energies

Article
Definition and Experimental Validation of a
Simplified Model for a Microgrid Thermal Network
and its Integration into Energy Management Systems
Andrea Bonfiglio 1, *, Massimo Brignone 1 , Federico Delfino 1 , Alessandro Nilberto 2
and Renato Procopio 1
1 Department of Naval, Electrical, and ICT Engineering, University of Genova, I-16145 Genoa, Italy;
massimo.brignone@gmail.com (M.B.); federico.delfino@unige.it (F.D.); renato.procopio@unige.it (R.P.)
2 Mechanical, Energy, Management and Transportation Engineering Department, University of Genova,
I-16145 Genoa, Italy; a.nilberto@unige.it
* Correspondence: a.bonfiglio@unige.it; Tel.: +39-010-353-2730

Academic Editor: Brian Agnew

Received: 29 August 2016; Accepted: 2 November 2016; Published: 4 November 2016

Abstract: The present paper aims at defining a simplified but effective model of a thermal network
that links the thermal power generation with the resulting temperature time profile in a heated
or refrigerated environment. For this purpose, an equivalent electric circuit is proposed together
with an experimental procedure to evaluate its input parameters. The paper also highlights the
simplicity of implementation of the proposed model into a microgrid Energy Management System.
This allows the optimal operation of the thermal network to be achieved on the basis of available
data (desired temperature profile) instead of a less realistic basis (such as the desired thermal
power profile). The validation of the proposed model is performed on the Savona Campus Smart
Polygeneration Microgrid (SPM) with the following steps: (i) identification of the parameters involved
in the equivalent circuit (performed by minimizing the difference between the temperature profile,
as calculated with the proposed model, and the measured one in a set of training days); (ii) test of
the model accuracy on a set of testing days (comparing the measured temperature profiles with the
calculated ones); (iii) implementation of the model into an Energy Management System in order to
optimize the thermal generation starting from a desired temperature hourly profile.

Keywords: energy management; optimization algorithm; parameter identification; smart grids;

thermal network

1. Introduction
One of the main focuses of microgrid research has been the development of optimal operational
strategies to reach the maximum potential of their promising features in the energy management
context. Such strategies are typically implemented in so-called Energy Management Systems (EMS).
Due to their importance, EMSs have been deeply investigated in various forms and
architectures [1,2]. The most common architecture for EMSs is the multilevel hierarchical
architecture [3], in which the higher level accounts for the dispatch of active and reactive power of the
programmable production units, while the lower ones perform a more detailed control of the single
units. The highest level of this architecture recalls the principles of optimization of traditional electricity
transmission networks, achieved by classical Optimal Power Flow (OPF) algorithms. However,
the presence of dynamic components such as storage devices, together with the uncertainties that
characterize load request and renewable energy sources (RES), and the combined presence of electric
and thermal infrastructure make this optimization problem a very hard task from a computational
point of view. To overcome this difficulty, many different papers have been published with the aim of

Energies 2016, 9, 914; doi:10.3390/en9110914 www.mdpi.com/journal/energies

Energies 2016, 9, 914 2 of 14

developing accurate and efficient algorithms for microgrid EMSs, addressing either the integration of
RES and storage devices, the so-called demand side management (DSM), or again the satisfaction of
electric network constraints [46].
In a recent paper [7], the attention of the authors has been concentrated on efficient electric
network modeling for optimization purposes that has reached a good compromise between accuracy
and reduction in the Computer Processing Unit (CPU) effort. However, the thermal network model
consisted only of a pure thermal power balance equation. This is quite unrealistic, as it assumes
knowledge of the current and future thermal load request expressed in terms of a thermal power rather
than a temperature profile and does not consider the dynamic transients between the instant in which
the thermal source is switched on and the moment in which the temperature reaches the desired value.
The proposed paper aims at bridging this gap by defining an equivalent electric circuit for a thermal
network that is able to account for both these issues.
In the literature, many studies have dealt with the modeling of thermal networks from
different point of views (as an example [810]) focusing either on optimization or simulation aspects.
The advantages of district heating networks based on combined heating and power (CHP) technologies
have been shown in [11,12] and, especially in [11], a model for the optimal daily operation of a district
system with a CHP plant has been proposed.
As highlighted by Larsen et al. [13], models including a full physical representation of thermal
networks, in terms of mass flow rates, pressure drops and pipe dimensions, are computationally
intensive and, as a consequence, are not suitable to be inserted into EMSs; to this end, they propose a
simplified method which permits them to study a district heating network by reducing its topological
complexity. On the other hand, Kuosa et al. [14] focus their attention on the differences between
radial and meshed networks and compare a traditional network with a ring one characterized by an
innovative flow rate control method. Furthermore, the same authors introduced in [8] an equivalent
thermal circuit to represent the heat transfer process in a heat exchanger and applied this methodology
to study a district heating system.
More simplified models of thermal networks are usually implemented within Mixed-Integer
Linear Programming (MILP) tools [1517], where district heating networks are optimally designed
and operated in order to minimize different objectives, such as capital and daily operational costs.
In the aforementioned models, energy balance equations, operational constraints and correlations
for the calculation of heat losses and pressure drops are reported. In any case, what is necessary
for an EMS is basically to provide a relationship between the thermal power generation (which is
typically a decisional variable) and the temperature time profile along the district fed by the thermal
network (which is typically the desired output). Such a relationship is inserted into the EMS as a
set of constraints in order to find the best thermal power generation profile that allows the desired
temperature to be reached.
As the physical relationship between power and temperature is complicated and involves
knowledge of many quantities and parameters, the present paper proposes a different strategy with
respect to the previous publications: it postulates that the thermal network can be represented by means
of an equivalent electric circuit and determines the numerical values of the circuit parameters by means
of an optimization algorithm that aims at minimizing the difference between the calculated temperature
time profile and the measured one starting from the measurement of the thermal power injection.
The equivalent electric circuit proposed in the present contribution is tailored on the thermal
network of the Savona Campus Smart Polygeneration Microgrid (SPM), but it is representative
of a general approach that can be adopted to study heat distribution systems in a simplified and
effective way.
The paper is structured as follows: Section 2 describes the theoretical part of the work with the
definition of the equivalent circuit of the thermal network, describing the meaning of all the terms in
detail and its dynamic behavior (Section 2.1) and the procedure for parameter estimation (Section 2.2),
while Section 2.3 describes the insertion of the proposed model into a microgrid EMS. Section 3
Energies 2016, 9, 914 3 of 14
Energies 2016, 9, 914 3 of 15

