Applied Energy 329 (2023) 120168

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Distributed MPC-based energy scheduling for islanded multi-microgrid

considering battery degradation and cyclic life deterioration
Arshad Nawaz a , Jing Wu a ,∗, Jun Ye b , Yidi Dong b , Chengnian Long a
a
Department of Automation and Key Laboratory of System Control and Information Processing, Ministry of Education of China, Shanghai Jiao Tong
University, Shanghai, China
b
POWERCHINA Shanghai Electric Power Engineering Co., Ltd., China

ARTICLE INFO ABSTRACT

Keywords: In this paper, distributed model predictive control (MPC) based energy scheduling problem is presented for
Multi-micorgrids islanded multi-microgrids. The objective is to achieve supply–demand balance in an individual microgrid
Distributed energy management through energy coordination and reduce the battery degradation for its extended cycle life. In order to
Battery degradation
achieve desired objective, the battery state of charge is taken as zone control optimization problem and new
Lagrange coordination
slack variable is introduced as optimized variable to restrict the battery state of charge in optimal range
Model predictive control
with considering battery degradation. Furthermore, accelerated distributed augmented Lagrangian (ADAL)
based distributed coordination strategy is presented, which can improve power supply reliability through
coordinated cooperation among MGs for energy exchanges. The system framework is modeled as mixed-
integer dynamic model which switches between different operating conditions. A mixed-integer quadratic
programming approach is addressed to solve the MPC optimization problem. The effectiveness of the proposed
scheme is validated through comparative performance with existing model. Finally, convergence analysis and
comparative fast convergence performance of the proposed scheme is provided.

1. Introduction 1.1. Background and motivation

Renewable energy sources (RESs), like Wind, photovoltaic (PV) The MMGs can be grid connected or can operate autonomously in is-
panels, and fuel cells, are nowadays extensively utilized due to their landed mode [7]. In grid-connected mode of MMGs operation, different
economical, ecological and political benefits [1]. By integrating them operators (for instance, distributed network operator (DNO), Market-
together with energy storage system in controlled environment known operator (MO) etc.) are involved in facilitation of energy exchange
as a microgrid (MG) [2], their power quality, supply reliability and the among MGs and main-grid [8]. Individual MG in grid-connected MMGs
energy efficiency can be improved with reduced carbon emissions [3]. acts as a controlled entity and tracks the control signal from upper level
However, due to the intermittent nature of renewable resources and controller to achieve local objectives (i.e. system stability, power supply
time varying load demand, it possess great challenge to the energy reliability, and supply–demand balance). In islanded mode, each MG in
management system (EMS) of a MG for optimal scheduling and reliable MMG network has limited local supply power and acts autonomously.
power delivery [4]. Furthermore, the amount of distributed generation Individual MG relies on its own local power resources and coordi-
that can be integrated into a single MG is finite which also limit its nates with other MGs in MMGs environment for cooperative energy
ability to deal with disturbances and ensuring power reliability [5]. exchange to achieve its local objectives [9]. Since the control output
Therefore, the concept of multi-MGs (MMGs) is emerged as a promising of individual MG is influenced by limited neighbors MGs, an efficient
solution to the aforementioned problem, which coordinates the energy power and information exchange for system stability in achieving local
sharing among MGs as well as achieves reliable power and optimal and system-wide objectives are still challenging and matter of concern
power scheduling according to time-varying loads and intermittent in MMG environment [10]. The key issues related to islanded MMG
renewable resources [6]. include, but are not limited to, achieving supply–demand balance with

∗ Corresponding author.
E-mail address: [email protected] (J. Wu).

https://doi.org/10.1016/j.apenergy.2022.120168
Received 1 July 2022; Accepted 15 October 2022
Available online 11 November 2022
0306-2619/© 2022 Elsevier Ltd. All rights reserved.
A. Nawaz et al. Applied Energy 329 (2023) 120168

for efficient economic schedule to achieve supply–demand balance in

Nomenclature
individual MG, it is important to consider both efficient ESS utilization
and lifetime degradation of the ESS in energy scheduling problem.
Sets/Indices

{𝑖, 𝑗} ∈ 𝑛 Indices for MGs 1.2. Literature review and contribution

𝑘 Index for time
𝑁𝑖 Binary set of 𝑀𝐺𝑖 ’s neighbor MGs In recent past, with improvement in computational processing and
communications, model predictive control (MPC) is extensively uti-
Variables lized for energy management of MMGs due to its inherent features
of model prediction [19], receding optimization and handling of con-
𝛥𝑃𝑖𝐿 (𝑘) Controllable loads
straints [20]. In the context, to fill the gap between supply and demand
𝑃𝑖𝐿 (𝑘) Load demand
in interconnected MMGs, many researchers have proposed MPC-based
𝛼𝑖𝑐ℎ,𝑑𝑐ℎ Binary variable for charging and discharg-
energy management schemes in literature [21,22]. Centralized MPC-
ing of battery
based schemes were proposed to achieve the optimal scheduling of the
𝜆 Lagrange multiplier charge/discharge of each local storage system and power exchanges
𝑏𝑎𝑡𝑖
𝐶𝑎𝑝𝑟𝑎𝑡𝑒𝑑 Battery rated capacity among MGs [23,24]. However, centralized method suffers from single-
𝑏𝑎𝑡
𝑃𝑖 (𝑘) Power exchanged with the energy storage point of failure and computational stresses. A hierarchical MPC-based
𝑃𝑖𝐶𝐿 (𝑘) Critical loads EMS was proposed in [25] for scheduling the generation of units in
𝑃𝑖𝑗𝑆 (𝑘) Amount of power received/sent from/to each MG through local controllers and handled the interaction between
neighbors of MG the main grid and MMGs. Authors in [26] proposed a hierarchical
𝑃𝑖PV (𝑘) PV𝑖 output power at time k stochastic energy management strategy for a MMGs using chance-
𝑃𝑖WT (𝑘) WT𝑖 output power at time k constrained framework. A robust MPC based two-level hierarchical
𝑆𝑜𝐶𝑖 (𝑘) State of energy storage of 𝑖𝑡ℎ battery EMS was proposed in [27] for enhancing utilization of its local RESs
𝑣𝑖𝑛,𝑗𝑖 Interacting neighbors inputs variable to and cost minimization of the energy purchasing from the main grid.
𝑀𝐺𝑖 Recently, energy management schemes based on distributed model pre-
dictive control (DMPC) integrated with cooperative and coordination
𝑣𝑜𝑢𝑡,𝑗𝑖 Interacting neighbors output variable from
algorithm have got great attention and reported in literature. In [28],
𝑀𝐺𝑖
a cooperative DMPC scheme was proposed for energy management
𝑃𝑖RES (𝑘) RES𝑖 power at time k
of interconnected MMGs. DMPC based economic scheduling with op-
Parameters timal procedure of a network of interconnected MGs was proposed
in [29] to reduce the degradation of storage system and maximum
𝜂𝑖ch ∕𝜂𝑖dch Charging/discharging efficiencies of bat-
economical benefit of the MG within relative constraints. An energy
tery storage
dispatch optimization problem for grid connected MGs through DMPC
𝛥𝑃𝑖𝐿 (𝑘) Upper limits of curtailed load was addressed in [30], where the cost of entire system is iteratively
𝑃𝑖𝑗𝑆 Maximum power exchange limit among minimized through decomposition method. In [31], the electricity cost
MGs minimization was achieved through coordinated optimal dispatch of
𝐿
𝑃𝑖 Upper limits of load demand ESS and optimal sharing of power among grid and MGs. Likewise,
𝑃𝐿𝑖 Lower limits of load demand a stochastic bi-level problem based control strategy for coordinated
𝑧𝑖 ∕𝑧𝑖 Proposed Lower/Upper limits of battery operation of MMGs to achieve reduction in overall system operation
storage costs was suggested in [32]. In this regard, due to potential benefits
𝜉𝑏𝑎𝑡 Battery degradation cost of distributed cooperative and coordinated operation of MMGs, it is
𝜉𝐿 Load curtailment cost getting attention of researchers for further improvement. However, the
𝑏𝑎𝑡 existing distributed EMS methods for MMG framework are not entirely
𝐶𝑎𝑝ref 𝑖 (𝑘) Battery maximum capacity reference
RES distributed in dealing with its neighbor node information handling.
𝑃𝑖 Upper limits of RESs𝑖 output power The findings of previous studies suggest that, compared to single
𝑆𝑜𝐶𝑖 ∕𝑆𝑜𝐶𝑖 Lower/Upper limits of battery storage microgrid, interconnecting multi-MGs with mutual cooperation has
significant advantages and benefits. Furthermore, as battery is an es-
sential and key source, the battery SoC must be optimized to extend
its life and performance efficiency. However, most existing work are
inadequate and intermittent energy resources, optimal energy storage focusing on the safety restrictions of the battery when doing energy
and cooperative energy exchange among neighboring MGs for power management and few of them considered the relationship of lifetime
reliability. versus performance efficiency of battery as well as rely on distributed
Furthermore, the energy storage system (ESS) is utilized in modern network operator for information exchange.
power systems to provide fast active power compensation [11] and In this paper, in order to highlight the main contribution, we have
also load-frequency control improvement [12], which are essential in defined energy scheduling problem for MMGs based on zone-MPC,
addressing shortage or surplus power due to mismatch between supply where the battery SoC is restricted within safe operating zones and em-
and demand in MGs and maintaining system reliability [13,14]. As one bedded in the optimal problem to extend cyclic life-time performance.
of ESS key part, the battery lifetime is dependent on its overcharging By penalizing the charging and discharging process for reducing the
or undercharging frequency [15] and its storage capacity degradation. degradation of ESS and restricting its SoC in proposed zone, the battery
Due to reduced output voltage, the batteries with very low state of cyclic life-span is increased and consequently the battery degradation
charge (SoC) are not capable of providing power to the MG system [16]. is reduced. An accelerated distributed augmented Lagrangian (ADAL)
Moreover, as battery efficiency is dependent on SoC, it also affects the based coordination scheme is utilized to guarantee local and system-
optimization objective function for optimal energy scheduling [17]. wide supply–demand balance. The convergence of the proposed algo-
The cycle based aging of a battery at a high depth of discharge rithm and its comparative computational analysis against alternating
(DoD) will result in a substantial reduction in its life [18]. Therefore, direction method of multiplier (ADMM) method are performed.

