the predicted disturbances and actual ones do not match is consid-
Q energy [J] ered. The simulations show that model predictive control provides
T water temperature [◦ C] the best control solution for the efficient management of ice CTES
Tair air temperature [◦ C] systems.
Tc control horizon [min] The paper is organised as follows. In Section 2 the structure of
Tp prediction horizon [min] the cooling plant with ice CTES and its model are presented. In
V volume [m3 ] Section 3, a TES charging simulation example is provided. In Sec-
VA-CD charging/discharging valve tion 4 extensive simulation results for TES discharging are given:
VA-TES modulating control valve conventional control strategies and the MPC approach are detailed
Z part load factor [–] and compared through dynamic simulation examples. Conclusions
˛ tuning coefficient [D K−2 ] are drawn in Section 5.
ˇ moving window weight factor [–]
 difference operator [–]
2. Plant description and model
 efficiency [–]
 specific latent heat [J kg−1 ]
In Fig. 1 a typical HVAC system of a commercial building with
 VA-TES valve opening [–]
ice storage, in chiller upstream configuration, is shown. The system
 water density [kg m−3 ]
is composed of two parallel air-condensed chillers, an ice stor-
 loss coefficient [K−1 s−1 ]
age, a temperature-modulating control valve, a diverting valve that
 time [min]
allows charging/discharging operations, and an air-conditioning
Subscripts system that can be a conventional central system. One chiller is
ch chiller used during the night in order to charge the TES; during the day
e electrical the other one and the ice CTES are used together to satisfy cooling
f water tank or piping demand.
g generation The following scenario is an example of a partial-storage sys-
i inlet tem at nominal conditions. During off-peak nighttime hours the
k block index nighttime-chiller (ChN) works as ice maker: a glycol solution (e.g.
l load 25% ethylene glycol) is pumped through the chiller coils and the
n nominal CTES in the chilled-water loop (VA-CD = 0, Fig. 1). The −5 ◦ C eth-
o outlet ylene glycol produced by the chiller freezes the water contained
s accumulation or storage inside the CTES and charges it for use during the next day’s cooling.
TES thermal energy storage Ice-making has the effect of de-rating the nominal chiller capacity
tran transition by approximately 30–35%. Compressor efficiency, however, varies
v VA-TES valve only slightly because lower nighttime temperatures result in cooler
w wastage condensation temperatures and help to keep the unit operating effi-
ciently (reducing the compressor pressure ratio). A full charging
cycle of an ice tank requires approximately 6–12 h, depending on
its size. During the discharge cycle in the following day (VA-CD = 1,
resorting to a lumped formulation of the conservation equa- Fig. 1), the glycol solution pre-cooled by the daytime-chiller (ChD) is
tions and a simulation environment based on Matlab/Simulink® further cooled by the ice CTES. The temperature-modulating valve,
is designed accordingly. The contribution of the paper is twofold. in the bypass loop around the TES, permits a sufficient quantity of
The first contribution regards modeling of the ice CTES. Differently glycol solution to bypass the storage, mixed with solution arriving
from what has been proposed so far in the literature, the ice CTES from the CTES, and allows us to achieve the desired temperature of
is modelled as a hybrid system, where the water phase transitions the supplied glycol solution that is distributed to the cooling devices
(solid–melting–liquid and liquid–freezing–solid) are described by (e.g. air-handler coils). The solution leaving the building re-enters
combining continuous and discrete dynamics, so that both latent the chiller and is cooled again.
and sensible heats are considered. It is worth noting that the ice The thermal behaviour of the plant can be usefully analysed by
latent heat represents a large portion of the total storage energy a lumped formulation of the conservation equations. The elements
(e.g. 90%). However it is appropriate to consider also the sensible of the plant are simulated in Matlab/Simulink® through blocks
heat portion in order to adequately evaluate the energy efficiency (chiller, piping, storage, valves, tank and thermal load). At each time
of the HVAC system when operated with different control strate- , the mass and energy balance equations for the k-th block are:
gies. The second contribution in the paper is the comparison, in
the developed simulation environment, of three standard control ṁi,k − ṁo,k = 0, ṁk := ṁi,k = ṁo,k , (1a)
strategies (constant-proportion control, chiller-priority control and
storage-priority control) and an advanced control strategy based Q̇g,k + ṁi,k cpl Ti,k − ṁo,k cpl To,k − Q̇w,k − Q̇s,k = 0. (1b)