performs a validation
the SPM. Finally, of the
Section proposed
4 provides model
some on the basis
conclusive of the
remarks onmeasurements
the work done acquired on field
and possible in
future
the SPM. Finally, Section
developments on the topic.4 provides some conclusive remarks on the work done and possible future
developments on the topic.
2. Results
2. Results
2.1. Equivalent
2.1. Equivalent Electric
Electric Circuit
Circuit
In order
In order to
to be
be integrated
integrated into
into the
the EMS,
EMS, the
the proposed
proposed thermal
thermal network
network model
model must
must have
have the
the
following properties:
following properties:
(1) It has to be simple in order to be used in synergy with an EMS. In other words, the equations
(1) It has to be simple in order to be used in synergy with an EMS. In other words, the equations
that constitute such a model have to be inserted as new constraints in the EMS. So, to limit the
that constitute such a model have to be inserted as new constraints in the EMS. So, to limit the
computational effort of the optimisation algorithm encoded in the EMS, such a model must be
computational effort of the optimisation algorithm encoded in the EMS, such a model must be
described by few and simple equations;
described by few and simple equations;
(2) It has to be accurate in modeling the delay time between the instant in which the thermal source
(2) is It switched
has to be accurate
on and the in one
modeling
in whichthe the
delay time between
temperature the instant
reaches in which
the desired valuetheatthermal
the usersource
side;
is switched on and the one in which the temperature reaches the
(3) It has to provide a reliable link between the desired temperature time profiles and the desired value at the user side;
(3) corresponding
It has to provide thermala reliable link has
power that between the desired
to be produced temperature
by thermal powertime profiles and the
generators.
corresponding thermal power that has to be produced by thermal power generators.
A simple schematic representation of the thermal circuit at the basis of this research work is
proposed
A simple in Figure 1. In order
schematic to meet allofthese
representation goals, one
the thermal can consider
circuit an equivalent
at the basis linearwork
of this research electric
is
circuit since
proposed in it is described
Figure 1. In orderby ato
linear
meetdynamic
all thesesystem; thecan
goals, one topology
considerof such a circuit can
an equivalent linearbe derived
electric
startingsince
circuit fromit the well-known
is described by aparallelism
linear dynamic between electric
system; and thermal
the topology devices.
of such According
a circuit to this
can be derived
starting from the well-known parallelism between electric and thermal devices. According to and
approach, typically resistors model the thermal losses, capacitors account for thermal inertias this
current (voltage)
approach, typicallysources represent
resistors model thethe power
thermal(temperature) generation.
losses, capacitors accountInfororder to meet
thermal the first
inertias and
model requirement,
current (voltage) sources the model accounts,
represent as depicted
the power in Figure
(temperature) 1, for a thermal
generation. In orderring
to meet (TR)theand
firsta
user (U). The first one involves the thermal sources and the thermal
model requirement, the model accounts, as depicted in Figure 1, for a thermal ring (TR) and a network until the buildings,
while the
user (U). second
The firstoneone takes
involvesintotheaccount
thermalthe buildings
sources and the and their network
thermal interface until
withthe thebuildings,
outdoor
environment.
while the second This
onelatter
takesrepresents
into accountthethe
temperature
buildings and mean request
their of an
interface witharray of homologous
the outdoor end-
environment.
userslatter
This (a higher numberthe
represents of temperature
rooms) to be considered
mean request unique
of antoarray
avoidofahomologous
significant increase
end-users in the
(a system
higher
complexity.
number of rooms) to be considered unique to avoid a significant increase in the system complexity.

Figure 1. Schematic representation of the thermal circuit considered for the proposed methodology:
methodology:
the thermal ring (TR) involves the thermal sources and the thermal network while the user (U)
represents the thermal loads fed
fed by
by the
the network.
network.

The choice of using an equivalent electric representation instead of a purely thermal one is
driven by the fact that several EMS for microgrid applications are typically designed to manage the
electric part of the grid. For this reason, if an electric equivalent of the system is provided, its
Energies 2016, 9, 914 4 of 14

The choice of using an equivalent electric representation instead of a purely thermal one is driven
by the fact that several EMS for microgrid applications are typically designed to manage the electric
part of 2016,
Energies the grid.
9, 914For this reason, if an electric equivalent of the system is provided, its integration could 4 of 15
be performed in a larger set of EMS tools and would also allow the management of thermal power to
integration
be upgradedcould be performed
by retrofitting in a larger
existing set of EMS tools and would also allow the management of
EMS structures.
thermal
The proposed equivalent circuit is depictedexisting
power to be upgraded by retrofitting EMS
in Figure structures.
2 and the circuital elements can be described
The proposed equivalent circuit is depicted in Figure 2 and the circuital elements can be
as follows:
described as follows:
The current source I0 represents the thermal power generation;
The capacitance
The current source C0 Iaccounts
0 represents the thermal power generation;
for the thermal inertia responsible for the time delay necessary to
The capacitance C 0 accounts for the thermal inertia responsible for the time delay necessary to
obtain the desired temperature in the heat generator circuit;
obtain the desired temperature in the heat generator circuit;
The conductance G0 models the heat generator circuit losses;
The conductance G0 models the heat generator circuit losses;
The voltage source V 0 models the fact that, without any thermal source, the heat generator
The voltage source V0 models the fact that, without any thermal source, the heat generator
network temperature approaches the ground temperature;
network temperature approaches the ground temperature;
The switch B1 models the possible disconnection between the generation and the load (encoded in
The switch models the possible disconnection between the generation and the load (encoded
a Boolean variable S, true when B1 is open);
in a Boolean variable S, true when B1 is open);
The diode D1 models
The diode modelsthe the fact
fact thatthat
the the
useruser
cannotcannot transfer
transfer heat toheat to theinwater
the water ingenerator
the heat the heat
generator
circuit; circuit;
The conductance G
The conductance G11 accounts for the pipe thermal losses in the thermal circuits of the buildings;
11 accounts for the pipe thermal losses in the thermal circuits of the buildings;
The capacitance C accounts
The capacitance C11 accounts for for the
the thermal
thermal inertia
inertia responsible for the
responsible for the time
time delay
delay necessary
necessary toto
obtain
obtain the
the desired
desired temperature
temperature at at the
the user
user side;
side;
The conductance G G11 represents the thermal losses, which depend on the user user insulation;
insulation;
The voltageV 1Vmimics
The voltage 1 mimics the the
fact that,
fact without any thermal
that, without source, the
any thermal user temperature
source, approaches
the user temperature
the outdoor the
approaches ambient
outdoor one.ambient one.

In the proposed
proposed model,
model, V VC0
C0 is
isthe
thevoltage
voltage between
between terminals
terminals A
A and
and B
B and
and represents
represents the thermal
while V
ring temperature, while VC1
C1,, which
which is is the voltage between terminals A11 and B11,, represents the user
mean temperature.

2. The
Figure 2. The proposed
proposed equivalent
equivalent electrical
electrical circuit of the Smart Polygeneration Microgrid (SPM)
thermal network
thermal network(the(thegreen
green
lineline denotes
denotes whatwhat has been
has been labelled
labelled as TR
as TR and theand
bluethe
one blue
is theone is the
boundary
boundary
of the userof
U).the user U).

