1 s2.0 S037877532301220X Main

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Journal of Power Sources 591 (2024) 233844

Contents lists available at ScienceDirect

Journal of Power Sources


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

Optimal multi-layer economical schedule for coordinated multiple mode


operation of wind–solar microgrids with hybrid energy storage systems
Muhammad Bakr Abdelghany a,b ,∗, Ahmed Al-Durra a , Zhou Daming c , Fei Gao d
a
Advanced Power and Energy Center, EECS Department, Khalifa University of Science and Technology, Abu Dhabi, United Arab Emirates
b
Computer and Systems Engineering Department, Faculty of Engineering, Minia University, Minia, Egypt
c
School of Astronautics, Northwestern Polytechnical University, Xi’an, 710072, China
d
School of Energy and Computer Sciences, University of Technology of Belfort-Montbeliard (UTBM), France

HIGHLIGHTS

• Implementation of a sophisticated hybrid energy storage system dynamic model.


• Design of a multi-layer MPC aimed at efficiently fulfilling hydrogen and electricity demands.
• Management of uncertainties of RESs through the stochastic MPC methodology.
• Integration of the MPC with a lab-scale microgrid benchmark.

ARTICLE INFO ABSTRACT

Keywords: The aim of this paper is the design and implementation of an advanced model predictive control (MPC) strategy
Re-electrification facilities for the management of a wind–solar microgrid (MG) both in the islanded and grid-connected modes. The MG
Demand-side management includes energy storage systems (ESSs) and interacts with external hydrogen and electricity consumers as an
Software-defined control
extra feature. The system participates in two different electricity markets, i.e., the daily and real-time markets,
Hydrogen applications
characterized by different time-scales. Thus, a high-layer control (HLC) and a low-layer control (LLC) are
Optimal economical schedule
developed for the daily market and the real-time market, respectively. The sporadic characteristics of renewable
energy sources and the variations in load demand are also briefly discussed by proposing a controller based
on the stochastic MPC approach. Numerical simulations with real wind and solar generation profiles and spot
prices show that the proposed controller optimally manages the ESSs, even when there is a deviation between
the predicted scenario determined at the HLC and the real-time one managed by the LLC. Finally, the strategy
is tested on a lab-scale MG set up at Khalifa University, Abu Dhabi, UAE.

1. Introduction In contrast, hydrogen-ESSs (HESSs) have emerged as promising long-


term ESSs because of their high energy density, ease of storage, and
Microgrids (MGs) offer a viable solution to ensure the resilience environmental compatibility [4].
of power systems in the emerging era of renewable energy. Indeed, An MG has the capability to function in two primary operational
in recent years, the integration of renewable energy sources (RESs), modes: the islanded mode and the grid-connected mode. In the is-
including solar and wind sources, has grown exponentially, as part landed mode, its aim is to uphold power equilibrium for local loads
of a global effort to reduce carbon emissions [1]. However, the in- autonomously, without relying on external grid assistance [5]. Con-
termittent nature of these sources has a substantial impact on both versely, in the grid-connected mode, the MG is configured to ac-
the quality of power and the profitability of the systems that incor- tively participate in electricity markets with the objective of optimizing
porate both RESs and fossil fuels. In order to overcome these chal- revenue generation whenever feasible. Moreover, in order to further
lenges, the integration of energy storage systems (ESSs) with RESs
improve economic benefits, in the local market, an MG can exchange
is imposed [2]. Battery-ESSs (BESSs) exhibit self-discharge and low
energy with external agents, i.e., fuel cell electric vehicles (FCEVs)
energy density, rendering them unsuitable for long-term storage [3].

∗ Corresponding author at: Advanced Power and Energy Center, EECS Department, Khalifa University of Science and Technology, Abu Dhabi, United Arab
Emirates.
E-mail addresses: abdelghany.muhammad@gmail.com (M.B. Abdelghany), ahmed.aldurra@ku.ac.ae (A. Al-Durra), daming.zhou@nwpu.edu.cn (Z. Daming),
fei.gao@utbm.fr (F. Gao).

https://doi.org/10.1016/j.jpowsour.2023.233844
Received 23 September 2023; Received in revised form 27 October 2023; Accepted 7 November 2023
Available online 16 November 2023
0378-7753/© 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

and battery electric vehicles (BEVs), by implying that the MG can strategies, however, include simple models for the operations of hydro-
operate in two additional modes [6]. This research study provides gen devices and batteries by excluding aspects, e.g., the cold-start and
control strategies according to the model predictive control (MPC) warm-start processes.
approach [7] for the management of a wind–solar MG operating in the In order to fill the several gaps reported above, hierarchical ap-
four modes, by taking into account operating and maintenance costs, proaches that account for several tasks of the devices involved in MGs
physical constraints, and degradation aspects. The operations of the have been proposed. For instance, the authors in [29] have integrated
ESSs and the interaction with the utility grid are described using the a cascaded MPC approach in a grid-connected MG powered by RESs.
mixed logical dynamical (MLD) framework [8]. The upper MPC optimizes the MG’s scheduling, while the lower con-
The islanded mode, where the MG operates autonomously, can ef- troller adeptly oversees revenue generation by selling excess power in
fectively facilitate the maintenance of power balance for the requested the energy market, taking into account the efficient use of RESs and
demands, improve the system’s resilience, optimize energy efficiency, ESSs. Furthermore, the authors in [30] have proposed a hierarchical
and mitigate the associated costs [5,9]. In [10,11], the MPC and MPC enhanced by data-driven algorithms to improve the operational
heuristic methods for the energy management of an islanded MG, which performance of MGs paired with hybrid ESSs. In order to increase the
includes RESs and hybrid ESSs, have been proposed with the goal of number of EVs and their effect on the infrastructure, a multi-layer
enhancing the operational of MG efficiency. The authors in [12] have economic MPC framework for the operation of an MG with EVs has
proposed some MPC strategies to guarantee the stability of islanded been developed in [31]. Additionally, a multi-layer coordinated MPC
MGs including ESSs. The controllers can mitigate the lack of inertia of a standalone MG integrated with EVs has been presented in [32]
under transitory conditions and enhance the dynamic properties of volt- to simultaneously charge and discharge EVs. However, the mentioned
age and frequency fluctuations under loading transition. An MPC-based works do not include the impact of constraints, such as the transient
strategy has been proposed in [13] for the management of islanded MGs states of the devices and the minimum working cycles, which can have
including photovoltaic (PV). However, the strategies mentioned above an adverse effect on the devices’ lifespan and short-time scale dynamics.
only focus on islanded wind–solar MGs and do not take into account the Therefore, this study presents a novel approach that addresses these
potential benefits of the grid-connected mode or the energy exchange constraints and evaluates their effect.
with external entities. An efficient control strategy for the management of MGs requires
The primary goal of an MG in the grid-connected mode is the max- the consideration of operating and maintenance costs. Control strate-
imization of power sold revenues. The authors in [14] have developed gies, which take into account the HESS acquisition costs, the degra-
an MPC strategy for an MG composed of a BESS and PV panels in order dation of the devices’ lifespan, the operation and maintenance costs,
to maximize profits. Furthermore, an optimal method that combines the and the costs of the energy markets, have been proposed in [33] to
𝜀−variable approach with the switched-MPC has been proposed in [15] provide the electric reference requested by users. The results have
to mitigate operational costs and maximize revenues of grid-connected shown that the controllers significantly minimize the running costs
MGs. A sophisticated rule-based MPC approach for a grid-connected of the MG. The effectiveness of a superconducting magnetic-ESS for
MG composed of ESSs has been developed in [16] to track the fast power smoothing has been studied in [34]. In order to increase the
dynamics. In [17], an MPC strategy, involving a grid-connected MG in lifespan of a BESS, a strategy that reduces its charge/discharge cycles
conjunction with a HESS, has been integrated to facilitate an advanced has been proposed in [35]. However, the authors do not include in
demand-side management solution aimed at efficiently catering to the their computations the initial capital costs of the equipment and the
energy requirements of both local and contractual loads. In [18], an decrease in efficiency resulting from switching between different states
MPC approach based on an online optimization problem has been (e.g., ON/OFF). In [36], the short- and long-term optimal schedules of
implemented for the management of an MG that incorporates ESSs an MG consisting of RESs and ESSs have been taken into account by
and exchanges electricity in the energy market through the utility grid. an MPC strategy to satisfy the electrical load demand and minimize
An MPC strategy for the participation of hydrogen-based MGs in the operating costs. In order to maximize the lifespan of the electrolyser
intraday energy market has been presented in [19] by taking into and the fuel cell, an MPC architecture that controls an MG composed
account economic and environmental aspects. However, the mentioned of RESs has been proposed and validated in [37] by considering the
strategies do not consider the possible benefits of including a hierar- degradation of the devices. An MPC technique that considers costs and
chical control to compensate for any deviations which can occur in degradation issues has been presented in [38]. The method restricts the
the real-time management of a wind–solar MG paired with HESSs and excessive usage of the electrolyser and the fuel cell by operating the
BESSs. devices to run within a fixed power range. A mixed integrate linear
In the local market, an MG can also exchange energy with external programming approach and a hierarchical control method have been
agents, e.g., other MGs, BEVs, or FCEVs, to further improve economic proposed in [39] to minimize degradations of the fuel cell and the
benefits [20]. The problem of the optimal schedule for an MG that in- battery and maintenance and operation costs in the applications of
teracts with external consumers has been studied in [21] by considering marine transportation.
the predictions of both energy prices and RESs. Similarly, in [22], an The sporadic nature of RESs and the necessity to predict energy
optimal scheduling strategy for charging a BEV using vehicle-to-grid prices and electric loads require the implementation of sophisticated
connectivity has been provided to minimize costs and charging time. energy management systems. A sophisticated MPC strategy has been
Moreover, a two-phase strategy that optimizes the charge has been introduced in [40] to enhance the efficiency of isolated MGs by ef-
proposed in [23], and a strategy for the management of an electric fectively addressing uncertainties in RESs and electric load profiles.
vehicle as ESS without including the possible energy exchange with Furthermore, the authors in [41] have proposed a stochastic program-
external consumers has been presented in [24]. In order to coordinate ming method for optimal scheduling of a multi-energy MG with RESs
the day-ahead schedule and the real-time scenario, a two-stage MPC and ESSs by considering uncertainties and incorporating contracts be-
of a solar MG that charges plug-in BEVs and refills FCEVs has been tween producers and consumers to decrease the risk of high operating
developed in [25]. Moreover, the frequency stabilization of an MG costs. A stochastic multi-objective optimization for an MG with RESs
used for the charge/discharge of electric vehicles has been controlled and ESSs has been developed in [42] in order to reduce investment
in [26]. In [27], a control approach involves the integration of wind costs. The authors in [43] have introduced the stochastic framework
energy into the grid by coupling wind farms with HESSs to produce to maximize the benefits of MGs while enhancing their flexibility and
green hydrogen as a fuel for FCEVs and for electric and contractual responsiveness. Moreover, the management of MGs has been addressed
demands. The game theory approach has been proposed in [28] to in [44] using the chance-constrained MPC approach by accounting for
evaluate the prices of electric vehicle charging stations. The reported the uncertainties in load demands. In contrast to the prior literature,