Please cite this article in press as: A. Beghi, et al., Energy efficient control of HVAC systems with ice cold thermal energy storage, J.
Process Control (2014), http://dx.doi.org/10.1016/j.jprocont.2014.01.008
G Model
JJPC-1712; No. of Pages 9 ARTICLE IN PRESS
A. Beghi et al. / Journal of Process Control xxx (2014) xxx–xxx 3

Fig. 1. Plant structure of the HVAC system.

2.1. Chiller From (1b), we can depict the ice CTES as a heat exchanger, where
the variation of the state of charge can be expressed as the sum of
In the hypothesis of using air condensed chillers, the nominal the exchanged powers:
cooling power and the nominal electric power consumption, at full
(100%) load conditions, are expressed (by using a multiple linear Q̇s,TES = ṁTES cpl (Ti,TES − To,TES ) − Q̇w,TES . (4)
regression) as a function of the inlet chiller water temperature,
external air temperature and water mass flow rate. Moreover, the The first term on the right side of (4) represents the power
part load operation influence is taken into account by multiplying exchange between the water solution and the ice CTES during pro-
the full cooling capacity by the part load ratio (PLR, defined as the cess operations, which results:
chiller cooling load divided by its maximum capacity for the given
ṁTES cpl (Ti,TES − To,TES ) = ṁTES cpl (Ti,TES − TTES ). (5)
values of air and glycol inlet temperature) as well as by multiplying
the full load power consumption by a part load factor Z, which is According to the lumped formulation, TTES is the average tem-
calculated as a function of PLR and air temperature [11]: perature of the water/ice in the CTES. To take into account the
Q̇g,ch = PLR · Q̇g,ch100% , Q̇e,ch = Z · Q̇e,ch100% . (2) efficiency of the heat exchange, we have added in (5) the coefficient
, i.e. the ratio of the actual rate of heat transfer to the maximum
For the particular system under consideration, chiller water con- possible rate. This coefficient varies during the charging and dis-
tent can be neglected and the energy balance equation, for the charging processes (due to the change of heat transfer coefficient
chiller block, at time , can be written as follows: and transfer area) and can be expressed as a function of the nominal
Q̇g,ch = ṁch cpl (To,ch − Ti,ch ). (3) capacity, the state of charge of the TES and the difference between
the glycol and ice temperatures.
2.2. TES Due to the storage dimension and the high difference between
water/ice temperature in the storage and the external air temper-
The ice CTES is modelled as a hybrid system, where the water ature, it is important to consider also the energy dispersion in the
phase transitions are described by combining continuous and dis- TES. In order to take into account the heat transfer to the ice from
crete dynamics. Although the ice latent heat represents a large the surroundings, the second term on the right side of (4), involves
portion of the total storage energy in the TES (e.g. 90%), we con- TES energy wastage:
sider here also the sensible heat portion in order to evaluate the
energy efficiency of the HVAC system when operated with different Q̇w,TES = Qn,TES (TTES − Tair ). (6)
control strategies.
The loss coefficient  is related to the degree of isolation of the
In the CTES an event changes the phase from solid to melting,
storage with respect to the external environment.
thereby also changing the continuous dynamics of the water: the
absorbed thermal energy melts ice and the temperature of the
water does not change during melting since water is in a two- 2.3. Modulating valve
phase condition. The change from a single-phase condition (liquid,
solid) to a two-phase condition (melting, freezing) happens when a In the bypass loop around the TES, the temperature-modulating
given temperature level is reached (e.g. the saturation temperature valve (VA-TES, Fig. 1), by manipulating its opening , allows a quan-
corresponding to the water pressure). Instead, a transition from tity of glycol (from chiller) to bypass the ice storage and to be mixed
a two-phase condition to a single-phase condition occurs when with the outlet TES glycol, achieving the chilled glycol set-point
a given energy (e.g. latent heat fusion energy) is added or sub- value for the building. At each time , material and energy balance
tracted to the water. Fig. 2 shows the different phases (and their equations yield:
corresponding continuous dynamics) and the transitions between
them. ṁv cpl To,v = (1 − )ṁv cpl To,ch + ṁv cpl To,TES . (7)

Please cite this article in press as: A. Beghi, et al., Energy efficient control of HVAC systems with ice cold thermal energy storage, J.
Process Control (2014), http://dx.doi.org/10.1016/j.jprocont.2014.01.008
G Model
JJPC-1712; No. of Pages 9 ARTICLE IN PRESS
4 A. Beghi et al. / Journal of Process Control xxx (2014) xxx–xxx

Fig. 2. Ice CTES hybrid system (HS): the phases of the water.

