You are on page 1of 18

Energy Conversion and Management: X 17 (2023) 100339

Contents lists available at ScienceDirect

Energy Conversion and Management: X


journal homepage: www.sciencedirect.com/journal/energy-conversion-and-management-x

A hierarchical Building Management System for temperature’s optimal


control and electric vehicles’ integration
Giovanni Bianco a, Federico Delfino a, Giulio Ferro b, *, Michela Robba b, Mansueto Rossi a
a
Department of Naval, Electrical, Electronic and Telecommunication Engineering, University of Genova, Genova 16145, Italy
b
Department of Informatics, Bioengineering, Robotics and Systems Engineering, University of Genoa, Genoa, Italy

A R T I C L E I N F O A B S T R A C T

Keywords: A bi-level building management system for optimal temperature control and scheduling of electric vehicles in a
Building Management System smart building is proposed. The architecture allows considering automated decisions, both for operational
Optimization management and real-time control, minimizing costs and the dissatisfaction of different requirements (electric
Temperature control
vehicles’ charging, demand response). A particular feature of the building management system is that it includes
Electric vehicles charging
a model for electric vehicles’ charging (both for vehicles-to-grid and classical electric vehicles) that can consider
three-phases unbalanced systems and the inputs required by the Charging Stations, i.e., the set point for the
current for each plugin use. Moreover, a distributed approach for the optimal control of fan coils is proposed that
allows adjusting the temperature in each room. The developed building management system has been tested at
the Savona Campus in a building characterized by a photovoltaic field, a geothermal heat pump, EVs charging
infrastructure, thermal storage, and a fan coil circuit. The tests carried out at the Campus show that the proposed
algorithm allows a reduction of the daily costs of about 20% with respect to a simple heuristic, without
compromising the fulfillment of the charging requests of the electric vehicles. The savings can be increased till
close to 35% if a demand response is also exploited.

Introduction a potential energy saving of about 30 % [1]. The EMSs for the case of
buildings are integrated into Building Management Systems (BMSs) or
The attention towards sustainable use of resources has necessitated even Building Automation Systems (BASs). A BMS is a computer-based
changing how energy systems are managed. In particular, the digitali­ control system installed in a building that controls and monitors the
zation of energy information, IoT (Internet of Things) sensors, and building’s mechanical and electrical equipment such as ventilation,
remote control, together with the need to make real-time decisions lighting, power systems, and air-conditioning (HVAC). Optimization
automatically, have led to the development of new Energy Management models give the intelligence within a BMS and control algorithms that
Systems (EMSs) for microgrids, buildings, distribution grids, and are integrated into the overall platform, need information related to
aggregators. These tools aim to improve energy efficiency, minimize energy demand and renewables forecasting, and provide the commands
costs and emissions, satisfy the requests of the grid operators, and satisfy to equipment in the field. Generally, BMSs require models (to represent
the needs of recharging electric vehicles (EVs). In this paper, attention is the thermal and electrical behavior of the building) that are fast enough
focused on buildings, which in Europe are responsible for to be embedded in optimization models (that generally work in the short
approximately-two-thirds of the total primary energy consumption with term and under a Model Predictive Control (MPC) scheme). Indeed, the

Abbreviations: BMS, Building Management System; EV, Electric Vehicle; V2G, Vehicle-to-grid (i.e., charging stations able to provide power to the electrical grid);
V1G, Vehicle-one-grid (i.e., charging stations that can take only power from the electrical grid and here characterized by multi-socket charging stations); CS,
Charging station; EMS, Energy Management System; BAS, Building Automation System; HVAC, Heating, ventilation and air conditioning; MPC, Model Predictive
Control; PV, photovoltaics; d-TC, distributed temperature controller; CSC, charging system controller; dADMM, distributed alternating direction method of multi­
pliers; AC, alternating current; DC, direct current; H-Lvl, high level; L-Lvl, low level; SOC, State of charge; BSP, Balance Service Provider; COP, Coefficient of per­
formance; RC, Resistor-Capacitor; POD, point of delivery.
* Corresponding author.
E-mail addresses: giovanni.bianco@edu.unige.it (G. Bianco), federico.delfino@unige.it (F. Delfino), giulio.ferro@unige.it (G. Ferro), michela.robba@unige.it
(M. Robba), mansuseto.rossi@unige.it (M. Rossi).

https://doi.org/10.1016/j.ecmx.2022.100339
Received 24 June 2022; Received in revised form 28 November 2022; Accepted 30 November 2022
Available online 5 December 2022
2590-1745/© 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

Fig. 1. The bi-level control system architecture.

thermal behavior of buildings can be represented by complex mathe­ algorithms, i.e. the Charging Stations (CSs) Charging System Controller
matical models and tools [2]; however, system models’ approximations (CSC) and the distributed Temperature Controller (d-TC). It is important
for BMSs under an MPC scheme are widely accepted in literature [3]. In to note that the high level provides set points for a time interval of 1 h
[4] MPC is applied to a domestic microgrid, while in [5,6], and [7] to (and over an optimization horizon of 24 h), which are tracked by the
buildings. In [5] MPC for the operation of building cooling systems with low-level that works over a time interval of 5 min and an optimization
thermal energy storage the focus is buildings equipped with a water tank horizon of one hour.
used for actively storing cold water produced by a series of chillers. In In the literature discussion, several articles are present concerning
[6], a survey highlights the relevance of MPC for energy management in control techniques for the power management in CSs, the scheduling of
buildings (because of its ability to consider constraints, prediction of EVs, and their integration in EMSs and BMSs. In the following, a dis­
disturbances, and multiple conflicting objectives, such as indoor thermal cussion and a comparison with the proposed models and architecture
comfort and building energy demand). In [7] instead building energy have been reported.
simulation is conducted in accordance with which energy consumption Power management in CSs consists of control algorithms and rules
could be predicted and a MPC framework in order to minimize the that allow power modulation among different EVs and CSs. Generally,
modelling errors and prediction. Home energy management systems algorithms are embedded in CSs and modulate power in real-time when
could be also used to minimize the home’s daily electricity cost by a vehicle is present and starts charging, guaranteeing the respect of
adopting peak shaving and load shifting strategies [8]. Indeed, different maximum power that can be taken from the external grid. This is the
kinds of optimization problems occur in microgrids and buildings: a) typical situation of a charging park where there is no knowledge about
planning problems (i.e., those related to long-term decisions such as the private EVs that arrive at the charging station; this is the case considered
choice of the kind, size, and location of technologies such as charging in the low level (CSC). More advanced controllers can be developed
stations over a territory [9], transformers [10], or energy storage system when arrivals, due dates, and state of charge are known and are
[11]); b) operational management decision problems (i.e., when all generally included in BMSs and Energy Management Systems (EMSs),
plants and storage systems are already installed and the matter is how to and the charging station controller should communicate with the overall
schedule plants and components during the day); c) real-time operation platform of the building or microgrid (that corresponds to the high level
(i.e., when time scales are faster, forecasted demands and renewables in the present work). An example of a power management algorithm is
could differ from the expected ones, and some planned schedules cannot presented in [12], and it is applied, as a test case, to an actual car park
be changed). located in the UK and equipped with EVs fast charging stations. The
In the present paper, a new BMS is presented that integrates EVs main differences with the approach proposed in the present paper for the
charging considering both classical and Vehicle to Grid (V2G) modes), low-level are that the two strategies presented in [12] are heuristic ap­
HVAC, solar and geothermal renewables. The BMS is characterized by proaches and that they rely on data that are hardly available in practical
different optimization models for day-ahead, intra-day, and real-time situations, namely the power demand and State of Charge (SOC) char­
operations. The BMS is in operation at the Smart Energy Building acteristic of the connected vehicle. The possibility of single-phase
(SEB) at the Savona Campus (University of Genova), a net-zero energy chargers is not accounted and setpoints are expressed in terms of
building with a surface of 750 m2, twenty-two rooms, a PV plant, a power rather than currents. The same considerations apply for [13], in
dedicated EV charging infrastructure, and a HVAC system fed by a which a commercial parking lot is considered, and a two-stage proced­
geothermal heat pump. In particular, the proposed BMS (see Fig. 1) is ure is proposed to minimize the park owner costs and to allow demand
characterized by a bi-level hierarchical architecture that includes high response. Approximate dynamic programming is used for long-term
and low-level controllers, which are differentiated by different time estimation, while a hybrid Big Bang Big Crunch algorithm is devel­
scales under an MPC approach. oped for the short-term optimization. In [14], the authors deal with a use
The high level is devoted to operational management, and the opti­ case concerning a hotel that exploits EVs as electric storage to help
mization model is used under a receding horizon control scheme. Spe­ reduce the hotel energy costs. Two optimization algorithms are pro­
cifically, firstly building load and production from renewables are posed (i.e., for optimal sizing and daily management) but no mention of
forecasted over the optimization horizon, and the temperature is the charging infrastructure’s actual topology and the considered
measured. Then, the optimization problem is run, but only the optimal charging stations (single-phase or three-phase) is made. In [15], a model
schedule for the first-time interval is retained. The main purpose of the for the sizing of resources in a microgrid is modified to account for the
high-level is to define the operation setpoints for the low-level possibility of hosting a recharging infrastructure for electric vehicles. An

2
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

Fig. 2. The high level decision problem.