Applying the
Applying the Kirchoffs
Kirchoffs laws
laws toto the circuit of
the circuit of Figure
Figure 2,
2, one
one obtains
obtains the
the following
following set
set of
of two
two
ordinary differential
differential equations
equations in
in the
the state
state variables
variables VVC0 and VC1:
ordinary C0 and VC1 :
dV ( t ) G0 + G11 G 1 G G0
VC 0 ( t ) +G1111 VC1 ( t ) + 1I 0 ( t ) + 11GV11B1 ( t ) +
)0+ 0 ( 0)V0 (t)
dVC0 (t) C 0 = + G
G V Gt (1)
= 0
dt
11
C0
VC0 ( t ) + C 0
V C1 ( t ) +C 0
I0 ( t ) +
C 0
V B1 ( t
C (1)
dt C0 C0 C0 C0 C0
dVC 1 ( t ) G11 G + G11 G G
= VC 0 ( t ) 1 VC 1 ( t ) 11 VB1 ( t ) + 1 V1 ( t ) (2)
dt C1 C1 C1 C1

with initial conditions V C 0 ( t 0 ) = V C*0 and V C 1 ( t 0 ) = V C*1 at time t0. Here the voltage VB1 at the switch B1
is given by:
Energies 2016, 9, 914 5 of 14

dVC1 (t) G G + G11 G G

= 11 VC0 (t) 1 VC1 (t) 11 VB1 (t) + 1 V1 (t) (2)
dt C1 C1 C1 C1
and V ( t ) = V at time t . Here the voltage V
with initial conditions VC0 (t0 ) = VC0 C1 0 C1 0 B1 at the switch
B1 is given by:
(
VC0 (t) VC1 (t) i f S (t) = true
VB1 (t) = (3)
0 i f S (t) = false
The switch is therefore simulated using the Boolean variable S which is equal to one if the switch
is open (e.g., during the night period when the TR is disconnected) and is equal to zero if the switch
is closed. The numerical solution of (1), (2) allows finding out the time profile of both the TR and U
temperatures, given the thermal power generation (which meets the aforementioned requirements of
the model).

2.2. Equivalent Electric Circuit Parameters Estimation

Once the topological layout of the network has been defined, it is necessary to define a procedure
to evaluate the five parameters proposed by the model: three resistances and two capacitances.
One possibility consists of considering the physical layout of the plant and proceeding with a
characterization of the elements that compose the system. Nevertheless, such a procedure is
computationally complex and affected by uncertainties on the phenomena to be considered. Moreover,
the model of the system varies in accordance to the specific thermal network, making difficult to
obtain a general procedure to be extended for a generic microgrid asset. To overcome these problems,
the methodology adopted in the present paper for parameters estimation relies on an optimization
procedure that aims at minimizing the difference between the temperature time profile calculated
with (1)(3) and the measured one. Under this assumption, the only a priori hypothesis is the topology
of the circuit, which, by the way, can be judged in an ex-post evaluation comparing the calculated and
the measured outputs. The aim of the present section is to present the proposed optimization approach.
As measured temperatures and thermal powers are supposed to be known on a uniform sampling
of the interval [t1 , t2 ] according to a step t and resulting in N = (t2 t1 ) /t + 1 samples, the whole
problem will be presented in a discrete form.
As far as the cost function is concerned, the 2 function has been chosen, in order to have the
possibility of subjecting the obtained trends to a statistical evaluation (i.e., the so called chi-squared
test [18] typically used to establish if a theoretical model is incompatible with a set of experimental
observations). With specific reference to the available measurements, the cost function for the problem
under consideration can be written as:
N meas ( t ) V ( t ))2
(VC0 meas ( t ) V ( t ))2
(VC1

i C0 i i C1 i
2calc = 2
+ 2
(4)
i =1 i,C0 i,C1

meas
being i,C0 and i,C1 the standard deviation of the measured temperature of the thermal ring VC0
meas , respectively.
and of the user VC1
The constraints of the problem are the discretized versions of (1) and (2), i.e.,
 
VC0 (ti+1 ) 1 G0 + G11 G11 1 G11 G0
t = VC0 (ti ) t C0 VC0 (ti ) + C0 VC1 ( ti ) + C0 I0 ( ti ) + C0 VB1 ( ti ) + C0 V0 ( ti ) (5)

and  
VC1 (ti+1 ) G 1 G + G11 G11 G
= 11 VC0 (ti ) + 1 VC1 (ti ) V (t ) + 1 V1 (ti ) (6)
t C1 t C1 C1 B1 i C1
The five unknown parameters must belong to reasonable ranges that can be estimated with
physical considerations (e.g., the initial estimation of capacitance C0 can be obtained by the water
specific heat multiplied by the overall water mass flowing in the thermal ring).
Energies 2016, 9, 914 6 of 14

Moreover, the defined optimization procedure does not require the parameters to be time invariant.
As a matter of fact, it seems to be reasonable that they may vary with time in accordance with the
specific usage conditions of the rooms (e.g., the number of occupants in a room or the people flowing
from one place to another have an impact on the efficiency and inertia of the system). This suggests the
possibility of dividing the optimization horizon into time slots, according to the habits of the district
fed by the considered thermal network.
Once the optimization problem has been set up, it is necessary to define an algorithm to identify
the optimal solution (parameters) to fit the measured data with the proposed model. This is not an easy
task, as the cost function (4) that has to be minimized is not a polynomial (the two capacitances appear
at the denominator after having inserted (5) and (6) into (4)). To address this problem, a stochastic
optimization algorithm has to be preferred in order to avoid possible local minima: the Ant Colony
Optimization algorithm (ACO) [19] encodes this characteristic in its nature. In the ACO algorithm,
two degrees of freedom are left to the users in order to increase the velocity of convergence and/or
emphasize its accuracy. However, to reduce the effect of an inappropriate choice, a different version of
the ACO algorithm has here been adopted: the Multivariate Ant Colony Algorithm for Continuous
Optimization (MACACO) [20]. MACACO is an extension of ACO based on a covariance matrix
instead of a simple variance vector (adopted in ACO).
The updating step for each ant is made such that the new distribution of the solutions points at the
promising region is supposedly indicated by the candidate solutions at the previous iteration. This goal
can be achieved recalculating this matrix using just 70% of the best solutions found at the present
iteration (the 70% value is defined on empirical basis). By doing that, the obtained distribution will go
on until the covariance tends to almost zero and is going to be centered at the best optima found.

