A Dual Distributed Optimal Energy Management Method for

Distribution Grids With Electric Vehicles

Hoang, P. H., Ozkan, G., Badr, P. R., Papari, B., Edrington, C. S., Zehir, M. A., Hayes, B., Mehigan, L., Kez, D.
A., & Foley, A. M. (2021). A Dual Distributed Optimal Energy Management Method for Distribution Grids With
Electric Vehicles. IEEE Transactions on Intelligent Transportation Systems. Advance online publication.
https://doi.org/10.1109/TITS.2021.3126543

Published in:
IEEE Transactions on Intelligent Transportation Systems

Document Version:
Peer reviewed version

Queen's University Belfast - Research Portal:

Link to publication record in Queen's University Belfast Research Portal

Publisher rights
Copyright 2021, IEEE.
This work is made available online in accordance with the publisher’s policies. Please refer to any applicable terms of use of the publisher.

General rights
Copyright for the publications made accessible via the Queen's University Belfast Research Portal is retained by the author(s) and / or other
copyright owners and it is a condition of accessing these publications that users recognise and abide by the legal requirements associated
with these rights.

Take down policy

The Research Portal is Queen's institutional repository that provides access to Queen's research output. Every effort has been made to
ensure that content in the Research Portal does not infringe any person's rights, or applicable UK laws. If you discover content in the
Research Portal that you believe breaches copyright or violates any law, please contact openaccess@qub.ac.uk.

Open Access
This research has been made openly available by Queen's academics and its Open Research team. We would love to hear how access to
this research benefits you. – Share your feedback with us: http://go.qub.ac.uk/oa-feedback

Download date:25. Mar. 2024

 IEEE Intelligent Transportation Systems Transactions

A Dual Distributed Optimal Energy Management Method for

Distribution Grids with Electric Vehicles

Journal: Transactions on Intelligent Transportation Systems

Manuscript ID T-ITS-20-09-1875.R1

Manuscript Type: Regular Papers

Date Submitted by the

26-Apr-2021
Author:

Complete List of Authors: Hoang, Phuong; Clemson University, Electrical and Computer
Engineering
Ozkan, Gokhan; Clemson University, Electrical and Computer
Engineering
Ramezani Badr, Payam; Clemson University, Electrical and Computer
Engineering
Papari, Behnaz ; University of North Carolina at Charlotte, Search
Results Web Result with Site Links Electrical and Computer Engineering
Edrington, Christopher; Clemson University, Electrical and Computer
Engineering
Zehir, Aparslan; University College Cork, Energy Policy and Modelling
Group, MaREI Centre, Environmental Research Institute
Mehigan, Laura; University College Cork, Energy Policy and Modelling
Group, MaREI Centre, Environmental Research Institute
Hayes, Barry; University College Cork, School of Engineering
Kez, Dlzar; Queen's University Belfast, School of Electronics, Electrical
Engineering and Computer Science
Foley, Aoife; Queen's University Belfast, School of Mechanical and
Aerospace Engineering and School of Electronics, Electrical Engineering
and Computer Science

distributed energy management, electric vehicle, controller-hardware-in-

Keywords:
the-loop simulation, real-time simulation, distribution grid

Future distribution grids are expected to face an increasing penetration

of electric vehicles (EVs) and heterogeneous distributed energy
resources (DERs). This demands a distributed energy management (EM)
to manage power generation and delivery of energy sources to maintain
power quality under the impact of EV charging, to save operating costs,
and to enhance resiliency. However, the global optimality of the
distributed EM's optimization problem is still an issue in existing work
because of the non-convex nature of the optimization problem. In this
Abstract:
paper, a distributed EM strategy for grid-connected distribution networks
is proposed. In particular, the EM strategy is composed of two steps. In
the first step, some conditions of the EM optimization task are relaxed to
apply an algorithm converging to the global optimality. The results of the
first step are used to reconfigure constraints of the full optimization
problem in Step 2. The proposed scheme is validated by implementing
the real-time controller-hardware-in-the-loop (CHIL) experimentation on
the IEEE 33 bus system. To study the impact of EV charging, EV data is

PLEASE KEEP CONFIDENTIAL

Page 1 of 12 IEEE Intelligent Transportation Systems Transactions

1
2
3
collected from the market and the literature to generate realistic EV load
4 profiles to demonstrate the effectiveness of the proposed strategy on
5 saving operating costs and maintaining power quality.
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 PLEASE KEEP CONFIDENTIAL
 IEEE Intelligent Transportation Systems Transactions Page 2 of 12

1
2
3 A Dual Distributed Optimal Energy Management
4
5
6
Method for Distribution Grids with Electric Vehicles
7 Phuong H. Hoang∗ , Student Member, IEEE, Gokhan Ozkan∗ , Member, IEEE,
8 Payam Ramezani Badr∗ , Student Member, IEEE, Behnaz Papari† , Member, IEEE,
9 Christopher S. Edrington∗ , Senior Member, IEEE, Mustafa Alparslan Zehir‡ , Member, IEEE,
10
Barry Hayes⊥ , Senior Member, IEEE, Laura Mehigan‡ , Student Member, IEEE,
11
12
Dlzar Al Kez§ , Student Member, IEEE, Aoife M. Foley¶ , Member, IEEE
13
14
15
16 Abstract—Future distribution grids are expected to face an in- N,R Set of all buses and set of all regions in the
creasing penetration of electric vehicles (EVs) and heterogeneous network, respectively.
17 distributed energy resources (DERs). This demands a distributed
18 Rr Set of buses in region r.
energy management (EM) to manage power generation and
19 delivery of energy sources to maintain power quality under the Nr Set of neighboring regions of region r.
20 impact of EV charging, to save operating costs, and to enhance M G, M T, P V Main grid, microturbine, and photovoltaic
21 resiliency. However, the global optimality of the distributed EM’s system, respectively.
optimization problem is still an issue in existing work because EV, L Electric vehicle and non-EV load, respec-
22 of the non-convex nature of the optimization problem. In this
23 tively.
paper, a distributed EM strategy for grid-connected distribution
24 networks is proposed. In particular, the EM strategy is composed MT Set of buses which contain a microturbine.
25 of two steps. In the first step, some conditions of the EM a, a Upper and lower limits of a, respectively.
26 optimization task are relaxed to apply an algorithm converging Vn , Vnre , Vnim Complex bus voltage at bus n, its real part,
to the global optimality. The results of the first step are used to and its imaginary part, respectively.
27 reconfigure constraints of the full optimization problem in Step 2. re im
28 Vr,n , Vr,n Real part and imaginary part of bus voltage
The proposed scheme is validated by implementing the real-time
29 controller-hardware-in-the-loop (CHIL) experimentation on the at bus n optimized at region r, respectively.
30 IEEE 33 bus system. To study the impact of EV charging, EV PnA , QAn Active and reactive power generated, sup-
31 data is collected from the market and the literature to generate plied, or consumed by A at bus n, respec-
realistic EV load profiles to demonstrate the effectiveness of the tively.
32 proposed strategy on saving operating costs and maintaining
33 PrD Dispatchable active power of region r.
power quality.
34 PnS Sum of non-dispatchable active power and
Index Terms—Electric vehicles, distributed energy manage- active load at bus n.
35 ment, distribution grid, controller-hardware-in-the-loop, real-
36 time simulation.
37 I. I NTRODUCTION
38
N OMENCLATURE
39
40 n, m
r, s
Bus index
Region index
E Vs are widely considered as a replacement for internal
combustion engine vehicles because they are environ-
mentally friendly and beneficial to sustainable energy goals.
41
42 N, R Number of buses and regions, respectively. However, the increasing of EV penetration and EV charging
43 capacities negatively impact distribution grids. For example,
∗ P. H. Hoang, G. Ozkan, P. R. Badr, and C. S. Edrington
44 it is shown in [1] that a distribution grid fails to maintain
are with the Real-Time COntrol and Optimization Laboratory
45 (RT-COOL)-Holcombe Department of Electrical and Computer required voltages when the EV penetration level reaches 30%.
46 Engineering-Clemson University, Clemson, SC 29634, USA. E-mails: Besides, it is expected that distribution grids will be popu-
{huuphuh, gokhano, prameza, cedring}@clemson.edu. lated with heterogeneous DERs [2]. Therefore, a distributed
47 † B. Papari is with Energy Production & Infrastructure Center (EPIC),

