Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

3568 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO.

4, JULY 2023

Robust Multi-Objective Congestion Management in


Distribution Network
Omniyah Gul M Khan , Graduate Student Member, IEEE, Amr Youssef , Senior Member, IEEE,
Magdy Salama , Life Fellow, IEEE, and Ehab El-Saadany , Fellow, IEEE

Abstract—Increased penetration of heavy loads is expected to Parameters


lead to congestion in distribution networks. The distribution net- β Price Sensitivity Coefficient Matrix.
work operator can use Demand Side Management (DSM) to moti- δ Percentage of the maximum branch limit.
vate consumers to shift their load from peak to off-peak times. In
this paper, multi-objective optimization is utilized to schedule flex- η, γ Step size.
ible load to alleviate potential congestions. The proposed scheme τ Small constant.
minimizes consumers’ electricity cost and decreases the peak to ah,t Availability of EV h at time t.
average ratio of the load curve to a required level that alleviates Chi Internal heat capacity of house h.
existing congestion. This results in a consumer load schedule that Chs Surface heat capacity of house h.
is economical and does not require the imposition of congestion tar-
iffs. However, the success of the proposed congestion management Ch Coefficient of Performance of HP of house h.
scheme relies on the accuracy of the consumer load consumption. D Power Transfer Distribution Factor.
Hence, in this paper, uncertainty analysis of consumers’ flexible E Customer to bus mapping matrix.
load schedule is executed to ensure the desired robustness of the Fmax
b
Maximum power of branch b.
power flowing in the distribution network to changes in uncertain k Iteration Number.
variables. The results obtained are compared with the existing
congestion management scheme demonstrating the advantage of Phevmax Maximum EV charging power.
the proposed multi-objective framework in terms of decreasing Pmax
hp
Maximum HP output limit.
price and flattening the load curve while alleviating congestion. Phreq Total required EV battery charge of house h.
Index Terms—Multi-Objective, Pareto, valley-filling, price-
pt Day-ahead electricity price.
based, Demand Side Management. Rhie Thermal resistance between interior and exterior of
house h.
Rhis Thermal resistance between interior and surface of
NOMENCLATURE house h.
Rhse Thermal resistance between surface and exterior of
Sets and Indices house h.
b Branch Number. Tte External temperature at time t.
h Consumer Number. Thmax Maximum internal temperature requirement.
Na Total number of aggregators. Thmin Minimum internal temperature requirement.
Nb Total number of network branches.
Variables
Nd Total number of load buses.
α Pareto Weight.
Nh Total Number of consumers.
λ Lagrange multiplier.
NT Total number of time slots.
μh Water level of consumer h.
t Time Slot.
fin ith normalized objective function.
fi ith objective function.
Manuscript received 23 February 2022; revised 5 July 2022 and 29 July Ft b
Power flowing in branch b at time t.
2022; accepted 8 August 2022. Date of publication 22 August 2022; date of
current version 22 June 2023. This work was supported in part by Khalifa L(x, λ) Lagrangian Function.
f
University, UAE under Grant CIRA-013-2020 and in part by NSERC, Canada. Ph,t Flexible Load of consumer h at time t.
Paper no. TPWRS-00256-2022. (Corresponding author: Omniyah Gul M Khan.)
Omniyah Gul M Khan and Magdy Salama are with the Electrical and Com-
Ph,t
ev
EV Power of consumer h at time t.
hp
puter Engineering Department, University of Waterloo, Waterloo, ON N2L 3G1, Ph,t HP Power of consumer h at time t.
Canada (e-mail: ogulmkhan@uwaterloo.ca; msalama@uwaterloo.ca). nf
Amr Youssef is with the Concordia Institute for Information Systems En- Ph,t Non-Flexible Load of consumer h at time t.
gineering, Concordia University, Montreal, QC H3G 1M8, Canada (e-mail: Pi,t Aggregator i total load requirement at time t.
youssef@ciise.concordia.ca).
Ehab El-Saadany is with the Advanced Power and Energy Cen-
Th,t
i
Internal temperature of house h at time t.
ter, EECS Department, Khalifa University, Abu Dhabi, UAE (e-mail: Th,t
s
Surface temperature of house h at time t.
ehab.elsadaany@ku.ac.ae). xti
Aggregator i multi-objective variables at time t.
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TPWRS.2022.3200838. zi ∗ Ideal Point of aggregator i.
Digital Object Identifier 10.1109/TPWRS.2022.3200838 zinadir Nadir Point of aggregator i.