2
A. Nawaz et al. Applied Energy 329 (2023) 120168

Fig. 1. General illustration of MMGs structure.

1.3. Paper outline Let the load demand of 𝑖𝑡ℎ MG at time k be denoted as 𝑃𝑖𝐿 (𝑘) and
modeled with related constraints as follows:
The organization of rest of the paper is as follows. Section 2 presents
𝑃𝑖𝐿 (𝑘) = 𝑃𝑖𝐶𝐿 (𝑘) + 𝛥𝑃𝑖𝐿 (𝑘) (1)
the system description and modeling. In Section 3, problem formulation
𝐿
and proposed strategy for energy management of MMG are presented. 𝑃𝐿
𝑖 ≤ 𝑃𝑖𝐿 (𝑘) ≤ 𝑃𝑖 , ∀𝑖 ∈ 𝑛, (2)
In Section 4, simulations are provided to show effectiveness of proposed 0≤ 𝛥𝑃𝑖𝐿 (𝑘) ≤ 𝛥𝑃𝑖𝐿 (𝑘), ∀𝑖 ∈ 𝑛 (3)
method. Finally, conclusion is drawn in Section 5.
𝐿
Here 𝑃 𝐿
𝑖 and are the lower and upper limits of the load demand
𝑃𝑖
2. System description and modeling respectively. The constraint (3) represents the upper and lower bounds
of controllable load.
In this section, a general structure of MMGs and the dynamic
modeling of MG’s components along with their relevant constraints are 2.3. Distributed generation model
provided.
We have considered two types of renewable sources for power
generation in microgrid. The renewable sources like wind and PV
2.1. System architecture
are not being dispatchable and the actual generation from renewable
sources can be represented as:
A MMG network comprised of 𝑛 MGs is depicted in Fig. 1, where
each MG in network can be regarded as one subsystem. It is comprised 𝑃𝑖RES (𝑘) = 𝑃𝑖WT (𝑘) + 𝑃𝑖PV (𝑘) ∀𝑖 ∈ 𝑛 (4)
of RES, ESS and electrical loads. The balance of supply and demand in
Here 𝑃𝑖RES (𝑘), 𝑃𝑖WT (𝑘)
and 𝑃𝑖PV (𝑘)
indicate the total actual RES genera-
𝑖𝑡ℎ MG depends on its RES, ESS and exchange of surplus power with
tion, wind and PV power generation at time 𝑘, respectively. The renew-
neighbors through mutual coordination.
able infeed is limited due to its intermittent nature and is expressed as:

2.2. Load model RES

0 ≤ 𝑃𝑖RES (𝑘) ≤ 𝑃 𝑖 , ∀𝑖 ∈ 𝑛 (5)
We classify loads into two categories as follows: RES
where 𝑃𝑖 is the maximum output power produced by the renewable
(I) Critical loads: refers to the loads which cannot be shed and source.
power supply to them is always assured. For instance, refrigera-
tor, hospital emergency operations, security system, and police 2.4. Energy storage model
office, etc., can be considered as critical loads. Critical load are
represented as 𝑃𝑖𝐶𝐿 (𝑘). In microgrid networks, the storage system (batteries, super ca-
pacitor, etc.) is of prime importance as it helps in smoothing the
(II) Controllable loads: means the load which can be shed, shifted
intermittent power production from RES and also provides support
in time or curtailed at any time to smooth load curve. These
in shaving peak power loads. Due to high energy density and low
types of load are represented as 𝛥𝑃𝑖𝐿 (𝑘). The controllable loads
self-discharge properties, Lithium-ion battery is selected as an energy
are like electric heating/central air-conditioning system whose
storage in this work. The dynamic energy storage model of 𝑀𝐺𝑖 at time
power consumption can be changed if the resulting thermal
𝑘 can be represented by its state of charge (SoC) as:
comfort is within a range specified by the consumer, and plug-in
hybrid electric vehicles (PHEV), whose battery charging rate can ⎡ 𝜂 ch 𝛼 ch (𝑘) 𝛼𝑖dis (𝑘) ⎤
be adjusted as long as the charging process is completed before 𝑆𝑜𝐶𝑖 (𝑘 + 1) = 𝑆𝑜𝐶𝑖 (𝑘) + ⎢ 𝑖 𝑏𝑎𝑡 𝑖
+ ⎥ ⋅ 𝑃 𝑏𝑎𝑡 (𝑘)𝛥𝑡 (6)
⎢ 𝐶𝑎𝑝 𝑖 (𝑘) 𝜂 dis ⋅ 𝐶𝑎𝑝 𝑖 (𝑘) ⎥ 𝑖
𝑏𝑎𝑡
the vehicle departure, etc. ⎣ ref 𝑖 ref ⎦

3
A. Nawaz et al. Applied Energy 329 (2023) 120168

0 < 20%𝑆𝑜𝐶𝑖 ≤ 𝑧𝑖 < 𝑆𝑜𝐶𝑖 (𝑘) ≤ 𝑧𝑖 < 100%𝑆𝑜𝐶𝑖 (7)

𝑃𝑖𝑏𝑎𝑡 (𝑘) = 𝑃𝑖𝑐ℎ,𝑏𝑎𝑡 (𝑘) − 𝑃𝑖𝑑𝑐ℎ,𝑏𝑎𝑡 (𝑘) (8)