office building is used as a test case and the building’s BMS controls the batteries. In [23] a control strategy to manage a hybrid energy storage
available resources to minimize costs. Differently from the present work, system (super capacitor and traditional energy storage). The system is
the considered infrastructure is single-phase, and the problem of sharing implemented to provide the sudden high power request due to the
the limited feeder capacity is not dealt with in detail. Authors in [16] beginning of one or more EVs charging sessions and avoid battery
propose an “energy management strategy” for a charging station suit­ degradation. In [24], an iterative process is proposed to solve a non-
able for a residential installation with photovoltaic and an energy stor­ cooperative Stackelberg game to define the optimal routes, where to
age system. The strategy aims to provide an instantaneous power charge the EVs, and the optimal scheduling of the charging station.
allocation to each electric vehicle connected to the station, considering Regarding distributed control, in smart buildings some interesting
its initial state of charge, the desired final state of charge, the expected contributions can be found in literature. In [25] the authors define a
departure time, and the rated capacity of the vehicle battery. However, distributed real-time BMS to control the HVAC system. Indeed, the
differently from the present work, the grid is in DC and no phases are temperature and energy loads, both electrical and thermal, are not
considered. In [17], the authors propose a control scheme specifically considered differently from the presented work that employs a distrib­
designed to manage electric vehicle charging infrastructures (i.e., smart uted MPC BMS. The authors of [26] define a bi-level distributed BMS
parking) for peak shaving and demand response. The system model is that controls the HVAC system but do not consider the whole thermal
simple and not characterized by controllable loads such as electric heat plant and the building’s electrical loads or production units (PV units,
pumps and renewables, and uncertainties are considered through a two- EV charging stations, and heat pumps).
stage stochastic optimization. In [18], the authors present a BMS to Like in the proposed approach, in [27] the centralized optimization
supply energy to a microgrid in which loads are ranked by different problem for home energy management is decomposed into a two-level
priority levels and can be unsatisfied, while, in the present work, loads optimization problem. In [28] a framework for minimizing energy
are optimally scheduled and modulated to avoid this possibility. In [19], costs through adaptive energy scheduling in smart buildings is proposed
an EMS has been developed to perform demand response and manage an and solved through a token bucket algorithm. In contrast to [27] and
energy system made up of distributed generation units, an energy stor­ [28], the proposed approach is more precise from an electrical point of
age, and a bidirectional EVs charging infrastructure. Differently from the view (three-phase modeling is adopted), EVs are not considered, and the
present work, the several kinds of EVs charging stations (single-phase, distributed algorithm is not heuristic, but the proposed algorithm leads
three-phase, DC) and unbalanced loads are not considered. In [20], the to the optimal solution.
authors develop a controller (both for day-ahead and real-time man­ The authors of [29] and [30] focus their analysis from a more
agement) to optimally exploit the system flexibility, buy the energy computational point of view. In [29] distributed scheme for structured
during the day-ahead market, and fulfill the gap between the day-ahead multi-agent decision-making problems over a time-varying communi­
scheduling and real-time demand. In [21], the authors develop a multi- cation network, with the specific case study of smart buildings in a
objective optimization problem that works under an MPC control cooling network. Instead, [39] investigates a distributed model predic­
scheme for an energy system connected to the main grid and charac­ tive control framework for multi-zone building control with various
terized by different kinds of CSs and EVs, a hydrogen feed fuel cell, an EV applications.
charging infrastructure, a PV field and wind turbine generation units. Although the theoretical dissertation of [29] is very interesting and
The purpose of this controller [21] is to maximize the charge of the EVs precise, the range of applicability is limited due to specific mathematical
by using energy from renewable sources. As in the BMS proposed in this models and network requirements. Instead, in the proposed approach,
paper, the MPC-based controller of [21] can dynamically change the the real applicability is shown through real case study testing and
setpoints of the system to update the control strategy at each time step. detailed modeling of the components. Finally, as regards [30], only the
However, the controller does not consider the three-phase system, and thermal features of the building are considered without mentioning the
thus the unbalanced loads are not considered. Similar considerations electrical part. This leads to inaccurate modeling of the system,
related to the fact that the system model does not include phases, a furthermore, the building control is performed by dividing the system
bilevel optimization, and a distributed algorithm are reported in [22], into zones, whereas, in the proposed approach, each room has its
[23] and [24]. In [22], a real-time controller for charging infrastructure controller leading to a more capillary control.
is presented. A mixed-integer-quadratic decision problem is formalized In summary, the main contributions of the paper, with respect to the
(and solved by a heuristic) to manage the flexibility of charging stations discussed literature, are:
using EVs as deferrable loads and preserving the state of health of

3
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

• T is the time horizon of the high-level algorithm (with t ∈ T);


• T dr is the demand response time interval (with t ∈ Tdr and Tdr ⊂T);
• T bms is the time horizon of the d-TC algorithm (with t ∈ T bms );
• J is the set of all the EVs (with j ∈ J);
• T rc is the time interval during which the reference temperature is
required within the building (with t ∈ T rc and T rc ⊂T).

Forecasted data

The high-level optimization problem requires in input the fore­


casting of PV active power, i.e.PPV
p,t , which is calculated for each phase
and for each time interval based on meteorological and electrical models
Fig. 3. The building’s electrical sub-system, connected to the main grid by a and considering that each phase generates one-third of the total power.
single Point Of Delivery (POD). The imaginary (Ip,tPV,Im PV,Re
) and real (Ip,t ) parts of the PV current are
calculated by solving the linear equation system expressed in (1)-(2),
• A new bi-level BMS for temperature control and EVs’ scheduling that considering that, in a single bus bar approximation, VpPV,Re and VpPV,Im are
is applied to a real case study and in operation at the Savona Campus. the known real and imaginary parts of the PV voltage, and that reactive
The architecture allows considering automated decisions, both for
power (QPV
p ) is assumed to be zero for the PV plant. That is
operational management and real-time control, minimizing costs
and the dissatisfaction of different requirements (EVs’ charging, PPV PV,Re PV,Re
p,t = Vp Ip,t + VpPV,Im Ip,t
PV,Im
∀t ∈ T ∀p ∈ P (1)
demand response);
• A model for EVs’ charging (both for V2G and classical modes) that QPV PV,Im PV,Re
− VpPV,Re Ip,t
PV,Im
(2)
p = Vp Ip,t ∀t ∈ T ∀p ∈ P
considers multiphase settings and the technological requirements of
the CSs (i.e., input in current). This aspect is particularly important Similar equations can be written for the building’s electrical load,
because it allows considering different kinds of CSs and EVs; where active power (PBp,t , non-necessarily equals on each phase) and
• A distributed approach for optimal control of fan coils (d-TC) in voltages (VpB,Im ,VpB,Re ) are a priori known, and the reactive power (QBp,t ) is
which information is exchanged only between adjacent rooms. assumed equal to zero.

The remaining part of the article is organized as follows: Section II The power balance
describes the decision problems for the high-level and the low-level.
Results obtained by applying the proposed algorithms to an actual The electrical system is modeled as a single bus bar, and the power
test-case are reported and discussed in Section III. Finally, conclusions balance equation is considered for each phase p ∈ P, both for the active
are drawn in Section IV. Appendix I reports the list of abbreviations used and reactive power and in the time interval (t, t + 1). In particular, the
in this paper, while in Appendices II and III are reported some features of
power balance includes: the power supplied by the PV plant (PPV PV
p,t , Qp,t ),
the higher level and lower level models, respectively. Finally, Appendix
the building demand (PBp,t , QBp,t ), the power (unrestricted in sign)
IV describes the models used for the comparison of the d-TC.
exchanged with the grid (PGp,t , QGp,t ), the dynamic load due to the HVAC
Methods system (PHVAC
p,t , QHVAC
p,t ), the power absorbed by the V1G charging station
(PEVSE−
p,t
V1G
, QEVSE−
p,t
V1G
) and the power (unrestricted in sign) exchanged
In this Section, the decision problems for the high-level and the low-
with the V2G charging station (PEVSE−
p,t
V2G
, QEVSE−
p,t
V2G
). That is,
level are outlined, detailing the mathematical models employed to
describe the various devices and the problems’ objective functions. PPV + PGp,t − PHVAC − PBp,t − PEVSE− V2G
− PEVSE− V1G
= 0 ∀t ∈ T ∀p ∈ P (3)
p,t p,t p,t p,t
A. The high level decision problem
The high level decision problem (Fig. 2) aims to define the sched­ QPV G HVAC
− QBp,t − QEVSE− V2G
− QEVSE− V1G
(4)
p,t + Qp,t − Qp,t = 0 ∀t ∈ T ∀p ∈ P
uling of production plants and storage systems, minimizing operational
p,t p,t

costs and the dissatisfaction of different requirements (EVs’ charging, where PPV B HVAC G
,Pp,t , PEVSE− V1G
p,t , Pp,t are forecasted inputs, while Pp,t p,t and
demand response), taking as inputs the in field measurements and
PEVSE− V2G
are control variables.
forecasts. The proposed BMS is designed to control the operation of a p,t

generic company equipped with its own EVs fleet. Indeed, the EVs’ The active and reactive power exchanged with the main grid (QGp,t ,
scheduling and the battery state of charge, both initial and final re­ PGp,t ) are given by the difference of power that is bought (QG,in G,in
p,t , Pp,t )
quirements, are known to the BMS. In the following, the system model is minus power that is sold (QG,out G,out
p,t , Pp,t ), as defined in (5)-(6):
firstly presented, and then the overall optimization problem is
formalized. PGp,t = PG,in G,out
p,t − Pp,t ∀t ∈ T ∀p ∈ P (5)
1) The building’s electrical and thermal sub-systems.
The electrical system (see Fig. 3) is characterized by: a PV plant, a QGp,t = QG,in G,out
p,t − Qp,t ∀t ∈ T ∀p ∈ P (6)
connection with the external grid, a geothermal heat pump that con­
sumes electrical power to produce thermal power, a charging station The power exchange with the main grid is bounded by employing:
with multiple sockets (called V1G, which has power management ca­ − PG,max ⩽PGp,t ⩽PG,max ∀t ∈ T ∀p ∈ P (7)
pabilities without V2G services), a non-deferrable electrical demand due
p p

to the building load, and a V2G charging station that can both absorb − QG,max ⩽QGp,t ⩽QG,max ∀t ∈ T ∀p ∈ P (8)
and inject power.
p p