48 University of North Carolina, 8700 Phillips Rd, Charlotte, NC 28223, USA. EM coordinating dispersed energy sources to gain economic
49 E-mail: bpapari@uncc.edu. objectives while maintaining power quality under the impact
‡ M. A. Zehir and L. Mehigan are with Energy Policy and
of EV penetration is valuable.
50 Modelling Group, MaREI Centre, Environmental Research Institute,
51 University College Cork, Old Lee Road, Cork, Ireland. E-mails:
52 {alparslan.zehir, laura.mehigan}@ucc.ie.
⊥ B. Hayes is with School of Engineering and Architecture, University A. Literature Review
53 College Cork, College Road, Cork, Ireland. E-mail: {barry.hayes}@ucc.ie.
54 § D. A. Kez is with School of Electronics, Electrical Engineering and EM is one of the constituents of future distribution grids’
55 Computer Science, Queen’s University Belfast, Belfast, BT9 5AH, United control and management system [3]. The primary function of
56
Kingdom. E-mail: dalkez01@qub.ac.uk. EM is to optimally allocate power generation and delivery
¶ A. M. Foley is with School of Mechanical and Aerospace Engineering
57 to sources of energy in a well-defined sense while adhering
and School of Electronics, Electrical Engineering and Computer Science,
58 Queen’s University Belfast, Belfast, BT9 5AH, United Kingdom. E-mail: to system constraints [4–10]. Distributed EM is considered
59 a.f oley@qub.ac.uk. superior to the centralized counterpart in terms of resiliency
60 PLEASE KEEP CONFIDENTIAL
1
Page 3 of 12 IEEE Intelligent Transportation Systems Transactions

1
2 and scalability. Yet, the global optimality of the distributed an ADMM-based algorithm in the second step seeks solutions
3 EM’s optimization problem is still an issue. in a narrowed region around the global optimality. Besides, the
4 The alternating direction method of multipliers (ADMM) reconfiguration of the reactive constraints ensures power factor
5 [11] is a popular method to solve the EM’s optimization (PF) requirements are satisfied. EV charging load profiles, gen-
6 problem [10, 12–14]. The popularity of the ADMM is at- erated through a novel approach for a realistic representation
7 tributed to its efficiency and straightforwardness as it is applied of daily cases in EV integrated distribution grids, are used
8 to various applications with three clear updating steps. It is to test the proposed strategy. Slow and fast charging sessions
9 worth noting that the theoretical guarantee of convergence of a large number and a wide range of cars available in the
10 of the ADMM is confined to convex optimization problems. US market are considered together in the explored scenarios.
11 However, the EM’s optimization problem for AC networks Based on a thorough literature review, there is no existing
12 is known to be non-convex that the ADMM fails to deliver work implementing distributed EM via a CHIL setup. This
13 the convergence guarantee [10]. Recently, the authors in work is one step further as a CHIL experiment is constructed
14 [15] apply the augmented Lagrangian alternating direction and real-time experimental evidence is shown.
15 inexact Newton method to the EM’s optimization problem.
16 The method overcomes the convergence issue for non-convex C. Paper Organization
17 problems, yet global optimality is still not ensured. In addition,
The remainder of this paper is organized as follows. In Sec-
18 each iteration of the method generally has higher complexity
tion II, the EM problem is derived and formulated. For ease of
19 and computation than the ADMM. There are papers that do
understanding, common notations are presented in the section.
20 not consider the bus voltage constraints and power limits on
transmission lines in the EM’s optimization problem [16]. Section III is dedicated to present the two-step distributed EM
21
However, it is likely that these constraints are violated with strategy. The CHIL demonstration of the proposed scheme is
22
the adoption of DERs and EVs. discussed in Section IV. Section V concludes the paper and
23
discusses future directions.
24 Realistic representation of EV charging demand is needed
25 to accurately examine the effectiveness of innovative EM
26 approaches. The studies in the literature are mainly based II. P ROBLEM F ORMULATION
27 on either direct use of charging profile data collected from Consider a grid-connected radial distribution network con-
28 field demonstrations [17] or deterministic charging scenarios sisting of heterogeneous DERs dispersed over the network:
29 built on general assumptions [18]. However, these conven- microturbine(s) (MT), wind turbine(s) (WT), and photovoltaic
30 tional methodologies have some drawbacks. Charging data (PV) system(s). It is desired that the grid 1) operates at
31 from a field pilot usually has limited representation (due economically optimal points, 2) maintains voltage satisfaction
32 to a low number of participants, short testing period, and for all the buses, and 3) adheres to power generation and
33 cars from a few brands and constrained availability to other delivery capacity limits while meeting the power balance
34 researchers. Furthermore, assumptions about using the same constraint.
35 charging demand and energy storage capabilities, considering
36 starting and ending times from a limited time period of the
day, straightforward determination of the number of charging A. System Modeling and Notations
37
38 sessions, initial and final State of Charge (SoC) values, dis- Let an undirected graph (N , E) represent the grid, where
39 tinguished consideration of slow and fast charging profiles are N = {1, ..., N } is the set of nodes and E = {(n, m)} ⊆
40 considerably far from observations in the field [19]. N × N is the set of lines. The terms bus and node are used
41 For a better representation of the EV charging demand of interchangeably in this paper. The bus connecting directly with
42 a large group of customers, a novel and easily replicable the substation is indexed 1. Let MT ⊂ N be the set of nodes
43 methodology is developed and used in this paper. It is based that have MTs installed. The system is divided into R regions;
44 on the combined use of characteristics of a wide range of cars each has a dispatchable source of energy, either a substation
45 available on a country level, charging starting and ending time or an MT. Assume that each bus in a region does not have
46 probabilities, SoC value probabilities, and driving statistics. more than one connection with buses outside the region. Let
47 A unique feature of the methodology is consecutive relations R = {1, ..., R} and Rr be the set of nodes in region r. Denote
48 built between slow and fast charging sessions. Nr as the set of neighboring regions of region r.
49 Let T r(.), (.)∗ , (.)T , and (.)H be the trace operator, complex
50 conjugate, transposition, and complex conjugate transposition
B. Statement of Contributions operators, respectively. Let [aij ]l×k be a l×k matrix. Through-
51
52 In this paper, a two-step distributed EM scheme is proposed. out this paper, there are some common notations. P and Q
53 The first step is to search for the global optimum by relaxing accordingly indicate active and reactive power. Subscripts m
54 some constraints in the EM’s optimization problem to reduce and n are the bus index and subscripts r and s are the region
55 it to a convex one. In doing so, an algorithm that secures index. Moreover, superscripts re, im, M G, M T , EV , P V ,
56 globally optimal solutions can be applied. Its optimized results and L accordingly indicate the real part, the imaginary part,
57 are used to reconfigure dispatchable active and reactive power the main grid, MT, EV, PV, and non-EV load. For example,
58 constraints in the second step in which the full problem is with these notations, PnM T and QM n
T
are active and reactive
59 considered. Reconfiguring the active power constraints makes power generated from an MT at bus n, respectively. Overline
60 PLEASE KEEP CONFIDENTIAL
2
 IEEE Intelligent Transportation Systems Transactions Page 4 of 12