0 ≤ 𝛼𝑖ch (𝑘) + 𝛼𝑖dis (𝑘) ≤ 1 (9)
𝑃𝑖𝑏𝑎𝑡,min 𝛼𝑖ch (𝑘) ≤ 𝑃𝑖𝑐ℎ,𝑏𝑎𝑡 (𝑘) ≤ 𝑃𝑖𝑏𝑎𝑡,max 𝛼𝑖ch (𝑘) (10)
𝑃𝑖𝑏𝑎𝑡 (𝑘) − 𝑃𝑖𝑏𝑎𝑡,max (1 − 𝛼𝑖ch (𝑘)) ≤ 𝑃𝑖𝑐ℎ,𝑏𝑎𝑡 (𝑘) (11)
( )
𝑃𝑖𝑐ℎ,𝑏𝑎𝑡 (𝑘) ≤ 𝑃𝑖𝑏𝑎𝑡 (𝑘) − 𝑃𝑖𝑏𝑎𝑡,min 1 − 𝛼𝑖ch (𝑘) (12)
Here 𝑆𝑜𝐶𝑖 (𝑘) indicates the state of energy storage of 𝑖𝑡ℎ battery at time
𝑘 and 𝑆𝑜𝐶𝑖 and 𝑆𝑜𝐶𝑖 are the lower and upper safety bound of battery
storage, respectively. 𝑧𝑖 and 𝑧𝑖 represents the lower and upper zones
limits of SOC, respectively. Note that zone limits [𝑧𝑖 𝑧𝑖 ] we introduced
here are to minimize overcharge or deep-discharge of battery and avoid
depth of discharge which significantly deteriorates its life and perfor-
mance efficiency. 20%𝑆𝑜𝐶𝑖 is generally consider in practical system
𝑏𝑎𝑡
as lower bound on SoC. 𝐶𝑎𝑝ref 𝑖 is the reference for battery maximum
storage capacity. 𝑃𝑖𝑏𝑎𝑡 (𝑘) represents the net power exchanged between
the energy storage and microgrid. 𝜂𝑖ch and 𝜂𝑖dch are charging and dis-
charging efficiencies for demonstrating charging/discharging losses, Fig. 2. Neighbor information exchange through Lagrange coordination.
respectively. The charging/discharging process cannot occur simultane-
ously. Therefore, two logical variables 𝛼𝑖ch and 𝛼𝑖dis are used for selecting
and avoiding simultaneous operation of battery charging and discharg-
ing power, by which we can convert logical variables into inequalities 2.5. Power balance
easily as constraints (9)–(12) in energy storage modeling [33].
Since each microgrid utilizes its local RES, battery storage, and
2.4.1. ESS degradation exchanges power with its neighboring microgrid to balance its supply
The degradation of batteries performance is one of the most eco- demand, the power balance dynamics of 𝑖𝑡ℎ microgrid can be expressed
nomical issue that has to be included in the energy model in order to as:
∑
take into account its extended operation life. Battery degradation refers 𝑃𝑖RES (𝑘) − 𝑃𝑖𝐿 (𝑘) =𝑃𝑖𝑏𝑎𝑡 (𝑘) + 𝑎𝑖𝑗 𝑃𝑖𝑗𝑆 (𝑘) − 𝛥𝑃𝑖𝐿 (𝑘), ∀𝑖 ∈ 𝑛 (18)
to the reduction in the reference capacity of battery for storing certain 𝑗∈𝑁𝑖
amount of energy, or the amount of throughput power it can deliver.
The degradation of battery is usually determined by its state of health 𝑃𝑖𝑗𝑆 (𝑘) ≤ 𝑃𝑖𝑗𝑆 (19)
(SOH) and mathematically can be expressed as [34]: Here 𝑃𝑖𝑗𝑆 (𝑘), ∀𝑗
= 1, 2, … , 𝑛|𝑗 ≠ 𝑖 is the interconnecting variable and
𝐶𝑎𝑝𝑏𝑎𝑡
ref
(𝑘) represent the amount of power received/sent from/to neighbors of 𝑖𝑡ℎ
SOH(𝑘) = , (13) MGs at time 𝑘, respectively. The 𝑃𝑖𝑗𝑆 is the maximum power exchange
𝐶𝑎𝑝𝑏𝑎𝑡
𝑟𝑎𝑡𝑒𝑑
limit among MGs in accordance to transmission line capacity, and 𝑎𝑖𝑗
SOH𝑖 (𝑘) ≥ SOH𝑚𝑖𝑛 (14) is the coordinating factor. The neighbor MGs of 𝑀𝐺𝑖 is defined by the
binary set as 𝑁𝑖 = {(𝑖, 𝑗), 𝑗 ∈ N|𝑗 ≠ 𝑖}. The interacting inputs/outputs
Here 𝐶𝑎𝑝𝑏𝑎𝑡
ref
(𝑘) is the reference for battery maximum capacity storage
variable for 𝑀𝐺𝑖 is shown in Fig. 2 and can be defined as:
at time 𝑘 and 𝐶𝑎𝑝𝑏𝑎𝑡 𝑟𝑎𝑡𝑒𝑑
is the rated capacity of battery. The SOH𝑖 is the
𝑖𝑡ℎ battery state of health relative to degradation of battery which is 𝑣𝑖𝑛,𝑗𝑖 (𝑘|𝑘) = 𝑎𝑖𝑗 𝑢𝑗 (𝑘 − 1|𝑘) (20)
limited by constraint (14). The prime cause of battery degradation is
due to cyclic and calender aging. The cyclic aging is battery capacity 𝑣𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘) = 𝑎𝑖𝑗 𝑢𝑖 (𝑘 − 1|𝑘) (21)
loss which is linearly consequent of SoC variations related to battery where 𝑣𝑖𝑛,𝑗𝑖 (𝑘|𝑘) and 𝑣𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘) are the predicted interconnecting input
discharge operation, where calender aging is related to manufacture and output variable, respectively. 𝑢𝑖 and 𝑢𝑗 are the set of control
design, material, operating temperature and other environmental fac- variables for local and neighbor MGs. These variables represent the
tor. These two aging phenomena are not interrelated. Therefore, in power exchange with neighbor MGs. The interacting variable can be
this work, calendar aging is considered absent and not contributing to mathematically presented as follows:
battery aging. The reference of battery storage is affected by cycling
aging with discharge operation of battery. The capacity reference of 𝑣𝑖𝑛,𝑖 (𝑘|𝑘) = [𝑣𝑇𝑖𝑛,1𝑖 (𝑘|𝑘), 𝑣𝑇𝑖𝑛,2𝑖 (𝑘|𝑘), … , 𝑣𝑇𝑖𝑛,𝑛𝑖 (𝑘|𝑘)]𝑇 (22)
battery after discharge at each step size can be obtained as:
𝑣𝑜𝑢𝑡,𝑖 (𝑘|𝑘) = [𝑣𝑇𝑜𝑢𝑡,1𝑖 (𝑘|𝑘), 𝑣𝑇𝑜𝑢𝑡,2𝑖 (𝑘|𝑘), … , 𝑣𝑇𝑜𝑢𝑡,𝑛𝑖 (𝑘|𝑘)]𝑇 (23)
𝑏𝑎𝑡 𝑏𝑎𝑡 𝑏𝑎𝑡
𝐶𝑎𝑝ref 𝑖 (𝑘) =𝐶𝑎𝑝ref 𝑖 (𝑘 − 1) − 𝛥𝐶𝑎𝑝ref 𝑖 (𝑘) (15) According to the conservation of energy, the common interacting vari-
𝑏𝑎𝑡 𝑏𝑎𝑡𝑖 [ ]
𝛥𝐶𝑎𝑝ref 𝑖 (𝑘) =𝐶𝑎𝑝𝑟𝑎𝑡𝑒𝑑 × 𝑍 ⋅ 𝑆𝑜𝐶𝑖 (𝑘 − 1) − 𝑆𝑜𝐶𝑖 (𝑘) (16) able must be equal. Therefore, from 𝑀𝐺𝑖 point of view,

Putting back Eqs. (15) and (16) in Eq. (13), the SOH of battery as a 𝑣𝑖𝑛,𝑗𝑖 (𝑘|𝑘) = 𝑣𝑜𝑢𝑡,𝑖𝑗 (𝑘|𝑘) (24)
result of cycling aging at time step k can be obtained as:
𝑣𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘) = 𝑣𝑖𝑛,𝑖𝑗 (𝑘|𝑘) (25)
𝑏𝑎𝑡
𝐶𝑎𝑝ref 𝑖 (𝑘 − 1) [ ]
SOH𝑖 (𝑘) = 𝑏𝑎𝑡
− 𝑍 ⋅ 𝑆𝑜𝐶𝑖 (𝑘 − 1) − 𝑆𝑜𝐶𝑖 (𝑘) . (17)
𝑖
𝐶𝑎𝑝𝑟𝑎𝑡𝑒𝑑 2.6. Predictive model

Here 𝑍 represent the coefficient of linear aging and its value depends Based on dynamic models of microgrid component, power bal-
on the battery material. Eq. (16) is always positive as it is computed ance constraints, and interaction with its neighbor MGs, the dynamic
only in discharge operation. The new capacity of reference calculated behavior of 𝑖𝑡ℎ microgrid can be modeled as follow:
by Eq. (15) is considered in the Eq. (6) of energy storage model such
as aging involves reduction in charge storing capacity of ESS. 𝑥𝑖 (𝑘 + 1|𝑘) =𝐴𝑖 𝑥𝑖 (𝑘|𝑘) + 𝐵𝑖 𝑢𝑖 (𝑘|𝑘) (26)