The following notation is used:


where PG,max
p and QG,max
p are the maximum active and reactive power that
can be injected to or withdrawn from the main grid.
• P is the set of the different phases (with p ∈ P);
The higher-level optimization problem provides setpoints for the

4
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

real-time and the intra-day controllers of two different sub-modules of EV):


the BMS in terms of currents. For this reason, it is necessary to define the ∑
PEV− V2G,out
= PEVSE− V2G,out
∀t ∈ T (15)
relation between power (PGp,t , QGp,t ) and AC and voltage both for imagi­ t p,t
p∈P
G,Im G,Re
nary (Ip,t , VpG,Im ) and real (Ip,t , VpG,Re ) parts like in (1)-(2). Similar

considerations are valid for the geothermal heat pump that absorb PEV−
t
V2G,in
= PEVSE−
p,t
V2G,in
∀t ∈ T (16)
powers (PHVAC
p,t , QHVAC
p,t ) to satisfy the building’s thermal needs and thus, p∈P

when fed with given phase voltages (VpHVAC,Im , VpHVAC,Re ), absorbs thee The State Of Charge (SOC) of the EV battery is given by [31]:
HVAC,Im HVAC,Re
phase currents (Ip,t , Ip,t ). PEV− V2G,in
PEV− V2G,out
EV− V2G
Since the HVAC system is a balanced load, i.e., the three-phase SOCt+1 = SOCtEV− V2G
+ ηEV− V2G t t
Δt − EV− V2G Δt ∀t
CAPEV− V2G η CAPEV− V2G

currents it absorbs have the same r.m.s. value, the following


∈T
constraint is added:
⃒ ⃒ ⃒ ⃒ ⃒ ⃒ (17)
⃒ HVAC ⃒ ⃒ HVAC ⃒ ⃒ HVAC ⃒
⃒I1,t ⃒ = ⃒I2,t ⃒ = ⃒I3,t ⃒ ∀t ∈ T (9) The SOC of the EVs is bounded between a minimum (SOCEV− V2G,min
)
and a maximum (SOCEV− V2G,max ) value:
Equation introduces an additional nonlinearity, which can be avoi­
ded using the same variable for the currents considering the three SOCEV− V2G,min EV−
⩽SOCt+1 V2G
⩽SOCEV− V2G,max
∀t ∈ T (18)
[ 2 2 ]
different phase shifts φHVAC p = 0 − π π . Then, the active and The bound constraints of the PEV− V2G,in and PEV− V2G,out are expressed in
3 3
reactive power absorbed by the geothermal heat pump become: (19)-(21), where PEV− V2G,out,max and PEV− V2G,in,max are the maximum dis­
( ) ( HVAC ) charging and charging value allowed for the V2G charging station, a and
PHVAC
p,t = V HVAC,Re
p I HVAC,Re
t + V HVAC,Im
p I HVAC,Im
t Re eiφp + b are known coefficients to take into account the fact that PEV− t
V2G,in
( ) ( HVAC )
(10) decreases linearly with the SOC [34].
− − V HVAC,Re I HVAC,Im + V HVAC,Im I HVAC,Re Im eiφp
p t p t

∀t ∈ T ∀p ∈ P 0⩽PEV−
t
V2G,in
⩽PEV− V2G,in,max
∀t ∈ T (19)
( ) ( HVAC )
QHAVC = − V HVAC,Re I HVAC,Re + V HVAC,Im I HVAC,Im Im eiφp + 0⩽PEV−
t
V2G,in
⩽aSOCtEV− V2G
+b ∀t ∈ T (20)
p,t p t p t
( ) ( HVAC )
+ − VpHVAC,Re HVAC,Im
It + VpHVAC,Im HVAC,Re
It Re e iφp (11) 0⩽PEV− V2G,out
⩽PEV− V2G,out,max
∀t ∈ T (21)
t

∀t ∈ T ∀p ∈ P
The vehicle-one-grid charging infrastructure
The vehicle-to-grid charging infrastructure
The V1G CS is a “standard” AC charging station equipped with four
The V2G Electric Vehicles CS operates in AC three-phase from the sockets (two three-phase and two single-phase). The V1G CS operates in
building side and DC from the EV side. The power converter is AC, and the current charging profile can be modulated during the
embedded in the CS, and only one socket is present. The active power charging session. The three-phase sockets can charge both single and
PEVSE− V2G
, appearing in the power balance, is a function of the power three-phase electric vehicles.
p,t
Since the EV and the socket can be both three-phase or single-phase,
injected into the system (PEVSE− V2G,out
) and the power absorbed from the
p,t it is necessary to define the relationship between the overall currents
system (PEVSE−
p,t
V2G,in
). That is, EVSE−
absorbed by the CS on each phase (Ip,t V1G,Re
) and the currents
1 absorbed by each single EV. Therefore, the binary parameter δEV−
j,p
V1G
is
PEVSE− V2G
= PEVSE− V2G,in
− ηInv,V2G PEVSE− V2G,out
∀t ∈ T ∀p ∈ P th th
introduced; this parameter couples the j vehicle with the p phase:
p,t p,t p,t
η Inv,V2G

(12) δEV−
j,p
V1G
it is equal to 1 if the considered EV is connected to the pth phase,
if the EV and phase are not coupled, the parameter is equal to zero. This
PEVSE−
p,t
V2G,in
PEVSE−
p.t
V2G,out
=0 ∀t ∈ T ∀p ∈ P (13) parameter is an a priori known input. The equation defines the relation
EV− V1G,Re
for the real part of the current (Ij,p,t ), while the equation considers
where ηInv,V2G is the inverter efficiency, and the equation is necessary to
EV− V1G,Im
guarantee that the power flow is only in one direction (i.e., that in each the imaginary part (Ij,p,t ) considering the same phase shift φEV−
p
V1G

instant either PEVSE− V2G,out


or PEVSE− V2G,in
is greater than zero, but not of the HVAC system.
p,t p,t
both). EVSE− V1G,Re
∑[ EV− V1G,Re
( EV− V1G )
Ip,t = Ij,t Re eiφp
In the power exchanged between the system and the CS V2G is j∈J
bounded: ( EV− V1G ) ]
− Ij,tEV− V1G,Im
Im eiφp δEV−
j,p
V1G
∀t
− PEVSE− V2G,max
⩽PEVSE− V2G
⩽PEVSE− V2G,max
∀t ∈ T ∀p ∈ P (14)
(22)
p,t
∈ T ∀p ∈ P
where PEVSE− V2G,max is the maximum power that can be withdrawn or
∑[ EV− ( EV− V1G )
injected into the system. EVSE−
Ip,t V1G,Im
= Ij,t V1G,Re
Im eiφp
As previously done, it is necessary to relate PEVSE−
p,t
V2G
and reactive j∈J
( EV− V1G ) ]
power QEVSE−
p,t
V2G
with phase currents and voltages, both for imaginary + Ij,tEV− V1G,Im
Re eiφp δEV−
j,p
V1G
∀t
EVSE− V2G,Im EVSE− V2G,Re
(Ip,t VpEVSE− V2G,Im ) and real
, VpEVSE− V2G,Re ) parts.
(Ip,t , ∈ T ∀p ∈ P (23)
Since the inverter of the CS V2G is balanced, the same approach already
introduced for the HVAC in (10)-(11) is used. The relation between the CS V1G active (PEVSE−
p,t
V1G
), and reactive
The EV side of the CS operates in DC, and it is necessary to relate the (QEVSE−
p,t
V1G
) power, current, and voltage both for the imaginary
AC power variables (PEVSE−
p,t
V2G,in
, PEVSE−
p,t
V2G,out
) to the ones at the DC side EVSE− V1G,Im
(Ip,t , EVSE− V1G,Re
VpEVSE− V1G,Im ) and real parts (Ip,t , VpEVSE− V1G,Re ) are the
EV− V2G,in EV− V2G,out
(i.e., the charging (Pt ) and (Pt ) discharging powers of the same as in (1)-(2). The power, active and reactive, absorbed by the CS

5
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

Table 1
Electric vehicles’ features.
Electric Vehicles Charging type Plug Capacity [kWh] Arrival Time Departure Time SoC0 [%] SoCref [%]

EV1 DC V2G 100 1 am 12 am 20 90


EV2 P1/P2/P3 T2/1 80 1 am 9 am 20 80
EV3 P1/P2/P3 T2/1 80 10 am 8 pm 20 90
EV4 P1/P2/P3 T2/2 70 11 am 9 pm 20 85
EV5 P2 3A/1 80 3 pm 9 pm 20 90
EV6 P2 3A/1 60 7 am 1 pm 20 90
EV7 P3 3A/2 60 1 am 4 pm 20 85