2.4. Water tank and piping : x( + 1) = f(x(), u(), d()), (12a)

The water tank is modelled as two separate parts connected in y() = h(x()), (12b)
series. In the first part a well-mixed condition is assumed and the
x(0) = x0 , (12c)
energy balance equation results:
dT f () ≥0, (12d)
ṁf cpl Ti,f () = ṁf cpl Tf () + hVf cpl , (8)
d
(12e)
where h is the well-mixed section fraction of the tank total volume. y ∈ Y := [ymin , ymax ] ⊂ R,
In the second part a perfect stratification condition is considered:
u ∈ U := [umin , umax ] ⊂ Rdim(u) , (12f)
ṁf cpl To,f () = ṁf cpl Tf ( − c ), (9)
d ∈ D := [dmin , dmax ] ⊂ Rdim(d) , (12g)
where  c is the tank section time constant defined as:
Vf x ∈ X := [xmin , xmax ] ⊂ R2 , (12h)
c = (1 − h) . (10)
ṁf
where x, u, d are the vectors of state, control variables and dis-
The same approach is used to model piping blocks, although turbances, respectively, y is the output variable, f is a non-linear
water mixing is negligible due to the low water velocity. More map, h is a simple map that selects the first element of x. The sets
details about the derivation of the previous equations can be found Y, U, D, X are the feasible domains.
in [12].
3. TES charging
2.5. Building thermal load
During TES charging, only the part load ratio is used in order to
For the building load block, at time , the energy balance equa- control the outlet chiller water temperature and all the glycol solu-
tion results: tion coming from the chiller enters the storage to charge it (Table 1).
In detail, a PI-controller with feed-forward is used: knowledge
Q̇g,l = ṁl cpl (To,l − Ti,l ). (11)
about the external air temperature can be fed forward and com-
bined with the PI-controller to improve the system performance;
2.6. Model summary
the reference trajectory for the outlet chiller water temperature is
set to −5 ◦ C. Fig. 3 shows temperature inside the CTES and the nor-
The overall discrete dynamical system (one-minute samp-
malised TES energy during a 12-h charging cycle (from 06 pm to
ling time) during charging/discharging operations (according to
06 am). The normalised TES energy is set to 0 when storage water
Table 1) can be summarised as follows:
temperature is 7 ◦ C while it is set to 1 when storage ice temperature
is −5 ◦ C. Since the energy price is low during the night, it is useful to
Table 1 consider also the ice/water sensible heat and to charge the storage
Charging/discharging operations. until its temperature reaches the outlet chiller water temperature.
Chiller Charging Discharging
ChN ChD
4. TES discharging
VA-CD 0 1
Eqs. (2)–(10), HS (2)–(11), HS
In order to satisfy cooling demand during working hours,
 1 [0, 1]
y To,ch Ti,l standard control strategies and a non-linear model predictive con-
u PLR [PLR ; ] trol approach are compared. In this section, at first the non-linear
d Tair [Q̇g,l ; Tair ] predictive control problem is described, then simulation results for
x [To,ch ; Qs,TES ] [Ti,l ; Qs,TES ] TES discharging are provided.

Please cite this article in press as: A. Beghi, et al., Energy efficient control of HVAC systems with ice cold thermal energy storage, J.
Process Control (2014), http://dx.doi.org/10.1016/j.jprocont.2014.01.008
G Model
JJPC-1712; No. of Pages 9 ARTICLE IN PRESS
A. Beghi et al. / Journal of Process Control xxx (2014) xxx–xxx 5

Fig. 4. Example of cooling load forecasting of a commercial building.