2
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

the present study treats RESs as exogenous signals beyond the control warm-start and cold-start processes are included in such models to
of the strategy, since it is assumed that the owners of the RES systems capture the slow responses of the devices. However, unlike [27], where
provide the day-ahead schedule that best fits the energy generation. two virtual states are introduced for the modeling of such processes, in
However, a stochastic MPC (SMPC) with two scenarios is proposed to our paper they are included through the definition of linear constraints,
show how the proposed control can handle the uncertainties when the thus reducing the number of decision variables in the controller. The
owners are unable to deliver the expected day-ahead schedule. main novelties of this paper are:
Heuristic and meta-heuristic approaches [45], including particle
(i) the use of an MLD-based model to describe the complex dynam-
swarm optimization (PSO), differential evolution, and genetic algo-
ics of the storage devices, including their states, state transitions,
rithm, have been also proposed in the literature for MG management.
and waiting actions (cold-start and warm-start) required for
For instance, the authors in [46] have presented a power management
some state switches;
of a PV-wind MG paired with a BESS in order to meet dynamic load de-
(ii) the design of a control architecture that reduces the interactions
mands in both the grid-connected and islanded modes using an integer
with the grid, minimizes working operation costs, and man-
control approach. In [47], a new energy management system for a BESS
ages the overall system in four modes (islanded, grid-connected,
based on two stages of optimization has been developed to manage both
exchange of energy with BEVs and FCEVs);
active and reactive power flows. In [48], the authors have proposed
(iii) the management of two electricity markets characterized by
several deterministic and stochastic optimization approaches based on
different time-scales using a multi-layer MPC scheme which
the heuristic and fuzzy logic frameworks for solar–wind islanded MGs.
addresses any deviation between the forecasted and real-time
The authors in [49] have proposed a strategy to find the optimal
scenarios;
size and configuration of hybrid wind–solar MGs incorporating BESSs.
(iv) the management of uncertainties due to the sporadic nature of
This has been achieved through a hybrid intelligent computational
RESs through the implementation of an SMPC which considers
technique given by the combination of an improved PSO algorithm
an optimistic scenario and a pessimistic one;
with a sophisticated differential evolution model. Furthermore, in [50],
(v) the integration of the control architecture with an ad hoc lab-
the authors have designed a consensus-based distributed control strat-
scale system which includes emulators of the grid, loads, RESs,
egy for multiple ESSs to improve the system’s reliability. Unlike the
ESSs, and BEVs.
mentioned literature, this study employs numerical commercial solvers
to address mixed-integer quadratic programming (MIQP) optimization The rest of this study is organized as follows: Section 2 presents
problems, by avoiding the use of evolutionary algorithms. notations and preliminaries; the characteristics of the system under
The literature has also addressed the economic aspects of renewable analysis are described in Section 3; the model of its main operations is
energy production and its applications, such as hydrogen energy in given in Section 4; the control strategies for the four operation modes
industries like iron, steel, pipelines, and glass [51,52]. For example, are presented in Section 5; the validation of such strategies through
the authors in [53] have examined the use of hydrogen to mitigate numerical analysis and an experimental setup is shown in Section 6;
emissions in the iron and steel industry, with a specific focus on closing remarks and future directions are given in Section 7.
cost-effective and environmentally favorable applications, especially
in contexts of reduced electricity prices or in the presence of RESs. 2. Notation and mathematical preliminaries
Numerous countries are dedicating investments to renewable hydrogen
production to fulfill the objectives achieved in the Paris Agreement, as The notation and assumptions that will be applied consistently
highlighted by the International Energy Agency [51]. In this regard, the throughout the paper are presented in this section. Each mode is
authors in [54] have delineated four distinct production pathways for denoted by 𝑚 ∈ {1, 2, 3, 4}, which corresponds to islanded, on-grid,
the time span from 2015 to 2035. These pathways serve to elucidate interaction with BEVs, and interaction with FCEVs, respectively. More-
the consequences of consistent, increasing, and decreasing production over, in this paper, the two different approaches, namely the MPC
layers, as well as variations in production processes within Germany. and the SMPC, used for the controller, are denoted by 𝑠𝑙 = 1 and
The authors in [55] have examined the shift from fossil fuels to RESs 𝑠𝑙 = 2, respectively. In order to distinguish between the different
in the context of the Russia-Ukraine crisis for multiple industries. devices’ operations, the subscripts 𝑒, 𝑓 , ch, and dc are used to refer
This transition has also highlighted the importance of robust interna- to the hydrogen production through the electrolyser, the hydrogen
tional communication regarding energy security. As further industry consumption through the fuel cell, the battery charge, and the battery
applications, the authors in [56] have investigated the feasibility of discharge, respectively, and a generic operation is denoted by 𝑑. The
using heat pumps for industrial preheating, demonstrating potential operation set is  = {𝑒, 𝑓 , ch, dc}. The MPC strategies rely on automata-
cost savings of around 10% and a 20% reduction in hydrogen fuel based models, where the states are OFF, STB, and ON. Therefore, the
requirements. While commercial deployment awaits further develop- state set is defined as  = {OFF, STB, ON}, while the transition set is
ment, it holds promise for cost and emissions reduction. With respect to given by  = {(𝛼, 𝛽) ∣ 𝛼, 𝛽 ∈ , 𝛼 ≠ 𝛽}, with (𝛼, 𝛽) representing the
techno-economic analysis, the authors in [57] have provided a detailed switch from the state 𝛼 to the state 𝛽. Moreover, 𝛶𝑑𝛼 , 𝛥𝛽𝛼,𝑑 , 𝛤𝑑𝛼 , 𝛬𝛼𝑑 ,
≥𝜉
analysis of the effects of existing RES policies in India, including the 𝜁𝑑 𝑚 , and 𝜁𝑑≤𝜉𝑚 ∈ {0, 1} are Boolean variables. The NOT, AND, and OR
newly implemented hybrid policy. The study has evaluated how these operators are denoted by ¬, ∧, and ∨, respectively. Table 1 lists all the
policies have impacted the techno-economic aspects of ongoing projects abbreviations used throughout the paper.
and addressed challenges faced by developers, all aimed at accelerating A multi-layer architecture is employed in this study, and in par-
the deployment of wind and solar power. A comprehensive list of risk ticular the high-layer control (HLC) and the low-layer control (LLC)
factors within the green hydrogen supply chain in the EU has been are denoted by the superscripts ℎ and 𝓁, respectively; when the two
given in [58]. This seeks to proactively address and mitigate challenges controllers are not distinguished, the generic superscript 𝑛 is used. Ac-
and obstacles that impede the establishment of green hydrogen as a cording to the multi-layer MPC, two constrained optimization problems
prominent energy carrier in EU markets. are solved at each time-step, where each problem is characterized by
To the best of the authors’ knowledge, all the studies reported above a different sample time. In particular, 𝜏 ℎ = 1 h and 𝜏 𝓁 = 10 min are
do not address the short-term and long-term optimal schedules for the the sample times for the HLC and the LLC, respectively. The prediction
HESS and the BESS which include degradation issues, working cycles horizons used by the two controllers are also different and denoted
of the devices, constraints on the ESSs’ operations, and operation and by 𝑇 ℎ and 𝑇 𝓁 , respectively. In this research study, the formulations
management costs. In our research paper, the MLD framework is used are presented using the variables ℎ and 𝓁 to denote the discrete-
for modeling the operations of the BESS and the HESS, and the so-called time-scales for the HLC and the LLC, respectively. The corresponding

3
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

Table 1
Abbreviations.
Acronyms Abbreviations
BESS Battery energy storage system
BEV Battery electrical vehicle
ESS Energy storage system
FCEV Fuel cell electric vehicle
HESS Hydrogen energy storage system
HLC High-layer control
MG Microgrid
MIQP Mixed-integer quadratic programming
MLD Mixed logical dynamical
MPC Model predictive control
LLC Low-layer control
PV Photovoltaic
PSO Particle swarm optimization
RES Renewable energy source
SMPC Stochastic model predictive control
STB Standby
Fig. 2. Control architecture.

2. the grid-connected mode, represented as 𝑚 = 2, which involves


the supplying power to the loads while also engaging in energy
exchange (either selling or purchasing) with the utility grid;
3. the interconnection with BEVs, denoted by 𝑚 = 3, in which the
power plant can exchange power to meet the energy demands of
BEVs;
4. the interconnection with FCEVs, indicated by 𝑚 = 4, in which the
power plant can exchange power to meet the hydrogen demands
of FCEVs.

In order to address the specific objectives of the four operating modes,


Fig. 1. MG with external agents.
the control system is decomposed into four configurations with differ-
ent objectives, as shown in Fig. 2. Each control strategy is activated
mutually exclusively when the plant is running. In particular, the
continuous time by means of 𝑇 ℎ (𝑇 𝓁 ) is obtained by setting 𝑇 ℎ = ℎ𝜏 ℎ selection of the operating mode is entirely controlled by a system
(𝑇 𝓁 = 𝓁𝜏 𝓁 ) for the HLC (LLC). When referring to both the LLC and operator, who can also monitor the system performance using user-
the HLC, the time instants and the sample times are denoted by 𝑘 and friendly interfaces in desktop and web-based applications. Then, the
𝜏 𝑛 , respectively, even though the equations may pertain to different current mode is activated through a switched function linked to the se-
time-scales. lection of the operator. The operator’s role in mode selection is crucial
since it involves the consideration of factors, such as forecasted data,
economic circumstances, environmental objectives, resilience needs,
3. Operations and functions
and user preferences.
The requirements and objectives of the scenario under investigation
The system under analysis uses wind and solar energies as its are pursued at two different time-scales, and then two correspond-
primary sources. As depicted in the conceptual representation in Fig. 1, ing controllers are developed. In particular, the larger and shorter
the system comprises PV panels, a wind farm, a BESS, a HESS, BEVs, time-scales are addressed by the HLC and the LLC, respectively. The
FCEVs, and local demands. The HESS is comprised of an electrolyser, reasoning behind this approach involves using long-term forecasts to es-
a fuel cell, and a tank, which allow the storage and conversion of tablish an initial schedule on larger time-scales. This initial schedule is
hydrogen. In Fig. 1, the wind and solar powers are indicated by 𝑃𝑤 then subject to real-time adjustments for shorter time-scales. Although
and 𝑃pv , respectively, 𝑃ev is the power generated by BEVs, the power both controllers pursue analogous objectives and constraints, these are
input supplied to the electrolyser is denoted by 𝑃𝑒 , the power generated customized to their specific time-scales.
by the fuel cell is 𝑃𝑓 , the power stored in the battery is indicated In this study, the RES forecasts are considered as exogenous signals,
by 𝑃𝑏 , and 𝑃𝑙 is the electric demand. Moreover, 𝑃𝑔 and 𝑃𝑠 denote and then their availability is not predicted by our control algorithm.
the grid and available powers, respectively, and 𝐻𝑙 is the hydrogen Indeed, it is assumed that the owners of the RES systems directly
requested by the hydrogen facilities. Load requests are satisfied by the provide the best day-ahead schedules obtained through sophisticated
methods. However, in order to take into account the rare case of
available electricity and any excess of RES energy is used as a backup
unavailability of the owners, we develop also a scenario-based SMPC
source by the electrolyser for hydrogen production, or by the BESS
which can be integrated into the system to handle the uncertainties.
to charge its battery. When there is limited or negligible wind/solar
In this study, the uncertainties are integrated into the forecast process
energy availability, the fuel cell re-electrifies the stored hydrogen, or by incorporating an uncertainty band to the deterministic profile. This
the stored battery energy is used for loads. results in the definition of two scenarios: the optimistic one and the
The main objectives of the system include providing a reliable and pessimistic one. Similarly to the selection of the modes, this option is
consistent power supply to the electric demand in both the islanded activated by the system operator, based on the availability of forecasts
and grid-supported modes and facilitating energy exchange when it from the owners of wind farms and/or PV panels.
interacts with BEVs and FCEVs. Thus, four possible modes are taken
into account: 4. System modeling

1. the autonomous mode, marked as 𝑚 = 1, in which the power The development of the multi-layer control strategies for the four
plant can fulfill the required load using the available system operations modes requires, as a first step, the definition of the model
power without relying on the support of the utility grid; of the system under investigation. In order to simplify the notation, the

4
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

The decision variables 𝛶𝑑𝛼,𝑛 , with 𝑑 ∈ , 𝑛 ∈ {ℎ, 𝓁}, and 𝛼 ∈ , are


connected to the power corresponding to each state, as follows

𝑃𝑑𝑛 (𝑘) = 0 ⟺ 𝛶𝑑OFF,𝑛 (𝑘) = 1, (1a)


𝑃𝑑𝑛 (𝑘) = 𝑃𝑑STB ⟺ 𝛶𝑑STB,𝑛 (𝑘) = 1, (1b)
𝑃𝑑𝑛 (𝑘) ∈ [𝑃𝑑𝑚 , 𝑃𝑑𝑀 ] ⟺ 𝛶𝑑ON,𝑛 (𝑘) = 1, (1c)

where 𝑃𝑑𝑛 is the operation power, and 𝑃𝑑𝑚 (𝑃𝑑𝑀 ) is the lower (upper)
bound of 𝑃𝑑𝑛 . The condition 𝑃𝑑𝑛 (𝑘) ∈ [𝑃𝑑𝑚 , 𝑃𝑑𝑀 ] in (1c) limits the powers
of the ESSs when their state is ON to reduce devices’ damages. In order
to solve the controller through numerical tools, the constraints in (1)
are equivalently expressed with mixed-integer inequalities, according
to the MLD framework [60]. To do this, the left-side expressions in (1)
are rewritten by introducing 𝜁 ≥𝜏𝑚 and 𝜁 ≤𝜏𝑀 defined as
{
≥𝜏𝑚 ,𝑛 1 𝑃𝑑𝑛 (𝑘) ≥ 𝜏𝑚 ,
Fig. 3. Sequential graph (automaton) for the ESSs’ operations. 𝜁𝑑 (𝑘) = (2a)
0 𝑃𝑑𝑛 (𝑘) < 𝜏𝑚 ,
{
≤𝜏 ,𝑛 0 𝑃𝑑𝑛 (𝑘) > 𝜏𝑀 ,
𝜁𝑑 𝑀 (𝑘) = (2b)
same time variable 𝑘 and layer 𝑛 are used in the MLD formulation of the 1 𝑃𝑑𝑛 (𝑘) ≤ 𝜏𝑀 ,
devices, although the model will be included in both the HLC and the
LLC, which addresses two different time-scales. The model presented with 𝜏𝑚 ∈ {0, 𝑃𝑑STB , 𝑃𝑑𝑚 } and 𝜏𝑀 ∈ {0, 𝑃𝑑STB , 𝑃𝑑𝑀 }. Based on the MLD
below derives from the one proposed in [59], where it is included in a framework, the expressions in (2) can be written as
single-layer control.
𝑃𝑑𝑛 (𝑘) − 𝜏𝑚 < 𝑀𝜁 ≥𝜏𝑚 ,𝑛 (𝑘), (3a)
≥𝜏𝑚 ,𝑛
4.1. Model of energy storage systems −𝑃𝑑𝑛 (𝑘) + 𝜏𝑚 ≤ 𝑀(1 − 𝜁 (𝑘)), (3b)