V1G is bounded as follows: geothermal heat pump, thermal storage, and fan coil circuit. The ther­
mal behavior of the building is simulated through an equivalent RC
0⩽PEVSE− V1G
⩽PEVSE− V1G,max
∀t ∈ T; ∀p ∈ P (24)
p,t circuit model based on the analogy between Ohm’s and Fourier’s laws,
where resistors are used to model the thermal resistances, and capacitors
0⩽QEVSE−
p,t
V1G
⩽QEVSE− V1G,max
∀t ∈ T; ∀p ∈ P (25) are introduced to consider the accumulation of thermal energy within
the building [33]. The sub-system model is detailed in [2] and reported
where PEVSE− V1G,max and QEVSE− V1G,max are the allowed maximum in Appendix II.
charging active and reactive power of the CS V1G. Equation identifies 2) The Demand Response Operation
for each EV the relation between the active power, current, and voltage, In this work, it is assumed that the building is part of an aggregate of
both for imaginary (QEV−j,p,t
V1G
) and real parts (PEV−
j,p,t
V1G
), while in it is users that collaborate to provide Demand Response services to the local
defined the reactive power for the V1G sub-system, also in this case the Distribution System Operator (DSO) [32,35]. The Balance Service Pro­
reactive power is considered zero. Similarly to the approach detailed for vider (BSP) is the entity that coordinates and manages the aggregate as
the HVAC system in the equation (10)-(11), it is possible to write: mentioned above and that acts as an interface between the DSO and final
users. Indeed, during the day ahead, the BSP receives the baseline
PEV− V1G
=
(
j,p,t
) ( EV− V1G ) consumption and the flexibility assessments by each final user. It asso­
= VpEV− V1G,Re EV− V1G,Re
I j,t EV− V1G,Im EV− V1G,Im
+ Vp I j,t Re eiφp δEV−
j,p
V1G
+ ciates this information and provides the DSO an aggregated consump­
(
EV− V1G,Re EV− V1G,Im EV− V1G,Im EV− V1G,Re
) ( EV− V1G )
iφp EV− V1G
tion baseline of the consumer’s cluster related to their overall potential
− − Vp I j,t + Vp I j,t Im e δj,p flexibility. The day after, in case of the necessity of a DR service, the DSO
∀t ∈ T ∀p ∈ P ∀j ∈ J communicates to the BSP a variation of the baseline consumption of the
(26) entire cluster to the overall allowed flexibility. Finally, the BSP shares
among the aggregate participants the DSO requirement concerning the
QEV−
j,p,t
V1G
= baseline and the flexibility assessment of each user. This is made explicit
(
EV− V1G,Re EV− V1G,Re EV− V1G,Im EV− V1G,Im
) ( EV− V1G ) through a reference value provided by the BSP to the consumers in terms
= − Vp I j,t + Vp I j,t Im eiφp δEV− V1G
+ G,ref G,ref
of absorbed power, both for active and reactive power (Pt , Qt ).
j,p
( ) ( EV− V1G )
+ − V EV−
p
V1G,Re EV− V1G,Im
I j,t + V EV−
p
V1G,Im EV− V1G,Re
I j,t Re eiφp δEV−
j,p
V1G
The proposed BMS aims to manage the operation phase of generic
∀t ∈ T ∀p ∈ P ∀j ∈ J demand response events. The flexibility assessment phase is assumed to
(27) be already carried out by a day-ahead algorithm. Thus, two additional
constraints are considered during the demand response event (t ∈ T dr ):
The EV batteries are in DC, so it is necessary to transform the AC ∑
power component (PEV− j,p,t
V1G
) into DC power (PEV−
j,t
V1G,DC
). The three-phase PG,ref
t ⩽ PGp,t ⩽PG,ref
t (1 + θ) ∀t ∈ T dr (33)
power components are summed to define the DC power in:
p∈P

∑ ∑
PEV−
j,t
V1G,DC
= PEV−
j,p,t
V1G
∀t ∈ T ∀j ∈ J (28) QG,ref
t ⩽ QGp,t ⩽QG,ref
t (1 + θ) ∀t ∈ T dr (34)
p∈P p∈P

The EV state of charge is represented through the dynamic state As defined in the above constraints and, the absorbed power limits
equation: vary according to the reference values in this way, the actual power
absorbed by the network is bound to be equal to the reference value.
EV− V1G EV− V1G
PEV−
j,t
V1G,DC
Moreover, to guarantee the feasibility of the operations, the constraints
SOCj,t+1 = SOCj,t + ηEV− V1G
Δt ∀t ∈ T ∀j ∈ J
that relate the actual absorbed power to the reference value can be
j EV− V1G
CAPj
(29) violated according to θ, a percentage value defined at the user’s
discretion.
where ηEV−
j
V1G
is the efficiency of the jth EV and CAPEV−
j
V1G
is the 3) The overall optimization problem
nominal capacity of the battery. The BMS is defined for a generic company owns an EV fleet. It can be
Similarly to the V2G CS, the following relations hold: assumed that the presence of electric vehicles at the charging stations is
a priori known and that the SOC value at the arrival of each vehicle is
SOCjEV− V1G,min EV−
⩽SOCj,t V1G
⩽SOCjEV− V1G,max
∀t ∈ T ∀j ∈ J (30) also available data, as reported in Table 1. Moreover, the objective
function is composed of three different objectives:
0⩽PEV−
j,t
V1G,DC
⩽PEV−
j
V1G,max
∀t ∈ T ∀j ∈ J (31)
• the minimization of the operating costs (f1 );
0⩽PEV−
j,t
V1G,in EV−
⩽aSOCj,t V1G
+b ∀t ∈ T ∀j ∈ J (32) • the reference temperature tracking (f2 );
• the satisfaction of the EV charging requirements (f3 ).

The building thermal sub-system

The building thermal sub-system includes three main components: a

6
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

Table 2 terms of state of charge) at a specific time instant tjref . To compare


Daily cumulated energy contribution for the winter sunny day over the three
different objectives, a normalization through Nadir-Utopia values has
scenarios [kwh].
been adopted [34]:
PV PMG PV2G PHP PB PV1G
∑ fl − flutopia
Scenario 1 52 353 67 36 107 195 J= ωl (38)
Scenario 2 52 372 69 49 107 199 l∈L flnadir − flutopia
Scenario 3 52 426 70 101 107 200

ωl = 1 (39)
[ ] [ ( ) ] l∈L
∑ ∑( ) ∑ ∑
f1 = cbuy PG,in − c sell G,out
P Δt − cdr
P G,BL
− P G
p,t Δt
t∈T
p,t
p∈P
p,t t
t∈T dr p∈P
where L is the set of the three above-defined objective functions, flutopia
(35) and flnadir are utopia and nadir reference value for the l-th objective
function; more precisely these values are calculated minimizing and
∑∑( )2
f2 = RM
Th,t RM,ref
− Th,t (36) maximizing, respectively flutopia andflnadir , each objective (35)-(37) sub­
t∈T rc h∈H ject to the constraints of the problem. Furthermore, ωl are weights to
privilege a specific objective with respect to the others. The presented
∑( ⃒ )2
optimization problem is a MINLP (Mixed Integer Non Linear Program­
V1G ⃒
f3 = E−
SOCj,t ⃒ − SOCjEV− V1G,ref

j∈J
ref
t=tj ming) problem and it is given by the minimization of (38) subject to
( ⃒ )2 constraints (3)–(34) (Table 2).
+ SOCtEV− V2G ⃒
t=t ref
− SOCEV− V2G,ref
(37) B. The low-level decision problem.
The low level is composed by two different modules, i.e., CSC and d-
Equation represents the operation costs and benefits related to the
TC. These modules operate independently from each other, and inputs
energy exchanges with the external grid and the demand response ser­
are the setpoints provided by the high-level.
vice, where cbuy is the unit energy price (considered constant in this 1) The charging system controller sub-module.
study), csell the unit selling price and cdr is the selling price for the The high-level communicates to the CSC (Fig. 4) the overall power
modulation power integrated over time (T dr is the time interval of the that can be used for charging, the current magnitude, and the power
demand response event, Tdr ⊂T). Indeed, PG,BL
t is the baseline consump­ factor to the EVs charging station, while for the d-TC the high-level
tion of the system. Equation refers to the tracking of a reference tem­ communicates the temperature of the thermal storage and the power
perature. Finally, equation allows considering the energy demand (in profile of the HVAC plant. The CSC algorithm is embedded with a
commercial CS, developed in collaboration between the CS provider and

Fig. 4. The charging system controller sub-module.

Fig. 5. The distributed temperature controller module.

7
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

Fig. 6. Power balance of the system in daily operation scenario (Scenario 1); MG: power from the main grid; PV: power from the PV system; HP and Building: power
absorbed by the geothermal heat pump and remaining part of the building loads; V2G: power exchanged by the V2G charging station; V1G: power absorbed by the
traditional charging stations.

defined time horizon and time interval, and its main goal is to define the
setpoints of the fan coils installed in each room of the building. Its inputs
are the available thermal power from the high-level algorithm, the
forecasted external temperature, and the external temperature mea­
surement at the initial time step. In this application, the control system is
implemented in field devices that are low-cost microcontrollers (i.e.,
digital thermostats).
Specifically, the d-TC is a distributed MPC algorithm based on the
distributed Alternating Direction Method of Multipliers (dADMM) al­
gorithm. A centralized approach can be inconvenient in real life appli­
cations since the robustness of the control architecture is essential. Then,
using a distributed algorithm allows the controller to operate even in
case of failures of the communication between the single devices and the
central controller. The convenience of using a dADMM algorithm grows
as increases the number of the state variables of the problem [35].
The d-TC solves the following optimization problem:
[ ]
∑ ∑( )2
min IN IN,ref
Th,t − Th,t (40)
t∈T bms − 1 h∈H

s.t.
Fig. 7. The overall power balance over the three phases is summarized to ⎛( A ) ⎞
visualize the exchanged power at the point of delivery. Tt − Th,t IN
IN CO
⎜ + Q̇h,t + Q̇h,t + ⎟
⎜ R E,IN ⎟ ll
⎜ ⎟ Δt
the authors of the present paper and in operation at the Savona Campus
h
IN
Th,t+1 IN
= Th,t +⎜
⎜ ( ) ⎟
⎟ CIN ∀t ∈ T bms , ∀h ∈ H
site. The algorithm’s main concepts are reported here, but equations are ⎜ ∑ T IN − T IN
⎝ k,t h,t SG
⎟ h

avoided for intellectual property preservation. The CSC is a real-time + Q̇h,t
RI,IN
controller that shares the overall available current among all the con­ k∈K h k,h

nected CSs, maximizing currents to vehicles under the constraints of the (41)
maximum current that the external grid can provide. The CSC works
both with an event and a time-triggered activation mode: it is activated ThIN,min ⩽Th,t
IN
⩽ThIN,max ∀t ∈ T bms , ∀h ∈ H (42)
when a particular event occurs (i.e., the connection of a new EV) or after
(43)
IN IN,max
a pre-determined time interval. 0⩽Q̇h,t ⩽Q̇h ∀t ∈ T bms , ∀h ∈ H
2) The distributed temperature controller module.
The d-TC (Fig. 5) is a receding horizon-based algorithm with a

8
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

Fig. 8. EV charging behavior in daily operation scenario (Scenario 1).