2.3. Equivalent Electric Circuit Parameters Estimation

The typical thermal network characterization inserted in a microgrid EMS consists of providing
the optimization procedure with a constraint stating that the overall thermal power generation must be
at least equal to the thermal request (expressed in terms of thermal power) over the optimization time
horizon. Unfortunately, this implies to know in advance the time profile of the thermal power request,
which is typically predicted by means of the persistency method [21]. This produces a significant
amount of uncertainty affecting the overall EMS result.
One of the most interesting applications of the proposed model, for a cogenerative microgrid,
is the possibility of integration into an EMS in order to make it able to process the request in terms
of temperature profile, which seems much easier to forecast. It is intuitive that for a given load
temperature profile, the optimal solution from an economic point of view is given by achieving the
minimum amount of primary energy (current in the equivalent modeling) able to guarantee that the
room temperature is equal or greater than the desired one.
Recalling the meaning of the electric-thermal equivalency of circuital quantities, it is possible to
express the cost function in terms of the I0 current. In other words, the optimal solution corresponds to
the values of I0 (t) that minimize the overall thermal energy in a finite time horizon Nt , i.e.,

Nt
min I0 (ti ) t (7)
i =1

constrained by (5) and (6) and such that:

des
VC1 (ti ) VC1 ( ti ) i = 1, . . . , Nt (8)

des ( t ) being the desired value of the load temperature at time t .

VC1 i i
Energies 2016, 9, 914 7 of 15

two Combined Heat and Power Gas Turbines (CHP-GT) Capstone C65 model, able to generate
65 2016,
Energies kW (electric
9, 914 power) and 112 kW (thermal power); 7 of 14
one CHP-GT Capstone C30 able to generate 30 kW (electric power) and 55 kW (thermal power);
two gas boilers characterized by a rated thermal power of 500 kW per boiler.
3. Discussion
Figure 3 sketches the SPM map, where a green line groups the TR (composed by the network
connections, in red, and
3.1. Model Validation theSPM
in the thermal sources, in orange) and the set of blue lines denotes the U (the
different buildings). The thermal network is fed by the three cogeneration gas turbines and two gas
This section reports the results obtained by applying the proposed model to the Savona Campus
boilers. The pipes are installed underground and are properly insulated in order to minimize heat
SPM. A complete description of the SPM has been provided in [7,22], but, from a pure thermal point of
losses; the pipe diameters are in the range between DN80 and DN125. Figure 3 does not show the
view, it is worth recalling the SPMs main features:
pipes that distribute heat inside each building, where hot water is moved by on/off centrifugal pumps
heat
and twoisCombined
transferred to the
Heat andindoor
Powerair Gas byTurbines
means of radiatorsCapstone
(CHP-GT) or fan coils.
C65During
model,winter
able todays, the
generate
Campus 65 kWthermal load
(electric can and
power) be up112 to
kW900 kW ofpower);
(thermal peak request and the thermal energy yearly
consumption
one CHP-GT is about 1300 MWh.
Capstone C30 able to generate 30 kW (electric power) and 55 kW (thermal power);
The twoheating network
gas boilers is monitored
characterized in real
by a rated time by
thermal the SPM
power control
of 500 kW per room by means of on-field
boiler.
sensors that allow the following variables to be measured: the water mass flow rate, the supply and
returnFigure 3 sketches thethe
lines temperatures, SPM map, where
thermal power aoutput
green ofline groups
each the TR and
gas turbine (composed
boiler andby the
the outdoor
network
connections, At
temperature. in present,
red, andthe thebuildings
thermal (U) sources,
are notin equipped
orange) and the
with anysetsensor
of blue lines denotes
measuring the U
the indoor
conditions as well as the water temperature and flow rate, so that it is possible to evaluate only two
(the different buildings). The thermal network is fed by the three cogeneration gas turbines and the
gas boilers. The pipes are installed underground and are properly insulated
overall thermal load of the Campus but not the single building ones. In the future, each building willin order to minimize
heat
be losses; the
equipped with pipe diameters
sensors in order aretoindevelop
the range between
more DN80
accurate and DN125.
analyses Figure 3 does
and to highlight not show
criticalities of
the pipes that distribute heat inside each building, where hot water is moved
the thermal network. However, the rooms temperatures have been recorded in the data collection by on/off centrifugal
pumpsbyand
period heatofisportable
means transferredindoorto the indoor airmonitoring
temperature by means systems
of radiators or fan
(Elitech RC-4coils.
MiniDuring winter
Temperature
days, the Campus thermal load can be up to 900 kW of peak request
Data Logger by Jiangsu Jingchuang Electronics Co. Ltd., Xuzhou, China, 2015 [23]). and the thermal energy yearly
consumption is about 1300 MWh.

TheSPM
Figure3.3.The
Figure SPMlayout
layoutwith
withthe
thehighlight
highlightofofthe
thethermal
thermalring
ring(the
(theunion
unionof
ofthe
thethermal
thermalsources
sources
and the thermal network) and the user (the set of buildings involved).
and the thermal network) and the user (the set of buildings involved).

As
Thementioned, in order
heating network to attain in
is monitored thereal
goals
timeofbythe
thework here presented,
SPM control a proper
room by means set of
of on-field
experimental data had
sensors that allow to be collected,
the following on the
variables base
to be of whichthe
measured: identifying
water mass theflow
model
rate,parameters
the supplyand and
performing
return linesthe consistency the
temperatures, estimation.
thermal power output of each gas turbine and boiler and the outdoor
Furthermore,
temperature. as specified
At present, before, the
the buildings representation
(U) of the
are not equipped thermal
with network
any sensor has beenthe
measuring based on
indoor
the introduction
conditions of an
as well as equivalent unique end-user
the water temperature and flow U, despite
rate, so of theitlarge
that number
is possible toof actual physical
evaluate only the
loads.
overallAs a consequence,
thermal load of thethe proper
Campus butchoice
not theofsingle
the rooms in ones.
building whichIntothe carry on each
future, the temperature
building will
measurements
be equipped withhadsensors
to be identified.
in order toIndevelop
our case,moretheaccurate
logic was that ofand
analyses performing the criticalities
to highlight temperature of
survey withinnetwork.
the thermal rooms that, even ifthe
However, included
roomsintemperatures
buildings with non-perfectly
have been recorded superimposable thermal
in the data collection
period by means of portable indoor temperature monitoring systems (Elitech RC-4 Mini Temperature
Data Logger by Jiangsu Jingchuang Electronics Co. Ltd., Xuzhou, China, 2015 [23]).
Energies 2016, 9, 914 8 of 14

As mentioned, in order to attain the goals of the work here presented, a proper set of experimental
data had to be collected, on the base of which identifying the model parameters and performing the
consistency estimation.
Furthermore, as specified before, the representation of the thermal network has been based on the
introduction of an equivalent unique end-user U, despite of the large number of actual physical loads.
As a consequence, the proper choice of the rooms in which to carry on the temperature measurements
had to be identified. In our case, the logic was that of performing the temperature survey within rooms
that, even if included in buildings with non-perfectly superimposable thermal characteristics, could be
considered as homologous under the temperature requirements of the occupants (offices, apartments,
classrooms). In the choice of the rooms to be monitored, particular attention has been devoted in
evaluating how the human behaviors (windows opening, room thermostat adjustments, etc.) could
affect the thermal profiles within the rooms, trying to minimize these uncontrollable variables.
Operatively, along the survey period the indoor temperature, within the selected rooms, has been
acquired continuously and the obtained trends have been recorded weekly. By averaging the data
collected in the selected users, a set of day-by-day temperature trends (and the related standard
deviations), each of which assumed as representative of the typical user (i.e., room) served by the
thermal network, has been obtained.