1
2 and underline indicate lower and upper limits of a quantity, final SoC values and charging times from car characteristics,
3 respectively. Let Vn = Vnre + jVnim be the bus voltage at bus the charging duration for each car is calculated. Taking into
4 n. The sign of active power Pn (reactive power Qn ) at node n account the assigned charge starting times and the calculated
5 is defined as follows: if Pn > 0 (Qn > 0), then node n injects charging durations, charge ending times are derived. Addi-
6 active (reactive) power to the network, otherwise it absorbs tional idle waiting times/parking times are assigned based on
7 active (reactive) power. the statistics provided in [22]. A charging model that is based
8 on the constant current charging up to around 90% SoC, and
9 B. Cost Functions then the linear limit of power for most of the vehicles provided
10 in [23] is used as part of the profile generation tool. The details
The active power of the dispatchable sources can be adjusted
11 of this model can be found in the reference. A transitional
to gain economic objectives formed in the following function:
12 period follows the slow charging session, considering only the
13 cars that are charged more than once a day. During this period,
X
f1 (P1M G ) + fn (PnM T ) (1)
14 n∈MT
cars are driven, spend the energy stored in their batteries, and
15 park based on the comprehensive statistics provided in [24] on
The first term in (1) is the cost of buying power from the distance traveled and time spent between charging sessions.
16
main grid which is a multiplication of the market price cM G
17 The fourth stage is fast charging in public charging stations.
($/kW h) and the amount of active power supplied from the
18 Considering the departure times from the previous charging
main grid. The second term is the cost of power generated by
19 the MTs which has the following form
session and adding on the time spent during the transitional
20  
period, fast charge starting times are calculated. In a similar
21 fn (PnM T ) =cF αn (PnM T )2 + βn PnM T + γn , αn > 0 (2) manner, considering the final SoC values from the previous
22 charging sessions and the energy spent/SoC reduced (calcu-
23 where cF ($/gallon) is the cost of fuel. lated according to the distance traveled and energy consumed
24 per kilometer) between two charging sessions, the initial SoC
25 values for the fast charging session are derived. Since full fast
C. EV Charging charging times (ranging from 20 to 80 minutes, with around
26
27 A probabilistic charging profile generation methodology that 40 minutes on average) are below typical charging durations,
28 combines characteristics and statistics from a range of data all the cars are assumed to be fully charged up to 90% SoC
29 resources is employed. The profile generator builds relations in fast charging sessions. Considering the derived fast charge
30 between slow and fast charging sessions to represent the EV starting times, initial SoC values, and minimum charging up to
31 charging profile of a large number of cars, including residential 90% SoC, fast charge ending times are calculated. As in stage
32 and public charging cases. 2, an additional time for parking or idle waiting is assigned,
33 In the first stage, characteristics of the 23 cars that are and departure times are calculated.
34 available in the US market over the last decade are considered
35 based on the information available in [20] and [21]. Slow
36 and fast charging demand and time, energy storage capability
37 and energy consumption per kilometer are imported into the
38 charging profile generation methodology as the main car
39 characteristics. The slow charging demand of the considered
40 cars ranges between 3.6 to 16.5 kW with 3 to 14 hours full
41 charging time. The fast charging demand of the considered
42 EVs ranges between 22 to 110 kW with 20 to 80 minutes
43 charging time up to around 90% SoC level. Depending on the
44 number of customers defined in a scenario, a random EV from
45 the determined pool of cars is assigned to each customer.
46 In the next stage, probabilities for slow charging sessions are
47 defined based on [19]. The detailed probabilities are available
48 in the cited publication. These consist of the overall share of
49 customers who charge their car at least once a day, proba-
50 bilities of charge starting time (in 15-minute ranges), initial
51 and final SoC values (in 8.33% ranges -due to SoC recording
52 resolution used in the considered field pilot-, depending on
53 the period of starting time, called as morning peak between Fig. 1. EV load profile generation methodology.
54 06:00 AM and 09:00 AM, evening peak from 03:00 PM to
55 09:00 PM and the rest of the day). After a range is selected If a charging session ends on the next day, all the hours
56 for a customer’s charging session, a random number inside with charging demand in the next day are also considered.
57 that range is chosen as the exact starting time. A similar This is required to make clear comparisons between the cases
58 approach is followed to assign initial and final SoC values without and with distributed EM. Random assignment of cars
59 for the considered sessions. Using the assigned initial and to each customer and random selection of the exact charge
60 PLEASE KEEP CONFIDENTIAL
3
Page 5 of 12 IEEE Intelligent Transportation Systems Transactions

1
2 starting times and exact initial SoC values for the first slow is defined as the n−element of en is 1 and the others are 0.
3 charging session allow generation of different individual and The following transformation can be obtained [25, 26]:
4 aggregated daily charging profiles at each run.
Yn := en eTn Y (11a)
5 2 H
6 |Vn | = Tr(VV ) (11b)
D. Constraints 1
7

Pn := Yn + (Yn )H (11c)
8 An MT unit has lower and upper generation capacity limits 2
9 on both active and reactive power as Pn = Tr(Pn VVH ) (11d)
10 j 
PnM T ≤ PnM T = (Pn − PnL − PnEV ) ≤ PnM T (3a) Qn := Yn − (Yn )H (11e)
11 2
12 QM
n
T
≤ QM
n
T
= (Qn − QL
n − QEV
n )≤ QM
n
T (3b) Qn = Tr(Qn VVH ) (11f)
13
14 Similarly, the main grid is imposed these constraints as Pn and Qn can be expressed as functions of V, so do the
15 cost function and the constraints in (10). It is well known that
P1M G ≤ P1M G = (P1 − P1L − P1EV ) ≤ P1M G (4a) the optimization problem in (10) is non-convex [10]. In the
16
17 QM G
≤ QM G
= (Q1 − QL − QEV )≤ QM G (4b) following, a two-step distributed strategy is proposed to solve
1 1 1 1 1
18 the non-convex optimization problem.
19 For a node that has a PV installed, the sum of all injections
20 and withdrawals at the bus equal 0; that is, III. D ISTRIBUTED EM S TRATEGY
21 Pn − (PnP V + PnL + PnEV ) = 0 (5) In this section, a distributed strategy, which consists of
22 two steps, is introduced and discussed. In the first step, the
23 It is assumed that PV systems actively participate in regulating optimization task is relaxed by neglecting the voltage and
24 bus voltages by adjusting reactive power, which is confined in reactive power constraints; i.e., (10) is relaxed to
25 following range:  X 
26 min f (PD ) = f1 (P1M G ) + fn (PnM T ) (12a)
27 QP
n
V
≤ QP
n
V
= (Qn − QL EV PV
n − Qn ) ≤ Qn (6) n∈MT

28 s.t. (3a), (4a), (5), (7a). (12b)