Fig. 3. TES temperature and normalized TES energy.

limited data set of standard measures available in HVAC installa-


4.1. NLMPC problem formulation
tions like the one considered here. In particular, it is assumed that
load forecasting for the next day is made at midnight. In the first
The controller consists of three fundamental elements: a pre-
step, training of the ANN on data from previous days is performed.
dictor that predicts the outputs based on the model and process
Then, input predictions are used as the inputs of the ANN model to
information, a cost function, and an algorithm to solve a constrained
forecast the coming 24-h cooling load profile. Only cooling devices
non-linear optimisation problem. The aim of the proposed non-
(e.g. Air Handling Unit) schedule, external dry-bulb and wet-bulb
linear MPC (NLMPC) is to minimize electricity costs during the
temperatures are used as inputs of the ANN model to obtain a cool-
building operating hours, while keeping the cooling demand satis-
ing load forecasting. In fact, in practical situations, information on
fied. We define the electricity bill as:
internal loads (lighting, computers, and occupancy) is not available.
B() = C()Q̇e,ch (). (13) However since internal gains are highly correlated to the cooling
devices schedule, a satisfactory load prediction performance can be
The model predictive control action, at time , is obtained by solving achieved: e.g. in Fig. 4 actual and predicted load data are compared
the optimisation problem: for two weeks in a typical commercial building.
find argminJ(x, u, d, B, Tc , Tp ), (14a) Forecasting of the external air temperature, T̂air , is made with a
u weighted moving average, since we suppose that weather forecast
is not available:
t+Tp
  2
with J = B() + ˛y() − r() , (14b)
1 
=t T̂air,j+1 () = ˇj Tair,j () + ˇj−1 Tair,j−1 ()
N
s.t. x( + 1) = f(x(), u(), d()), (14c) 
+· · · + ˇj−N+1 Tair,j−N+1 () , (15)
y() = h(x()), (14d)
where Tair,j refers to the temperature of the j-th day. Fig. 5 shows
u() ∈ U , ∀ ∈ [t, t + Tc ], (14e) an example of air temperature prediction for two weeks.
u() = u(t + Tc ), ∀ ∈ [t + Tc , t + Tp ], (14f)

d() ∈ D, ∀ ∈ [t, t + Tp ], (14g)

x() ∈ X, ∀ ∈ [t, t + Tp ], (14h)

y() ∈ Y, ∀ ∈ [t, t + Tp ], (14i)

where Tp and Tc are the prediction and the control horizon with
Tc ≤ Tp ; ˛ is a weight coefficient. The term r refers to the reference
trajectory. The vectors of state, input, disturbances and output (x,
u, d, y) are stated in Table 1. In order to distinguish the real system
from the system model used to predict the future within the con-
troller, we use a bar to denote the internal variables in the controller
(e.g. x). The control action sent to the plant is the first element of
the optimal sequence u.

4.1.1. Disturbances forecasting


To solve (14a), disturbances forecasting (cooling load and exter-
nal air temperature) are required. Therefore, we implement the
strategy proposed in [13] where an ANN (Artificial Neural Network)
algorithm allows us to obtain accurate predictions by exploiting a Fig. 5. Example of temperature prediction.

Please cite this article in press as: A. Beghi, et al., Energy efficient control of HVAC systems with ice cold thermal energy storage, J.
Process Control (2014), http://dx.doi.org/10.1016/j.jprocont.2014.01.008
G Model
JJPC-1712; No. of Pages 9 ARTICLE IN PRESS
6 A. Beghi et al. / Journal of Process Control xxx (2014) xxx–xxx