3.2. Parameter Characterization

The parameters characterization was conducted during the winter season in the period between
January 2016 and March 2016. Measurement campaigns were carried out three days per week
(Monday and Friday have been avoided because the thermal behavior of the system is influenced in
those days by the fact that during weekend the heating system of the whole campus is switched off).
The SPM thermal behavior was divided into three different time frames (each one object of parameter
identification):

(a) Time Frame 1: from 10 p.m. to 6 a.m. when the thermal ring stops working and peoples presence
inside the buildings is extremely reduced;
(b) Time Frame 2: from 6 a.m. to 5 p.m. when the building pumping system works and a significant
number of people are present inside the buildings;
(c) Time Frame 3: from 5 p.m. to 10 p.m. when some loads are disconnected from the thermal
network and gradually fewer people are present.

As explained before, this suggests considering the circuit parameters as piecewise constant
functions throughout the day, i.e.,

Gi,1

if 6 a.m. t < 5 p.m.
Gi (t) = Gi,2 if 5 p.m. t < 10 p.m. i {0, 1, 2} (9)


G
i,3 if 10 p.m. t 5 a.m.

Ci,1

if 6 a.m. t < 5 p.m.
Ci (t) = Ci,2 if 5 p.m. t < 10 p.m. i {0, 1} (10)


C
i,3 if 10 p.m. t 5 a.m.
Moreover, condition S(t) becomes true if:

0 a.m. t 6 a.m. or 10 p.m. t 12 p.m., in order to encode condition (a);

VC0 (t) < VC1 (t), in order to encode the presence of the diode D1 ;
VC1 (t) > VC1,max (t), where VC1,max (t) is the user maximum temperature allowed at the instant t.

Once all the measurements were acquired, two thirds of the measurement data were used to train
the model (identifying the parameters) while the other third has been used to test the performances
 Energies 2016, 9, 914 9 of 14

of the parameters found. In more detail, data acquired on Tuesday and Thursday of each week were
used to choose the optimal values of the parameters, while Wednesdays data were used for the
verification. This choice was made in order to have uniformity in both the identification and validation
on the results. The tests have been performed taking the input measurements (ground temperature,
environmental temperature and thermal power from the heating ring), calculating the U temperature
using the model and then comparing the calculated results with the measured ones.
The optimal values of the circuit parameter (which minimize the cost function) are reported in
Table 2016,
Energies 1 (all9,the
914data used to find the parameter reported in Table 1 are available at Savona Campus
9 ofand
15
can be provided at request).
Table 1. Optimal Parameters Values.
Table 1. Optimal Parameters Values.
Parameter 6 a.m.5 p.m. 5 p.m.10 p.m. 10 p.m.6 a.m.
G0 Parameter
(kW/K) 62.58
a.m.5 p.m. 3.72 p.m.
5 p.m.10 100.40
p.m.6 a.m.
CG0 (kJ/K)
0 (kW/K)
0.93 10 5
2.58 2.26 3.72
10 5 1.73 0.40
105
G1 (kW/K)
C0 (kJ/K) 28.76
0.93 105 24.03
2.26 10 5 50.01
1.73 105
CG 1 (kW/K)
1 (kJ/K) 3.91 28.76
106 3.17 24.03
106 3.11 50.01
106
6 6 6
C 1 (kJ/K)
G11 (kW/K) 3.62 10
3.91 3.17
20.00 10 4.01 10
3.11
G11 (kW/K) 3.62 20.00 4.01

As one can see from Table 1, the results are sensibly different from frame to frame, highlighting
As one can
the importance of see from Table
defining 1, the
a suitable results are sensibly
differentiation different assets
of the operation from frame
of the to frame, highlighting
network. Figures 49
the importance
report the resultsofofdefining
the testsaperformed
suitable differentiation
on Wednesday, of the operation
20 January assets
2016. of the network.
In particular, Figures
Figures 4 and49
5
plot the user and thermal ring temperature time profiles during nighttime showing an excellent5
report the results of the tests performed on Wednesday, 20 January 2016. In particular, Figures 4 and
plot the user
agreement and thermal
between ring and
calculations temperature time profiles
measurements. Figuresduring
6 and nighttime
7 report the showing
comparisonan excellent
of the
agreement
user and thermalbetween
ring calculations
temperatureand measurements.
during Figuresfrom
the daytime period, 6 and 7 report
6 a.m. to 10the
p.m. comparison
Once again,ofthe
the
user and thermal ring temperature during the daytime period, from 6 a.m. to
estimated temperature is very close to the measured one with some deviations, especially in the early 10 p.m. Once again,
the estimated
morning in thetemperature
graph of theis ring
very temperature.
close to the measured one with
Nevertheless, the some
maximumdeviations,
error is especially in the
around 13%,
early morning
which is anywayinathe goodgraph of the ring temperature.
approximation. It should be Nevertheless,
observed that the maximum
for the error is around
EMS integration, 13%,
the most
which is anyway a good approximation. It should be observed that for the EMS integration,
important quantity to be identified is the user temperature, as this is the real user request. Finally, the most
important
Figures 8 and quantity
9 provideto be
theidentified is the user
model validation intemperature,
the last time as this is
frame, the 5real
from p.m.user request.
to 10 p.m. AsFinally,
one
Figures 8 and 9 provide the model validation in the last time frame, from 5 p.m.
can see, the thermal ring temperature is slightly over estimated but the relative error is always lower to 10 p.m. As one
can 10%.
than see, the thermal ring temperature is slightly over estimated but the relative error is always lower
than 10%.
User Temperature ( o C)

Usertemperature
Figure4.4.User
Figure temperature(V
(VC1C1 ) profile:measured
) profile: measured dataagainst
data againstmodel
modelresults
resultsfrom
from1010p.m.
p.m.toto6 6a.m.
a.m.
Energies 2016, 9, 914 10 of 15
Energies
Energies2016,
2016,9,9,914
914 10
10ofof15
14
Energies 2016, 9, 914 10 of 15