WTs are similar to PVs as
29 where P =D
[..., PrD , ...]T ,
r ∈ R, is a vector consisting of
30 Pn − (PnW T + PnL + PnEV ) = 0 (7a) the dispatchable active power of all the regions, and PrD is
31 P1M G if region r has the substation and PnM T if region r has
32 QW
n
T
≤ QW
n
T
= (Qn − QL EV WT
n − Qn ) ≤ Qn . (7b)
an MT at bus n.
33 Let bus 1 be the slack bus. The voltage magnitude at bus The goal of the first step is to find the optimal allocation
34 n ∈ N \ {1} are desired to be confined in a bounded region of active power for the dispatchable sources. Its optimized
35 defined as outcomes are then fed into the second step in which constraints
36 (3) and (4) are reconfigured based on the inputs. The active
|Vn | ≤ |Vn | ≤ |Vn | (8)
37 power inequality constraints are reconfigured such that the box
38 constraints are narrowed to the region around the solutions of
Typically, |Vn | = 0.95 per units (pu) and |Vn | = 1.05 pu
39 (12), which are denoted as P M G,∗ and PnM T,∗ , as
are used. If phase angles of voltages at the buses are confined,
40 power transmitted on lines are bounded. Therefore, a constraint
41 max(PnM T,∗ − ∆PnM T , PnM T ) ≤ PnM T ≤
is set on the voltage phase angle at bus n ∈ N \ {1} as
42 min(PnM T,∗ + ∆P1M T n , PnM T ) (13a)
43 V im
44 −θn ≤ θn = arctan( nre ) ≤ θn (9) max(P M G,∗
− ∆P1M G , P M G ) ≤ P1M G ≤
Vn
45 min(P1M G,∗ + ∆P1M G , P1M G ) (13b)
46
47
E. EM’s Optimization Task where ∆PnM T = σ|PnM T,∗ | and ∆P1M G = σ|P1M G,∗ |, where
48 Given the costs and constraints presented in the previous 0 < σ < 1, and σ = 0.1 is selected in this paper.
49 subsections, the optimization task is stated as For the reactive power inequality constraints at buses where
50  X  MTs are installed, a minimum PF of 0.95 with active power
51
MG
min f1 (P1 ) + MT
fn (Pn ) (10a) referred to solutions of (12) are allowed. In the second step,
52 n∈MT the optimization task (10) is considered. In [16], only Step
53 s.t. (3), (4), (5), (6), (7), (8), (9). (10b) 1 is considered. However, with the adoptions of EVs and
54 DERs, the voltage constraints and power transmission limits
55 The problem (10) can be transformed into an equivalent one on transmission lines are likely violated. There are works
56 which has voltages as the only variables [25, 26]. Denote that consider only Step 2 and utilize the ADMM to solve
57 the bus voltage vector by V = [V1 , ..., VN ]T and the bus the EM’s optimization problem [12–14]. But (10) is a non-
58 admittance matrix by Y = [Ylk ]N ×N . Let {en }n∈N form convex optimization problem, and the convergence and global
59 standard basis vectors in Rn , in which en := [0, ..., 1, ..., 0]T optimality are not ensured.
60 PLEASE KEEP CONFIDENTIAL
4
 IEEE Intelligent Transportation Systems Transactions Page 6 of 12

1
2 A. Step 1 where x and z are vectors of variables, c is a constant vector,
3 Denote the sum of non-dispatchable active power and loads Cx and Cz are their constraint sets, and A and C are matrices.
4 at bus n by PnS = PnP V + PnW T + PnL + PnEV , ∀n ∈ N . The Suppose the vectors and matrices have appropriate dimensions.
5 augmented Lagrangian of (12) is The associated augmented Lagrangian is
6 X X  L2 (x, y, z) =f (x) + g(z) + yT (Ax + Bz − c)
7 L1 = f (PD ) + µ PrD + PnS ρ2 (18)
8 r∈R n∈N + k Ax + Bz − c k22
2
9
X X
+ λmax
r (PrD − PrD ) + λmin
r (−PrD + PrD ) (14) where ρ2 > 0. The ADMM has three repetitively updating
10 r∈R r∈R steps as
11 +
ρ1
k
X
PrD +
X
PnS k2
12 2 xk+1 = argmin L2 (x, zk , yk ) (19a)
r∈R n∈N
13 x∈Cx

14 where µ, λmax r , and λminr are Lagrange multipliers, and zk+1 = argmin L2 (xk+1 , z, yk ) (19b)
z∈Cz
15 ρ1 > 0. In [27] and [28], a distributed algorithm based on
16 the singular perturbation method is proposed aiming to solve a yk+1 = y + ρ2 (Axk+1 + Bzk+1 − c)
k
(19c)
17 general optimization problem. The underlying idea of the algo- The ADMM’s configuration is now specified to fit the
18 rithm is constructing a two-time-scale dynamical system which optimization problem. Applying the ADMM for consensus
19 has an equilibrium point satisfying the Karush–Kuhn–Tucker optimization problem [11], let g(z) = 0 and c = 0. Define a
20 conditions: the first set of equations is called the fast dynamic re
voltage vector Vr = [..., Vr,n re
, ..., Vr,m im
, ..., Vr,n im
, ..., Vr,m , ...]T ,
21 layer and the second set is called the slow dynamic layer. The where bus n is in region r and bus m is not region r but it has
22 fast dynamic layer can be derived as a neighbor in region r. Let x = [V1T , ..., VR T T
] and decompose
23 X X
f (x) as
ξ˙rh = − ξrh − (ξrh − ξsh ) − (ζrh − ζsh )
24
X
s∈Nr s∈Nr f (x) ≡ fn (Vr ) (20)
25  X  n∈MT ∪{1},n∈Rr
26 + PrD + PnS (15a)
27 n∈Rr Define the matrix A = [aij ]4l×k such that it is full row rank
28
X and l is the number of edges connecting two nodes of two
ζ̇rh = (ξrh − ξsh ) (15b)
29 different regions and k is the dimension of x. Additionally,
s∈Nr re im
30 X X aij = |κ| if j−th element of x, which is either Vr,n or Vr,n
ξ˙rµ = − ξrµ − (ξrµ − ξsµ ) − (ζrµ − ζsµ ) + µr , (15c) with bus n has a neighbor not in region r. Otherwise, aij = 0.
31 s∈Nr s∈Nr
32 X The matrix B is defined as B = −I4l×4l , where I4l×4l is the
33 ζ̇rµ = (ξrµ − ξsµ ), (15d) 4l × 4l identity matrix. In the second updating step, z k+1 is
34 s∈Nr the solution of the following equation
35 of ( r∈R PrD + ∆L2 (xk+1 , yk , z)
P
It is noted that (15a) is to estimate the averageP
36
P S h 1
P D S |z∈Cz = 0 (21)
n∈N Pn ), i.e., ξr −→ R ( r∈R Pr + n∈N Pn ). The ∆z
37 following is the slow dynamic layer which can be expressed as
38  ∂f  1 k
39 ẋr = −krx
r
(xr ) + ξrµ + λmax
r − λmin
r + ρ1 ξrh (16a) zk+1 |z∈Cz = P( y + Axk+1 ) (22)
40  ∂xr X  ρ2
41 µ̇r = kr ξrh −
µ
(µr − µs ) (16b) If y has a zero initialization and z0 = P(Ax0 ), then zk+1 =
42 s∈Nr P(Axk+1 ), where P is the orthogonal projection operator on
43 λ̇max = krλ
max
(xr − PrD ) (16c) Cz . The orthogonal projection matrix P = [pij ]4l×4l is selected
r
44 min such that pii = 12 , pij = − 12 if both the i−th and j−th
45 λ̇min = krλ (−xr + PrD ) (16d)
r elements of Axk+1 are either the real part or the imaginary
46 While   1 is a real positive number which aims to have (16) part of a bus voltage optimized in two regions, otherwise pij =
47 max
having a slower dynamics than that of (15), krx , krµ , krλ , and 0. With the specified configuration, voltages of a bus (both the
48 krλ
min
∈ R>0 are to cope with differences in the dynamics of real part and imaginary) optimized in two different controllers
49 xr , µr , λmax , and λmin . reach a consensus as (Axk − zk ) −→ 0 when k −→ ∞.
r r
50
51 IV. CHIL I MPLEMENTATION
52 B. Step 2
A. CHIL Experimental Setup
53 The ADMM algorithm [11] is briefly presented before de-
54 1) CHIL Setup: The IEEE 33 bus system in [29] with
riving its application to the considered optimization problem.
55 additions of DERs is used to validate the proposed scheme.
Consider the following optimization problem:
56   The system is geographically divided into 6 regions, with each
57 min f (x) + g(z) (17a) corresponding to an EM agent. These DERs’ locations, param-
58 eters, and block diagrams are shown in Fig. 2a). Particularly,
s.t. Ax + Bz = c and x ∈ Cx , z ∈ Cz (17b) Rowen’s model is used to simulate MTs’ engines [30], and the
59
60 PLEASE KEEP CONFIDENTIAL
5
Page 7 of 12 IEEE Intelligent Transportation Systems Transactions