Fig. 3 shows that the operations of the ESSs, i.e., the hydrogen −𝑃𝑑𝑛 (𝑘) + 𝜏𝑀 < 𝑀𝜁 ≤𝜏𝑀 ,𝑛 (𝑘), (3c)
≤𝜏𝑀 ,𝑛
production/consumption and the battery charge/discharge, are repre- 𝑃𝑑𝑛 (𝑘) − 𝜏𝑀 ≤ 𝑀(1 − 𝜁 (𝑘)), (3d)
sented by automata whose nodes and edges denote the states and the
switches, respectively. In particular, the model includes four automata, with 𝑀 (−𝑀) being the upper (lower) limit of 𝑃𝑑𝑛 − 𝜏𝑀 (𝑃𝑑𝑛 − 𝜏𝑚 ). Each
one for each operation. constrain defined in (1) can be now recast as
The three possible states for the operations are ON, OFF, and STB ≥𝜏𝑚 ,𝑛
(1 − 𝛶𝑑𝛼,𝑛 (𝑘)) + 𝜁𝑑 (𝑘) ≥ 1, (4a)
(standby). If the electrolyser (fuel cell) is producing (consuming) hy-
≤𝜏 ,𝑛
drogen, and the BESS is being charged or discharged, the corresponding (1 − 𝛶𝑑𝛼,𝑛 (𝑘)) + 𝜁𝑑 𝑀 (𝑘) ≥ 1. (4b)
automaton is set to ON. Conversely, the state is OFF when the hydrogen ∑
devices are not producing or consuming hydrogen, and the battery is Finally, the variables 𝛶𝑑𝛼,𝑛
are mutually exclusive, i.e., 𝛼,𝑛
𝛼∈ 𝛶𝑑 (𝑘) =1
not charging or discharging. Finally, STB depicts a scenario where the for all 𝑑 ∈  and 𝑛 ∈ {ℎ, 𝓁}.
corresponding device absorbs power, but is not operating. Moreover,
due to the devices’ slow response, two state switches necessitate a 4.1.2. State transitions
waiting time period: the switch (OFF, STB) is enabled after a cold- The variables 𝛶𝑑𝛼,𝑛 determine the values of 𝛤𝑑𝛼,𝑛 and 𝛥𝛽,𝑛 𝛼,𝑑
at each
start process, and similarly the transition (STB, ON) is activated after time-step. In particular, for short time-scales, the switches (OFF, STB)
a warm-start process. Such processes disallow the direct switch (OFF, and (STB, ON) are allowed after a time interval required for the execu-
ON), which is represented by a dashed edge in Fig. 3. The cold-start and tion of the cold-start and the warm-start, respectively. These constraints
warm-start processes take some minutes, and then they are neglected are introduced by determining the variables 𝛥STB,𝑛
OFF,𝑑
ON,𝑛
and 𝛥STB,𝑑 as
in the HLC, which is characterized by a time-scale of the order of hours.
Logical and continuous variables can be introduced to describe the 𝛥STB,𝑛
OFF,𝑑
(𝑘) = 𝛶𝑑STB,𝑛 (𝑘 − 𝜏 𝑐 ) ∧ ⋯ ∧ 𝛶𝑑STB,𝑛 (𝑘) (5a)
automata, according to the MLD framework [60]. Thus, the decision ∧ 𝛤𝑑OFF,𝑛 (𝑘 − 𝜏 𝑐 ) ∧ ⋯ ∧ 𝛤𝑑OFF,𝑛 (𝑘 − 1),
variables 𝛤𝑑𝛼,𝑛 and 𝛥𝛽,𝑛 , with 𝑑 ∈ , 𝑛 ∈ {ℎ, 𝓁}, and 𝛼, 𝛽 ∈ , refer to
𝛼,𝑑
the states and the switches of the automaton, respectively. Specifically, 𝛥ON,𝑛
STB,𝑑
(𝑘) = 𝛶𝑑ON,𝑛 (𝑘 − 𝜏 𝑤 ) ∧ ⋯ ∧ 𝛶𝑑ON,𝑛 (𝑘) (5b)
𝛤𝑑𝛼,𝑛 is introduced such that 𝛤𝑑𝛼,𝑛 (𝑘) = 1 if the state of the automaton ∧ 𝛤𝑑STB,𝑛 (𝑘 − 𝜏 𝑤 ) ∧ ⋯ ∧ 𝛤𝑑STB,𝑛 (𝑘 − 1),
describing the operation 𝑑 is 𝛼 at the time instant 𝑘, and 𝛤𝑑𝛼,𝑛 (𝑘) = 0
otherwise. Furthermore, 𝛥𝛽,𝑛 is such that 𝛥𝛽,𝑛 (𝑘) = 1 if the transition with 𝜏 𝑐 (𝜏 𝑤 ) being the time interval for the cold (warm) start process.
𝛼,𝑑 𝛼,𝑑
(𝛼, 𝛽) occurs at the time 𝑘 for the automaton related to the operation 𝑑, However, in the HLC the waiting actions are neglected because of its
and 𝛥𝛽,𝑛 (𝑘) = 0 otherwise. large time-scales. Thus, the waiting intervals at the HLC are set equal
𝛼,𝑑
to zero, i.e., 𝜏 𝑐 = 𝜏 𝑤 = 0. Following the MLD equivalences, (5a) can be
4.1.1. State actions rewritten as
As discussed above, the state transitions (STB, ON) and (OFF, STB) 𝛥STB,𝑛
OFF,𝑑
(𝑘) ≤ 𝛶𝑑STB,𝑛 (𝑘 − 𝜏 𝑐 ),
are characterized by their corresponding waiting processes because of
the slow response of the equipment. Then, these actions are considered ⋮
by introducing the logical variable 𝛶𝑑𝛼,𝑛 , with 𝑑 ∈ , 𝑛 ∈ {ℎ, 𝓁}, and 𝛥STB,𝑛
OFF,𝑑
(𝑘) ≤ 𝛶𝑑STB,𝑛 (𝑘 − 1),
𝛼 ∈ , such that 𝛶𝑑𝛼,𝑛 (𝑘) = 1 if the operation 𝑑 intends to be in 𝛼 at the
𝛥STB,𝑛 (𝑘) ≤ 𝛤𝑑OFF,𝑛 (𝑘 − 𝜏 𝑐 ),
time instant 𝑘, and 𝛶𝑑𝛼,𝑛 (𝑘) = 0 otherwise. For instance, 𝛶𝑑STB,𝑛 (𝑘) = 1 OFF,𝑑
(6)
and 𝛤𝑑OFF,𝑛 (𝑘) = 1 represents the case where the operation 𝑑 is currently ⋮
in OFF and is intended to transition to STB. However, due to the cold- 𝛥STB,𝑛
OFF,𝑑
(𝑘) ≤ 𝛤𝑑OFF,𝑛 (𝑘 − 1),
start, it remains in OFF without making the shift to STB. Conversely,
𝛶𝑑STB,𝑛 (𝑘) = 1 and 𝛤𝑑STB,𝑛 (𝑘) = 1 represents that the operation 𝑑 is in 𝛥STB,𝑛
OFF,𝑑
(𝑘) ≥ 𝛤𝑑OFF,𝑛 (𝑘 − 1) + ⋯ + 𝛤𝑑OFF,𝑛 (𝑘 − 𝜏 𝑐 )
STB and remains unchanged, without any intended state transitions. + 𝛶𝑑STB,𝑛 (𝑘 − 1) + ⋯ + 𝛶𝑑STB,𝑛 (𝑘 − 𝜏 𝑐 ) + 2𝜏 𝑐 − 1,

5
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

and (5b) can be recast in a similar way. by the HLC and passed as a reference to the LLC to mitigate possible
Since there are no waiting actions involved in the other transitions, mismatches in the grid power. The definition (13) can be equivalently
one can define rewritten as

𝛥𝛽,𝑛
𝛼,𝑑
(𝑘) = 𝛤𝑑𝛼,𝑛 (𝑘 − 1) ∧ 𝛶𝑑𝛽,𝑛 (𝑘) (7) [𝜗𝑛𝑔 (𝑘) = 1] ⟺ [𝑃𝑔𝑛 (𝑘) < 𝜏𝑔 ] ∨ [𝑃𝑔𝑛 (𝑘) > 𝜏𝑔 ], (14)

for (𝛼, 𝛽) ∈ {(STB, OFF), (ON, STB), (ON, OFF)}. The definition (7) is that can be formulated through the definition of the variables
rewritten as
[𝜗𝑛𝑏 (𝑘) = 1] ⟺ [𝑃𝑔𝑛 (𝑘) > 𝜏𝑔 ], (15a)
𝛥𝛼,𝑛
𝛽,𝑑
(𝑘) ≤ 𝛤𝑑𝛽,𝑛 (𝑘 − 1), (8a)
[𝜗𝑛𝑠 (𝑘) = 1] ⟺ [𝑃𝑔𝑛 (𝑘) < 𝜏𝑔 ]. (15b)
𝛥𝛼,𝑛
𝛽,𝑑
(𝑘) ≤ 𝛶𝑑𝛼,𝑛 (𝑘), (8b)
These definitions can be also written as
𝛥𝛼,𝑛
𝛽,𝑑
(𝑘) ≥ 𝛤𝑑𝛽,𝑛 (𝑘 − 1) + 𝛶𝑑𝛼,𝑛 (𝑘) − 1. (8c) {
𝑛 1 𝑃𝑔𝑛 (𝑘) > 𝜏𝑔 ,
Additionally, the model incorporates the inadmissibility of the switch 𝜗𝑏 (𝑘) = (16a)
ON,𝑛 0 𝑃𝑔𝑛 (𝑘) ≤ 𝜏𝑔 ,
(OFF, ON) by setting 𝛥OFF,𝑑 (𝑘) = 0 for all 𝑘, 𝑑 ∈ , and 𝑛 ∈ {ℎ, 𝓁}. {
Finally, for each operation 𝑑 ∈ , the constrain 0 𝑃𝑔𝑛 (𝑘) ≥ 𝜏𝑔 ,
∑ 𝛽,𝑛 𝜗𝑛𝑠 (𝑘) = (16b)
1 𝑃𝑔𝑛 (𝑘) < 𝜏𝑔 ,
𝛥𝛼,𝑑 (𝑘) ≤ 1 (9)
(𝛼,𝛽)∈
which are equivalent to
is defined because at most one switch can be activated at each time
instant. 𝑃𝑔𝑛 (𝑘) − 𝜏𝑔 ≤ 𝑀𝑔 + 𝜗𝑛𝑏 (𝑘),
(17a)
−𝑃𝑔𝑛 (𝑘) + 𝜏𝑔 ≤ 𝑀𝑔 (1 − 𝜗𝑛𝑏 (𝑘)),
4.1.3. State selection −𝑃𝑔𝑛 (𝑘) + 𝜏𝑔 ≤ 𝑀𝑔 𝜗𝑛𝑠 (𝑘),
The final stage in modeling the automata involves determining the (17b)
state variables 𝛤𝑑𝛼,𝑛 from the variables 𝛥𝛽,𝑛 , i.e., 𝑃𝑔𝑛 (𝑘) − 𝜏𝑔 ≤ 𝑀𝑔 (1 − 𝜗𝑛𝑠 (𝑘)),
𝛼,𝑑
⋁ ( ⋁ ) where 𝑀𝑔 (−𝑀𝑔 ) is the upper (lower) bound of 𝑃𝑔𝑛 − 𝜏𝑔 . The auxiliary
𝛤𝑑𝛼,𝑛 (𝑘) = 𝛥𝛼,𝑛
𝛽,𝑑
(𝑘) ∨ ¬ 𝛥𝛾,𝑛
𝛽,𝑑
(𝑘) ∧ 𝛤𝑑𝛼,𝑛 (𝑘 − 1) . (10) variables 𝜁𝑏𝑛 and 𝜁𝑠𝑛 are used to represent the connection with the grid,
𝛽∈⧵{𝛼} (𝛽,𝛾)∈
i.e.,
Based on this definition, the state of the automaton is 𝛼 at the time-
step 𝑘 in two cases: first, the state was not 𝛼 at the time instant 𝑘−1, and 𝜁𝑏𝑛 (𝑘) = (𝑃𝑔𝑛 (𝑘) − 𝜏𝑔 )𝜗𝑛𝑏 (𝑘), (18a)
one of the switches leading to 𝛼 becomes active at the time 𝑘; second,
the state was set to 𝛼 at the time-step 𝑘 − 1, no switches are activated at 𝜁𝑠𝑛 (𝑘) = −(𝑃𝑔𝑛 (𝑘) − 𝜏𝑔 )𝜗𝑛𝑠 (𝑘). (18b)
the time-step 𝑘, and the state remains unaltered. The expression (10)
is equivalent to The definition (18a) is rewritten as
∑ ( ∑ 𝛾,𝑛 )
𝛤𝑑𝛼,𝑛 (𝑘) = 𝛥𝛼,𝑛 (𝑘) + (1 − 𝛥𝛽,𝑑 (𝑘))𝛤𝑑𝛼,𝑛 (𝑘 − 1) . (11) 𝜁𝑏𝑛 (𝑘) ≥ −𝑀𝑔 𝜗𝑛𝑏 (𝑘), (19a)
𝛽,𝑑
𝛽∈⧵{𝛼} (𝛽,𝛾)∈
𝜁𝑏𝑛 (𝑘) ≤ 𝑀𝑔 𝜗𝑛𝑏 (𝑘), (19b)
The product of decision variables introduces a nonlinearity in the
𝜁𝑏𝑛 (𝑘) ≥ (𝑃𝑔𝑛 (𝑘) − 𝜏𝑔 ) − 𝑀𝑔 (1 − 𝜗𝑛𝑏 (𝑘)), (19c)
integrated system, leading to the definition of the decision variables