Fig. 9. The power balance of the system in DR scenario (Scenario 2).

∑ ∑ IN IN
(44) control variables are the heat gains Q̇h,tbms due to the HVAC plant (pos­
FC
0⩽ Q̇h,t Δtll ⩽Q̇ Δthl
t∈T bms h∈H itive during the winter season and negative in summer).
The problem aims to minimize the distance from Th,t
IN
(the internal Constraint describes the temperature dynamic behavior in the hth
CO
temperature of the hth room) and the temperature setpoint Th,t
IN,ref
. The room, considering TtA the external ambient temperature, Q̇h,t the

9
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

form:

1 T
min x Qx + pT x
x 2
{
Gx = b (45)
s.t.
Mx⩽c

where x is the decision variables vector, andp,Q,G,M, b and c are suitable


vectors and matrices. Proof is reported in Appendix.
In the following, it is shown a distributed algorithm to solve.

Problem decomposition

First of all, problem is decomposed into sub-problems, then the


distributed algorithm is applied. It is assumed that the communication
network coincides with the structure of the building (i.e., the graph
defined by the adjacency matrix of the rooms). Therefore, the number of
agents corresponds to the number of building network nodes. It is
considered the graph associated with the building network V = (H, E),
where H = {1, ..., n} is the number of the agents (i.e., rooms of the
Fig. 10. The power balance at the POD of the system in DR scenario (Sce­
nario 2).
building) and E is the set of the edges. For any h, let A(h) = {k|(h, k) ∈
E } ∪ {h} denote the set of its neighbours, defining dh as the degree of a
SG
nodeh, i.e., |A(h)| = dh + 1.
internal heat gains, Q̇h,t the solar radiation gain and the heat exchange To decouple into |H| sub-problems, it is necessary to introduce an
between adjacent rooms, RE,IN
h is the thermal resistance between the augmented set of decision variables. The new vector is defined as yh =
external ambient and the room, RI,IN x h }, where xh is the “owned” variables vector of the agent h, the
k,h is the thermal resistance between
{xh ∪ ̂
th internal temperature of each room, and ̂ x h is the vector of the “copied”
two adjacent rooms, and CIN h is the thermal capacity of the h room.
variables that the agent h needs from its neighbors, more precisely the
Constraints and are the bound of the internal rooms temperature and the
internal temperature of the adjacency rooms, to solve the associated sub-
allowed power injected into each room considering the technology
problem. By introducing these new “copied” variables, the optimization
limits. Constraint states that the total amount of thermal energy injected
problem through the agents can be decoupled, generating a set of in­
into the building over the low-level time horizon must be equal or lower
dependent sub-problems. A new decision vector is defined y =
than the thermal energy fed to the building during one time step of the { }
col yh ∀h ∈ H , and a new constraint is added to the optimization
high-level algorithm.
problem:
Proposition I. The problem (40)-(44) can be rewritten in standard QP

Fig. 11. EV charging behavior in demand response scenario (Scenario 2).

10
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

Fig. 13. Power balance of the system using a heuristic controller scenario
(Scenario 3).
Fig. 12. Power balance enlargement Scenario 1 vs Scenario 2.

xh = ̂x h ∀h ∈ H (46) Table 3
Operation costs for the company in the different cases.
In the variables and “copy” are set equal to reach the same final
condition and can be rewritten as follows: Operation costs Costs S1 € Costs S2 € Costs S3 €

Winter sunny 90,5 75,5 109,9


By = 0 (47) Winter cloudy 95,6 74 114,6
Summer sunny 72,6 58 93,9
where B ∈ R|̂x |×|y| is the adjacency matrix of the direct communication Winter cloudy 80,4 66,6 100,7

graph (for more details, see [35–37]). Then, the augmented form of the
problem becomes:
⎧ The distributed alternating direction method of multipliers algorithm
1 T̃ ⎨
min y Q y + ̃ T
p ys.t. ̃y=̃ ̃ c
bBy = 0My⩽̃ (48)
y 2 ⎩
G To solve subproblems, μh and νh are introduced as the Lagrange

multipliers for constraints G
̃ h yh = ̃
bh and
k∈A(h) Bkh yk = 0, respectively.

where ̃ p,Q,̃ G,
̃ M,
̃ ̃b and ̃c are the augmented version of p,Q,G,M, b and c The Lagrange function of the hth agent is:
considering the new decision variables vectory. It is easy to see that the 1 ̃
( ) ∑
Lh (yh , μh , ν) = yTh Q pTh yh + μTh Q̃ h yh − ̃ νTh Bkh yh
optimal solution of it is the same problem. Problem can be decomposed 2 h yh + ̃ bh + ∀h ∈ H
into |H| sub-problems, where the hth agent corresponds to an indepen­
k∈A(h)

dent sub-problem with its sub-matrices and sub-vectors respectively (50)

p ,Q
̃h
̃ h ,G
̃ h ,M̃ h, ̃
bh and ̃ch : The dADMM algorithm with a step size ρ is given by the following
⎧ steps (for each hth agent) [38] and [40]:
}

⎨ ∑
min
{ yh ⩽̃ch 1 T̃
pTh yh s.t.
y Q yh + ̃ ̃ h yh = ̃
G bh Bk,h yk = 0 ∀h
yh ∈̃h
M 2 h h ⎪
⎩ k∈A(h)

∈H
(49)

11
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

⎧ ⎫
⎪ Li (yi , μi [τ], ν[τ] )+ ⎪

⎪ ⎪

⎪ ⎪


⎪ ρ⃦ ⃦∼

∼ ⃦2 ⎪


⎨ + ⃦Gh yh − bh ⃦ + ⎬
yh [τ + 1] = argmin 2 2
(51)
⎪ ⃦
yi ∈{M h yh ⩽ c h }⎪
∼ ∼ ⃦2 ⎪

⎪ ⎪


⎪ ρ⃦
⃦ ∑ ⃦
⃦ ⎪


⎪+ ⃦
⎩ [Bkh (yh − yh [τ] ) + wk [τ] ] ⃦ ⎪

2⃦ k∈A(h)

2

( )
̃ h yh [τ + 1] − ̃
μh [τ + 1] = μh [τ] + ρ G b (52)

(communicate{yh [τ + 1] ∀h ∈ H(h) }) (53)


( )∑
1
wh [τ + 1] = Bhk yk [τ + 1] (54)
1 + dh k∈A(h)

(communicate{wh [τ + 1] ∀hA(h) }) (55)

νh [τ + 1] = νh [τ] + ρwh [τ + 1] (56)

Fig. 14. Convergency ratio of the dADMM algorithm, with a logarithmic scale. (communicate{vh [τ + 1] ∀hA(h) }) (57)
th
The algorithm is initialized ad follows (for each h agent):

yh [0] ∈ Rdh (58)


( )
μh [0] = ρ G̃ h yh [0] − ̃
b (59)

( )∑
1
wh [0] = Bhk yh [0] (60)
1 + dh k∈A(h)

νh [0] = ρwh [0] (61)


Algorithm (51)-(57) allows each agent to store only three sub-vectors
of variables, i.e., yh [τ], wh [τ] and νh [τ], which are updated at each iteration
and the results are shared only with the neighbors to solve the

Fig. 15. Comparison of the three control strategies.