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43 Fig. 2. CHIL experimentation: a) IEEE bus system simulation model b) CHIL experimental setup.
44
45
control structure in [31] is applied to integrate the MTs with is also added to the testbed for sending starting and stopping
46
the grid through power electronics converters. Furthermore, signals and collecting experimental data from the controllers.
47
the type 4 wind turbine model in [32] is used, and the control Experimental data from the model running in the DRTS is
48
model of the PVs is taken from [33]. Additionally, PV panels collected by OPAL-RT’s OpWriteFile modules. A 1-Gbps
49
are modeled by the double-diode model with parameters ethernet network connecting these devices is established by
50
extracted from the Siemens SM50 solar panel’s specifications. a TP-LINK T1600G-52TS switch. The DRTS communicates
51
Because the main focus of this paper is EM, average models with the controllers by the User Datagram Protocol (UDP),
52
for power electronics converters of the MTs, and PVs, and while the communication protocol among the controllers is
53
WT are sufficient to demonstrate the EM strategy. the RTI Data Distribution Service (DDS). Moreover, the HMI
54
55 Fig. 2b) illustrates the CHIL experimental setup. The system exchanges data with the controllers by shared variables in
56 is simulated by MATLAB/Simulink and deployed into a dig- LabVIEW. EM’s results are passed to the model running in the
57 ital real-time simulator (DRTS) OPAL-RT OP4510. The EM DRTS as setpoints. The MTs have a duty of controlling active
58 strategy is realized by LabVIEW and deployed into NI sbRIO and reactive power injected into the network and bus voltages
59 9627 controllers. A human-machine interface (HMI) computer at which the MTs are installed, while the WT and the PVs are
60 PLEASE KEEP CONFIDENTIAL
6
 IEEE Intelligent Transportation Systems Transactions Page 8 of 12

1
2 controlled to reach the commanded reactive power from EMs.
3 Fig. 3a) illustrates the EM program with a detail of the use
4 of LabVIEW’s Timed Loop and While Loop blocks. First,
5 the program starts in a Timed Loop to initialize variables
6 and buffers and to listen to starting signals from the HMI.
7 After receiving the starting signal, there are two main blocks
8 running in parallel: Optimization Block and Communication
9 Block. The former is to implement the two-step optimization
10 process while the latter exchanges data among the controllers
11 and the simulator with the controllers. NI sbRIO 9627 boards
12 are powered by a dual-core ARM Cortex-A9 microprocessor
13 which allows the parallel running of the two blocks. Every 15
14 minutes, information of non-EV load, EV load, and weather
15 data are fed into EMs; each EM can only receive this infor-
16 mation of buses which it manages. Although these values can
17 be generated from forecasting engines, they are stored in the
18 DRTS and are sent to the controllers by the UDP protocol
19 as this work primarily studies EM. Each controller uses this
20 information and local system data of the region to form a
21 LabVIEW cluster called sysInf o. Local system data includes
22 the local admittance matrix and the WT and PVs’ system
23 parameters to estimate their power generation outputs given
24 weather data. The information of cost functions and constraints
25 is also in sysInf o.
26 A synchronous updating mechanism is designed in this
27 work. In particular, the iteration index k is added at the end
28 of the exchanging data frame. A circular buffer reads the data
29 coming from neighboring EMs, and the buffer is updated by
30 the first-in-first-out rule if it discerns a new iteration index;
31 otherwise, it ignores the receiving data. Every iteration, the
32 controller reads the circular buffer and matches the iteration
33 index in order to have the same k for all involved EMs. This
34 applies to both Step 1 and Step 2 as both steps use only the
35 same circular buffer and RTI DDS’s writer and reader block.
36 As an illustration of exchanging data between two EMs, Fig.
37 3b) shows the exchange of data between EM1 and EM5. Buses
38 that are managed in Region 1 are 1, 2, 3, and 4, and buses
39 that are managed in Region 5 are 23, 24, and 25. These buses
40 are illustrated by circles. There is a connection between nodes
41 3 and 23. Although Region 1 does not contain bus 23, EM1
re im
42 has V1,23 and V1,23 as variables in the optimization process.
43 Note that the first subscript indicates the region index, and
44 the second indicates the bus index. Bus 23 in Region 1 is
45 illustrated by a pentagon. Region 5 contains bus 23 and its bus
re im
46 voltage can be represented by [V5,23 , V5,23 ]. The optimization
re im
47 process aims to reach a consensus between [V1,23 , V1,23 ] and
re im
48 [V5,23 , V5,23 ]. This applies to the other buses which have
49 connections between two regions.
50 After Step 1, active and reactive power constraints are
51 reconfigured by modifying corresponding fields in sysInf o,
52 and then fed into Step 2. To solve (19a), the LabVIEW’s
53 Constrained Nonlinear Optimization module, which utilizes
54 the sequential quadratic programming algorithm, is used. To
55 obtain the information of the optimization problem, this mod-
56 ule accesses sysInf o via a Formula Node with C++ syntax Fig. 3. Distributed EM: a) Diagram of EM’s LabVIEW program b) Illustration
57 structure code deployed inside. Although it is observed that of exchanging data between EM1 and EM5.
58 each iteration of Step 1 has a computational time fewer than
59 60 ms by implementing on the NI sbRIO 9627 controllers,
60 PLEASE KEEP CONFIDENTIAL
7
Page 9 of 12 IEEE Intelligent Transportation Systems Transactions

1
2 the time for each iteration of Step 1 is set to 60 ms, which is is specified in Texas Instrument’s design recommendation
3 higher than 40 ms of the Comm. w. EMs block in Fig. 3a) to [36]. With the assumption of 3715 households, three levels
4 ensure the communication time is less than the computational of EV penetration are considered: 500 EVs, 1000 EVs, and
5 time of Step 1. Step 2 is more computationally expensive than 1500 EVs. The number of EVs assigned to each bus is also
6 Step 1 because of the nonlinear constraint solver; therefore, a proportional to the value of the original load. The fast-charging
7 duration of 3 seconds is set for each iteration. The maximum profile generation feature of the EV profile generation tool
8 number of iterations is the stopping condition for both Step 1 for the public charging stations outside the analysis area is
9 and Step 2. The stopping conditions of the Comm w/ DRTS taken into account to better represent the car arrival times
10 block and the Comm w/ EMs block are designed such that and starting SoC levels of the second charging sessions for
11 either Step 1 or Step 2 runs in parallel with the Comm w/ the residential chargers in the considered network. While the
12 EMs block. CREST model can be used to generate weather data for PVs,
13 wind speed generation is not included in the model; therefore,
14 4000 it is generated by the Weibull distribution. Fig. 4 shows data
15 generated by using the methodology described above for the
16 day of January 15. Furthermore, the market price of electricity
2000
17 is taken from [37], and the fuel price is cF = 5.807 ($/gallon)
18 which is taken from [38] at the time of the experiment. The
19 0
values of αn , βn , and γn of MTs at Buses 8, 24, and 30
20 0 4 8 12 16 20 24 are 0.12, 3.507, and 149.977, respectively. Those of MTs at
21 30 Buses 14 and 20 are 0.097, 4.269, and 92.293, respectively.
22 Additionally, the the market price of electricity is taken from
20
23 [37].
m/s