𝛬𝛼,𝑛
𝑑
(𝑘) = (1 − (𝛽,𝛾)∈ 𝛥𝛾,𝑛
𝛽,𝑑
(𝑘))𝛤𝑑𝛼,𝑛 (𝑘 − 1). Using the MLD equivalences, 𝜁𝑏𝑛 (𝑘) ≤ (𝑃𝑔𝑛 (𝑘) − 𝜏𝑔 ) − 𝑀𝑔 (1 − 𝜗𝑛𝑏 (𝑘)), (19d)
such variables can be equivalently rewritten as
∑ 𝛾,𝑛 and (18b) as
𝛬𝛼,𝑛 (𝑘) ≤ (1 − 𝛥𝛽,𝑑 (𝑘)), (12a)
𝑑
(𝛽,𝛾)∈
𝜁𝑠𝑛 (𝑘) ≤ −𝑀𝑔 𝜗𝑛𝑠 (𝑘), (20a)

𝛬𝛼,𝑛 (𝑘) ≤ 𝛤𝑑𝛼,𝑛 (𝑘 − 1), (12b) 𝜁𝑠𝑛 (𝑘) ≥ 𝑀𝑔 𝜗𝑛𝑠 (𝑘), (20b)
𝑑

𝛼,𝑛
𝛬𝑑 (𝑘) ≥ 𝛤𝑑𝛼,𝑛 (𝑘 − 1) − 𝛥𝛾,𝑛 (𝑘). (12c) 𝜁𝑠𝑛 (𝑘) ≥ (𝑃𝑔𝑛 (𝑘) − 𝜏𝑔 ) − 𝑀𝑔 (1 − 𝜗𝑛𝑠 (𝑘)), (20c)
𝛽,𝑑
(𝛽,𝛾)∈
𝜁𝑠𝑛 (𝑘) ≤ (𝑃𝑔𝑛 (𝑘) − 𝜏𝑔 ) − 𝑀𝑔 (1 − 𝜗𝑛𝑠 (𝑘)). (20d)
In contrast to the two hydrogen operations, which can happen
simultaneously, the battery charging and discharging are mutually 4.3. Hydrogen dynamics
ON,𝑛 ON,𝑛
exclusive, meaning that the constraint 𝛤ch + 𝛤dc ≤ 1 is introduced.
The hydrogen in the tank 𝐻 𝑛 evolves according to
4.2. Utility grid model 𝜁𝑓𝑛 (𝑘)𝜏 𝑛
𝐻 𝑛 (𝑘 + 1) = 𝐻 𝑛 (𝑘) − 𝐻loss
𝑛
(𝑘) + 𝜂𝑒 𝜁𝑒𝑛 (𝑘)𝜏 𝑛 − , (21)
The system is developed to enable bidirectional energy flow, thus 𝜂𝑓
facilitating both energy purchase from and energy sale to the grid. 𝑛 is an
where 𝜂𝑒 (𝜂𝑓 ) is the efficiency of the electrolyser (fuel cell), 𝐻loss
This dynamic interaction with the utility grid is rigorously formulated 𝑛 𝑛 ON,𝑛 𝑛 𝑛 ON,𝑛
export term, 𝜁𝑒 (𝑘) = 𝑃𝑒 (𝑘)𝛤𝑒 (𝑘), and 𝜁𝑓 (𝑘) = 𝑃𝑓 (𝑘)𝛤𝑓 (𝑘).
through a logical variable denoted as 𝜗𝑛𝑔 . Specifically, 𝜗𝑛𝑔 is assigned a
The efficiencies 𝜂𝑒 and 𝜂𝑓 included in (21) are tuned through
value of 1 to indicate the grid-connected state and a value of 0 to denote
exhaustive numerical analysis. However, they can be adjusted over
the islanded state. The formal representation is given by
{ time by introducing, e.g., models that capture how they are affected
1, 𝑃𝑔𝑛 (𝑘) ≠ 𝜏𝑔 , by degradation as time progresses. An example of such a model is
𝜗𝑛𝑔 (𝑘) = (13) ( )
0, 𝑃𝑔𝑛 (𝑘) = 𝜏𝑔 , deg
𝜂𝑜 (𝑘 + 1) = 1 − 𝑀 𝑜 𝜁𝑜𝑛 (𝑘) 𝜂𝑜 (𝑘), (22)
where 𝑃𝑔𝑛 is the power that determines if the energy is being purchased 𝑃𝑜 HY𝑜
from (when 𝑃𝑔𝑛 (𝑘) > 𝜏𝑔 ) or sold to (when 𝑃𝑔𝑛 (𝑘) < 𝜏𝑔 ) the utility where 𝑜 ∈ {𝑒, 𝑓 }, deg𝑜 is the efficiency degradation of the device 𝑜, and
grid according to the set point references, 𝜏𝑔 = 0 for 𝑛 = ℎ (HLC), HY𝑜 denotes the operation hours of 𝑜. The model above emphasizes
𝜏𝑔 = (𝑃𝑔ℎ (𝑘))∗ for 𝑛 = 𝓁 (LLC), (𝑃𝑔ℎ (𝑘))∗ is the optimal schedule computed that degradation depends on both the power and operating hours (ON

6
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

state). However, this dynamic occurs over longer time-scales than those the four modes. The control architecture distinguishes between the
in this research study for MG management, and then the efficiencies can MG in the electricity markets without any energy/hydrogen exchange
be set as constants to avoid increasing the numerical complexity of the with external agents and the MG dispatched with BEVs or FCEVs,
controller. This does not preclude the option of periodically updating where further costs are introduced to incorporate the optimal track-
these values if a substantial deviation from the current adopted values ing of the power/hydrogen demand by external consumers. For each
is detected. mode, dedicated strategies are developed, optimizing operations such
as power-price forecasts and uncertains, and taking into account the
4.4. Electrical dynamics corresponding constraints. In particular, in the islanded mode, the
controller is implemented to distribute the energy such that a global
The energy soc𝑛 in the BESS evolves according to the state at cost function of the MG is minimized. For the grid-connected mode, the
the previous time instant and the power associated to charging and controller is implemented to find the optimal power schedule profiles
discharging, thus resulting in for the devices to deliver power to the grid and the load, minimize
𝜁𝑑𝑛 (𝑘)𝜏 𝑛 ESSs’ operational costs, and maximize power sold revenues. As regards
soc𝑛 (𝑘 + 1) = soc𝑛 (𝑘) + 𝜂𝑐 𝜁𝑐𝑛 (𝑘)𝜏 𝑛 − , (23) the interaction with BEVs, the controller addresses both the charge
𝜂𝑑
management during a parking time and the vehicle to the grid where
where 𝜂𝑐 (𝜂𝑑 ) is the efficiency of BESS charging (discharging), 𝜁𝑐𝑛 (𝑘) =
ON,𝑛 𝑛 = 𝑃 𝑛 (𝑘)𝛤 ON,𝑛 (𝑘), the power exchanged with the the battery of a vehicle can interact with the ESSs to maximize benefits.
−𝑃𝑏𝑛 (𝑘)𝛤ch (𝑘), 𝜁𝑑𝑐 𝑏 dc In the case of the interaction with FCEVs, the proposed control tracks
BESS is given by 𝑃𝑏𝑛 (𝑘) = 𝜁dc 𝑛 (𝑘) − 𝜁 𝑛 (𝑘), and the Boolean variables
ch
ON,𝑛 ON,𝑛 the requested hydrogen demand by FCEVs with the highest priority,
𝛤ch and 𝛤dc model the ON state of charging and discharging of the
and then satisfies the load and contracted loads. The control framework
BESS, respectively. The latters are related to the sign of 𝑃𝑏𝑛 (positive
relies on a switching signal that, upon a choice made by a system
(negative) for discharging (charging)), i.e., operator, determines the control strategy related to the selected mode.
ON,𝑛
[𝛤dc (𝑘) = 1] ⟺ [𝑃𝑏𝑛 (𝑘) ≥ 0], (24a)
ON,𝑛 5.1. Cost functions
[𝛤ch (𝑘) = 1] ⟺ [𝑃𝑏𝑛 (𝑘) < 0]. (24b)

According to the MLD framework, these definitions can be expressed The control architecture is developed according to the MPC ap-
equivalently using linear inequalities. Finally, it is important to high- proach [60], in which the cost function is minimized at each time 𝑘
light that similarly to the efficiencies of the electrolyser and the fuel and an optimal input sequence is obtained for 𝑗 = 0, … , 𝑇 𝑛 − 1 steps
cell, 𝜂𝑐 and 𝜂𝑑 in (23) can possibly be adjusted in the case of a ahead, where 𝑇 𝑛 is the generic horizon. Then, the first entry is applied,
substantial deviation from the currently adopted values. and the window is shifted one step ahead. In what follows, it is useful
to specify the sets 𝑇 ℎ = {1, 2, … , 24} and 𝑇 𝓁 = {1, 2, … , 6} containing
4.5. System constraint the samplings considered in the HLC and the LLC, which are derived
from the sample time of 1 h and 10 min, respectively.
The equality constraint

𝑃re𝑛 (𝑘) + 𝑃𝑏𝑛 (𝑘) − 𝑃ℎ𝑛 (𝑘) + 𝑃𝑔𝑛 (𝑘) − 𝜁ev


𝑛
(𝑘) = 𝑃𝑠𝑛 (𝑘) (25) 5.1.1. Grid cost functions
The grid cost functions are defined as
is introduced to balance the available power in the MG. Such constraint
𝑚,𝑛
includes the RES generated power (𝑃re𝑛 = 𝑃𝑤𝑛 + 𝑃pv 𝑛 ), the battery power 𝐽grid (𝑘 + 𝑗) = [−𝑐𝑠𝑚,𝑛 (𝑘 + 𝑗)𝜁𝑠𝑚,𝑛 (𝑘 + 𝑗) (27)
(𝑃𝑏𝑛 = 𝜁dc𝑛 − 𝜁 𝑛 ), the net-hydrogen power (𝑃 𝑛 = 𝜁 𝑛 − 𝜁 𝑛 ), the grid
ch ℎ 𝑒 𝑓 + 𝑐𝑏𝑚,𝑛 (𝑘 + 𝑗)𝜁𝑏𝑚,𝑛 (𝑘 + 𝑗)]𝑇𝑠𝑛 ,
power (𝑃𝑔𝑛 ), and the system power (𝑃𝑠𝑛 ). Moreover, in order to take
into account BEVs in the model, one defines the auxiliary variable 𝜁ev 𝑛 where 𝜁𝑠𝑚,𝑛 and 𝜁𝑏𝑚,𝑛 are given as in Eq. (18), and 𝑐𝑠𝑚,𝑛 and 𝑐𝑏𝑚,𝑛 are the
that is given by the product between the EV power consumption 𝑃ev 𝑛 corresponding energy prices. For 𝑚 = 1, the islanding system operations
ON . The latter is such that 𝛿 ON (𝑘) = 1 if an EV is in 1,𝑛
and a variable 𝛿ev ev
are not considered, and then it is 𝐽grid = 0, and the corresponding prices
the charge state at the instant 𝑘, and 𝛿ev ON (𝑘) = 0 otherwise. Note are all set to zero.
that the EV battery follows a similar modeling approach to the BESS
in (23), with the inclusion of 𝛿ev ON . However, the requested load from
5.1.2. Load tracking cost function
BEVs represents an exogenous uncontrollable signal of the controller. The costs corresponding to the load tracking are expressed as
Indeed, it is either provided by a charging station or determined by the
owner’s load planning. 𝐽ld𝑚,𝑛 (𝑘 + 𝑗) = (𝑃𝑠𝑚,𝑛 (𝑘 + 𝑗) − 𝑃𝑙𝑚,𝑛 (𝑘 + 𝑗))2 , (28)
In order to guarantee the stability of the system, the power of each
where 𝑃𝑠𝑚,𝑛 denotes the available power and 𝑃𝑙𝑚,𝑛 is the load demand.
device, the hydrogen level in the tank, and the storage energy in the
battery must be maintained within specific minimum and maximum
bounds, i.e., 5.1.3. BEVs tracking cost function
The cost connected to the interaction with BEVs is introduced
𝑃𝑑𝑚 ≤ 𝑃𝑑𝑛 (𝑘) ≤ 𝑃𝑑𝑀 , (26a) to track the requested power 𝑃exc , which is provided some hours in
𝐻 𝑚 ≤ 𝐻 𝑛 (𝑘) ≤ 𝐻 𝑀 , (26b) advance and does not involve real-time analysis. Therefore, such cost
function is included only in the HLC and given by
soc𝑚 ≤ soc𝑛 (𝑘) ≤ soc𝑀 , (26c)
3,ℎ 3,ℎ 3,ℎ
𝐽exc (ℎ + 𝑗) = (𝜁ev (ℎ + 𝑗) − 𝑃exc (ℎ + 𝑗))2 . (29)
for 𝑑 ∈  and 𝑛 ∈ {ℎ, 𝓁}, where 𝑃𝑑𝑚 , 𝑃𝑑𝑀 , 𝐻 𝑚 , 𝐻 𝑀 , soc𝑚 , and soc𝑀 are
the lower and upper bounds of the powers, the hydrogen level, and the
state of charge, respectively. 5.1.4. Hydrogen tracking cost function
Similar to (29), at the HLC, the cost function
5. Control strategy 4,ℎ
𝐽𝐻 (ℎ + 𝑗) = (𝐻𝑙4,ℎ (ℎ + 𝑗) − 𝐻loss
4,ℎ
(ℎ + 𝑗))2 (30)

This section presents the architecture that controls the hydrogen is introduced to track the hydrogen demand 𝐻𝑙 from FCEVs with the
devices and the battery to provide the system’s optimal operation in decision variable 𝐻loss .