12
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

optimization problem. The convergence proof is reported in [36]. absorbed from the main grid. The following Figure (Fig. 6) represents
the power balance over the three different phases of the system over this
Results and discussion scenario, whereas, in Fig. 7 the overall power balance at the Point Of
Delivery (POD) is depicted for Scenario 1.
The developed BMS is applied to a real case study represented by the In Fig. 6, the geothermal heat pump operates during the first hours of
Smart Energy Building (SEB) [38], a net-zero emission building located the day to heat the thermal storage, and the BMS will optimally deploy
at the Savona Campus of the University of Genoa; its surface is about 750 the thermal inertia stored in the tank to provide heat to the whole
square meters organized in 22 rooms over two floors for a total heigh of building over the remaining part of the day. The EV charging stations
ten meters (mean electricity consumptions about 40 MWh for year). The charge the EVs in such a way to fulfill the SOC requirement within the
SEB is equipped with dedicated generation units from renewable energy scheduled time and not considering any power absorption limits due to
sources, i.e., a photovoltaic field with rated peak power of 21 kW (mean the DR. Furthermore, the V2G vehicle does not discharge its battery at
production 25 MWh/year), a reversible geothermal heat pump capable any time interval since there is not the necessity to provide additional
of generating up to 45 kW of thermal power and 40 kW of cooling power. energy to the system. Indeed, the additional cost due to the round-trip
In this model the Coefficient Of Performance (COP) varies on two efficiency would not justify this behavior in the considered scenario.
different values, one for the cooling and one for the heating operations, In Fig. 7 the results over the three phases are summarized to give an
this assumption is justified as the energy source is the geothermal heat, all-round view of the energy balance at POD level.
unlike the heat pumps that use air as a source, the source’s temperature In Fig. 8, the V1G charging station scheduling is provided. It is
is constant throughout the various seasons of the year. Furthermore, possible to see how the different EVs are charged over the time horizon
during operations, the delivery and return temperatures of the thermal and how much energy they will receive at every time interval from each
vector liquid are comparable to constant values thanks the thermal phase.
inertia provided by the storage tank. EVs charging infrastructure, a In Scenario 2, the BMS manages the system’s operation during a
thermal storage, and a fan coil circuit. As regards charging stations, demand response event. The respective power balance is depicted in
there is a bidirectional V2G charging station with a single socket (the Fig. 9.
maximum power is 10 kW both in charge and discharge), which operates In Fig. 9 it is underlined how during the DR event (from 3 pm until 5
in DC. Moreover, there is a V1G AC charging station with four sockets, pm), the V1G CS reduces the absorption of energy to match the power
two single-phase plugs for small EV or hybrid vehicles (7 kW), and two G,ref
requirementPt . Although the values of the absorbed power from the
T2 with a maximum power of 22 kW [39]. The EVs charging infra­ grid related to the three different phases are lower than in Scenario 1,
structure serves seven EVs during the whole day (as reported in Table 1), they are slightly different from each other. This is due to the building
three EVs are connected to the AC T2 sockets, one to the DC bidirectional electrical energy demand that is modeled as a three-phase unbalanced
charging station, and the remaining three are coupled to the single- load. Indeed, the total amount of power absorbed at each time step from
phase plugs. The technical features of the and the required state of the grid corresponds to the reference value, this is clear from Fig. 10 in
charge SoC EVs are defined in Table 1. The reference temperature within which the overall power exchange at the POD is depicted.
the building rooms is 21 ◦ C (to be preserved from 9:00 am until 6:00 From Fig. 10, it is clear that the programmable loads are shifted from
pm). the demand response time period. Indeed, since the BMS is governed by
The high-level algorithm operates under an MPC scheme with a time a weighted objective function (49) that considers all the different
interval of an hour and a time horizon of 24 h. At the lower level, the objective together, the extra-savings due to the remuneration of the DR
CSC works in real-time while the d-TC operates in MPC with a time service allows the deployment of more resources for the EV charge and
interval of 5 min and a time horizon of one hour. to reach the set-point temperature within the rooms of the building.
Three different scenarios are considered: In Fig. 11 the V1G charging station scheduling is presented. Con­
cerning Scenario 1 the charging process of EVs 3 and 4 is interrupted
• Scenario 1: the daily standard operation through the BMS. In this during the DR in such a way that the reference power required by the
scenario, constraints are automatically not considered forcing T dr = grid operator is respected.
∅; indeed the possibility of providing a demand response service is Fig. 12 is an enlargement of the POD power balance between time
not considered; steps 14 and 18 to underline better the differences among the first and
• Scenario 2: the BMS operational management during a demand second scenarios over the time interval of DR events.
response event. In this scenario, the baseline absorption (PG,BL t ) In Fig. 12, it is possible to see that the absorbed power from the grid
profile is assumed as the value of the energy absorbed from the grid is reduced during the demand response event in Scenario 2 with respect
in Scenario 1 over the same time interval, and the reference value of to Scenario 1 according to the absorbed power reference requirement;
G,ref
the absorbed grid power (Pt ) is imposed as equal to 15 kW. furthermore, in the second scenario the EVs charging operation are
• Scenario 3: the BMS is not involved in the operation. The manage­ almost totally stopped, with an exception of EV 5, to cope the limitations
ment of the system is carried out using heuristic rules that charge the imposed on the energy absorbed by the grid.
EVs at the maximum power (without the possibility provide the DR Finally, to validate the operation of the BMS’s operational efficiency,
service) and that activate (or deactivate) the heat pump for tem­ a third scenario in which the energy system will be controlled by means
perature control only when the temperature of the thermal storage is of a heuristic controller. For brevity, it is chosen to present just the
lower (or higher) than a certain threshold. power balance of the system depicted in Fig. 13.
Fig. 13 highlights that the main differences with scenarios one and
Each scenario is performed over four different days of the year, two two are related to the operation of the geothermal heat pump controlled
days for the winter season and two for summer, one in a sunny weather with a threshold logic. The daily cumulated energy contribution for the
condition and one in cloudy weather. three scenarios of the presented day are reported in Table 2.
A. High-level results In scenario 1, the consumption of energy withdrawn from the
In this sub-section, the results of three scenarios are presented for the network assumes its minimum value; furthermore the energy absorbed
winter sunny day, while the results for all the investigated days and by the heat pump and the vehicles being recharged also reach their
scenarios are summarized in Table 3. minimum values. This behavior is the result of the minimization of the
Scenario 1 represents the daily operation of the BMS. In this scenario, three different objectives contrasting, respectivelyf1 , f2 andf3 . In Sce­
since the DR event does not occur, there is no limitation on the energy nario 2, the management of the demand response service allows to

13
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

further minimize the operating costs of energy; this also reflects on the 18) from 8 am until 6 pm for a single day with a discretization step of 5
other objectives, and allows to withdraw more energy from the grid min.
while maintaining minimal management costs. In the last scenario, the It is clear from Fig. 15 that the d-TC is better performing with respect
lower efficiency of the control system based on heuristic rules signifi­ to the other controllers regarding the tracking of the setpoint tempera­
cantly increases the energy absorbed from the grid and the consumption ture (that is 21 ◦ C for this specific example). In order to evaluate the
of the heat pump and consequently the related management costs. overall performance of the d-TC the value of the optimal cost is adopted
In Table 3, the overall daily operational costs, calculated as in as KPI for each control strategy; the overall cost of the TC-FP is 2266.45,
equation, for the company in the different cases are reported. while TC-OP is 567.64 and for the d-TC is 39.01.
The difference in costs between Scenario 3 and 1 is due to the more
efficient operation of the BMS concerning the heuristic controller, more Conclusion and Future developments
precisely in the first two scenarios the geothermal heat pump operates
with MPC logic considering the meteorological forecast and his opera­ In this paper a new bi-level BMS is proposed that is able to control
tion time horizon. Moreover, in Scenario 2 BMS shifts the EV charging rooms’ temperature and EVs’ charging in a smart building characterized
operation to meet the power reference value imposed by the grid by a heat pump, renewables, a storage system, and charging stations for
operator, indeed the gap between Scenario 1 and Scenario 2 is justified electric vehicles (both traditional and V2G). The two levels work in a
by the extra gains due to the revenue received for DR service. hierarchical architecture at different time scales, and the low level re­
B. The low-level distributed temperature controller ceives the optimal results of the high level as inputs. In particular, the
The BMS controls the HVAC system to reach the temperature in each high level is devoted to operational management, and the optimization
room of the building using an MPC approach through the information of model is used under a receding horizon control scheme. Specifically,
the maximum power available from the storage tank coming from the firstly building load and production from renewables are forecasted over
higher level decision layer. the optimization horizon, and the temperature is measured. Then, the
Firstly, the convergence trajectory of the dADMM algorithm to the optimization problem is run, but only the optimal schedule for the first-
optimal solution is presented (see Fig. 14). This rate is defined as the time interval is retained. The main purpose of the high-level is to define
difference between the solution calculated by means the vector of the the operation setpoints for the low-level algorithms, i.e. the Charging
actual decision variables at the iteration τ and the solution calculated by Stations (CSs) Charging System Controller (CSC) and the distributed
means of the optimal decision variables vector. Temperature Controller (d-TC). The main novelty related to the high
The convergence is reached after a thousand iterations. Considering level is related to the system model of the electrical components because
that the average solver time for each iteration is equal to 4 ms, the different kinds of CSs are considered and phases have been included in
overall computational time for the solution is equal to 4 s. Instead, the the overall power balance. This approach guarantees the possibility of
equivalent centralized problem is equal to 1.2 s. This result could considering the unbalanced loads since the approximation of the
discourage the use of a distributed algorithm, but if the computational equivalent single-phase circuit is neglected. Another novelty is related to
times of the two algorithms are compared with the time interval and the the low level. In fact, a distributed algorithm has been developed for the
good reaction times of the application, the time gap between the solu­ low-level temperature controller. The BMS is tested in a real environ­
tion of the two algorithms can be assumed as negligible. Moreover, as ment (Savona Campus) and is able to work in standard operational
the rooms number of the building increases, the computation time of the management conditions and during demand response events. The tests
centralized solution will increase, as reported in [40]. It is also carried out at the Savona Campus show that the proposed algorithm
emphasized that the distributed solution is more resilient as it is not allows a reduction of the daily costs of about 20 % with respect to a
necessary to have information on all the system agents to allow the simple heuristic, without compromising the fulfillment of the charging
resolution of the sub-problems. requests of the electric vehicles. The savings can be increased till close to
In order to show the performances of the proposed controller the d- 35 % if a demand response is also exploited. Future developments could
TC has been compared with the following traditional controllers that be oriented to use an accelerated distributed algorithm for the low-level
are: algorithm and to the definition of robust approaches to face
uncertainties.
• Thermostatic control with fixed set point (TC-FP)
• Thermostatic control with optimized set point (TC-OP) Declaration of Competing Interest

Both mathematical models of the TC-FP and the TC-OP are reported The authors declare that they have no known competing financial
in Appendix IV. TC-FP is a classical controller implemented in buildings interests or personal relationships that could have appeared to influence
that represents the state of the art from a practical point of view, in the work reported in this paper.
which the thermal power is controlled by on/off relay thermostats which
keep the indoor temperature around the setpoint (under a specified Data availability
bandwidth). While the TC-OP is an advanced version of the TC-FP that
defines optimal time varying set point for the relay of the thermostat by Data will be made available on request.
minimizing an objective function subject to the thermal model of the
building. The d-TC proposed in this paper allows to exploit the whole Acknowledgement
functionality of the smart building including the continuous regulation
of the fan coil perform control that minimizes temperature set point This work was partially supported by Foster BIPE project from the
tracking. University of Genoa, and from Region Liguria in the project “Programma
In Fig. 15 a comparison between the three control methods (TC-FP, Operativo” POR FSE Regione Liguria 2014-2020, RLOF18ASSRIC/13/1
TC-OP and d-TC) is reported for four sample rooms (rooms 2, 5, 12 and A portion of the work has been funded by RLOF18ASSRIC/3/1.