24 10
25 0 B. CHIL Experimental Results and Discussion
26 0 4 8 12 16 20 24
27 1.04
200
28 500 EVs
W/m2

29 100
1.03 1000 EVs
1500 EVs
30 1.02
31 0
32 0 4 8 12 16 20 24 1.01

33 1
34 4
pu

0.99
35
C°

3
36 2 0.98
37
0 4 8 12 16 20 24 0.97
38
39 0.96
40 0.95
41 Fig. 4. Simulated operational data used in the experiment: a) load profile and
renewable generation, b) wind speed, c) solar irradiance, and d) temperature.
42
1 5 12 19 23 26 33
43 Bus Number
44 2) Operational Data: The original load is replaced by the
45 sum of EV load profiles generated from the method presented Fig. 5. Voltage profile of Scenario 1.
46 in Subsection II-C and non-EV load profiles generated from
47 the CREST model [34]. 100 non-EV load profiles are gen- 1) Scenario 1: The system without distributed EM is set
48 erated from the model, and they are averaged. It is observed up. MTs are simulated in the grid following mode that they
49 that the peak value of the 15-min averaged load per individual generate active power to serve the load locally in their regions
50 profile is around 1 kW. Originally, the IEEE 33 bus system until they reach limits. The reactive power of the three types of
51 has a fixed 3.715 MW of load in total. Assuming the original DERs is set to 0. With this setup, the impact of EV charging
52 load as the peak total demand and 3715 households in the under the three levels of EV penetration at the peak load,
53 residential area, load profiles are generated for each house- which is at 19:30, is studied. Fig. 5 shows bus voltages of
54 hold. The load profiles are randomly distributed to the buses the three testing cases. As can be seen, while voltages for the
55 such that the number of load profiles is proportional to the cases of 500 and 1000 EVs are in the range [0.95,1.05] pu,
56 original load values at the buses. The CREST model does not buses 16, 17, and 18 have voltages less than 0.95 pu when the
57 consider reactive power, so it is assumed PF follows uniform system populates 1500 EVs.
58 distribution in the range [0.95,0.99], which is close to that of 2) Scenario 2: This scenario follows the setup described in
59 [35]. For EV reactive load, a PF of 0.99 is assumed which IV-A1 with 1500 EVs populated, and EMs are deployed. Three
60 PLEASE KEEP CONFIDENTIAL
8
 IEEE Intelligent Transportation Systems Transactions Page 10 of 12

1
2000
2 Step 1 + Step 2
3 Step 2
4 1500
Without EM

5
6 1000

$
7
8
9 500

10
11 0
00:00 12:45 19:30
12 Time
a)
13 1.05 0.03
b)
14 1.04 00:00
15 1.03 0.025 12:45
19:30
16 1.02
0.02
1.01
17
1
18 0.99
0.015
19 0.98 0.01
20 0.97 00:00
21 0.96 12:45
19:30
0.005
22 0.95
0
0.94
23 1 5 12 19 23 26 33 0 50 100 150 200
Iteration
24 c) d)
25
26 Fig. 6. Experimental results of Scenario 2: a) power generation allocation b) cost of generation c) bus voltages d) .
27
28
29 cases are tested: at 00:00 when the load is the lowest, at 12:45 the first step enhances the global optimality searching for the
30 when the renewable generation is max, and at 19:30 when the second step. In addition, information after running Step 1 is
31 total load is at peak. The simulation results are shown in Fig. used to reconfigure constraints to ensure that the PF limits
32 6. Particularly, Fig. 6a) shows the amount of power generated of generators and the substation are respected. A realistic EV
33 or delivered at the 6 dispatchable power suppliers. For the case charging profile generation methodology is also presented in
34 of 19:30, the MTs generate at full capacity. This is because this work. The IEEE 33 bus system is used to investigate the
35 the grid’s electricity price is more expensive than the price if impact of EVs. The CHIL implementation of the strategy is
36 power generated by the MTs; therefore, EMs allocates power also reported, and the experimental evidence is shown. Ex-
37 generation to the 5 MTs to meet the power load balance. The perimental results show the economic and voltage regulation
38 total cost of generation is shown in Fig. 6b) in which the case benefits of deploying the distributed EM.
39 that generation allocated to the 6 energy sources proportionally There are potential directions to leverage this work in
40 to their maximum capacities is also computed; this case is the future. First, energy storage technologies can be added
41 without EM. It can be seen that deploying EM with both Step to the current version. Second, adopting the vehicle-to-grid
42 1 and Step 2 is economically beneficial in the 12:00 AM and (V2G) mode into the system is another extension. Third,
43 12:45 PM cases. For the case of applying only Step 2, only the future work will consider adding distribution locational
44 local optimality is found. In Fig. 6c), voltages are in the range marginal pricing of the power grid and the traffic flow of the
45 of [0.95, 1.05] pu. Compared to the case without EM, bus transportation systems to the model.
46 voltages are maintained under the impact of 1500 EVs. Define
47 the following quantity R EFERENCES
48  = max |Vr,n − Vs,n | (23) [1] G. A. Putrus, P. Suwanapingkarl, D. Johnston, E. C. Bentley,
49 n∈Ω and M. Narayana. “Impact of electric vehicles on power dis-
50 where Ω is the set of buses which have voltage optimized in
tribution networks”. 2009 IEEE Vehicle Power and Propulsion
Conference, pp. 827–831.
51 both region r and region s.  is shown in Fig. 6d) and it can [2] M. Henderson, D. Novosel, and M. L. Crow. “Electric power
52 be seen that voltages converge after 60 iterations for all three grid modernization trends, challenges, and opportunities”.
53 cases. 2017.
54 [3] B. P. Hayes, M. Prodanovic. “State forecasting and opera-
55 tional planning for distribution network energy management
V. C ONCLUSION AND F UTURE W ORK systems”. IEEE Transactions on Smart Grid, vol. 7, no. 2
56 (2016), pp. 1002–1011.
57 In this paper, a distributed EM to efficiently manage dis-
58 tribution grids with the penetration of heterogeneous DERs
59 and EV is presented. The strategy includes two steps that
60 PLEASE KEEP CONFIDENTIAL
9
Page 11 of 12 IEEE Intelligent Transportation Systems Transactions