4
A. Nawaz et al. Applied Energy 329 (2023) 120168

𝑦𝑖 (𝑘|𝑘) =𝐶𝑦𝑖 𝑢𝑖 (𝑘|𝑘) (27)

where

𝑥𝑖 (𝑘|𝑘) = [𝑆𝑜𝐶𝑖 (𝑘|𝑘)]

[ ]𝑇
𝑢𝑖 (𝑘|𝑘) = 𝑃𝑖𝑏𝑎𝑡 (𝑘) 𝛥𝑃𝑖𝐿 (𝑘|𝑘) 𝑃𝑖1𝑆 (𝑘|𝑘) 𝑃𝑖2𝑆 (𝑘|𝑘), … , 𝑃𝑖𝑛𝑆 (𝑘|𝑘)
𝑦𝑖 (𝑘|𝑘) = [𝑃𝑖RES (𝑘) − 𝑃𝑖𝐿 (𝑘)]
𝐴𝑖 = [1],
[ ( ) ]
𝜂𝑖ch 𝛼𝑖ch (𝑘) 𝛼𝑖dis (𝑘)
𝐵𝑖 = + 𝛥𝑡 0, … , 0
𝐶𝑎𝑝𝑏𝑎𝑡
ref
(𝑘) 𝜂𝑖dch .𝐶𝑎𝑝𝑏𝑎𝑡
ref
(𝑘)
𝐶𝑦𝑖 = [1 −1 𝑎𝑖1 , 𝑎𝑖2 , … , 𝑎𝑖𝑛 ].

Here, 𝐴𝑖 , 𝐵𝑖 , and, 𝐶𝑦𝑖 are the relevant system matrices of dynamic

predictive model.

2.7. Control objective

The control objective in our paper is to balance local and global

supply and demand by coordinating local RES, battery storage and sur-
plus/deficit power with/from neighbors MG as well as minimizing bat-
tery’s performance degradation and extending life span by maintaining
battery SoC within safe zone to minimize.
Fig. 3. Proposed distributed energy management scheme for Multi-microgrid.

3. Problem formulation
Therefore, the local distributed zone MPC optimization problem for
To meet the above objectives, an optimal problem is formulated for 𝑀𝐺𝑖 can be expressed as:
distributed zone MPC of individual MGs. The proposed scheme is shown
in Fig. 3. The coordination problem related to coupling variables for min 𝐽Local,i (⋅) (30)
𝛿𝑖 (𝑘+𝓁|𝑘),𝑢𝑖 (𝑘|𝑘)
exchanging information of surplus and deficit power with neighbor-
S.t: 𝑋𝑖 (𝑘 + 1|𝑘) = 𝑆𝑥 𝑋𝑖 (𝑘|𝑘) + 𝑆𝑢 𝑈𝑖 (𝑘|𝑘) (31)
ing MGs is achieved using ADAL based distributed coordination. The
detail problem formulation of distributed zone MPC and distributed 𝑧𝑖 ≤ 𝛿𝑖 (𝑘 + 𝓁|𝑘) ≤ 𝑧𝑖 (32)
coordination cost functions are discussed in proceeding subsections.
and Constraints (2-3), (5), (8-12), (18-19), (24-25)
and (26-27)
3.1. Distributed zone-MPC
where the coefficient matrices in Eq. (31) are:
For 𝑀𝐺𝑖 at time 𝑘, the distributed zone MPC targets are to achieve
𝑋𝑖 (𝑘 + 1|𝑘) = [𝑥𝑖 (𝑘 + 1|𝑘), 𝑥𝑖 (𝑘 + 2|𝑘), … , 𝑥𝑖 (𝑘 + 𝑃 |𝑘)]𝑇
supply and demand balance while keeping battery charge level within
expected range [𝑧𝑖 𝑧𝑖 ]. Therefore, we introduce optimization variables 𝑈𝑖 (𝑘|𝑘) = [𝑢𝑖 (𝑘|𝑘), 𝑢𝑖 (𝑘 + 1|𝑘), … , 𝑢𝑖 (𝑘 + 𝑀 − 1|𝑘)]𝑇
𝛿𝑖 to satisfy the zone constraint of battery SoC subject to supply and
demand objective. ⎡ 𝐵𝑖 0 … 0 ⎤
Let 𝑥𝑖 (𝑘 + 𝓁|𝑘) represents the predicted value for battery SoC, then ⎡ 𝐴𝑖 ⎤ ⎢ ⎥
⎢ 2 ⎥ ⎢ ⎥
⎥ , 𝑆𝑢 = ⎢ 𝐴𝑖 𝐵𝑖 𝐵𝑖 … 0
the local performance index for 𝑀𝐺𝑖 can be expressed as follow: 𝑆𝑥 = ⎢ 𝐴𝑖 ⎥
[ ] ⎢ ⋮ ⎥ ⎢ ⋮ ⋮ ⋱ ⎥
∑𝑃
‖ ‖2 ⎢ 𝑃 ⎥ ⎢ ⎥
𝐽Local,i (⋅) = ‖𝛿𝑖 (𝑘 + 𝓁|𝑘) − 𝑥𝑖 (𝑘 + 𝓁|𝑘)‖ ⎣ 𝐴𝑖 ⎦ ⎢ 𝑃 −1 ⎥
‖ ‖ ⎣ 𝐴𝑖 𝐵𝑖 𝐴𝑃𝑖 −2 𝐵𝑖 𝐴𝑃𝑖 −𝑀 𝐵𝑖 ⎦
𝓁=1 ‖ ‖𝑄1(𝑖) …
⎡‖‖ 𝑏𝑎𝑡 ‖ ⎤‖2
⎢ ‖𝜉𝑏𝑎𝑡 ⋅ 𝑃𝑖 (𝑘 + 𝓁|𝑘)‖ ⎥
⎢ ‖ ‖𝑅 1(𝑖) ⎥ 3.2. Distributed coordination
∑⎢ ‖
𝑀−1
‖ 2
⎥
‖ 𝐿 ‖
+ ⎢+‖𝜉𝐿 ⋅ 𝛥𝑃𝑖 (𝑘 + 𝓁|𝑘)‖ ⎥ (28)
𝓁=0 ⎢ ‖ ‖𝑅2(𝑖) ⎥ In MMGs environment, individual MG interact with its neighboring
⎢ ‖ ‖2 ⎥
⎢ +‖ ‖ MG for possible cooperation as shown in Fig. 2. The individual MG
‖𝛥𝑢𝑖 (𝑘 + 𝓁|𝑘)‖ ⎥
⎣ ‖ ‖𝑅3(𝑖) ⎦ resolves distributed zone MPC problem based on its local information
where 𝛿𝑖 (𝑘 + 𝓁|𝑘) is the new predicted variable of battery SoC which and interacting neighbor information, which are the information about
is limited within a given range [𝑧𝑖 𝑧𝑖 ]. 𝜉𝑏𝑎𝑡 is a factor to penalize surplus and deficit power of neighbors. Therefore, in order to facilitate
charging and discharging process for reducing the degradation of ESS the information exchange among neighboring MGs, an ADAL based dis-
and 𝜉𝐿 is a factor to penalize the load curtailment. The 𝑄1(𝑖) ≥ 0 is tributed coordination is utilized. This formulation incorporates several
diagonal weight matrix to penalize the deviation between 𝛿𝑖 (𝑘 + 𝓁|𝑘) iterations, during which individual MG carry out its local computa-
and predicted value 𝑥𝑖 (𝑘 + 𝓁|𝑘). 𝑅1(𝑖) , 𝑅2(𝑖) 𝑅3(𝑖) are positive diagonal tions based on its own resources data and information received from
neighbors at previous time. The iteration continues until the difference
weight matrices for input variables, 𝑃 and 𝑀 are prediction horizon
between interacting variables among neighbor MG, computed by indi-
and control horizon respectively. The predicted control signal will be
vidual MG, is within expected tolerance. In order to relax the equation
zero after control horizon. The variable increment of inputs are:
constraints, the constraints (24) and (25) are added to the objective
𝛥𝑢𝑖 (𝑘|𝑘) = 𝑢𝑖 (𝑘|𝑘) − 𝑢𝑖 (𝑘 − 1|𝑘) (29) function in terms of additional linear cost terms and quadratic terms