14
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

Appendix I. The building thermal sub-system

GHP
Q̇t
PHVAC
t = ∀t ∈ T (62)
COP

PHVAC
t = PHVAC
p,t ∀t ∈ T (63)
p∈P

0⩽PHVAC
p,t ⩽PHVAC,max ∀t ∈ T ∀p ∈ P (64)
( ) hl
Δt
(65)
ST GHP FC
Tt+1 = TtST + Q̇t − Q̇t ∀t ∈ T
CST

TsST,min ⩽TtST ⩽TsST,max ∀t ∈ T ∀s ∈ S (66)


∑ RM
(67)
FC
Q̇t = Q̇h,t ∀t ∈ T
h∈H

( )
(68)
RM
Q̇h,t = εfc ṁair
h,t cp
air
TtST − Th,t
RM
∀t ∈ T ∀h ∈ H
( )
∑ Δthl
(69)
RM RM E AR RM CO SG
Th,t+1 = Th,t + Q̇h,t + Q̇k,h,t + Q̇h,t + Q̇h,t + Q̇h,t ∀t ∈ T ∀h ∈ H
k∈K h
ChRM

T RM,min ⩽Th,t
RM
⩽T RM,max ∀t ∈ T ∀h ∈ H (70)
( )
TtA − Th,t
RM

(71)
E
Q̇h,t = ∀t ∈ T ∀h ∈ H
RE,RM
h

( )
RM RM
Tk,t − Th,t
(72)
AR
Q̇k,h,t = ∀t ∈ T ∀h ∈ H ∀k ∈ K h
RI,RM
k,h

(73)
SG
Q̇h,t = Radh,t τw Awh,t ∀t ∈ T ∀h ∈ H

Constraints define the three-phase power absorbed by the Geothermal Heat Pump (GHP) as a function of the thermal power expressed by the
GHP
geothermal heat pump (Q̇t ) and its Coefficient Of Performance (COP) that is assumed as a constant parameter during each season for this appli­
cation. It is important to note that in our model the COP varies depending on the considered cases in the scenarios, one for the cooling and one for the
heating operations. This assumption is consistent since the energy source is the geothermal heat, unlike the heat pumps that use air as a source,
therefore source’s temperature is constant throughout the various seasons of the year. In the relation between the overall HVAC three-phase AC power
and the power of every single phase (PHVACp,t ) is represented. Constraint represents the dynamic behavior of the thermal storage internal temperature
GHP FC
(TtST ) by considering the thermal storage capacity (CST ) and the thermal fluxes in input Q̇t , and in output, Q̇t , being this latter the thermal power
RM FC
extracted by the fan coils circuit. The relation between the thermal power absorbed by every single room (Q̇h,t ) and the overall needed power (Q̇t ) is
RM
expressed in. Q̇h,t is defined in for the fan coil in each hth room as a function of the air temperature within the room (Th,t
RM
) and the temperature of the
water supplied to the fan coil circuit (TtST ), where εfc is the fan coil efficiency, cpair is the specific heat capacity of the air, both assumed as constant
values, whereas ṁair
h,t is the mass flow rate of the air that passes through the fan coil considered as a control variable of the system. This hypothesis
differs from [2] in which the fan coil mass flow rate was considered as a constant value and the control was made by means a binary variable to manage
the fan coil with an on/off logic; this new assumption is made to leverage the possibility, in this specific application, of controlling the fan coil mass
flow rate with all the values in range from 0 % to 100 %. Constraint defines the dynamic behavior of the internal room temperatures (Th,t RM
),
E
considering: the heat flux given from the difference of temperature between the room and the external environment (Q̇h,t ), both through the wall and
AR RM
the windows, the heat exchange between adjacent rooms (Q̇k,h,t ), the heat supplied through the fan coil circuit within each room (Q̇h,t ), the internal
SG
heat generation given by the occupant and the electronic devices (assumed as input data), finally, the solar radiation gain (Q̇h,t ), differently from [2] in
E
which this specific gain was neglected. Equations (71)-(73) express the evaluate the heat exchange with the external ambient (Q̇h,t , being TtA the
AR SG
ambient temperature), with the adjacent rooms (Q̇k,h,t ), the solar radiation gain (Q̇h,t ) and the internal heat gain that is considered as known a priori,
CO
since the aim of this work is not to define a model that simulate this issue (Q̇h,t ). RE,RM
h represents the thermal resistances between the internal air of
each room and the external air, it considers both the walls and the surfaces of the window, whereas RI,RM
k,h is the thermal resistance between two
adjacent rooms. Furthermore, τw is the transmittance of the windows considered as the same constant value for each window of the building, Awh,t is the
surface of the windows of each room, and Radh,t is the solar radiation that reaches each room during the various hours of the day, this quantity is
assumed given a priori as an external input to the system. Equations, and represent bound constraints.

15
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

Appendix II

Proof of Proposition I: posing:


[ ]
(74)
IN IN
T tbms = col Th,t bms , h ∈ H

[ ]
(75)
A
T tbms = col TtAbms , h ∈ H
[ ]
(76)
IN*
T tbms = col TtIN*
bms , h ∈ H

[ ]
(77)
IN IN
Qtbms = col Q̇h,tbms , h ∈ H

[ ]
(78)
CO CO
Qtbms = col Q̇h,tbms , h ∈ H

[ bms ]
Δt
KC = diag , h ∈ H (79)
ChIN
[ ]
1
KE = diag E,IN , h ∈ H (80)
Rh
⎧ ∑
⎪ 1

⎪ − if k = h

⎪ I,IN
h Rk,h

⎪ k∈K

KI = 1 (81)

⎪ if k ∕
= h and (k, h) ∈ H

⎪ RI,IN

⎪ k,h


0 otherwise
Thanks to previous statements, state equation can be written in state-space form:

(82)
IN IN IN
T t+1 = AT t + BQt + ̃
ht ∀t ∈ T bms

where
A = [I + KC ( − KE + KI ) ] (83)

B = KC (84)
( CO )
(85)
A
̃
ht = KC KE T t + Qt ∀t ∈ T bms

Thanks to the previous relations, it is possible to write the objective function as a quadratic form of the state vector
∑ [( )T ( IN )]
min TtIN − TtIN* I Th,t − TtIN* (86)
t∈T bms

That is equivalent to
∑ [( )T ( )T ]
min TtIN TtIN − 2 TtIN* TtIN (87)
t∈T bms

It is necessary to transform the dynamic optimization problem into a static one using the following property. As a matter of fact, in the generic
IN
instant t, the state T tbms can be written as

t− 1 ( )
(88)
IN IN IN
T tbms = At T 0 + An BQt− n− 1 + ̃
ht− n− 1 ∀tbms ∈ T bms
n=0

[ ]T
IN [ ]T IN [ ]T
Thus posing T = TIN
1 ...
IN
TTbms ,Q = QIN
1 ...
IN
QTbms and ̃
h= ̃ hTbms the standard QP form of problem (40)-(44) is obtained
h1 ...̃

1 T
min x Qx + pT x
x 2
{
Gx = b (89)
s.t.
Mx⩽c

where
[ IN IN ]
x = T ;Q (90)
Q=I (91)

16
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

p = − TtIN*
bms (92)
[ ]
I 0
G= ̂ (93)
0 B

0 0 0 0
⎡ ⎤
0 0 0 B
⎢ ⎥
̂=⎢
B ⎢ ... ... ... ...

⎥ (94)
⎣ bms −

0 B ... AT 2
B
bms −
B AB ... AT 1
B

̂ IN + C
b = AT 0
̂̃h (95)
⎡ ⎤
I 0 ... 0
⎢ 0 ⎥
̂=⎢0
A
A ... ⎥ (96)
⎣ ... ... ... ... ⎦
T bms
0 0 ... A
⎡ ⎤
0 0 ... 0
⎢ ⎥
̂ =⎢0
C
A ... 0 ⎥ (97)
⎣ ... ... ... ... ⎦
T bms − 1
0 0 ... A
⎡ ⎤
− I 0 0 0
⎢ 0 I 0 0⎥
M=⎢
⎣ 0
⎥ (98)
0 − I 0⎦
0 0 0 I
⎡ ⎤
IN,min
⎢− T ⎥
⎢ IN,max ⎥
⎢ T ⎥
c=⎢


⎥ (99)
⎢ 0 ⎥
⎣ IN,max ⎦
Q̇h

Appendix III

TC-FP model
The dynamical behavior of the thermostat can be modelled by introducing a binary state variablez ∈ {0, 1}. Then z = 1 represents the “on” state,
while z = 0 is to the “off” state. The state equation associated to the thermostat is given by the state-update equation coming from [40]
⎧ ( ( )) ( ( ))
⎨ 1 if z = 1 ∧ ¬ T IN ⩾r + γ IN
∨ zh,t = 0 ∧ ¬ Th,t ⩽r − γ
(100)
h,t
zh,t+1 = h,t
∀t ∈ T bms , ∀h ∈ H
⎩ 0 otherwise

where and r is the temperature setpoint and γ is half the width of the switching zone, ¬ denotes logic negation, ∧ is the logic conjunction and ∨ stands
for the logic disjunction.
Once the binary state of the thermostat is available, the thermal power injected into building is given by

⎨ IN,max
Q̇h if zh,t = 1
(101)
IN
Q̇h,t = ∀t ∈ T bms , ∀h ∈ H
⎩ 0 oherwise

TC-OP model
As regards the TC-OP model, in this case the temperature setpoint r becomes a decision variable of the following optimization problem:
[ ]
∑ ∑( )2
min IN
Th,t IN,ref
− Th,t (102)
t∈T bms − 1 h∈H

s.t.