1
2 [4] O. Ciftci, M. Mehrtash, and A. Kargarian. “Data-driven //www.anl.gov/es/light-duty-electric-drive-vehicles-monthly-
3 nonparametric chance-constrained optimization for microgrid sales-updates. (accessed: 22.08.2020).
energy management”. IEEE Transactions on Industrial Infor- [22] M. G. Flammini, G. Prettico, A. Julea, G. Fulli, A. Mazza, and
4 matics, vol. 16, no. 4 (2020), pp. 2447–2457. G. Chicco. “Statistical characterisation of the real transaction
5 [5] H. Kanchev, D. Lu, F. Colas, V. Lazarov, and B. Francois. data gathered from electric vehicle charging stations”. Electric
6 “Energy management and operational planning of a microgrid Power Systems Research, vol. 166 (2019), pp. 136–150.
7 with a PV-based active generator for smart grid applications”. [23] F. Pallonetto, M. Galvani, A. Torti, and S. Vantini. “A frame-
IEEE Transactions on Industrial Electronics, vol. 58, no. 10 work for analysis and expansion of public charging infrastruc-
8
(2011), pp. 4583–4592. ture under fast penetration of electric vehicles”. World Electric
9 [6] J. Lagorse, D. Paire, and A. Miraoui. “A multi-agent system Vehicle Journal, vol. 11, no. 1 (2020). ISSN: 2032-6653. URL:
10 for energy management of distributed power sources”. Renew- https://www.mdpi.com/2032-6653/11/1/18.
11 able Energy, vol. 35, no. 1 (2010), pp. 174–182. [24] P. Weldon, P. Morrissey, J. Brady, and M. O’Mahony. “An in-
12 [7] B. Papari, C. S. Edrington, I. Bhattacharya, and G. Radman. vestigation into usage patterns of electric vehicles in Ireland”.
“Effective energy management of hybrid AC–DC microgrids Transportation Research Part D: Transport and Environment,
13
with storage devices”. IEEE Transactions on Smart Grid, vol. 43 (2016), pp. 207–225.
14 vol. 10, no. 1 (2019), pp. 193–203. [25] J. Lavaei, S. H. Low. “Zero duality gap in optimal power flow
15 [8] B. Papari, C. S. Edrington, and D. Gonsoulin. “Optimal problem”. IEEE Transactions on Power Systems, vol. 27, no. 1
16 energy-emission management in hybrid AC-DC microgrids (2012), pp. 92–107.
17 with vehicle-2-grid technology”. Journal of Renewable and [26] S. H. Low. “Convex relaxation of optimal power flow—part I:
Sustainable Energy, vol. 11, no. 1 (2019). formulations and equivalence”. IEEE Transactions on Control
18 [9] B. Papari, C. S. Edrington, T. V. Vu, and F. Diaz-Franco. of Network Systems, vol. 1, no. 1 (2014), pp. 15–27.
19 “A heuristic method for optimal energy management of DC [27] P. H. Hoang, C. S. Edrington, B. Papari, G. Ozkan, and
20 microgrid”. 2017 IEEE Second International Conference on H. Ahn. “Distributed constrained optimization over networked
21 DC Microgrids (ICDCM), pp. 337–343. systems via a singular perturbation method and application to
22 [10] T. Mühlpfordt, X. Dai, A. Engelmann, and V. Hagen- economic dispatch”. 2020 Clemson University Power Systems
meyer. “Distributed power flow and distributed optimiza- Conference (PSC), pp. 1–6.
23 tion—Formulation, solution, and open source implementa- [28] P. H. Hoang, G. Ozkan, P. R. Badr, B. Papari, and C. S.
24 tion”. Sustainable Energy, Grids and Networks, vol. 26 (2021). Edrington. “Distributed constrained optimization by singular
25 [11] S. Boyd, N. Parikh, and E. Chu. “Distributed optimization perturbation method and its application to energy networks
26 and statistical learning via the alternating direction method of by real-time implementation”. Sustainable Energy, Grids and
27 multipliers”. Now Publishers Inc (2011). Networks, vol. 27 (2021).
[12] T. Erseghe. “Distributed optimal power flow using ADMM”. [29] M. E. Baran, F. F. Wu. “Network reconfiguration in distribution
28 IEEE Transactions on Power Systems, vol. 29, no. 5 (2014), systems for loss reduction and load balancing”. IEEE Transac-
29 pp. 2370–2380. tions on Power Delivery, vol. 4, no. 2 (1989), pp. 1401–1407.
30 [13] S. Mhanna, G. Verbič, and A. C. Chapman. “Adaptive ADMM [30] W. I. Rowen. “Simplified mathematical representations of
31 for distributed AC optimal power flow”. IEEE Transactions on heavy-duty gas turbines”. Journal of Engineering for Power,
32 Power Systems, vol. 34, no. 3 (2019), pp. 2025–2035. vol. 105, no. 4 (1983), pp. 865–869.
[14] Y. Zhang, M. Hong, E. Dall’Anese, S. V. Dhople, and Z. [31] S. Grillo, S. Massucco, A. Morini, A. Pitto, and F. Silvestro.
33 Xu. “Distributed controllers seeking AC optimal power flow “Microturbine control modeling to investigate the effects of
34 solutions using ADMM”. IEEE Transactions on Smart Grid, distributed generation in electric energy networks”. IEEE
35 vol. 9, no. 5 (2018), pp. 4525–4537. Systems Journal, vol. 4, no. 3 (2010), pp. 303–312.
36 [15] A. Engelmann, Y. Jiang, T. Mühlpfordt, B. Houska, and T. [32] A. D. Hansen, I. D. Margaris. “Type IV wind turbine model”.
37 Faulwasser. “Toward distributed OPF using ALADIN”. IEEE DTU Wind Energy (2014).
Transactions on Power Systems, vol. 34, no. 1 (2019), pp. 584– [33] N. Femia, G. Petrone, G. Spagnuolo, and M. Vitelli. “A
38 594. technique for improving P&O MPPT performances of double-
39 [16] S. Xia, S. Bu, C. Wan, X. Lu, K. W. Chan, and B. Zhou. stage grid-connected photovoltaic systems”. IEEE Transac-
40 “A fully distributed hierarchical control framework for co- tions on Industrial Electronics, vol. 56, no. 11 (2009),
41 ordinated operation of DERs in active distribution power pp. 4473–4482.
42 networks”. IEEE Transactions on Power Systems, vol. 34, no. 6 [34] I. Richardson, M. Thomson, D. Infield, and C. Clifford.
(2019), pp. 5184–5197. “Domestic electricity use: A high-resolution energy demand
43 [17] A. Supponen, A. Rautainen, J. Markkula, A. Mäkinen, P. model”. Energy and Buildings, vol. 42, no. 10 (2010),
44 Järventausta, and S. Repo. “Power quality in distribution net- pp. 1878–1887.
45 works with electric vehicle charging - a research methodology [35] J. Ponoćko, J. V. Milanović. “Forecasting demand flexibil-
46 based on field tests and real data”. 2016 Eleventh International ity of aggregated residential load using smart meter data”.
47 Conference on Ecological Vehicles and Renewable Energies IEEE Transactions on Power Systems, vol. 33, no. 5 (2018),
(EVER). 2016, pp. 1–11. pp. 5446–5455.
48 [18] R. Li, Q. Wu, and S. S. Oren. “Distribution locational marginal [36] B. Johnson. “Power factor correction design for on-board
49 pricing for optimal electric vehicle charging management”. chargers in electric vehicles”. Texas Instruments’ Application
50 IEEE Transactions on Power Systems, vol. 29, no. 1 (2014), Report (2018).
51 pp. 203–211. [37] J. Wang, G. R. Bharati, S. Paudyal, O. Ceylan, B. P. Bhattarai,
[19] J. Quiros-Tortos, L. F. Ochoa, and T. Butler. “How electric and K. S. Myers. “Coordinated electric vehicle charging with
52
vehicles and the grid work together: Lessons learned from one reactive power support to distribution grids”. IEEE Transac-
53 of the largest electric vehicle trials in the world”. IEEE Power tions on Industrial Informatics, vol. 15, no. 1 (2019), pp. 54–
54 and Energy Magazine, vol. 16, no. 6 (2018), pp. 64–76. 63.
55 [20] Electric Vehicle Database. All electric vehicles. URL: https : [38] Global petrol prices. United Kingdom gasoline price. URL:
56 //ev-database.org/. (accessed: 22.08.2020). https : / / www . globalpetrolprices . com / United - Kingdom /
[21] Argonne National Laboratory Energy Systems Division. Light gasoline prices/. (accessed: 24.08.2020).
57
duty electric drive vehicles monthly sales updates. URL: https:
58
59
60 PLEASE KEEP CONFIDENTIAL
10
 IEEE Intelligent Transportation Systems Transactions Page 12 of 12