5
A. Nawaz et al. Applied Energy 329 (2023) 120168

Algorithm 1 Zone DMPC for Energy Management 2𝛾𝑐 is used and coordination objective function is convex, the iteration
will converge to minimum for less 𝛾𝜀 .
1: Initialize values
2: for 𝑘 = 1 do Consequently, the termination condition for iteration ‖𝜆(𝑠+1)
𝑖𝑛,𝑗𝑖 (𝑘|𝑘) −
3: Get measurement of the current states and inputs 𝜆(𝑠) ≤ 𝛾𝜀 and ‖𝜆(𝑠+1)
𝑖𝑛,𝑗𝑖 (𝑘|𝑘)‖∞
(𝑠)
𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘) − 𝜆𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘)‖∞ ≤ 𝛾𝜀 can be satisfied.
4: for 𝑘 < 𝑁 do Finally, the over all cost function of 𝑖𝑡ℎ microgrid can be expressed as:
5: At iteration 𝑠 = 1, initialize Lagrange multiplier ∑
(𝑠) (𝑠) min 𝐽Local,i (⋅) + 𝐽coord,i (⋅) (34)
𝜆𝑖𝑛,𝑗𝑖 (𝑘), 𝜆𝑜𝑢𝑡,𝑗𝑖 (𝑘) arbitrarily 𝛿𝑖 (𝑘+𝓁|𝑘),𝑢𝑖 (𝑘|𝑘)
𝑗∈𝑁𝑖
6: for all 𝑠 < 𝑠𝑚𝑎𝑥 do
7: Solve the optimization problem (34) 𝑠.𝑡. ∶ Constraints (2-3), (5), (8-12), (18-19), (26-27)
8: Determine 𝛿𝑖(𝑠+1) (𝑘 + 𝓁|𝑘), 𝑢(𝑠+1)
𝑖 (𝑘), 𝑣(𝑠+1) (𝑠+1)
𝑖𝑛,𝑖 (𝑘), 𝑣𝑜𝑢𝑡,𝑖 (𝑘) Send and (31-32)
interconnection variables 𝑣(𝑠+1) (𝑠+1)
𝑖𝑛,𝑖 (𝑘), 𝑣𝑜𝑢𝑡,𝑖 (𝑘) to neighbors To make the optimization problem in Eq. (34) solvable, the local
controller 𝑗 ∈ 𝑁𝑖 objective is rewritten as
9: Update Lagrange Multipliers
𝜆(𝑠+1) (𝑠) (𝑠+1) (𝑠+1) ‖ ‖2
𝑖𝑛,𝑗𝑖 (𝑘) = 𝜆𝑖𝑛,𝑗𝑖 (𝑘) + 𝜏𝛾𝑏 (𝑣𝑖𝑛,𝑗𝑖 (𝑘) − 𝑣𝑜𝑢𝑡,𝑖𝑗 (𝑘)) 𝐽Local,i (⋅) = ‖ ‖
‖𝜌𝑖 (𝑘 + 1|𝑘) − 𝑋𝑖 (𝑘 + 1|𝑘)‖
𝜆(𝑠+1) (𝑠) (𝑠+1) (𝑠+1) ‖ ‖𝑄
𝑜𝑢𝑡,𝑗𝑖 (𝑘) = 𝜆𝑜𝑢𝑡,𝑗𝑖 (𝑘) + 𝜏𝛾𝑏 (𝑣𝑜𝑢𝑡,𝑗𝑖 (𝑘) − 𝑣𝑖𝑛,𝑖𝑗 (𝑘))
‖ ‖2 ‖ ‖2
10: Evaluate end-iteration conditions +‖ 𝑏𝑎𝑡 ‖ ‖ 𝐿 ‖
‖𝜉𝑏𝑎𝑡 ⋅ 𝑃𝑖 (𝑘|𝑘)‖ + ‖𝜉𝐿 ⋅ 𝛥𝑃𝑖 (𝑘|𝑘)‖
𝑠 ≥ 𝑠𝑚𝑎𝑥 ‖ ‖𝑅1 ‖ ‖𝑅2
‖𝜆(𝑠+1) (𝑠)
𝑖𝑛,𝑗𝑖 (𝑘|𝑘) − 𝜆𝑖𝑛,𝑗𝑖 (𝑘|𝑘)‖∞ ≤ 𝛾𝜀
‖ ‖2
+‖ ‖
‖Hi 𝑈𝑖 (𝑘|𝑘) + ℎ𝑖 ‖ (35)
‖𝜆(𝑠+1) (𝑠)
𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘) − 𝜆𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘)‖∞ ≤ 𝛾𝜀
‖ ‖𝑅3
11: if conditions are not satisfied then where,
12: 𝑠 ← 𝑠 + 1 and go to step 7.
⎡ 𝛿𝑖𝑇 (𝑘 + 1|𝑘) ⎤
13: else ⎢ ⎥ ⎡𝑄𝑖 (1) ⎤
14: Receive the future input control sequence ⎢ ⎥
𝜌𝑖 (𝑘 + 1|𝑘) = ⎢ 𝛿𝑖𝑇 (𝑘 + 2|𝑘) ⎥ , 𝑄 = ⎢ ⋱ ⎥,
15: end if ⎢ ⎥ ⎢ ⎥
⋮ ⎣ 𝑄𝑖 (𝑃 )⎦
16: end for ⎢ 𝑇 ⎥
⎣𝛿𝑖 (𝑘 + 𝑃 |𝑘)⎦
17: 𝑘 ← 𝑘 + 1 and proceed back to step 3
18: end for ⎡𝑅𝑖 (1) ⎤
19: end for 𝑅=⎢ ⋱ ⎥
⎢ ⎥
⎣ 𝑅𝑖 (𝑀 − 1)⎦
and
based on Lagrange multipliers. The augmented ADAL coordination cost ⎡ 𝐼(𝑛+2) ⎤
⎢−𝐼 𝐼(𝑛+2) ⎥
function can be expressed as: ⎢ (𝑛+2) ⎥
[ (𝑠) ]𝑇 [ (𝑠) ] Hi = ⎢ − 𝐼(𝑛+2) 𝐼(𝑛+2) ⎥
𝜆𝑖𝑛,𝑗𝑖 𝑣𝑖𝑛,𝑗𝑖 (𝑘|𝑘) ⎢ ⋱ ⎥
𝐽coord,i (⋅) = ⎢ ⎥
−𝜆(𝑠) 𝑣(𝑠) ⎣ −𝐼(𝑛+2) 𝐼(𝑛+2) ⎦
𝑜𝑢𝑡,𝑗𝑖 𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘) [ ]
[ (𝑠−1) ] ℎ𝑖 = −𝑢𝑇𝑖 (𝑘 − 1) 01×(𝑀−1) 01×(𝑀−1)
𝑇
𝛾 ‖ 𝑖𝑛,𝑖𝑗
𝑣 (𝑘|𝑘) − 𝑣(𝑠)
𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘) ‖
2
+ 𝑐‖ ‖
‖
‖
2‖ 𝑣 (𝑠−1) (𝑠)
‖2 Let the optimized variable be the followings:
𝑜𝑢𝑡,𝑖𝑗 (𝑘|𝑘) − 𝑣𝑖𝑛,𝑗𝑖 (𝑘|𝑘) [ ]
[ (𝑠) ]
𝛾 − 𝛾𝑐 ‖
(𝑠−1)
𝑣 (𝑘|𝑘) − 𝑣𝑖𝑛,𝑗𝑖 (𝑘|𝑘) ‖2 ̂ = 𝜌𝑖 (𝑘 + 1|𝑘)
X (36)
+ 𝑏 ‖ 𝑖𝑛,𝑗𝑖 ‖ (33) 𝑖
𝑈𝑖 (𝑘|𝑘)
2 ‖ (𝑠)
‖ 𝑣𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘) − 𝑣(𝑠−1)
‖
‖2
𝑜𝑢𝑡,𝑗𝑖 (𝑘|𝑘)
Substituting 𝑋𝑖 (𝑘 + 1|𝑘) to Eq. (34), the optimization performance
(𝑠)
where 𝜆𝑖𝑛,𝑗𝑖 and 𝜆(𝑠) 𝑜𝑢𝑡𝑗𝑖 are the Lagrange multipliers related to equality
index can be transformed into standard quadratic programming (QP)
constraint (24) and (25), 𝛾𝑏 , 𝛾𝑐 are both positive weighting scalars, as follow:
and 𝑣(𝑠−1) (𝑠−1) (𝑠−1) (𝑠−1) ∑ 𝛾𝑏 𝑇 𝑻
𝑖𝑛,𝑗𝑖 (𝑘), 𝑣𝑜𝑢𝑡,𝑖𝑗 (𝑘), 𝑣𝑖𝑛,𝑖𝑗 (𝑘), 𝑣𝑜𝑢𝑡,𝑗𝑖 (𝑘) are the interacting information ̂𝑇 𝜱 X
𝐽coord,i (⋅) = X ̂ 𝑇̂
𝐽Local,i (⋅)+ 𝑖 𝒊 𝑖 + 𝜳 𝑖 X𝑖 + 𝝀𝑨X + X 𝑨 𝑨X𝑠 (37)
variables computed at previous iteration 𝑠 − 1. 𝑗∈𝑁
2 𝑠
𝑖