4.1.2. Optimisation
To solve the non-linear optimisation problem (14a) a nature-
inspired, meta-heuristic algorithm (PSO) is used, [14]. The
preference for the PSO algorithm with respect to other kinds of opti-
misation is motivated as follows. At first, PSO algorithm makes few
or no assumptions about the optimisation problem to be solved; it
does not need gradient information of the objective function under
consideration, as required by classic optimisation methods such as
gradient descent and quasi-Newton methods, and it can search very
large spaces of candidate solutions. Even if in this particular case
the objective function could be minimized by classic optimisation
methods, we want to validate a method able to grant optimisation
for more general problems, since here we provide a particular case
for the modeling of a HVAC plant with TES, whereas other kinds
of modeling could be arduous for an optimisation task using gradi-
ent methods. Moreover, the PSO algorithm is suitable for avoiding
local minima. With respect to other stochastic optimisation tech-
niques (e.g. genetic algorithms) PSO is more advantageous because
there are few parameters to set. Furthermore, it can be easily imple-
mented and it is not computationally expensive, since its memory
and CPU speed requirements are low, [15].
The PSO algorithm searches the decision variable space by
adjusting the trajectories of individual particles as the piecewise
paths formed by positional vector in a quasi-stochastic manner.
The key idea in the PSO method is in the way to combine two dif-
ferent types of perspectives (individual and social perspectives) to
predict the best update for the position in the next iteration [16].
At each iteration, each particle combines information from the cur-
rent motion vector ṗ(t), individual correction (vector pP (t) − p(t)),
and social correction (vector pG (t) − p(t)). The contribution of these
components results in the prediction of the position vector in the
next iteration as given by p(t + 1). The positions pG (t) and pP (t) rep-
resent the global and local best particles positions at iteration t
respectively, Fig. 6. The aim is to find the global best solution among
all the current best ones until the objective function no longer
improves or after a certain number of iterations.
There are many PSO variants, and in this paper, to determine the
optimal future control inputs, a version of PSO algorithm which
deals with constraints is used, [10]. The sub-optimal solution for
the optimisation problem (14a) obtained by applying the PSO Fig. 7. Building cooling load and external air temperature.
algorithm has been validated with a brute-force search method.
The computations required by the optimisation process can be
performed within a fixed time length thus granting on-line imple- are available (until the end of the working hours [13]). The control
mentation. sampling time is set to 15 min. To reduce the computational bur-
den, a move blocking strategy is used by imposing inputs (PLR and
) to be constant over several time-steps in the prediction hori-
4.1.3. NLMPC
zon [17]. It is worth noting that to ensure feasibility for the NLMPC
The prediction and control horizons are set to a value depend-
problem, the storage is assumed to always have enough energy to
ing on the length of time window, where disturbances predictions
satisfy time-varying, uncertain building load demands.

4.2. Simulation results

To illustrate the performance of the non-linear MPC we compare


the predictive control with three standard control strategies:

(P) Constant-proportion control: the storage and the chiller meet


a constant fraction of the cooling load under all conditions (e.g.
Chiller 40% and ice CTES 60%).
(C) Chiller-priority control: chiller runs continuously to satisfy
cooling load. If cooling load becomes greater than chiller max-
imum capacity the storage provides the remaining cooling
power needed.
(S) Storage-priority control: this strategy requires melting as much
ice as possible during the on-peak period. After the discharge
of the CTES, chiller provides cooling power needed in the
Fig. 6. PSO, graphical illustration of position update. remaining time.

Please cite this article in press as: A. Beghi, et al., Energy efficient control of HVAC systems with ice cold thermal energy storage, J.
Process Control (2014), http://dx.doi.org/10.1016/j.jprocont.2014.01.008
G Model
JJPC-1712; No. of Pages 9 ARTICLE IN PRESS
A. Beghi et al. / Journal of Process Control xxx (2014) xxx–xxx 7

Fig. 8. Part load ratio. Fig. 9. VA-TES valve opening.