1
2 Phuong H. Hoang received his BS degree in Con- Christopher Shannon Edrington received his BS
trol and Automation Engineering in 2013 at Hanoi in Engineering from Arkansas State University in
3 University of Science and Technology, Vietnam. He 1999 and his MS and PhD in Electrical Engineering
4 then obtained an MS degree in Mechanical Engi- from the Missouri University of Science and Tech-
5 neering at Gwangju Institute of Science and Tech- nology in 2001 and 2004, respectively; where he was
nology in 2018. He researched distributed control both a DoE GAANN Fellow, NSF IGERT Fellow
6 and optimization applied to energy networks during and Grainger Foundation Fellow. He currently is
7 his master’s studies. After earning the MS degree, the Warren H. Owen Distinguished Professor of
8 he was accepted to the Ph.D. program in Electrical Electrical and Computer Engineering at Clemson
Engineering Fall 2018 at Florida State University University and is the lead for RT-COOL (Real-Time
9 where he worked as a Graduate Research Assistant COntrol and Optimization Laboratory) as well as the
10 in Center for Advanced Power Systems under the supervision of Dr. Edrington. co-director of the Smart Energy and Propulsion Focus Area in the new VIPR-
11 Currently, he is with Clemson University and also under the supervision of GS (VIrtual PRototyping for Ground vehicle Systems) Center. His research
Dr. Edrington in the Department of Electrical and Computer Engineering. interests include modeling, simulation, and control of electromechanical
12 His current research interests include real-time CHIL simulations, energy drive systems; applied power electronics; distributed control; integration of
13 management, and degradation forecasting for power systems. renewable energy, storage, and pulse power loads. Dr. Edrington has published
14 over 170 papers (including 2 IEEE Prize Awards and multiple conference
paper awards), has graduated 23 MS students and 16 PhD students (with 3
15 in process) and has 6 patents (in real-time stability and complexity metrics).
16
17 Gokhan Ozkan received his BS degrees in Teacher
18 Training in Electrical Field and Energy System
Engineering from Marmara University and Erciyes
19 University, Turkey in 2006 and 2014, and his MS in Mustafa Alparslan Zehir received his BS in 2010,
20 Energy System Engineering from Erciyes University, MS in 2013 and PhD in 2019 in Electrical En-
21 Turkey in 2016. He was a lecturer at Bozok Uni- gineering at Istanbul Technical University (ITU).
versity, Turkey. He completed his PhD in Electrical
22 and Computer Engineering at FAMU-FSU College
From 2012 to 2019, he worked as a research and
teaching assistant at ITU, while also taking part
23 of Engineering, and worked as a graduate research in a number of European projects. He is currently
24 assistant at the Center for Advanced Power Systems, a postdoctoral researcher at MaREI/Environmental
Florida State University. He is now working as a
25 Post-doctoral Research Associate for RT-COOL (Real-Time COntrol and Op-
Research Institute, University College Cork and he
is in the team of CREDENCE international project.
26 timization Laboratory) at Clemson University. His research interests include His research fields are electric power distribution,
27 renewable energy real-time modelling, simulation, control design, and electro- distributed energy resources, demand response, elec-
thermal management for power converters.
28 tricity markets, prosumers and microgrids.
29
30
31
32 Payam Ramezani Badr received his B.S. degree
Barry Hayes is a researcher and lecturer in Electri-
in electrical engineering from the University of
33 Lorestan, Iran, in 2019 and is currently pursuing his
cal and Electronic Engineering at UCC. His research
interests include the grid integration of sustainable
34 PhD degree as a direct-entry candidate in electrical
energy technologies, and the operation and planning
35 engineering with a focus on adaptive control systems
of future power systems.
at Clemson University. He works as a research
36 assistant for RT-COOL (Real-Time COntrol and Op-
Before joining UCC in July 2018, Barry was a
lecturer in Electrical and Electronic Engineering at
37 timization Laboratory) in Holcombe Department of
NUI Galway. From 2013 to 2016, Barry was a Marie
38 Electrical and Computer Engineering (ECE), Clem-
Sklodowska-Curie research fellow at IMDEA En-
son University under the supervision of Dr. Chris S.
39 Edrington. His research interests include microgrids,
ergy in Madrid, where he worked on European and
national research projects related to power systems
40 control, performance optimization, electro-thermal modelling of machines,
and smart grids. During his time in Madrid, he was awarded the Jose Castillejo
41 and real-time experimentation. Payam Badr is a committee member for IEEE
Student Branch Clemson University. He also currently serves as a senator
and Juan de la Cierva grants from the Spanish national research councils.
42 in the Graduate Student Government at Clemson University representing the
He holds a PhD in Electrical Power Systems Engineering from the
University of Edinburgh (2013), and has held visiting researcher positions
43 ECE department.
at National Grid UK (2011) and at the University of Tennessee (2016). Barry
44 has more than 30 peer-reviewed international publications in book chapters,
45 conferences, and journals including IEEE Transactions on Power Systems and
IEEE Transactions on Smart Grid. He is a Senior Member of the IEEE.
46
47 Behnaz Papari received her Ph.D. degree in elec-
48 trical and electronics engineering from Florida State
University, USA in 2018. She is currently an As-
49 sistant Professor in Power Electronics Engineering,
50 and manager of the Duke Energy Smart Grid Lab. Laura Mehigan is a final year PhD Researcher in
the Energy Policy and Modelling team in MAREI.
51 She is associated with the Energy Production and
Her research focuses on understanding the future
Infrastructure Center. She has expertise in power
52 systems with an emphasis on modeling, analysis, shape of electricity systems and more specifically the
53 control, planning, and optimization. Her specialties interplay between decentralization and interconnec-
tion as part of the CREDENCE project. She joined
54 are distributed controls and decision analysis under
in 2017 having spent the previous 12+ years working
uncertainty, applications on energy system modeling,
55 and Stochastic optimization. Her research interests include renewable energy in a variety of senior engineering roles with EirGrid,
56 sources, power and energy management, control of stand-alone and utility- the Irish Transmission System Operator. She holds a
postgraduate diploma in Project Management from
57 interactive energy systems, distributed control of smart grids, stochastic
Trinity College Dublin and a Degree in Electrical
analysis, and real-time power distribution system simulation and hardware-
58 in-the-loop instantiation. and Electronic Engineering from University College Cork.
59
60 PLEASE KEEP CONFIDENTIAL
11
Page 13 of 12 IEEE Intelligent Transportation Systems Transactions

1
2 Dlzar Al Kez is currently a PhD student at Queen’s
University Belfast. Before joining Queen’s Univer-
3 sity Belfast, he obtained an MS in Energy and
4 Sustainability with Electrical Power Engineering at
5 University of Southampton and a BS in Electrical
Engineering at University of Sulaimani. His research
6 interests include impact of distributed generation on
7 the power system dynamics, impact of renewable
8 energy on the power system dynamics, and Applica-
tion of energy storage, smart appliances, and FACTS
9 devices in power system stability.
10
11
12
13 Aoife M. Foley (M’06) received the B.Eng. (Hons.)
14 degree in civil engineering from the University Col-
lege Cork, Cork, Ireland, in 1996, the M.Sc. degree
15 in environmental & transportation engineering from
16 Trinity College, Dublin, Ireland, in 1999, and the
17 Ph.D. degree in energy engineering from the Univer-
sity College Cork in 2011. She worked in industry
18 until 2008. She is currently a Reader with the School
19 of Mechanical and Aerospace Engineering, Queen’s
20 University Belfast, Belfast, U.K. Her research in-
terests include wind power, energy markets, energy
21 storage, and electric vehicles. Dr. Foley has been a Chartered Engineer since
22 2001 and a Fellow of Engineers Ireland since 2012. She is the Editor-in-Chief
23 of Renewable and Sustainable Energy Reviews.
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 PLEASE KEEP CONFIDENTIAL
12