7
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

5.1.5. HESS cost functions


The HESS cost functions derive from the operation of the electrol-
yser and the fuel cell, and their optimization aims to achieve cost
efficiency and extend their lifespan. Therefore, they are given by the
combination of the component depreciation, the life cycle reduction,
and the energy consumption, i.e.,
(( 𝑐 rep )
𝑜
𝐽𝑜𝑚,𝑛 (𝑘 + 𝑗) = + 𝑐𝑜OM 𝛤𝑜ON,𝑛 (𝑘 + 𝑗)
NH𝑜

𝑚,𝑛 sp
+ 𝑃𝑜,𝛼 𝑐 (𝑘 + 𝑗)𝛤𝑜𝛼,𝑛 (𝑘 + 𝑗)𝜏 𝑛
𝛼∈⧵{OFF}
∑ ∑ )
𝑚,𝑛
+ 𝑐𝛼,𝛽 (𝑘 + 𝑗)𝛥𝛽,𝑛
𝛼,𝑜 (𝑘 + 𝑗)
𝛼∈ 𝛽∈
𝛼≠𝛽 Fig. 4. Multi-layer MPC diagram for multiple modes.

+ 𝛾𝐻 + 𝛾𝑃 , (31)
rep
for 𝑜 ∈ {𝑒, 𝑓 }, where 𝑐𝑜 is the cost replacement, NH𝑜 is the cycle 𝑚,𝑛
where 𝐽grid 𝑚,𝑛
(𝑘 + 𝑗) is defined in (27) and 𝐽com (𝑘 + 𝑗) is given by
𝑚,𝑛
lifespan, 𝑐𝑜OM is the maintenance cost, 𝑃𝛼,𝑜 is the power corresponding ∑
to the state 𝛼, the mode 𝑚, and the layer 𝑛, 𝑐 sp is the power spot 𝑚,𝑛
𝐽com (𝑘 + 𝑗) = 𝜔𝑑 𝐽ld𝑚,𝑛 (𝑘 + 𝑗) + 𝜔𝑏 𝐽𝑏𝑚,𝑛 (𝑘 + 𝑗)
𝑚,𝑛
price, 𝑐𝛼,𝛽 denotes the cost associated to the switch (𝛼, 𝛽). Note that 𝑏∈{ch,dc}

when the decision variables and 𝛤𝑜𝛼,𝑛 𝛥𝛽,𝑛
are not allowed for given 𝑚
𝛼,𝑜 + 𝜔𝑜 𝐽𝑜𝑚,𝑛 (𝑘 + 𝑗), (34)
𝑚,𝑛 𝑚,𝑛
and 𝑛, then 𝑃𝛼,𝑜 and 𝑐𝛼,𝛽 are assigned to zero. Moreover, the terms 𝑜∈{𝑒,𝑓 }

𝛾 𝐻 = 𝜔𝐻 (𝐻 𝑚,𝓁 (ℎ, 𝓁) − 𝐻 𝑚,ℎ (ℎ))2 and 𝛾 𝑃 = 𝜔𝑃 (𝜁𝑜𝑚,𝓁 (ℎ, 𝓁) − 𝜁𝑜𝑚,ℎ (ℎ))2 with 𝐽ld𝑚,𝑛 , 𝐽𝑜𝑚,𝑛 , and 𝐽𝑏𝑚,𝑛 defined in (28), (31), (32), respectively, and
represent the hydrogen and power reference trackings provided by the 𝜔𝑔 , 𝜔𝑜 , 𝜔𝑏 , and 𝜔𝑑 are suitable weights. The conditions 𝛩con 𝑛 = 1
HLC, respectively, where 𝜔𝑃 and 𝜔𝐻 are two penalized desired weights. 𝑛
and 𝛩con = 0 correspond to the grid-connected and islanded modes,
respectively.
5.1.6. BESS cost functions In the islanded mode of the MG (𝑚 = 1), any excess of energy after
The cost functions associated to the BESS are linked to the battery meeting the load demand is used to store hydrogen via electrolysis
lifetime and can be represented as a function of the total working hours and the battery charge, while the green hydrogen consumption and the
as battery discharge are set ON when RESs are null. Then, the optimiza-
(( bat ) ON tion problem is solved to track the local demand while minimizing the
𝑐
𝐽𝑏𝑚,𝑛 (𝑘 + 𝑗) = cyc + 𝑐𝑏
OM
𝛤𝑏 (𝑘 + 𝑗) ESSs’ operations. In particular, the optimization problem of the HLC is
2𝑁𝑏
∑ ∑ 𝑚,𝑛 ) defined as
+ 𝑐𝛼,𝛽 (𝑘 + 𝑗)𝛥𝛽𝛼,𝑏 (𝑘 + 𝑗) ℎ−1
𝑇∑
𝛼∈ 𝛽∈ 1,ℎ
𝛼≠𝛽 min 𝐽gs (ℎ + 𝑗)
̂ℎ 𝑗=0
+ 𝛾 soc + 𝛾 bat , (32)
s.t. Actions constraints (3)–(4),
cyc
where 𝑏 ∈ {ch, dc}, 𝑐 bat is the battery capital cost, 𝑁𝑏 is the bat- Switches constraints (5)–(9), (35)
( )2
tery working cycles, and 𝛾 soc = 𝜔soc soc𝑚,𝓁 (ℎ, 𝓁) − soc𝑚,ℎ (ℎ)
and Selections constraints (11)–(12),
( )2
𝛾 bat = 𝜔bat 𝜁𝑏𝑚,𝓁 (ℎ, 𝓁) − 𝜁𝑏𝑚,ℎ (ℎ) , according to the desired factors 𝜔soc
ESSs dynamics (21)–(23),
and 𝜔bat , accounts for the deviation of the real-time state of charge
soc𝑚,𝓁 (𝓁 + 𝑗) and the scheduled power 𝜁𝑏𝑚,𝓁 (𝓁) in the battery at 𝓁 and System constraints (25) − (26),
𝑇 𝓁 from soc𝑚,ℎ (ℎ) and 𝜁𝑏𝑚,ℎ (ℎ) computed by the HLC. The slack variable
where ℎ is the current time instant, ̂ℎ is the set of decision variables
𝜁𝑏𝑚,𝑛 (𝑘 + 𝑗) = 𝑃𝑏𝑚,𝑛 (𝑘 + 𝑗)𝛤𝑏ON (𝑘 + 𝑗) hides the non-linearity deriving of the HLC, and 𝐽gs1,ℎ
is given in (33). The optimization problem of the
by the product of two decision variables. The last two terms in (32) LLC, instead, is defined as
are introduced to prevent excessive stress during the charging and
𝓁−1
𝑇∑
discharging operations. 1,𝓁
min 𝐽gs (𝓁 + 𝑗)
̂𝓁 𝑗=0
(36)
5.2. MPC-based control approach s.t. Similar to (35),

The architecture of the multi-layer MPC strategy is illustrated in where 𝓁 is the current time-step and ̂𝓁 is the set of decision variables
Fig. 4. At the HLC, the following day predictions of the RES power 𝑃re of the LLC.
and the electric load 𝑃𝑙 are compared, and the schedule that optimizes For the grid-connected mode (𝑚 = 2), two optimization problems
the operations of the ESSs, the hydrogen 𝐻 ℎ in the tank, and the energy are solved to find the optimal power schedule profiles for the devices to
socℎ in the BESS for the next few hours are computed. Then, the LLC deliver power to the grid and the local load, minimize ESS operational
uses the optimal scheduling computed by the HLC to correct possible costs, and maximize power sold revenues in both the day-ahead and
mismatches between the real-time and predicated RESs and local loads. real-time markets. The optimization problem of the HLC is given by
In the scheme in Fig. 4, 𝑢𝑛0 , 𝑢𝑛in , and 𝑈 𝑛 are matrices including the ℎ−1
𝑇∑
initial conditions, inputs, and outputs, respectively, for the control layer 2,ℎ
min 𝐽gs (ℎ + 𝑗)
𝑛 ∈ {ℎ, 𝓁}. In order to achieve the objectives from the perspective of ̂ℎ 𝑗=0
the MG operating modes at each layer, the global cost function used (37)
s.t. Similar to (35),
for the optimization problem is
Grid constraints (17)–(19).
𝑚,𝑛 𝑚,𝑛 𝑚,𝑛 𝑛
𝐽gs (𝑘 + 𝑗) = 𝐽com (𝑘 + 𝑗) + 𝜔𝑔 𝐽grid (𝑘 + 𝑗)𝛩con , (33)

8
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

The optimization problem of the LLC for 𝑚 = 2 is defined as


𝓁−1
𝑇∑
Finally, the optimization problem of the LLC is defined as
2,𝓁
min 𝐽gs (𝓁 + 𝑗) 𝓁−1
𝑇∑
̂𝓁 (38) 4,𝓁
𝑗=0 min 𝐽gs (𝓁 + 𝑗)
̂ 𝓁 𝑗=0
(46)
s.t. Similar to (37).
s.t. Similar to (38).
For the interaction with BEVs (𝑚 = 3), three optimization problems
are solved: (i) to meet the requested demand from BEVs; (ii) to sell/buy The optimal input sequence for the island optimization problem
energy to/from the grid through the intraday market and to supply a in the HLC is ̂ℎ∗ = {ℎ∗ , ℎ+1
∗ , … , ∗

}, where ̂ℎ+𝑗
∗ is the opti-
ℎ+𝑇 −1
load after any excess of RESs has met the exchange requested demand;
mal value of ℎ at ℎ + 𝑗. The optimal input sequence in the LLC is
(iii) to minimize a cost function for the control of the participation in
̂𝓁∗ = {𝓁∗ , 𝓁+1
∗ , … , ∗
𝓁

}, where 𝓁+𝑗 has the similar meaning of
the real-time market and the tracking of the load. At the HLC, the two 𝓁+𝑇 −1
∗ . The other optimal input sequences are
the corresponding term ℎ+𝑗
optimization problems (i) and (ii) are optimized based on the sequential
optimization technique. In particular, the first problem is similarly defined. According to the MPC [60], the first sample is used,
ℎ−1 while the remaining samples are excluded.
𝑇∑
ℎ ∗ 3,ℎ
(𝜏ev ) = min 𝐽exc (ℎ + 𝑗)
̂ ℎ 𝑗=0
(39) 5.3. Scenario-based approach
s.t. Similar to (37),
For the controller proposed in the previous section, it is assumed
3,ℎ
with 𝐽exc defined in (29). The second optimization problem, which is that the owners of the RES systems directly provide the best day-ahead
solved in sequence to (39), is schedule, thus resulting in considering RESs as exogenous signals. How-
ever, in order to account for the unusual circumstance of the owners’
ℎ−1
𝑇∑
3,ℎ
unavailability, this section presents a scenario-based SMPC, which is
min 𝐽gs (ℎ + 𝑗) integrated into the system to handle the uncertainties. The inclusion
̂ ℎ 𝑗=0
(40) of the SMPC is only activated by a system operator. In the following,
s.t. Similar to (39), a brief description of the SMPC is presented, and we refer the reader
Sequential constraint, to [61] for more details.
In order to design the scenario-based SMPC strategy, it is important
where the sequential constraint according to BEVs can be defined as to derive the number of scenarios and the corresponding probability
ℎ−1
𝑇∑ coefficients. In particular, for the system under investigation, the SMPC
3,ℎ
𝐽exc ℎ ∗
(ℎ + 𝑗) ≤ (𝜏ev ) . (41) includes two scenarios of the RES power, called optimistic and pes-
𝑗=0 simistic scenarios. In the following, 𝑠 = {1, 2} is the scenario set,
Finally, the optimization problem (iii) at the LLC is defined as where 𝑛𝑠 = 1 (𝑛𝑠 = 2) denotes the optimistic (pessimistic) scenario.
𝑛
Moreover, 𝑃𝑟𝑒𝑠 denotes the RES power for the scenario 𝑛𝑠 ∈ 𝑠 . An
𝓁−1
𝑇∑
3,𝓁
uncertainty band for the SMPC, based on [62], is defined as
min 𝐽gs (𝓁 + 𝑗) √
̂ 𝓁 (42) √ 𝑁𝑚 𝑁ℎ ( )2
𝑗=0 √ 1 ∑ ∑
𝑃̃ = √ 𝑃re (𝑚, ℎ) − 𝑃remea (𝑚, ℎ) , (47)
s.t. Similar to (38). unc
𝑁𝑚 𝑁ℎ 𝑚=1 ℎ=1