where
Remark 1. The Lagrange multipliers, and𝜆(𝑠)
𝑖𝑛,𝑗𝑖 𝜆(𝑠)
can be arbitrarily
𝑜𝑢𝑡𝑗𝑖 [ ]
initialized. However, a faster convergence decision can be achieved by 𝑄 0
𝜱𝑖 = ,
initializing it with estimate close to optimal value. Consequently, using −2𝑆𝑢𝑇 𝑄 𝛩
values obtained from previous iteration for initialization of Lagrange
multiplier is always effective. 𝛩 = 𝑆𝑢𝑇 𝑄𝑆𝑢 + 𝐾1𝑇 𝜉𝑏𝑎𝑡
𝑇
𝑅1 𝜉𝑏𝑎𝑡 𝐾1
𝛾𝑐 is a positive scalar and it penalizes the deviation of exchange + 𝐾2𝑇 𝜉𝐿𝑇 𝑅2 𝜉𝐿 𝐾2 + 𝐇𝐓 𝑅 H
𝐢 3 i
variables from their previous iteration. Therefore, choosing higher
value for 𝛾𝑐 can result in faster convergence of exchange variable to [ ]𝑇 [ ]
−2𝑋𝑖 𝑆𝑥 𝑄 𝐴𝑖𝑗 0
satisfy equality constraints. However, it results in increased iterations. 𝜳 𝑇𝑖 = , 𝑨 = ,
Accordingly, choosing lower value of 𝛾𝑐 results in slow convergence for 2(ℎ𝑇𝑖 𝑅3 Hi + 𝑋𝑖𝑇 𝑆𝑥𝑇 𝑄𝑆𝑢 ) 0 𝐴𝑗𝑖
satisfying equality constraints and requires less iterations for reaching
optimality [35]. [ ] [ 𝑠 ]𝑇
𝝀 = 𝜆𝑠𝑖𝑛,𝑗𝑖 −𝜆𝑠𝑜𝑢𝑡,𝑗𝑖 , X = X
̂ ̂𝑠
X ,
𝑗 𝑖
Remark 2. Additional weighting positive scalar, 𝛾𝑏 , is used to provide and
less incentive for change in interacting variables between consecutive ⎡ ̂ 𝑠 ̂ 𝑠−1 ⎤
X − X𝑗 ⎥
iteration of agent 𝑀𝐺𝑖 . If 𝛾𝑏 > 𝛾𝑐 then the 𝑀𝐺𝑖 interacting variables X𝑠 = ⎢ 𝑗𝑠
deviation between current and last iteration is penalized. In case 𝛾𝑏 ≥ ⎢X̂ ̂ 𝑠−1 ⎥
⎣ 𝑖 − X𝑖 ⎦

6
A. Nawaz et al. Applied Energy 329 (2023) 120168

Fig. 4. Forecasted RES and load demand curve of (a) 𝑀𝐺1 (b) 𝑀𝐺2 (c) 𝑀𝐺3 .

By solving the above quadratic problem, optimal control reference is The local function, 𝐽𝐢 (𝐱̂𝐢 ) is convex and decision variable set are closed
received and therefore, each microgrid adjust its resources accordingly and bounded for ∀𝑖 ∈ 𝑛. Assume the Lagrange function has saddle point
as shown in Algorithm 1. so that:

𝐿(𝐗∗ , 𝜆) ≤ 𝐿(𝐗∗ , 𝜆∗ ) ≤ 𝐿(𝐗, 𝜆∗ ), ∀𝜆 ∈ R𝑚 .

3.3. Convergence of proposed algorithm
and all sub-problems are solvable at every iteration. We define resid-
ual vector 𝐫(𝐗𝐬 ) with all constraint violations with respect to primal
In order to prove the convergence of the proposed algorithm, we
variable (𝐗𝐬 ) and given as:
re-write the Eq. (37) in Lagrange formulation 𝐿𝑖𝛾 , for MG𝑖 as:
∑
𝑛
𝛾𝑏 𝐫(𝐗𝐬 ) = 𝐀X𝑠 (39)
̂𝐢 , 𝐗, 𝝀, 𝜸 𝒃 ) = 𝐽𝐢 (𝐗
𝐿𝑖𝛾 (𝐗 ̂𝐢 ) + ⟨𝜆, 𝐀𝐗⟩ + ‖𝐀X𝑠 ‖2 , (38)
2 𝑖=1

7
A. Nawaz et al. Applied Energy 329 (2023) 120168

Fig. 5. Energy management schedule for (a) 𝑀𝐺1 (b) 𝑀𝐺2 (c) 𝑀𝐺3 .

𝑠
Then auxiliary dual variable 𝜆 is defined as: Then
𝑠 ∑ 1 𝑠+1
𝜆 = 𝜆𝑠 + 𝛾𝑏 (1 − 𝜏)𝐫(𝐗𝐬 ) (40) 𝜙𝑠+1 (𝜆) = 𝛾𝑏 ‖𝐀(𝐗𝐬 − 𝐗∗ )‖2 + ‖𝜆 − 𝜆∗ ‖2 (43)
𝑖
𝛾𝑏
1
Here 𝜏 ∈ (0, is step time and 𝑞 is the sparsity of matrix 𝑨. Based
𝑞
) 𝑠
The auxiliary dual variable, 𝜆 , strictly decreases if the step 𝜏 satisfy
on the above assumption and definition, we define Lyapunov/Merit
0 ≤ 𝜏 ≤ 1𝑞 . As Lyapunov function is dependent on auxiliary dual
function, 𝜙𝑠 (𝜆), for Eq. (38) which strictly decreases with each iteration
variable, it also decreases with each iteration, which is key to getting
as:
convergence. Considering the earlier assumptions: if
∑ 1 𝑠
𝜙𝑠 (𝜆) = 𝛾𝑏 ‖𝐀(𝐗𝐬 − 𝐗∗ )‖2 + ‖𝜆 − 𝜆∗ ‖2 (41) 1 𝑠
𝑖
𝛾𝑏 𝐿(X, 𝜆∗ ) − 𝐿(X∗ , 𝜆∗ ) ≤ (𝜙 (𝜆) − 𝜙𝑠+1 (𝜆)) (44)
2𝜏
The dual update step is equivalent to the update rule then the following holds
𝑠+1 𝑠
̂ ∗ ) + {𝜆, 𝐫(X𝑠 )} ≤
̂ 𝑠 ) − 𝐽 (𝐗 1 𝑠
𝜆 = 𝜆 + 𝜏𝛾𝑏 𝐫(X𝑠 ). (42) 𝐽 (𝐗 𝑖 𝑖 (𝜙 (𝜆) − 𝜙𝑠+1 (𝜆)) (45)
2𝜏

8
A. Nawaz et al. Applied Energy 329 (2023) 120168

Fig. 6. Energy exchange in individual MG.