( C)
( o (C)C)
o o
Ring
Ring
Ring Temperature
Temperature
Temperature

Figure 5. Ring temperature (VC0) profile: measured data against model results from 10 p.m. to 6 a.m.
Figure 5. Ring temperature (VC0) profile: measured data against model results from 10 p.m. to 6 a.m.
Figure 5.
Figure Ring temperature
5. Ring temperature (V C0) )profile:
(VC0 profile:measured
measureddata
dataagainst
againstmodel
modelresults
resultsfrom
from10
10p.m.
p.m.toto66a.m.
a.m.
( o C)
Temperature
Temperature
Temperature
User
User
User ( o C)
( o C)

Figure 6. User temperature (VC1) profile: measured data against model results from 6 a.m. to 5 p.m.
Figure Usertemperature
6. User
Figure 6. temperature (V
(VC1C1
) )profile:
profile:measured
measureddata
dataagainst
againstmodel
modelresults
resultsfrom
from66a.m.
a.m.toto55p.m.
p.m.
Figure 6. User temperature (VC1) profile: measured data against model results from 6 a.m. to 5 p.m.
( o C)
Temperature
Temperature
Temperature
Ring
Ring
Ring ( o C)
( o C)

Figure 7. Ring
Figure 7. Ring temperature
temperature (V
(VC0 ) profile: measured data against model results from 6 a.m. to 5 p.m.
C0) profile: measured data against model results from 6 a.m. to 5 p.m.

Figure 7. Ring temperature (VC0) profile: measured data against model results from 6 a.m. to 5 p.m.
Figure 7. Ring temperature (VC0) profile: measured data against model results from 6 a.m. to 5 p.m.
Energies 2016, 9, 914 11 of 15

Energies 2016, 9, 914 11

11 of 15
14

( o C)( C)
o
User
User Temperature
Temperature

Figure 8. User temperature (VC1) profile: measured data against model results from 5 p.m. to 10 p.m.
Figure 8. User temperature (VC1 ) profile: measured data against model results from 5 p.m. to 10 p.m.
Figure 8. User temperature (VC1) profile: measured data against model results from 5 p.m. to 10 p.m.

70
70
( o C)( C)

65
o
Temperature

65
60
Temperature

60
55
55
RingRing

50 Measured temp.
50 Model temp.
Measured temp.
45
05:00 p.m. Model temp.10:00 p.m.
45
Time (h)
05:00 p.m. Time (h) 10:00 p.m.
Figure 9. Ring temperature (VC0 ) profile: measured data against model results from 5 p.m. to 10 p.m.
Figure 9. Ring temperature (VC0 ) profile: measured data against model results from 5 p.m. to 10 p.m.

Figure 9. Ring temperature (VC0) profile: measured data against model results from 5 p.m. to 10 p.m.
In order
In order to tohave
havean anoverall
overallevaluation
evaluation ofofthethe accuracy
accuracy of the
of the proposed
proposed model,
model, the the
value value of
of the
the chi-squared
chi-squared function
function has been
has been calculated
calculated for thefor shown
the shown
testingtesting
period period and compared
and compared to thetochi-
the
In order to have an overall evaluation of the accuracy of the proposed model, the value of the
chi-squared distribution at level for determining the goodness of fit [15].
squared distribution at level for determining the goodness of fit [15]. To determine the degrees ofTo determine the degrees of
chi-squared function has been calculated for the shown testing period and compared to the chi-
freedom (d.o.f.)
freedom (d.o.f.) ofofthe
thechi-squared
chi-squared distribution,
distribution,it is itnecessary to consider
is necessary the total
to consider thenumber of acquired
total number of
squared distribution at level for determining the goodness of fit [15]. To determine the degrees of
measurements
acquired (n) and subtracts
measurements (n) and the number
subtracts theofnumber
estimated of parameters (in our case(in
estimated parameters 5). our
Thecase
test statistic
5). The
freedom (d.o.f.) of the chi-squared distribution, it is necessary to consider the total number of
follows,
test approximately,
statistic a chi-squarea distribution
follows, approximately, with (n 5)
chi-square distribution degrees
with (n 5)ofdegrees
freedom. The moreThe
of freedom. the
acquired measurements (n) and subtracts the number of estimated parameters (in our case 5). The
chi-squared
more calculatedcalculated
the chi-squared is lower than the chi-squared
is lower distribution
than the chi-squared at level =at0.95,
distribution levelthe
=more
0.95,accurate
the more is
test statistic follows, approximately, a chi-square distribution with (n 5) degrees of freedom. The
the result.
accurate is Moreover,
the result. deciding
Moreover, whether
decidinga model
whether is in a accordance
model is in with the datawith
accordance is often summarized
the data is often
more the chi-squared calculated is lower than the chi-squared distribution at level = 0.95, the more
by the p-value.
summarized by According
the p-value.toAccording
[24], the more
to [24], p-value
thethe more theis close to 1isthe
p-value more
close to the
1 themodel
more is theaccurate.
model
accurate is the result. Moreover, deciding whether a model is in accordance with the data is often
Inaccurate.
is Table 2 theIn numerical
Table 2 thevalues of allvalues
numerical the introduced
of all theindicators
introduced areindicators
reported:are thereported:
fact that the
thereference
fact that
summarized by the p-value. According to [24], the more the p-value is close to 1 the more the model
chi-square
the reference value is always
chi-square greater
value than the
is always calculated
greater than the one calculated
and, in particular,
one and,that the p-valuethat
in particular, extremely
the p-
is accurate. In Table 2 the numerical values of all the introduced indicators are reported: the fact that
close to
value 1 indicates
extremely the high
close to 1 accuracy
indicatesofthe thehigh
model in describing
accuracy of thethe thermal
model in phenomena.
describing the thermal
the reference chi-square value is always greater than the calculated one and, in particular, that the p-
phenomena.
value extremely close to 1 indicates the high accuracy of the model in describing the thermal
Table 2. Statistics on the testing period.
phenomena.
Timeslot 2calc p-Value 2 (95%) dof
6 a.m.5 p.m. 14.8 ~1 39.8 56
5 p.m.10 p.m. 7.42 ~1 92.1 116
10 p.m.6 a.m. 9.00 ~1 68.3 89
 Timeslot p-Value (95%) dof
6 a.m.5 p.m. 14.8 ~1 39.8 56
5 p.m.10 p.m. 7.42 ~1 92.1 116
10 p.m.6 a.m. 9.00 ~1 68.3 89
Energies 2016, 9, 914 12 of 14