For the interaction with FCEVs (𝑚 = 4), the MPC strategy is where 𝑁𝑚 is the total number of days in a year, 𝑁ℎ is the number
based on similar concepts as those in the previous mode. Indeed, three of hours in a day, 𝑃re (𝑚, ℎ) and 𝑃remea (𝑚, ℎ) denote the forecasted and
optimization problems are solved: (i) to meet hydrogen requests from measured values of the RES power for the day 𝑚 and the hour ℎ. Thus,
FCEVs; (ii) to sell energy to the grid through the intraday market the RES powers corresponding to the optimal and pessimistic scenarios
and to supply loads after any excess of RESs has been used for the are defined as
hydrogen demand; (iii) to minimize a cost function for the control of
𝑃re1 (𝑘) = 𝑃re0 (𝑘) + 𝑃̃
unc , (48a)
the participation in the real-time market and the tracking of the load.
In particular, the first optimization problem of the HLC is described as 𝑃re2 (𝑘) = 𝑃re0 (𝑘) − 𝑃̃
unc , (48b)

ℎ−1
respectively, where 𝑃re0
is the RES power determined from the standard
𝑇∑
ℎ ∗
(𝜏𝐻 ) = min 4,ℎ
𝐽𝐻 (ℎ + 𝑗) MPC and 𝑃̃ unc is defined in (47).
̂ ℎ 𝑗=0
(43) When considering extreme situations that cannot be adequately
addressed by the predictions from the optimistic and pessimistic sce-
s.t. Similar to (39),
narios, supplementary scenarios are defined as
4,ℎ
with 𝐽𝐻 given by (30). The HLC prioritizes the hydrogen production 𝑛
𝑃re𝑠 (𝑘) = 𝑃re0 (𝑘) ± 𝑠 𝑃̃ (49)
unc ,
by adopting a sequential optimization approach. Consequently, the
optimal value of hydrogen layer must be considered as a constraint in where 𝑠 ∈ is the width of the band 𝑃̃
𝐑+ unc . However, increasing the
the second problem, i.e., number of scenarios enhances the robustness of the control approach
ℎ −1
𝑇∑
but at the cost of increased computational time with limited perfor-
4,ℎ
𝐽𝐻 ℎ ∗
(ℎ + 𝑗) ≤ (𝜏𝐻 ) , (44) mance improvements. Therefore, in this paper, only two scenarios are
𝑗=0 included, i.e., we assume 𝑠 = 1.
The main objective of the SMPC is to address uncertainties by
and then the second optimization problem is given by
solving optimization problems where the cost functions are determined
ℎ−1
𝑇∑ as the sum of the trajectories corresponding to the disturbances. In
4,ℎ
min 𝐽gs (ℎ + 𝑗) particular, both the optimistic and pessimistic scenarios are simulta-
̂ ℎ 𝑗=0
(45) neously executed, allowing the control system to adapt to the actual
s.t. Similar to (39), conditions and make informed decisions based on the range of pos-
Sequential constraint (44). sibilities. This simultaneous consideration of both scenarios enhances

9
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

the MG’s resilience and efficiency since it can effectively respond to Algorithm 1: Multi-layer MPC algorithm for multiple modes.
different variations in RESs. Input : 𝑚, 𝑠𝑙
The optimization problems (35)–(38), (40), (42), and (45)–(46) can begin
be recast in the SMPC framework as Load parameters;
𝑁𝑠 [ 𝑛 −1 ] 𝐻 ℎ (ℎ) ← 𝐻0ℎ ; 𝐻 𝓁 (𝓁) ← 𝐻0𝓁 ;
∑ ∑
min 𝜌𝑛 𝑠 𝑚,𝑛
𝐽gs,𝑛 (𝑘 + 𝑗) socℎ (ℎ) ← socℎ0 ; soc𝓁 (𝓁) ← soc𝓁0 ;
̂ 𝑘 𝑛𝑠 =1 𝑠
 𝑗=0
for ℎ ← 1 to  ℎ do
s.t. Similar to (35), if 𝑚 == 1 then
(50) ( )
Grid constraints (17)–(19), if 𝑚 ≠ 1, ̂ℎ∗ ← (𝑠𝑙 == 1) sol(35) ? sol(50) ;
Sequential constraints, ℎ∗ ← ̂ℎ∗ {1};
for 𝓁 ← 1 to  𝓁 do
if (𝑚 = 3 ∨ 𝑚 = 4) ∧ (𝑛 = ℎ), ( )
̂𝓁∗ ← (𝑠𝑙 == 1) sol(36) ? sol(50) ;
for 𝑚 ∈ {1, 2, 3, 4} and 𝑛 ∈ {𝓁, ℎ}, where 𝑛𝑠 denotes the scenario 𝓁∗ ← ̂𝓁∗ {1};
considered, 𝜌𝑛𝑠 represents the probability of the scenario 𝑛𝑠 , and 𝑁𝑠 𝓁 ← 𝓁 + 1;
is the number of scenarios. Similarly, the optimization problem (39) else if 𝑚 == 2 then
( )
can be recast in the SMPC framework as ̂ℎ∗ ← (𝑠𝑙 == 1) sol(37) ? sol(50) ;
𝑁𝑠 [  ℎ −1 ]  ← ̂ ∗ {1};

∑ ∑ ℎ ℎ
ℎ ∗ 3,ℎ
(𝜏ev ) = min 𝜌𝑛𝑠 𝐽exc,𝑛 (ℎ + 𝑗) for 𝓁 ← 1 to  𝓁 do
̂ ℎ 𝑛𝑠 =1
𝑠 (51) ( )
𝑗=0 ̂𝓁∗ ← (𝑠𝑙 == 1) sol(38) ? sol(50) ;
s.t. Similar to (39), 𝓁∗ ← ̂𝓁∗ {1};
𝓁 ← 𝓁 + 1;
and the SMPC of the problem in (43) is recast as
else if 𝑚 == 3 then
[ 𝑇 ℎ −1 ] ( )
𝑁𝑠
∑ ∑ ̂ ∗ℎ ← (𝑠𝑙 == 1) sol(39) ? sol(51) ;
ℎ ∗ 4,ℎ
(𝜏𝐻 ) = min 𝜌𝑛 𝑠 𝐽𝐻,𝑛 (ℎ + 𝑗) ∗ℎ ← ̂ ∗ℎ {1};
̂ ℎ 𝑛𝑠 =1 𝑗=0
𝑠 (52) ( )
̂ ∗ℎ ← (𝑠𝑙 == 1) sol(40) ? sol(50) ;
s.t. Similar to (43).
∗ℎ ← ̂ ∗ℎ {1};
for 𝓁 ← 1 to  𝓁 do
( )
̂ ∗𝓁 ← (𝑠𝑙 == 1) sol(41) ? sol(50) ;
5.4. Proposed integrated multi-layer MPC algorithm
∗𝓁 ← ̂ ∗𝓁 {1};
𝓁 ← 𝓁 + 1;
A description of the proposed controller, which includes both the
MPC and SMPC approaches, is provided in Algorithm 1. The optimiza- else
( )
tion problem solved at each time instant depends on the two inputs set ̂ ℎ∗ ← (𝑠𝑙 == 1) sol(43) ? sol(52) ;
by the system operator, i.e., 𝑚 ∈ {1, 2, 3, 4} for the mode and 𝑠𝑙 ∈ {1, 2} ℎ∗ ← ̂ ℎ∗ {1};
( )
for using the MPC or the SMPC. In particular, 𝑠𝑙 = 1 denotes the usual ̂ ∗ℎ ← (𝑠𝑙 == 1) sol(45) ? sol(50) ;
configuration using the MPC, where the RES generation predictions are ∗ℎ ← ̂ ∗ℎ ;
provided by the owners of the RES systems, while 𝑠𝑙 = 2 corresponds for 𝓁 ← 1 to  𝓁 do
( )
to the use of the SMPC in the particular cases where the predictions ̂ 𝓁∗ ← (𝑠𝑙 == 1) sol(46) ? sol(50) ;
are not available. For each mode selected by the input 𝑚, controllers 𝓁∗ ← ̂ 𝓁∗ ;
based on the MPC approach are developed to address the optimal load 𝓁 ← 𝓁 + 1;
tracking, the electricity market participation, and the satisfaction of
ℎ ← ℎ + 1;
the requests from external agents, as discussed above. Moreover, for
each 𝑚, the algorithm also selects the optimization problems to solve,
according to the values of 𝑠𝑙 . Note that in Algorithm 1 the abbreviated
expression (cond) A ? B corresponds to if(cond) A else B end,
where cond denotes a condition and A and B are generic statements. 6.1. Simulation setup
Instead, the functions sol(⋅) stands for ‘‘solve’’.
In this research study, the parameters are determined to ensure the
6. Results analysis validation of the integrated control. Consequently, for the day-ahead
market (HLC), simulations cover a 𝑇 ℎ = 24 h horizon with 𝜏 ℎ = 1 h
This section illustrates the validation of the proposed multi-layer sampling time. On the other hand, for the real-time market (LLC), the
MPC strategy based on the MG under investigation via numerical shorter time horizon 𝑇 𝓁 = 1 h and the sampling time 𝜏 𝓁 = 10 min are
simulations by using a power profile and spot prices of an MG located in used. Table 2 reports the list of the parameters and weighting factors
Abu Dhabi, UAE. Simulation results reveal that the proposed controller selected for the numerical analysis.
can effectively fulfill all the system constraints and control objectives. For the simulation analysis, the wind and solar profiles have been
Moreover, in order to simulate the operation of a real wind–solar MG provided by ADNOC, Abu Dhabi, UAE,2 and refer to a wind farm and
with hybrid ESSs, an experimental emulator has been developed at the solar panels situated in Abu Dhabi. Market prices have been furnished
Energy Systems and Control Optimization (ESCO) Lab, ADNOC, Khalifa by the energy market operator Iberian Market Operator and are those of
University.1 June 8, 2023. In order to achieve a proper balance between tracking
goal satisfaction and operational costs, the weights assigned to the cost

1
https://www.ku.ac.ae/facilities/energy-systems-and-control-
2
optimization-esco-lab. https://www.adnoc.ae/.

10
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

Table 2
Main characteristics of the devices and weights of the control strategy.
Electrolyser/Fuel cell parameters
Variable Parameter Value
𝑃𝑒min ∕𝑃𝑓min Minimum power of electrolyser/fuel cell 0.1/0.002 MW
𝑃𝑒max ∕𝑃𝑓max Maximum power of electrolyser/fuel cell 3/0.15 MW
𝑃𝑒STB ∕𝑃𝑓STB Standby consumption of electrolyser/fuel cell 0.003/0.001 MW
𝜂𝑒 ∕𝜂𝑓 Efficiency of electrolyser/fuel cell 0.015 kg∕kWh/22 kWh∕kg
deg𝑒 ∕deg𝑓 Efficiency degradation of electrolyser/fuel cell 4%∕3% year
NH𝑒 ∕NH𝑓 Cycles lifespan of electrolyser/fuel cell 4000/3000 h
NY𝑒 ∕NY𝑓 Operation hours of electrolyser/fuel cell 20 000/15 000 h
Tank parameters
𝐻0𝑛 Initial of hydrogen tank 0.6 p.u.
𝐻 min ∕𝐻 max Minimum/maximum of hydrogen tank 0.1/0.95 p.u.
Li-Ion battery parameters
𝑃𝑏𝑀 ∕𝑃𝑏𝑚 Minimum/maximum power 0.001/3 MW
𝜂𝑐 ∕𝜂𝑑 Charging/Discharging efficiency 0.90/0.95
𝑐 bat Battery capital cost 125 A
C∕kWh
deg deg
𝑐ch ∕𝑐dc Charging/Discharging degradation cost 10−9 A
C∕W2 h
cyc
𝑁𝑏 Battery working cycles 3000 h
soc𝑛0 Initial of stored energy 0.15 p.u.
socmin ∕socmax Minimum/maximum of stored energy 0.1/1 p.u.
Controller parameters
𝜏 ℎ ∕𝜏 𝓁 Sampling time for HLC/LLC 1 h/10 min
𝑇 ℎ ∕𝑇 𝓁 Simulation horizon for HLC/LLC 24/1 h
𝜔ℎ𝑔 ∕𝜔𝓁𝑔 Grid weight for HLC/LLC 4 × 106 /16 × 104
𝜔ℎ𝑒 ∕𝜔𝓁𝑒 Electrolyser weight for HLC/LLC 1 × 108 /1 × 1012
𝜔ℎ𝑓 ∕𝜔𝓁𝑓 Fuel cell weight for HLC/LLC 2 × 104 /8 × 108
𝜔ℎ𝓁 ∕𝜔𝓁𝓁 Load tracking weight for HLC/LLC 12 × 1010 /1 × 106

functions are thoughtfully chosen through a comprehensive numeri-


cal examination. The proposed controller is an MIQP problem with
multi-timescale rolling optimization. This problem can be solved using
methods such as the cutting plane and branch and bound algorithms,
which are available in commercial solvers, e.g., CPLEX and GUROBI.
In this study, the development and the simulation analysis have been
conducted in MATLAB by using YALMIP/GUROBI. The optimization
problems described in Section 5 with 24 h horizon are solved in 30 s
at most on a workstation with a Xeon (R) 𝑄 − 3395HQ 4.9 GHz with
Fig. 5. Data profiles.
memory size 256 GB.