Table 1 shortage and provide power to neighbors during interval 1am–4am

Simulation system parameters.
and 8pm−12 pm. Similarly, 𝑀𝐺2 coordinates with neighbors for re-
Sources MG1 MG2 MG3 ceiving power during intervals 1am–4am, 8pm and 10pm−12 pm. It
Min Max Min Max Min Max provide power to neighbors during interval 10am−5 pm. Likewise, the
Wind turbine (kW) 0 1000 0 1000 0 1000 coordination power exchange schedule for 𝑀𝐺3 can be observed in
PV panel (kW) 0 600 0 600 0 600 Fig. 5(c). The energy exchange among microgrid is illustrated in Fig. 6.
Battery (kWh) 40 300 40 300 40 300
The battery storage is utilized by each MG to partly compensate for
the difference between supply–demand shortage, the dynamics which
are shown in Fig. 7. It can be observed that 𝑀𝐺1 utilized its battery
The Eq. (44) suggests the convergence condition and shows that the storage to partly compensate for shortage power during interval 6am
sum of 𝐽 (𝐗 ̂ ∗ ) and residual 𝐫(𝐗𝑠 ) together decreases at the
̂ 𝑠 ) − 𝐽 (𝐗 to 9am and 2pm. Similarly, the SoC utilization of 𝑀𝐺2 and 𝑀𝐺3 can
𝑖 𝑖
rate of 𝑂(1∕𝑠). The strict decrease property implies the convergence of be observed in Figs. 7(b) and 7(c). It can be noticed that individual
optimal values of dual and primal variables. The detail proof of ADAL battery SoC is within constrained zone. The supply–demand balance of
convergence can be found in [36]. each MG can be derived with certain amount of load curtailment. The
detail load curtailment schedule is shown in Fig. 8, from which we can
4. Case study simulation see the load curtailment by our method is a little bit more than con-
ventional method, which is because of the zone constraints on battery
A MMG network, comprised of three MGs, is considered to illustrate SoC. Therefore, with proposed scheduling scheme, the supply–demand
the effectiveness of the proposed method. The MMGs are operating in balance of individual MG in MMG environment are is achieved and
islanded mode and individual microgrid includes wind turbine, PV, bat- battery storage degradation are reduced through constraint utilization
tery and local electrical loads. For MPC, the optimization horizon and of energy storage within optimal zone limits.
control horizon are set to be P = 24 and M = 4, respectively. The system
parameters are given in Table 1. In comparison to battery regular safe 4.2. ESS performance comparison
upper and lower limits of 20% and 90% respectively, the bound on
battery upper and lower limits are set to be 90% and 40% respectively. In order to render the effectiveness of our proposed scheme for
By doing so, the battery SoC is maintained in such a way to achieve improved cyclic life and reduce degradation of battery storage through
an improved battery life time. It is assumed that batteries are initially zone control, a comparative study is conducted against conventional
80% charged. The Islanded multi-microgrid scenario is established in DoD scheme for battery energy utilization in [39]. The effect of dif-
MATLAB 2021a environment using Intel core i7, 3.19 GHz CPU with ferent level of discharge on battery life is evaluated by Eq. (46) given
16 GB RAM memory. The Gurobi 9 solver of GAMS, integrated with in [40].
MATLAB is utilized for quadratic programming optimization.
𝐿𝑛(𝐿) = −0.795𝐿𝑛(𝐷) + 6.5425 (46)

4.1. Results and discussion where D is the DoD in percentage and L denotes the battery life
in number of cycles. The relationship between cycle life and charge
The RES power generation and consumer demand of three MGs utilization is logarithmic for the battery and is shown in Fig. 11. The
are obtained through utilization of artificial neural network (ANN) battery life is shown for different rates of DoD and it can be seen
forecasting technique [37] and depicted in Fig. 4. The data used in that with percent rate of increase in DoD, the battery life decreases
this work is obtained from [38]. It can be seen from Fig. 4(a–c) that exponentially. The battery average cycle life is about 2500 cycles when
𝑀𝐺1 has surplus power during time interval 1am to 5am and 6pm 20% DoD is selected for battery utilization which is about 20 times
to 12pm, and it suffers from power shortage for the rest of time. greater when 90% DoD is adopted. In our work, while with lower
Likewise, the 𝑀𝐺2 has surplus power during interval 9am to 6pm and it zone limits of 40% storage utilization bound, our battery cyclic life
suffers power shortage for the rest of time. The 𝑀𝐺3 has surplus power time is improved to 1500 cycles, which in comparison to ordinary DoD
during interval 9am to 6pm and power shortage during two intervals method, utilizes 80% of total battery storage, results in the reduction
(i.e., 1am–9am and 6pm–12am). of battery cyclic life to 840 cycles.
By our Algorithm 1, the energy management schedule for individual A comparative evolution of the battery degradation in terms of its
MG with power coordination and battery utilization is derived in Fig. 5. SOH due to cyclic aging is depicted in Fig. 9. The battery reduces its
It can be seen that 𝑀𝐺1 receive power from its neighbors through storage capacity with each charge and discharge operation and also
coordination during interval 9am−5 pm to compensate its load-supply depends on how deep its discharge its charge. It can be noticed that

9
A. Nawaz et al. Applied Energy 329 (2023) 120168

Fig. 7. Battery SoC relative to charging/discharging for (a) 𝑀𝐺1 (b) 𝑀𝐺2 (c) 𝑀𝐺3 .

SOH loss of battery per day with proposed algorithm is less (0.00016% Table 2
SOH loss per day) compared to conventional DOD method (0.0003% Computing time vs number of iterations with ADMM and ADAL.

SOH loss per day). Therefore, in terms of reduced battery aging and Method Computing time Iterations
increased operational battery cyclic-life, our proposed method provides per iterate (s) Average Maximum
better performance compared to conventional DoD method. ADAL 0.29 125 400
ADMM 0.28 398 800
4.3. Convergence time

We have compared the convergence of our proposed method against

the ADMM that has similar information exchange at each iterations. 4.4. Computation time
The termination condition of 𝜖 = 10−4 , and weighting scalar 𝛾 = 103 is
selected. The convergence rate in terms of iteration for both methods The per iterate computing time as well as the maximum average
are illustrated in Fig. 10, where it can be seen that our proposed method iterations for ADAL and ADMM are tabulated in Table 2. The comput-
take less iterations for convergence compared to ADMM method. ing time of both methods are almost the similar, however the ADAL

10
A. Nawaz et al. Applied Energy 329 (2023) 120168

Fig. 8. Load curtailment schedule for (a) 𝑀𝐺2 (b) 𝑀𝐺3 .

Fig. 9. Comparative analysis of Battery state of health.

requires 398 iterations which is compared to ADMM less than half the reduced battery degradation and increased cyclic life-span. A coordina-
number of iteration required for convergence. The computing time of tion strategy based on accelerated distributed augmented Lagrangian
ADAL for one control cycle is 115 secs where the ADMM takes 225 is adopted for power exchange among MGs in MMG environment
secs. Consequently, ADMM take longer time and information exchange to compensate power deficit and achieve individual supply–demand
compared to ADAL method. Therefore, the algorithm with fewer com- balance. With propose DMPC zone control method, the cyclic life of
munication exchange and lesser convergence time is preferred due to
battery storage is prolonged and reduced degradation loss is achieved
associated delays in physical communication networks.
as well as local and system-wide supply demand is ensured. More-
5. Conclusion over, with utilization of accelerated distributed Lagrange-based mixed-
integer quadratic programming for MPC problem solving guarantees
This paper proposes a distributed zone MPC-based energy schedul- the optimal and fast convergence of the proposed DMPC method. A case
ing for multi-MGs operating in isolated mode. A zone MPC scheme is study based on three interconnected MMGs verifies the effectiveness
formulated to restrain battery state of charge within optimal range for of proposed scheme. Finally, comparative study of battery aging with

11
A. Nawaz et al. Applied Energy 329 (2023) 120168

Fig. 10. Convergence comparison of proposed and ADMM algorithm.