3.3. Validation of the EMS Integration

3.3. Validation of the
Finally, the EMS Integration
proposed model has been implemented in the SPM control system in order to
integrate it into
Finally, the the microgrid
proposed model EMS.
hasThe
beenexperiment
implemented hasinbeen performed
the SPM controlproviding
system in the EMS
order with the
to integrate
measured temperature and assuming it as the desired temperature; the EMS produced,
it into the microgrid EMS. The experiment has been performed providing the EMS with the measured as output,
the thermal power to be provided by the heating units in order to meet the temperature
temperature and assuming it as the desired temperature; the EMS produced, as output, the thermal request and
minimizing
power to be the overallby
provided energy over the
the heating time
units inhorizon.
order to The
meetpower request profile
the temperature is then
request andcompared
minimizing to
the real power generation recorded by the SPM Supervisory Control And Data Acquisition
the overall energy over the time horizon. The power request profile is then compared to the real (SCADA)
system, managed in
power generation accordance
recorded to SPM
by the the following
Supervisoryrules based And
Control on the ringAcquisition
Data temperature T:
(SCADA) system,
managed
if T < in
49accordance
C, then both to the following
boilers on based on the ring temperature T:
switchrules
if 49 C <T < 53 C, then only one boiler switches on
if T < 49 C, then both boilers switch on
if T >53 C, then both boilers switch off
if 49 C < T < 53 C, then only one boiler switches on
Figure
if T > 5310 C,
provides
then boththeboilers
comparison
switchofoff
the thermal power obtained by the EMS (black one) and
the actual thermal power profile (red dashed line). As can be noted, the optimal power profile is
smotherFigureand10 more
provides the comparison
constant, providingof the thermalequal
a saving powertoobtained
1400 kWh by the EMSthe
along (black one) and
analyzed 24 the
h,
actual thermal power
corresponding profileof
to the 12.7% (red
thedashed line). As
total thermal can berequested
energy noted, theinoptimal power
the whole dayprofile is smother
(integrating the
and more
thermal constant,
power providing
red curve a saving
of Figure equal
10 one to 1400
obtains kWh kWh).
11,050 along the
From analyzed 24 h, corresponding
an economical to
point of view,
the 12.7% of the total thermal energy requested in the whole day (integrating
considering a mean cost of the thermal energy of 0.096 /kWh (which is the case of the SPM), it isthe thermal power red
curve of Figure
possible 10 onethe
to calculate obtains
daily 11,050
savingskWh). From
as 134 an economical
against pointofofthe
a total cost view, considering ascenario
non-optimized mean cost
of
of the thermal energy of
1061. Finally, Figure 11 shows: 0.096 /kWh (which is the case of the SPM), it is possible to calculate the daily
savings as 134 against a total cost of the non-optimized scenario of 1061. Finally, Figure 11 shows:
i. the measured temperature (red dotted line);
ii.
i. the
the temperature calculated (red
measured temperature feeding theline);
dotted model (1)(3) with the optimal thermal power profile
(black solid line);
ii. the temperature calculated feeding the model (1)(3) with the optimal thermal power profile
iii. the temperature
(black solid line);calculated feeding the model (1)(3) with the measured thermal power profile
(blue dash-dotted line).
iii. the temperature calculated feeding the model (1)(3) with the measured thermal power profile
(blue dash-dotted line).
Thermal Power (kW)

Figure
Figure 10.
10. Power
Power daily
daily profile:
profile: optimal
optimal profile
profile versus real profile.

This allows us to draw the following conclusions: (a) the comparison between (i) and (ii) shows
that the optimal solution is able to reproduce the desired temperature; (b) the good superposition
between (i) and (iii) validates once again the effectiveness of the equivalent circuit, as, when fed by the
real power, it is able to predict the correct time profile of the temperature; finally, (c) the comparative
analysis of (ii) and (iii) states that there can be many different power injections that produce very
similar temperature profiles, thus supporting the need to determine the most convenient one.
 Energies 2016, 9, 914 13 of 15
Energies 2016, 9, 914 13 of 14

User Temperature ( o C)

Figure
Figure 11.11. Roomtemperature
Room temperaturecomparison
comparison (measured
(measured against
againstcalculated
calculatedapplying
applyingthe model
the fedfed
model with
with
the optimal and the measured power profile).
the optimal and the measured power profile).

This allows us to draw the following conclusions: (a) the comparison between (i) and (ii) shows
4. Conclusions
that the optimal solution is able to reproduce the desired temperature; (b) the good superposition
The aim
between (i)of this
and contribution
(iii) validates once has again
been tothepropose a simple
effectiveness but equivalent
of the general model of aas,
circuit, thermal
when fed network
by
bythe
meansrealofpower,
an equivalent
it is ableelectric circuit.
to predict theThe aim time
correct of theprofile
presentofmodel is to be sufficiently
the temperature; finally, (c)accurate
the
butcomparative
also simpleanalysis
in orderofto(ii) beand
integrated into
(iii) states cogenerative
that there can bemicrogrid EMSs.power
many different The main task of
injections this
that
model is that
produce of linking
very similarthe thermal power
temperature request
profiles, thus of supporting
the load (usually provided
the need as an input
to determine thethe EMS)
most
to convenient one.
the user temperature (a more meaningful desired parameter for the user). The model proposed
is based on a linear equivalent electric circuit whose main parameters can be determined through a
4. Conclusions
measurement campaign applying a parameter identification optimal procedure. The proposed model
has been tested
The aim in of the
thisSavona Campus
contribution hasSmart
been Polygeneration
to propose a simple Microgrid thermalmodel
but general network,of a in order to
thermal
show the actual
network applicability
by means and robustness
of an equivalent of the procedure.
electric circuit. The aim of theThepresent
values model
of the parameters involved
is to be sufficiently
have been computed
accurate but also simpleby means of antooptimization
in order be integratedprocedure that, in practice,
into cogenerative microgridminimizes
EMSs. Thethe difference
main task
between
of this the
modeltemperature profilesthe
is that of linking of the thermal
thermal circuit
power and of the user
request load as calculated
(usually with as
provided theanproposed
input
the EMS)
model and the to measured
the user temperature (a more numerical
ones. The obtained meaningfulresults
desired parameter
show that thefor the user).
model The model
describes in a very
proposed is based on a linear equivalent electric circuit whose main parameters
accurate way the set of measured temperatures and, at the same time, can be easily used as a procedure can be determined
through a schedule
to optimally measurement campaign
the thermal power applying a parameter
production leading,identification
to significantoptimal
thermalprocedure.
energy savings. The
proposed model has been tested in the Savona Campus Smart Polygeneration Microgrid thermal
Author in order toRenato
Contributions:
network, show the Procopio, Federico Delfino
actual applicability andand Alessandro
robustness Nilberto
of the developed
procedure. the equivalent
The values of the
electric modeling
parameters of the thermal
involved have beennetwork; Andreaby
computed Bonfiglio
means of andanMassimo Brignone
optimization performed
procedure theinexperimental
that, practice,
validation and the EMS integration; Andrea Bonfiglio wrote the paper.
minimizes the difference between the temperature profiles of the thermal circuit and of the user as
Conflicts of Interest:
calculated with the The authors declare
proposed model no andconflict of interest.ones. The obtained numerical results show
the measured
that the model describes in a very accurate way the set of measured temperatures and, at the same
References
time, can be easily used as a procedure to optimally schedule the thermal power production leading,
1. to Palma-Behnke,
significant thermal energy savings.
R.; Benavides, C.; Lanas, F.; Severino, B.; Reyes, L.; Llanos, J.; Sez, D. A microgrid energy
management system based on the rolling horizon strategy. IEEE Trans. Smart Grids 2013, 4, 9961006.
Author Contributions: Renato Procopio, Federico Delfino and Alessandro Nilberto developed the equivalent
[CrossRef]
electric modeling of the thermal network; Andrea Bonfiglio and Massimo Brignone performed the experimental
2. Hatziargyriou, N. MicrogridsLarge Scale Integration of Micro-Generation to Low Voltage Grids; Paper C6-309 in
validation and the EMS integration; Andrea Bonfiglio wrote the paper.
Cigr Biennial Section; Cigr: Paris, France, 2006.
3. Conflicts of Interest:
Vasquez, The authors
J.C.; Guerrero, J.M.;declare
Miret, no
J.; conflict of M.;
Castilla, interest.
de Vicua, L.G. Hierarchical control of intelligent
microgrids. IEEE Ind. Electron. Mag. 2010, 4, 2329. [CrossRef]
References
4. Bonfiglio, A.; Brignone, M.; Delfino, F.; Invernizzi, M.; Pampararo, F.; Procopio, R. A technique for the
optimal control and operation of grid-connected photovoltaic production units. In Proceedings of the 2012
47th International Universities Power Engineering Conference (UPEC), London, UK, 47 September 2012;
pp. 16.
Energies 2016, 9, 914 14 of 14