In the hypothesis that the ice CTES is fully recharged by the chiller power and do not discharge the CTES. In this way, part
chiller during nighttime, we report simulation results during day- of the energy produced during the night is not used to satisfy
time, when CTES discharge occurs (from 06 am to 06 pm). We set cooling demand, causing energy wastage.
r() equal to 7 ◦ C for each . We introduce three time slots of elec- (H) It can be seen that proportion control and storage pri-
tricity price: F1 (high price, on-peak), F2 (medium price, on-peak) ority control do not satisfy cooling demand (Fig. 10(b)):
and F3 (low price, off-peak). Two building cooling load case studies the resulting discharging process turns out to be too fast
are considered: low load (L) and high load (H) conditions, Fig. 7(a) (Figs. 11(b) and 12(b)), and the chiller cannot satisfy cooling
and (b), respectively. demand. Instead, chiller control priority strategy and predic-
tive control always provide sufficient cooling power. However,
(L) Figs. 8(a) and 9(a) show the manipulated variables with in chiller-priority case complete discharge of the CTES is not
the different control strategies. Fig. 9(a) includes time slot achieved.
F3 meaning that TES energy has been produced during the Some comments on the NLMPC simulation results are in
night (off-peak period). In Fig. 10(a) the inlet load-side water order. From information given by disturbances forecasting,
temperature is reported. Finally, Figs. 11(a) and 12(a) show predictive control strategy can completely discharge the ice
temperature inside the CTES and the normalised TES energy, storage within the end of the working hours. During F1
respectively. It can be seen that in this case all control strate- time slot (high price) the NLMPC algorithm prefers using TES
gies meet cooling building load (Fig. 10(a)). However, NLMPC energy (produced during the night), while during F2 time slot
controller and storage priority controller satisfy demand by (medium price), NLMPC tends to exploit chiller power instead
using only storage energy, while the remaining strategies use of the storage energy (Figs. 8(b) and 9(b)). The increase of the

Please cite this article in press as: A. Beghi, et al., Energy efficient control of HVAC systems with ice cold thermal energy storage, J.
Process Control (2014), http://dx.doi.org/10.1016/j.jprocont.2014.01.008
G Model
JJPC-1712; No. of Pages 9 ARTICLE IN PRESS
8 A. Beghi et al. / Journal of Process Control xxx (2014) xxx–xxx

Fig. 10. Inlet load-side water temperature. Fig. 11. TES temperature.

VA-TES valve opening in the late afternoon is explained by the


need of exploiting of the last energy in the storage, since its can be seen that the NLMPC provides the lowest energy bill, always
temperature has exceeded 0 ◦ C. satisfying cooling demand.
The NLMPC controller gives the best performance due to the
In Table 2 the normalized electricity bills for all control strategies fact that we consider the actual disturbances equal to the predicted
are given, also including nighttime bill during the next charging. ones. In this study we also consider the case in which the actual load
Nighttime-chiller has a lower energy efficiency ratio (EER, defined and external air temperature differ from the forecast values. When
as the ratio of cooling capacity and total power absorption) with actual and forecast values do not match, we use dc () for each 
respect to the one used during the day due to the fact that cooler instead of d() where:
water temperature is needed to recharge the TES. However it is
convenient to fully recharge the TES to exploit low electricity price.
dc () = d() + q(), (16)
Bills are normalized on the bill of the chiller-priority strategy (100,
high case). Bills are not reported when the demand is not satisfied. It
and q is an appropriate function that allows smoothly connect-
ing the actual values of d(t) with its future values as previously
Table 2 predicted.
Electricity bills.
In the following example we consider an unpredictable load dis-
L H turbance from 01 pm to 03 pm, Fig. 13. The inlet load side water
Constant-proportion control 59 –
temperature trajectory performed by the NLMPC is depicted in the
Chiller-priority control 67 100 same figure. It can be seen that updating predictions results in good
Storage-priority control 41 – performances since the error from the reference trajectory (7 ◦ C) is
Non-linear model predictive control 41 94 negligible for typical HVAC building applications.

Please cite this article in press as: A. Beghi, et al., Energy efficient control of HVAC systems with ice cold thermal energy storage, J.
Process Control (2014), http://dx.doi.org/10.1016/j.jprocont.2014.01.008
G Model
JJPC-1712; No. of Pages 9 ARTICLE IN PRESS
A. Beghi et al. / Journal of Process Control xxx (2014) xxx–xxx 9

5. Conclusion

In HVAC systems, thermal energy storages can be used to shift


peak loads from on-peak to off-peak periods. The control operations
are crucial in order to ensure energy efficiency and load demand
satisfaction. In this paper a cooling system with ice storage is con-
sidered. From a modeling point of view, the ice CTES is described in
terms of a hybrid system, thus taking into consideration the dynam-
ics of both sensible and latent heat. Standard control strategies are
compared to an NLMPC-based approach. In particular, the NLMPC is
developed by using a predictor based on plant model and process
information, a cost function, and a PSO algorithm. The predicted
outputs of the plant are also obtained from cooling devices sched-
uling as a mean for inferring information on the internal loads,
which is in general not available in practice. Extensive dynamic sim-
ulations show that NLMPC exhibits better performance (in terms
of energy efficiency and demand satisfaction) than conventional
strategies.
The use of a stochastic algorithm like PSO (which does not use
derivative information) for the optimisation task gives the possi-
bility to easily extend the predictive control strategy to other plant
configuration (e.g. chiller downstream configuration) or inserting
extra terms in the performance index (e.g. comfort index). Future
developments include controller design for such extended cases
and the development of a model free approach (using experimental
data) for the prediction task.