17
G. Bianco et al. Energy Conversion and Management: X 17 (2023) 100339

⎛( A IN
) ⎞
Tt − Th,t IN CO
⎜ + Q̇h,t + Q̇h,t + ⎟
⎜ R E,IN ⎟ ll
⎜ h ⎟ Δt
IN
Th,t+1 IN
= Th,t +⎜
⎜ ( ) ⎟
⎟ CIN ∀t ∈ T bms , ∀h ∈ H (103)
⎜ ∑ T IN − T IN ⎟ h
⎝ k,t h,t SG ⎠
+ Q̇h,t
k∈K h
RI,IN
k,h

ThIN,min ⩽Th,t
IN
⩽ThIN,max ∀t ∈ T bms , ∀h ∈ H (104)

⎨ IN,max
Q̇h if zh,t = 1
(105)
IN
Q̇h,t = ∀t ∈ T bms , ∀h ∈ H
⎩ 0 oherwise
⎧ ( ( )) ( ( ))
⎨ 1 if z = 1 ∧ ¬ T IN ⩾r + γ IN
∨ zh,t = 0 ∧ ¬ Th,t ⩽r − γ
(106)
h,t
zh,t+1 = h,t
∀t ∈ T bms , ∀h ∈ H
⎩ 0 otherwise

References photovoltaics’ uncertainty and stochastic electric vehicles’ driving schedule. Appl
Energy 2018;210:1188–206.
[20] Wu D, Zeng H, Lu C, Boulet B. Two-stage energy management for office buildings
[1] Islam M, Mithulananthan N, Bhumkittipich K, Sode-yome A. EV charging station
with workplace EV charging and renewable energy, v 3, n 1, p 225-237, March
design with PV and energy storage using energy balance analysis.
2017. IEEE Trans Transport Electr PP (99):1–1.
[2] Bianco G, Bracco S, Delfino F, Gambelli L, Robba M, Rossi M. A Building Energy
[21] El-naggar MF, Elgammal AAA. Multi-Objective Optimal Predictive Energy
Management System Based for demand response in smart grids, IEEE International
Management Control of Grid-Connected Residential Wind-PV-FC-Battery Powered
Conference on Automation Science and Engineering, August 2020.
Charging Station for Plug-in Electric Vehicle, JEET , p 742-751, 2018.
[3] Oldewurtel F, Parisio A, Jones CN, Gyalistras D, Gwerder M, Stauch V, et al. Use of
[22] Rudnik R, Wang C. Student Member, L. Reyes-Chamorro, J. Achara, J. Y. Le
model predictive control and weather forecasts for energy efficient building
Boudec, and M. Paolone, Real-time control of an electric-vehicle charging station
climate control. Energy Build 2012;45:15–27.
while tracking an aggregated power-setpoint, IEEE Trans Ind Appl, 56 (5), 2020.
[4] Bruni G, Cordiner S, Mulone V, Sinisi V, Spagnolo F. Energy management in a
[23] Hu Y, Chen C, Member IEEE, He T, Fellow IEEE, He J, et al. Senior member, IEEE,
domestic microgrid by means of model predictive controllers. Energy 2016;108:
“proactive power management scheme for hybrid electric storage system in EVs:
119–31.
An MPC Method”. IEEE Trans Intellig Transport Syst 2019.
[5] Ma Y, Borrelli F, Hencey B, Coffey B, Bengea S, Haves P. Model predictive control
[24] Tookanlou MB, Pourmousavi SA, Marzband M. An optimal day-ahead scheduling
for the operation of building cooling systems. IEEE Trans Control Syst Technol
framework for E-mobility ecosystem operation with drivers preferences. IEEE
2011;20(3):796–803.
Trans Power Syst 2021.
[6] Killian M, Kozek M. Ten questions concerning model predictive control for energy
[25] Zhang X, Wenbo S, Bin Y, Ali M, Na L. Decentralized and distributed temperature
efficient buildings. Build Environ 2016;105:403–12.
control via HVAC systems in energy efficient buildings. IEEE Trans Control
[7] Kwak Y, Huh JH, Jang C. Development of a model predictive control framework
Network Syst, 2017.
through real-time building energy management system data. Appl Energy 2015;
[26] Yang Y, Srinivasan S, Hu G, Spanos CJ. Distributed control of multizone HVAC
155:1–13.
systems considering indoor air quality. IEEE Trans Control Syst Technol 2021.
[8] F. S. Tidjani, A. Hamadi, A. Chandra, Fellow IEEE, B. Saghir, B. Mounir, Energy
[27] Joo I, Choi D. Distributed optimization framework for energy management of
Management of Micro Grid based Electrical Vehicle to the Building (V2B) 2019.
multiple smart homes with distributed energy resources. IEEE Access 2017;5:
[9] Ferro G, Minciardi R, Parodi L, Robba M. Optimal planning of charging stations in
15551–60.
coupled transportation and power networks based on user equilibrium conditions.
[28] Lee S, Kwon B, Lee S. Joint energy management system of electric supply and
IEEE Trans Autom Sci Eng 2021.
demand in houses and buildings. IEEE Trans Power Syst 2014;29(6):2804–12.
[10] van den Berg MA, Lampropoulos I, AlSkaif TA. Impact of electric vehicles charging
[29] Belluschi F, Falsone A, Ioli D, Margellos K, Garatti S, Prandini M. Distributed
demand on distribution transformers in an office area and determination of
optimization for structured programs and its application to energy management in
flexibility potential. Sustainable Energy, Grids and Networks 2021.
a building district. J Process Control 2020;89:11–21.
[11] Wang S, Lu L, Han X, Ouyang M, Feng X. Virtual-battery based droop control and
[30] Hou X, Xiao Y, Cai J, Hu J, Braun J. Distributed model predictive control via
energy storage system size optimization of a DC microgrid for electric vehicle fast
Proximal Jacobian ADMM for building control applications. 2017 American
charging station, Appl Energy, v 259, 2020.
Control Conference (ACC). 2017.
[12] Wang Y, Gladwin DT. Power management of EV car parks. IECON 2019 – 45th
[31] Delfino F, Ferro G, Minciardi R, Robba M, Rossi M. Identification and optimal
Annual Conference of the IEEE Industrial Electronics Society, p 4316-4322,
control of an electrical storage system for microgrids with renewables. Sustainable
October 2019.
Energy Grids Networks 2019.
[13] Mostafa S, Amirhosein M, Mahdi AM. A daytime optimal stochastic energy
[32] Koliou E, Bartusch C, Picciariello A, Eklund T, Soder L, Hakvoort RA. Quantifying
management for EV commercial parking lots by using approximate dynamic
distribution-system operators’ economic incentives to promote residential demand
programming and hybrid big bang big crunch algorithm. Sustainable Cities Soc
response. Util Policy 2015;35:28–40.
2018;45:486–98.
[33] Ji Y, Xu P, Duan P, Lu X. Estimating hourly cooling load in commercial buildings
[14] Gamallo C, Álvaro R, Ardanuy JF, González J, Fuentes M. Evaluation of the
using a thermal network model and electricity submetering data. Appl Energy
Utilization of Electric Vehicles for Building Energy Management in Hotels,
2016;169:309–23.
International Conference on New Concepts in Smart Cities: Fostering Public and
[34] Kim IY, de Weck OL. Adaptive weighted sum method for multiobjective
Private Alliances (SmartMILE) Spain, 2013.
optimization: a new method for Pareto front generation. Springer Struct
[15] Momber I, Gómez T, Senior Member IEEE, Venkataramanan G, Senior
Multidiscip Optimiz 2006;31:105–16.
Member IEEE, Stadler M, et al. Plug-in electric vehicle interactions with a small
[35] Romvary JJ, Ferro G, Haider R, Annaswamy AM. A proximal atomic coordination
office building: an economic analysis using DER-CAM. Germany: IEEE PES General
algorithm for distributed optimization. IEEE Trans Autom Control 2021.
Meeting; 2010.
[36] Makhdoumi A, Ozdaglar A. Convergence rate of distributed ADMM over networks.
[16] Kouka K, Masmoudi A, Abdelkafi A, Krichen L. Dynamic energy management of an
IEEE Trans Autom Control 2016;62(10):5082–95.
electric vehicle charging station using photovoltaic power. Sustainable Energy,
[37] Boyd S, Parikh N, Chu E. Distributed optimization and statistical learning via the
Grids Networks 2020.
alternating direction method of multipliers, Now Publishers Inc, 2011.
[17] Khalkhali H, Hosseinian SH. Multi-class EV charging and performance-based
[38] Bracco S, Delfino F, Laiolo P, Morini A. Planning & open-air demonstrating smart
regulation service in a residential smart parking lot, Sustainable Energy, Grids and
city sustainable districts. Sustainability 2018;10(12):4636.
Networks, vol 22, 2020.
[39] Falvo MC, Sbordone D, Bayram IS, Devetsikiotis M. EV charging stations and
[18] Shoko K, Yoshihiko S, Atsushi I. A reactive management system for reliable power
modes: International standards, IEEE In 2014 International Symposium on Power
supply in a building microgrid with vehicle-to-grid interaction, Communications
Electronics, Electrical Drives, Automation and Motion, pp. 1134–1139, 2014.
and Computer Sciences, v E101A, n 8, p 1172-1184, 2018, IEICE Trans Fundam
[40] Drgoňa J, Klaučo M, Kvasnica M. MPC-based reference governors for
Electr Commun Comp Sci E101.A(8):1172–1184.
thermostatically controlled residential buildings. In: 2015 54th IEEE Conference on
[19] Dimitrios T, Olivier D, Ioakimidis, Christos S. Optimal operation of an energy
Decision and Control (CDC); 2015. p. 1334–9.
management system for a grid-connected smart building considering

18

You might also like