5. Nikkhajoei, H.; Iravani, R. Steady-state model and power flow analysis of electronically-coupled distributed
resource units. IEEE Trans. Power Deliv. 2007, 22, 721728. [CrossRef]
6. Palensky, P.; Dietrich, D. Demand side management: Demand response, intelligent energy systems and
smart loads. IEEE Trans. Ind. Inf. 2011, 7, 381388. [CrossRef]
7. Bonfiglio, A.; Barillari, L.; Bracco, S.; Brignone, M.; Delfino, F.; Pampararo, F.; Procopio, R.; Robba, M.;
Rossi, M. An Optimization Algorithm for the Operation Planning of the University of Genoa Smart
Polygeneration Microgrid. In Proceedings of the IREP 2013 Symposium-Bulk Power System Dynamics and
ControlIX, Rethymnon, Greece, 2530 August 2013.
8. Chen, Q.; Fu, R.; Xu, Y. Electrical circuit analogy for heat transfer analysis and optimization in heat exchanger
networks. Appl. Energy 2015, 139, 8192. [CrossRef]
9. Kmpf, J.H.; Robinson, D. A simplified thermal model to support analysis of urban resources flows.
Energy Build. 2007, 39, 445453. [CrossRef]
10. Fraisse, G.; Viardot, C.; Lafabrie, O.; Achard, G. Development of a simplified and accurate building model
based on electrical analogy. Energy Build. 2002, 34, 10171031. [CrossRef]
11. Aringhieri, R.; Malucelli, F. Optimal operations management and network planning of a district heating
system with a combined heat and power plant. Ann. Oper. Res. 2003, 120, 173199. [CrossRef]
12. Rezaie, B.; Rosen, M.A. District heating and cooling: Review of technology and potential enhancements.
Appl. Energy 2012, 93, 210. [CrossRef]
13. Larsen, H.V.; Palsson, H.; Bohn, B.; Ravn, H.F. Aggregated dynamic simulation model of district heating
networks. Energy Convers. Manag. 2002, 43, 9951019. [CrossRef]
14. Kuosa, M.; Kontu, K.; Makila, T.; Lampinen, M.; Lahdelma, R. Static study of traditional and ring networks
and the use of mass flow control in district heating applications. Appl. Ther. Eng. 2013, 54, 450459.
[CrossRef]
15. Haikarainen, C.; Pettersson, F.; Saxen, H. An MILP model for distributed energy system optimization.
Chem. Eng. Trans. 2013, 35, 295300.
16. Bracco, S.; Dentici, G.; Siri, S. Economic and environmental optimization model for the design and the
operation of a combined heat and power distributed generation system in an urban area. Energy 2013, 55,
10141024. [CrossRef]
17. Wakui, T.; Kinoshita, T.; Yokoyama, R. A mixed-integer linear programming approach for cogeneration-based
residential energy supply networks with power and heat interchanges. Energy 2014, 68, 2946. [CrossRef]
18. Banks, J.; Carson, J.S.; Nelson, B.L.; Nicol, D.M. Discrete-Event System Simulation, 5th ed.; Pearson:
Upper Saddle River, NJ, USA, 2010.
19. Dorigo, M.; Gambardella, L.M. Ant colony system: A cooperative learning approach to the travelling
salesman problem. IEEE Trans. Evol. Comput. 1997, 1, 5366. [CrossRef]
20. De Franca, F.O.; Coelho, G.P.; Von Zuben, F.J.; Attux, R.R. Multivariate Ant Colony Optimization in
Continuous Search Spaces. In Proceedings of the 10th Annual Conference on Genetic and Evolutionary
Computation, Atlanta, GA, USA, 1216 July 2008; ACM: New York, NY, USA, 2008; pp. 916.
21. Beyer, H.; Martnez, J.; Suri, M.; Torres, J.; Lorenz, E.; Mller, S.; Hoyer-Klick, C.; Ineichen, P. Report on
Benchmarking of Radiation Products. Sixth Framework Programme MESOR, Management and Exploitation
of Solar Resource Knowledge. 2009. Available online: http://www.mesor.org/docs/MESoR_Benchmarking_
of_radiation_products.pdf (accessed on 3 October 2016).
22. Barillari, L.; Bonfiglio, A.; Delfino, F.; Pampararo, F.; Procopio, R.; Rossi, M. The Smart Polygeneration
Microgrid Test-Bed Facility of Genoa University. In Proceedings of the 2012 47th International Universities
Power Engineering Conference (UPEC), London, UK, 47 September 2012; pp. 16.
23. Temperature Data Logger Feature. Available online: http://www.elitech.uk.com/news/Elitech_RC_4_
temperature_data_logger_usb_data_logger_67.html. (accessed on 3 October 2016).
24. Beaujean, F.; Caldwell, A.; Kollr, D.; Krninger, K. P-values for model evaluation. Phys. Rev. D 2011,
83, 012004. [CrossRef]

2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC-BY) license (http://creativecommons.org/licenses/by/4.0/).