[4] Bukar AL, Tan CW, Yiew LK, Ayop R, Tan W-S. A rule-based energy management
scheme for long-term optimal capacity planning of grid-independent micro-
grid optimized by multi-objective grasshopper optimization algorithm. Energy
Convers Manage 2020;221:113161.
[5] Wu P, Huang W, Tai N, Liang S. A novel design of architecture and con-
trol for multiple microgrids with hybrid AC/DC connection. Appl Energy
2018;210:1002–16.
[6] Zou H, Mao S, Wang Y, Zhang F, Chen X, Cheng L. A survey of energy
management in interconnected multi-microgrids. IEEE Access 2019;7:72158–69.
[7] Evangelopoulos VA, Georgilakis PS, Hatziargyriou ND. Optimal operation of
smart distribution networks: A review of models, methods and future research.
Electr Power Syst Res 2016;140:95–106.
Fig. 11. Effect of depth of discharge on battery cyclic life. [8] Du Y, Wu J, Li S, Long C, Onori S. Hierarchical coordination of two-
time scale microgrids with supply-demand imbalance. IEEE Trans Smart Grid
2020;11(5):3726–36.
[9] Karavas C-S, Arvanitis K, Papadakis G. A game theory approach to multi-agent
proposed control against depth of discharge scheme is also performed decentralized energy management of autonomous polygeneration microgrids.
to illustrate the improved battery life performance. Energies 2017;10(11):1756.
[10] Rahman M, Oo A. Distributed multi-agent based coordinated power management
and control strategy for microgrids with distributed energy resources. Energy
CRediT authorship contribution statement
Convers Manage 2017;139:20–32.
[11] Rajasekharan J, Koivunen V. Optimal energy consumption model for smart grid
Arshad Nawaz: Conceptualization, Methodology, Writing – origi- households with energy storage. IEEE J Sel Top Sign Proces 2014;8(6):1154–66.
nal draft. Jing Wu: Validation, Writing – review & editing, Supervi- [12] Kim AR, Seo HR, et al. Operating characteristic analysis of HTS SMES for
sion, Funding support. Jun Ye: Simulation. Yidi Dong: Simulation. frequency stabilization of dispersed power generation system. IEEE Trans Appl
Supercond 2010;20(3):1334–8.
Chengnian Long: Writing – review & editing, Supervision, Funding
[13] Koohi-Kamali S, Rahim NA. Coordinated control of smart microgrid during and
support. after islanding operation to prevent under frequency load shedding using energy
storage system. Energy Convers Manage 2016;127:623–46.
Declaration of competing interest [14] Hou J, Song Z, Hofmann H, Sun J. Adaptive model predictive control for
hybrid energy storage energy management in all-electric ship microgrids. Energy
Convers Manage 2019;198:111929.
The authors declare that they have no known competing finan- [15] Agarwal V, Uthaichana K, DeCarlo RA, Tsoukalas LH. Development and valida-
cial interests or personal relationships that could have appeared to tion of a battery model useful for discharging and charging power control and
influence the work reported in this paper. lifetime estimation. IEEE Trans Energy Convers 2010;25(3):821–35.
[16] Min Chen, Rincon-Mora GA. Accurate electrical battery model capable
of predicting runtime and I-V performance. IEEE Trans Energy Convers
Data availability
2006;21(2):504–11.
[17] Parvini Y, Vahidi A. Maximizing charging efficiency of lithium-ion and lead-acid
No data was used for the research described in the article. batteries using optimal control theory. In: 2015 American control conference.
2015, p. 317–22.
[18] Huang Y, Wang H, Khajepour A, He H, Ji J. Model predictive control power
Acknowledgment
management strategies for HEVs: A review. J Power Sources 2017;341:91–106.
[19] Morstyn T, Hredzak B, Agelidis VG. Control strategies for microgrids with
This work is supported by National Natural Science Foundation distributed energy storage systems: An overview. IEEE Trans Smart Grid
(NSFC) of China under Grants 62136006, 62273236, 62073215 and 2018;9(4):3652–66.
61873166. [20] Ouammi A, Dagdougui H, Dessaint L, Sacile R. Coordinated model predictive-
based power flows control in a cooperative network of smart microgrids. IEEE
Trans Smart Grid 2015;6(5):2233–44.
References [21] Christofides PD, Scattolini R, Muñoz de la Peña D, Liu J. Distributed model
predictive control: A tutorial review and future research directions. Comput
[1] Kiptoo MK, Lotfy ME, Adewuyi OB, Conteh A, Howlader AM, Senjyu T. Integrated Chem Eng 2013;51:21–41.
approach for optimal techno-economic planning for high renewable energy-based [22] Alvarado I, Limon D, Muñoz de la Peña D, Maestre J, Ridao M, Scheu H, et al.
isolated microgrid considering cost of energy storage and demand response A comparative analysis of distributed MPC techniques applied to the HD-MPC
strategies. Energy Convers Manage 2020;215:112917. four-tank benchmark. J Process Control 2011;21(5):800–15.
[2] Lidula N, Rajapakse A. Microgrids research: A review of experimental microgrids [23] Ouammi A, Dagdougui H, Dessaint L, Sacile R. Coordinated model predictive-
and test systems. Renew Sustain Energy Rev 2011;15(1):186–202. based power flows control in a cooperative network of smart microgrids. IEEE
[3] Karavas C-S, Arvanitis KG, Kyriakarakos G, Piromalis DD, Papadakis G. A novel Trans Smart Grid 2015;6(5):2233–44.
autonomous PV powered desalination system based on a DC microgrid concept [24] Olivares DE, Cañizares CA, Kazerani M. A centralized energy management system
incorporating short-term energy storage. Sol Energy 2018;159:947–61. for isolated microgrids. IEEE Trans Smart Grid 2014;5(4):1864–75.

12
A. Nawaz et al. Applied Energy 329 (2023) 120168

[25] Wang Y, Mao S, Nelms RM. On hierarchical power scheduling for the macrogrid [33] Bemporad A, Morari M. Control of systems integrating logic, dynamics, and
and cooperative microgrids. IEEE Trans Ind Inf 2015;11(6):1574–84. constraints. Automatica 1999;35(3):407–27.
[26] Bazmohammadi N, Tahsiri A, Anvari-Moghaddam A, Guerrero JM. A hierar- [34] Mendil M, Domenico D, Heiries AV, et al. Battery aging-aware energy manage-
chical energy management strategy for interconnected microgrids considering ment of green small cells powered by the smart grid. J Wirel Commun Netw
uncertainty. Int J Electr Power Energy Syst 2019;109:597–608. 2017;10(1):1–13.
[27] Marín LG, Sumner M, Muñoz-Carpintero D, Köbrich D, Pholboon S, Sáez D, et [35] Negenborn R, Schutter BD, Hellendoorn J. Multi-agent model predictive control
al. Hierarchical energy management system for microgrid operation based on for transportation networks: Serial versus parallel schemes. Eng Appl Artif Intell
robust model predictive control. Energies 2019;12(23):4453. 2008;21(3):353–66.
[28] Parisio A, Wiezorek C, et al. Cooperative MPC-based energy management for [36] Lee S, Chatzipanagiotis N, Zavlanos MM. Complexity certification of a distributed
networked microgrids. IEEE Trans Smart Grid 2017;8(6):3066–74. augmented Lagrangian method. IEEE Trans Automat Control 2018;63(3):827–34.
[29] Garcia-Torres F, Bordons C, Ridao MA. Optimal economic schedule for a [37] Kong W, Dong ZY, Jia Y, Hill DJ, Xu Y, Zhang Y. Short-term residential load
network of microgrids with hybrid energy storage system using distributed model forecasting based on LSTM recurrent neural network. IEEE Trans Smart Grid
predictive control. IEEE Trans Ind Electron 2019;66(3):1919–29. 2019;10(1):841–51.
[30] Zheng Y, Li S, Tan R. Distributed model predictive control for on- [38] Open, Source. Wind and solar data set. 2022, https://energydata.info/dataset?
connected microgrid power management. IEEE Trans Control Syst Technol vocab_topics=Renewableenergya.
2018;26(3):1028–39. [39] Du Y, Wu J, Li S, Long C, Onori S. Coordinated energy dispatch of au-
[31] Sandgani MR, Sirouspour S. Coordinated optimal dispatch of energy stor- tonomous microgrids with distributed MPC optimization. IEEE Trans Ind Inf
age in a network of grid-connected microgrids. IEEE Trans Sustain Energy 2019;15(9):5289–98.
2017;8(3):1166–76. [40] Zhou C, Qian K, Allan M, Zhou W. Modeling of the cost of EV battery
[32] Wang Z, Chen B, Wang J, Begovic MM, Chen C. Coordinated energy manage- wear due to V2G application in power systems. IEEE Trans Energy Convers
ment of networked microgrids in distribution systems. IEEE Trans Smart Grid 2011;26(4):1041–50.
2015;6(1):45–53.

13