6.2. Case studies


deficit scenarios, which are determined by the power flow towards or
Both the MPC and the SMPC are investigated to demonstrate the away from the ESSs. Any excess of RES power generated after fulfilling
optimal management of the system across the four different modes. For the load demand is directed towards the electrolyser for hydrogen
the first strategy, an ideal deterministic forecast scenario is taken into generation or the battery for its charge. In the case of a power deficit,
account by neglecting the uncertainties associated to RESs. The second the controller activates the fuel cell and the discharge battery to supply
approach solves the optimization problem for two scenarios including additional power to the MG and maintain energy balance. Fig. 6(b)
the uncertainties, i.e., the optimistic scenario and the pessimistic one. shows that the power of the MG devices consistently falls within the
predefined minimum and maximum bounds. Moreover, it is possible
6.2.1. Analysis of the MPC strategy to observe from Fig. 7(a) that the LLC tracks the reference schedule
The multi-layer MPC strategy is tested through simulations that provided by the HLC and successfully meets the electric demand at all
take into account the RES profile shown in Fig. 5. In the following, instances 𝑘. Additionally, Fig. 7(b) provides details on the hydrogen
the validation of the controller for the four modes (islanded, grid- device and the battery references determined by the HLC and their
connected, and exchange of energy with BEVs and FCEVs) is provided. tracking by the LLC.
In the grid-connected configuration, the controller manages the
The main goal of the MG in islanded mode, i.e., without the support possibility of buying and selling energy. Note that in the energy market
of the main grid, is to meet the load demand using RESs and stored operator, energy prices are provided hourly and resampled at the LLC,
hydrogen while simultaneously reducing the operational costs of the which operates at a sampling time of 10 min. Fig. 8 reports the energy
hybrid ESSs. As depicted in Figs. 6 and 7, the power available in the prices for the intraday market. Regarding the HLC, Fig. 9 shows that
system, denoted by 𝑃𝑠𝑛 , follows the load demand based on the control the power 𝑃𝑠ℎ in the systems tracks the requested load. The energy
layer. The proper functioning of the devices relies on power surplus or surpluses or deficits are correctly managed by the power flow with

11
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

Fig. 8. Energy price for the intraday market.

Fig. 6. Response of the HLC for the islanded mode.

Fig. 9. Response of the HLC for the grid-connected mode.

the references determined by the HLC, while maximizing revenues. The


corresponding plots are not reported to avoid repetitions.
Based on the figures discussed, it is evident that in all four modes,
the operations of the ESSs are executed optimally, leading to a min-
imization of operation and maintenance costs. In the grid-connected
mode and when the energy is exchanged with EVs, the maximization
of the profits is also successfully considered.
Fig. 7. Response of the LLC for the islanded mode.

6.2.2. Analysis of the SMPC strategy


Unlike the MPC strategy, the SMPC accounts for the deviations in
the ESSs. In particular, after the requested demand is met, any surplus the energy forecast. For the simulation analysis, the probability factors
of RES energy is routed either to the electrolyser for green hydrogen 𝑛
of the optimistic and pessimistic scenarios 𝑃re𝑠 , with 𝑛𝑠 ∈ 𝑠 , are set
generation or to the battery for its charge. In the power deficit scenario, to 𝜌𝑛𝑠 = 0.5, see (48). Fig. 13 shows the power profiles used in the
the controller either sets the fuel cell and the discharge to their ON numerical simulations for both scenarios.
state to satisfy the energy balancing equation. Fig. 10 shows that the First, we analyze the SMPC strategy for the grid-connected mode.
LLC tries to achieve the references computed by the HLC. Moreover, Fig. 14 shows the load tracking in both scenarios. It is shown that the
the revenues from sold power are maximized, according to the available available power adequately fulfills the load demand, with the exception
power and the energy prices. Specially, during the 24 h simulations with of the hours 13, 15, and 19 in the pessimistic scenario. The hydrogen
the exception of the hours 6, 9, 16–18, and 22, the control strategy tends and the energy stored in the ESSs are marked in Fig. 15. Similarly to
to maximize power sold revenues. Thus, the controller is able to both the MPC, it is possible to observe that the hybrid ESSs correctly manage
track users’ requests and supply loads. the power surpluses or deficits. The interaction with the utility grid is
In the MG dispatched with BEVs and FCEVs, the integrated con- presented in Fig. 16. It is evident that also for this strategy, the sold
trollers track the hydrogen/electric load requested by external con- power revenues are maximized, by tending to sell as much as possible.
sumers. Figs. 11–12 show that the exchange of power and hydrogen In the MG operating with external consumers, both the MPC and
fulfills the demands of BEVs and FCEVs. Moreover, the LLC achieves SMPC approaches exhibit the same behaviors concerning hydrogen

12
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

Fig. 14. Load tracking according to two scenarios.

Fig. 15. ESSs layers according to two scenarios.

Fig. 10. Response of the LLC for the grid-connected mode.

Fig. 16. Grid powers according to the two scenarios.

Fig. 11. Response of the HLC for MG dispatched with BEVs.

Fig. 17. Experimental emulator setup.

production, battery storage, and power exchange. Consequently, we


refer the reader to Figs. 11 and 12 for the SMPC approach when the
MG interacts with external consumers.
Fig. 12. Response of the HLC for the MG dispatched with FCEVs.
6.3. Experimental validation

The effectiveness of the proposed MPC strategy is also shown on


a lab-scale MG running in real-time, set up at the ESCO Lab. Fig. 17
is a picture of the experimental emulator developed to simulate the
operation of a real MG, while its scheme is shown in Fig. 18.
A PV emulator (marked by G in the figures) represents the RES in
the simulator. The PV emulator has a maximum power of 20 kW and
belongs to the XR160-12 family (Source: ‘‘Magna-Power’’). Instead, an
electronic load (marked by F) 20 kW that belongs to the Cinergia EL-15
family (Source: ‘‘Cinergia-EL’’) is used to simulate the loads. In turn,
Fig. 13. Data profiles for the SMPC. the MG emulator includes a battery emulator (B in the figures) with
maximum rated power 15 kW and operational capacity 5 kW that be-
longs to the Cinergia BE-15 family (Source: ‘‘Cinergia-BE’’). Regarding
the HESS unit, the lab-scale hydrogen system consists of an electrolyser

13
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

Fig. 21. BEVs tracking.

Fig. 18. Scheme of the emulator MG.

Fig. 22. ESSs layers.

Figs. 19–22 show the data profile used for the testing on the
lab-scale setup, the powers, and the state of the storage devices, respec-
tively. The results in the real-time emulator confirm that the proposed
MPC strategy successfully manages the MG, thus maximizing the profits
and minimizing the operational and maintenance costs.
Fig. 19. Profiles used for the testing on the lab-scale setup.

7. Conclusions

This paper has proposed a novel multi-layer control architecture


for the optimal management of a wind–solar MG that is coupled with
a HESS and a BESS and interacts with external agents. The proposed
control strategy minimizes energy exchange with the grid, reduces
operation costs, and manages the overall system in four modes, i.e., is-
landed, grid-connected, and exchange of energy with BEVs and FCEVs.
In order to model the ESSs’ operations, an MLD model including waiting
actions is developed. The proposed multi-layer MPC allows for the op-
Fig. 20. Load tracking. timal management of the devices, even in cases of deviations between
the predicted scenario computed by the HLC and the real-time one
addressed via the LLC, where tracking the references set by the HLC is
(denoted by C), a tank (denoted by D), and a Nexa-1200 fuel cell (E the main goal. Furthermore, in order to take into account the situations
in the figures) whose power is about 12 kW (Source: ‘‘Nexa-FC’’). The in which the predictions of the RESs are not available, an SMPC, which
setup includes also a grid emulator (A in the figures) with a maximum includes an optimistic scenario and a pessimistic one, is proposed to
capacity of 15 kW and belonging to the Cinergia GE-15 family (Source: manage the uncertainties due to the sporadic nature of RESs. The
‘‘Cinergia-GE’’). Moreover, the emulator devices described above are controller has been validated through numerical simulations. Finally, a
connected to a power electronics interface (denoted by I), which is further validation of the proposed control has been conducted through
responsible for controlling the power flow to the loads and for man- its integration with a lab-scale system which includes emulators of the
aging the voltage and frequency of the MG. In order to simulate the grid, loads, RESs, ESSs, and a BEV.
It is noteworthy to emphasize that the proposed method is not
interaction with external consumers, the system also includes a BEV
limited to a laboratory simulator. Indeed, it can be similarly integrated
(marked by H). Finally, the algorithm is embedded in a real-time digital
into real MGs for their management. The latter is the future direction
simulator (RTDS) (L in the figures) that belongs to the NovaCor family
of this research work. Indeed, the controller proposed in this paper
(Source: ‘‘NovaCor’’) and can also be used to program and control the
will be deployed into a wind–solar MG, which is under implementa-
real-time state of the system. The RTDS has different purposes: (i) it
tion in Abu Dhabi, UAE. Moreover, it is important to highlight that
simulates all components of the MG and their real-time operations; (ii)
the proposed approach can be applied to MGs of different sizes and
it represents the central processing unit that manages and dispatches configurations, including different energy storage and generation units.
all physical/simulated quantities; (iii) it serves as a communication However, the inclusion of extra units may result in a higher number of
gateway with MATLAB to solve the optimization problems, as explained decision variables. In this cases, it is possible to apply the proposed
in Section 5. In order to ensure the third purpose, in real-time the RTDS approach while considering the total capacity and managing multiple
sends the state variables of the system to MATLAB to be used as input units collectively.
for the optimization problems. The optimal schedule computed using Other future research can explore the extension of the proposed
MATLAB is then sent to each system component for their management. method to include more sophisticated and hierarchical distributed MPC
This type of connection between hardware components and RTDS is of interconnected MGs. Also, the SMPC can be well-explored to include
called hardware in the loop. more detailed scenarios or probabilistic approaches.