References

[1] I. Dincer, S. Dost, X. Li, Performance analyses of sensible heat storage systems for
thermal applications, International Journal of Energy Research 21 (12) (1997)
1157–1171.
[2] M. Rosen, I. Dincer, N. Pedinelli, Thermodynamic performance of ice thermal
energy storage systems, ASME – Journal of Energy Resources Technology 122
(4) (2000) 205–211.
[3] I. Dincer, M. Rosen, Thermal Energy Storage: Systems and Applications, John
Wiley & Sons, Ltd., West Sussex, United Kingdom, 2011.
[4] H. Akbari, O. Sezgen, Case Studies of Thermal Energy Storage (TES) Systems:
Evaluation and Verification of System Performance, Tech. Rep., 1992.
[5] G.P. Henze, M. Krarti, M.J. Brandemuehl, A simulation environment for
the analysis of ice storage controls, HVAC&R Research 3 (2) (1997)
128–148.
[6] G.P. Henze, M. Krarti, M.J. Brandemuehl, Guidelines for improved per-
formance of ice storage systems, Energy and Buildings 35 (2) (2003)
111–127.
[7] G.P. Henze, C. Felsmann, G. Knabe, Evaluation of optimal control for active and
passive building thermal storage, International Journal of Thermal Sciences 43
Fig. 12. Normalized TES energy. (2) (2004) 173–183.
[8] Y. Ma, F. Borrelli, B. Hencey, B. Coffey, S.C. Bengea, P. Haves, Model predic-
tive control for the operation of building cooling systems, IEEE Transactions on
Control Systems Technology 20 (3) (2012) 796–803.
[9] Y. Ma, A. Kelman, A. Daly, F. Borrelli, Predictive control for energy efficient build-
ings with thermal storage: modeling, stimulation, and experiments, Control
Systems, IEEE 32 (1) (2012) 44–64.
[10] A. Beghi, L. Cecchinato, G. Cosi, M. Rampazzo, A pso-based algorithm for opti-
mal multiple chiller systems operation, Applied Thermal Energy 32 (2012)
31–40.
[11] A. Beghi, L. Cecchinato, M. Rampazzo, A multi-phase genetic algorithm for the
efficient management of multi-chiller systems, Energy Conversion and Man-
agement 52 (3) (2011) 1650–1661.
[12] M. Albieri, A. Beghi, C. Bodo, L. Cecchinato, Advanced control systems for single
compressor chiller units, International Journal of Refrigeration 32 (5) (2009)
1068–1076.
[13] A. Beghi, L. Cecchinato, M. Rampazzo, F. Simmini, Load forecasting for the effi-
cient energy management of hvac systems, in: Proceedings of the 2nd IEEE
International Conference on Sustainable Energy Technologies, 2010.
[14] J. Kennedy, R. Eberhart, Particle swarm optimization, in: Proceedings of the IEEE
International Conference on Neural Networks, vol. 4, 1995, pp. 1942–1948.
[15] R. Eberhart, P. Simpson, R. Dobbins, Computational Intelligence PC Tools, Aca-
demic Press, Inc., Boston, MA, USA, 1996.
[16] S. Mikki, A. Kishk, Particle Swarm Optimization: A Physics-Based Approach,
Morgan and Claypool, Arizona State University, 2008.
[17] R. Cagienard, P. Grieder, E. Kerrigan, M. Morari, Move blocking strate-
gies in receding horizon control, Journal of Process Control 17 (6) (2007)
Fig. 13. Inlet load-side water temperature with load prediction update. 563–570.

Please cite this article in press as: A. Beghi, et al., Energy efficient control of HVAC systems with ice cold thermal energy storage, J.
Process Control (2014), http://dx.doi.org/10.1016/j.jprocont.2014.01.008