14
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

CRediT authorship contribution statement [18] A. Parisio, E. Rikos, L. Glielmo, A model predictive control approach to
microgrid operation optimization, IEEE Trans. Control Syst. Technol. 22 (5)
(2014) 1813–1827.
Muhammad Bakr Abdelghany: Conceptualization, Methodology,
[19] M. Daneshvar, B. Mohammadi-Ivatloo, K. Zare, S. Asadi, Transactive energy
Software, Validation, Writing – original draft. Ahmed Al-Durra: Inves- management for optimal scheduling of interconnected microgrids with hydrogen
tigation, Supervision, Review. Zhou Daming: Review and editing. Fei energy storage, Int. J. Hydrogen Energy 46 (30) (2020) 16267–16278.
Gao: Review, Editing, Investigation. [20] A. Shafiqurrahman, B.S. Umesh, N.A. Sayari, V. Khadkikar, Electric vehicle-to-
vehicle energy transfer using on-board converters, IEEE Trans. Transp. Electrif.
9 (1) (2023) 1263–1272.
Declaration of competing interest
[21] F. Garcia-Torres, C. Bordons, M.A. Ridao, Optimal economic schedule for a
network of microgrids with hybrid energy storage system using distributed model
The authors declare that they have no known competing finan- predictive control, IEEE Trans. Ind. Electron. 66 (3) (2019) 1919–1929.
cial interests or personal relationships that could have appeared to [22] Y. Dahmane, R. Chenouard, M. Ghanes, M. Alvarado-Ruiz, Optimized time step
influence the work reported in this paper. for electric vehicle charging optimization considering cost and temperature,
Sustain. Energy Grids Netw. 26 (2021) 100468.
[23] F. Baccino, S. Grillo, S. Massucco, F. Silvestro, A two-stage margin-based
Data availability algorithm for optimal plug-in electric vehicles scheduling, IEEE Trans. Smart
Grid 6 (2) (2014) 759–766.
Data will be made available on request. [24] G. Kakoulaki, I. Kougias, N. Taylor, F. Dolci, J. Moya, A. Jäger-Waldau, Green
hydrogen in Europe–A regional assessment: Substituting existing production
with electrolysis powered by renewables, Energy Convers. Manage. 228 (2021)
Acknowledgments
113649.
[25] X. Fang, Y. Wang, W. Dong, Q. Yang, S. Sun, Optimal energy management of
The authors gratefully acknowledge the financial support by Khalifa multiple electricity-hydrogen integrated charging stations, Energy 262 (2023)
University of Science and Technology (KUST), Abu Dhabi, UAE, under 125624.
[26] J. Pahasa, I. Ngamroo, PHEVs bidirectional charging/discharging and SoC control
grant agreement No. CIRA-2021-063.
for microgrid frequency stabilization using multiple MPC, IEEE Trans. Smart Grid
6 (2) (2014) 526–533.
References [27] M.B. Abdelghany, V. Mariani, D. Liuzza, O.R. Natale, L. Glielmo, A unified
control platform and architecture for the integration of wind-hydrogen systems
[1] N.A. El-Taweel, A. Ayad, H.E.Z. Farag, M. Mohamed, Optimal energy man- into the grid, IEEE Trans. Autom. Sci. Eng. (2023) 1–16.
agement for battery swapping based electric bus fleets with consideration of [28] W. Lee, L. Xiang, R. Schober, V.W. Wong, Electric vehicle charging stations with
grid ancillary services provision, IEEE Trans. Sustain. Energy 14 (2) (2023) renewable power generators: A game theoretical analysis, IEEE Trans. Smart Grid
1024–1036. 6 (2) (2014) 608–617.
[2] A.M. Moustafa, M.B. Abdelghany, A.-S.A. Younis, M. Moness, A. Al-Durra, J.M. [29] P.R. Mendes, L.V. Isorna, C. Bordons, J.E. Normey-Rico, Energy management
Guerrero, Software-defined control of an emulated hydrogen energy storage for of an experimental microgrid coupled to a V2G system, J. Power Sources 327
energy Internet ecosystems, Int. J. Hydrogen Energy (2023). (2016) 702–713.
[3] A. Foda, H. Abdelaty, M. Mohamed, E. El-Saadany, A generic cost-utility-emission [30] D. Yassuda Yamashita, I. Vechiu, J.-P. Gaubert, Two-level hierarchical model
optimization for electric bus transit infrastructure planning and charging predictive control with an optimised cost function for energy management in
scheduling, Energy 277 (2023) 127592. building microgrids, Appl. Energy 285 (2021) 116420.
[4] C. Bordons, F. Garcia-Torres, M.A. Ridao, Model Predictive Control of Microgrids, [31] J. Engel, T. Schmitt, T. Rodemann, J. Adamy, Hierarchical economic model
Springer, 2020. predictive control approach for a building energy management system with
[5] B. Hamad, A. Al-Durra, T.H.M. EL-Fouly, H.H. Zeineldin, Economically optimal scenario-driven EV charging, IEEE Trans. Smart Grid 13 (4) (2022) 3082–3093.
and stability preserving hybrid droop control for autonomous microgrids, IEEE [32] H. Faraji, A. Khorsandi, S.H. Hosseinian, Multi-level coordinated control of
Trans. Power Syst. 38 (1) (2023) 934–947. islanded DC microgrid integrated with electric vehicle charging stations with
[6] A. Pepiciello, C. Bernardo, J.L. Domínguez-García, Modeling of multi-energy fault ride-through capability, J. Clean. Prod. 420 (2023) 138372.
systems as multilayer networks, in: 2023 IEEE Belgrade PowerTech, IEEE, 2023, [33] P. García, J.P. Torreglosa, L.M. Fernández, F. Jurado, R. Langella, A. Testa, En-
pp. 01–06. ergy management system based on techno-economic optimization for microgrids,
[7] A. Serna, I. Yahyaoui, J.E. Normey-Rico, C. de Prada, F. Tadeo, Predictive control Electr. Power Syst. Res. 131 (2016) 49–59.
for hydrogen production by electrolysis in an offshore platform using renewable [34] Y. Zhai, J. Zhang, Z. Tan, X. Liu, B. Shen, T. Coombs, P. Liu, S. Huang, Research
energies, Int. J. Hydrogen Energy 42 (17) (2017) 12865–12876. on the application of superconducting magnetic energy storage in the wind power
[8] C. Bernardo, F. Vasca, A mixed logical dynamical model of the Hegselmann– generation system for smoothing wind power fluctuations, IEEE Trans. Appl.
Krause opinion dynamics, IFAC-PapersOnLine 53 (2) (2020) 2826–2831, 21st Supercond. 31 (5) (2021) 1–5.
IFAC World Congress. [35] B. Wang, G. Cai, D. Yang, Dispatching of a wind farm incorporated with dual-
[9] A. Latif, S.M.S. Hussain, A. Iqbal, D.C. Das, T.S. Ustun, A. Al-Durra, Coordination battery energy storage system using model predictive control, IEEE Access 8
of time delay and GCSC for frequency stabilization of dual-area interlinked (2020) 144442–144452.
microgrid using non-integer controller optimization, IET Renew. Power Gener. [36] M. Petrollese, L. Valverde, D. Cocco, G. Cau, J. Guerra, Real-time integration
17 (11) (2023) 2835–2852. of optimal generation scheduling with MPC for the energy management of a
[10] U.R. Nair, R. Costa-Castelló, A model predictive control-based energy manage- renewable hydrogen-based microgrid, Appl. Energy 166 (2016) 96–106.
ment scheme for hybrid storage system in islanded microgrids, IEEE Access 8 [37] F.J.V. Fernández, F. Segura Manzano, J.M. Andújar Márquez, A.J.
(2020) 97809–97822. Calderón Godoy, Extended model predictive controller to develop energy
[11] M.B. Abdelghany, M. Sheshzad, V. Mariani, D. Liuzza, L. Glielmo, Optimal management systems in renewable source-based smart microgrids with hydrogen
tracking of grid operated load demand with hydrogen-based storage system using as backup. Theoretical foundation and case study, Sustainability 12 (21) (2020)
model-based predictive control, in: Proceedings of WHEC, 2022, pp. 899–901. 8969.
[12] A. Saleh, H.M. Hasanien, R. A. Turky, B. Turdybek, M. Alharbi, F. Jurado, [38] M. Pereira, D. Limon, T. Alamo, L. Valverde, Application of periodic eco-
W.A. Omran, Optimal model predictive control for virtual inertia control of nomic MPC to a grid-connected micro-grid, IFAC-PapersOnLine 48 (23) (2015)
autonomous microgrids, Sustainability 15 (6) (2023) 5009. 513–518.
[13] X. Zhang, C. Huang, J. Shen, Energy optimal management of microgrid with [39] D. Pivetta, C. Dall’Armi, R. Taccani, Multi-objective optimization of hybrid
high photovoltaic penetration, IEEE Trans. Ind. Appl. 59 (1) (2023) 128–137. PEMFC/Li-ion battery propulsion systems for small and medium size ferries, Int.
[14] A.G. Li, M. Preindl, Assessing degradation-aware model predictive control for J. Hydrogen Energy 46 (72) (2021) 35949–35960.
energy management of a grid-connected PV-battery microgrid, in: 2022 IEEE [40] Y. Zhang, L. Fu, W. Zhu, X. Bao, C. Liu, Robust model predictive control for
Transportation Electrification Conference & Expo (ITEC), 2022, pp. 546–551. optimal energy management of island microgrids with uncertainties, Energy 164
[15] M. Cavus, A. Allahham, K. Adhikari, M. Zangiabadi, D. Giaouris, Energy (2018) 1229–1241.
management of grid-connected microgrids using an optimal systems approach, [41] M. Nasir, A. Rezaee Jordehi, M. Tostado-Véliz, S.A. Mansouri, E.R. Sanseverino,
IEEE Access 11 (2023) 9907–9919. M. Marzband, Two-stage stochastic-based scheduling of multi-energy microgrids
[16] T. Pippia, J. Sijs, B. De Schutter, A single-level rule-based model predictive with electric and hydrogen vehicles charging stations, considering transactions
control approach for energy management of grid-connected microgrids, IEEE through pool market and bilateral contracts, Int. J. Hydrogen Energy (2023).
Trans. Control Syst. Technol. 28 (6) (2020) 2364–2376. [42] V.S. Tabar, H. Banazadeh, M. Tostado-Véliz, A.R. Jordehi, M. Nasir, F. Jurado,
[17] M.B. Abdelghany, V. Mariani, D. Liuzza, L. Glielmo, Hierarchical model predic- Stochastic multi-stage multi-objective expansion of renewable resources and
tive control for islanded and grid-connected microgrids with wind generation electrical energy storage units in distribution systems considering crypto-currency
and hydrogen energy storage systems, Int. J. Hydrogen Energy (2023). miners and responsive loads, Renew. Energy 198 (2022) 1131–1147.

15
M.B. Abdelghany et al. Journal of Power Sources 591 (2024) 233844

[43] M. Asensio, P. Meneses de Quevedo, G. Muñoz-Delgado, J. Contreras, Joint [53] P. Marocco, M. Gandiglio, D. Audisio, M. Santarelli, Assessment of the role of
distribution network and renewable energy expansion planning considering hydrogen to produce high-temperature heat in the steel industry, J. Clean. Prod.
demand response and energy storage—Part I: Stochastic programming model, 388 (2023) 135969.
IEEE Trans. Smart Grid 9 (2) (2018) 655–666. [54] M. Arens, E. Worrell, W. Eichhammer, A. Hasanbeigi, Q. Zhang, Pathways to a
[44] X. Guo, Z. Bao, Z. Li, W. Yan, Adaptively constrained stochastic model predictive low-carbon iron and steel industry in the medium-term – the case of Germany,
control for the optimal dispatch of microgrid, Energies 11 (1) (2018). J. Clean. Prod. 163 (2017) 84–98.
[45] M.B. Abdelghany, A.M. Moustafa, M. Moness, Benchmarking tracking autopilots [55] S.E. Hosseini, Transition away from fossil fuels toward renewables: Lessons from
for quadrotor aerial robotic system using heuristic nonlinear controllers, Drones Russia–Ukraine crisis, Future Energy 1 (1) (2022) 2–5.
6 (12) (2022). [56] A.J. Pimm, T.T. Cockerill, W.F. Gale, Reducing industrial hydrogen demand
[46] M. Patel, S. Bohra, Power management of grid-connected PV wind hybrid system through preheating with very high temperature heat pumps, Appl. Energy 347
incorporated with energy storage system, Future Energy 2 (3) (2022) 7–19. (2023) 121464.
[47] R. Nebuloni, L. Meraldi, C. Bovo, V. Ilea, A. Berizzi, S. Sinha, R.B. Tamirisakan- [57] A. Das, H.K. Jani, G. Nagababu, S.S. Kachhwaha, A comprehensive review
dala, P. Raboni, A hierarchical two-level MILP optimization model for the of wind–solar hybrid energy policies in India: Barriers and recommendations,
management of grid-connected BESS considering accurate physical model, Appl. Renew. Energy Focus 35 (2020) 108–121.
Energy 334 (2023) 120697. [58] A.H. Azadnia, C. McDaid, A.M. Andwari, S.E. Hosseini, Green hydrogen supply
[48] S.A. Shezan, M.F. Ishraque, G. Shafiullah, I. Kamwa, L.C. Paul, S. Muyeen, R. chain risk analysis: A European hard-to-abate sectors perspective, Renew. Sustain.
NSS, M.Z. Saleheen, P.P. Kumar, Optimization and control of solar-wind islanded Energy Rev. 182 (2023) 113371.
hybrid microgrid by using heuristic and deterministic optimization algorithms [59] M.B. Abdelghany, A. Al-Durra, A coordinated model predictive control of grid-
and fuzzy logic controller, Energy Rep. 10 (2023) 3272–3288. connected energy storage systems, in: 2023 American Control Conference (ACC),
[49] P. Singh, M. Pandit, L. Srivastava, Multi-objective optimal sizing of hybrid 2023, pp. 1862–1867.
micro-grid system using an integrated intelligent technique, Energy 269 (2023) [60] A. Bemporad, M. Morari, Control of systems integrating logic, dynamics, and
126756. constraints, Automatica 35 (3) (1999) 407–427.
[50] L. Cheng, C. Yang, Distributed control for multiple hybrid energy storage systems [61] M.B. Abdelghany, A. Al-Durra, H. Zeineldin, F. Gao, Integrating scenario-based
using consensus algorithm in direct current power supply system, J. Power stochastic-model predictive control and load forecasting for energy management
Sources 588 (2023) 233701. of grid-connected hybrid energy storage systems, Int. J. Hydrogen Energy 48
[51] S.E. Hosseini, Hydrogen fuel, a game changer for the world’s energy scenario, (91) (2023) 35624–35638.
Int. J. Green Energy (2023) 1–17. [62] J. Garcia-Gonzalez, R.M.R. de la Muela, L.M. Santos, A.M. Gonzalez, Stochastic
[52] V. Javaheri, Steel pipeline for the hydrogen storage and delivery: metallurgical joint optimization of wind generation and pumped-storage units in an electricity
viewpoint for finnish ecosystem, Future Technol. 2 (1) (2023) 58–61. market, IEEE Trans. Power Syst. 23 (2) (2008) 460–468.

16

You might also like