0885-8950 © 2022 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ROBUST MULTI-OBJECTIVE CONGESTION MANAGEMENT IN DISTRIBUTION NETWORK 3569

Uncertainty Parameters distribution network. Utilization of DSM [7] to schedule flexible


ΔFt Change in line loading at time t. loads helps in reducing these challenges. To manage congested
ΔPreq Uncertainty in required EV charge. networks, DSM techniques, specifically load shifting and valley
ΔT e Uncertainty in ambient temperature. filling methods, have been developed. These techniques encour-
Δtar Uncertainty in arrival time. age consumers to adjust their electricity consumption profiles.
Δtdep Uncertainty in departure time. Dynamic Tariff (DT) [8] and Distributed DT (DDT) are price-
σT Resulting distribution total standard deviation. based load shifting methods [9] in which a charge is imposed
σTe Standard deviation of external temperature distribution. at congested times. Aggregators, representing consumers in the
ζPreq Sensitivity of line loading to changes in required EV electricity market, re-optimize their customers’ load schedules
charge. to meet their demand preferences while minimizing their costs.
ζT e Sensitivity of line loading to changes in ambient tem- On the other hand, incentive-based load shifting methods, such
perature. as subsidy-based [10] and peak time rebate [11], pay consumers
ζtar Sensitivity of line loading to changes in electric vehicle a rebate for reducing their peak load. Load shifting methods
arrival time. result in reducing both power consumption and energy costs. On
ζtdep Sensitivity of line loading to changes in electric vehicle the other hand, valley filling techniques have been used in the
departure time. literature to flatten the load curve through efficient scheduling
ai Gaussian Scaling Limits. of the flexible load in off-peak times. In [12], vehicle to grid
bi Gaussian density function means. peak shaving and valley filling control strategy was utilized,
ci Gaussian density function standard deviation. while [13], [14], [15] adopted the water-filling algorithm to flat-
R Robust Metric. ten the overall power consumption. In [16], a nested optimization
U Uncertainty variables. approach was adopted to schedule EV charging. The optimal EV
charging schedule was characterized to be a valley-filling profile.
Similarly, [17] developed a decentralized shrunken primal-dual
I. INTRODUCTION subgradient algorithm to effectively schedule charging of EVs
ATELY, the smart grid has observed heavy penetration of while flattening the demand profile. Table I lists other existing
L renewable energy sources (RES)s, such as solar and wind,
to help with peak demands. However, this causes bidirectional
literature in managing congestion in the distribution network.
Load shifting and valley filling DSM techniques contribute to
power flow in the grid and also introduces uncertainty due to postponing huge investments needed in reinforcing distribu-
their intermittent nature. Moreover, to decrease greenhouse gas tion network assets to deal with the increased peak demand
emissions, the smart grid drives towards the electrification of [18].
the transportation system, via electric vehicles (EVs), and space The existing price and incentive-based congestion manage-
heating, via heat pumps (HPs). This has increased the usage ment methods [8], [9], [10], [11] aims to alleviate congestion
of such heavy flexible loads and changed the operating condi- using an electricity cost minimization objective function. How-
tions of the distribution system [1]. The demand for electricity ever, this may result in new peaks as loads are scheduled at
appears to be increasing as HPs are replacing many traditional times when prices are low. Moreover, the peak to average ratio
heating/cooling units in order to reduce oil consumption and of load curves in such modeling is high. On the other hand,
heating costs, and EVs are replacing combustion engine vehicles valley filling techniques [12], [13], [14], [15], [16], [17] results
to reduce harmful exhaust emissions. Moreover, the dependence in flattening the load curve by filling the valleys in the load
of such technologies on uncontrollable factors, such as tem- profiles which is advantageous. However, as the price is not taken
perature, wind speed, etc., may result in potential congestion into consideration, consumers end up paying more compared to
problems in the distribution network which were not of concern load shifting methods. To gain the advantage of both worlds,
in the past. Congestion has been defined in [2] as an event multi-objective optimization is proposed to schedule flexible
that “occurs when planned or actual flows across a system loads by merging load shifting and valley filling methods. This
component exceed its safe design capacity”. Congestion in the would result in consumers minimizing their electricity costs as
distribution network occurs due to the high power consumption well as flattening the curve enough to alleviate any existing
of active loads and the weakening correlation between electricity congestion in the network.
prices and demand resulting from the increased penetration Accuracy of the proposed multi-objective congestion man-
level of intermittent RESs [3], [4]. Such congestion results in agement scheme, and any other existing technique, relies on the
voltage violations and/or thermal overloading as a result of accuracy of the modeling parameters. However, the scheduling
the power flow exceeding a network asset’s transfer capability, of flexible loads is generally reliant on uncertain data. For
possibly damaging devices such as distribution transformers example, planning the day-ahead operation of HPs relies on un-
and feeders [5]. Hence, reinforcement of overloaded network certain ambient temperature. As a result of load uncertainty, the
assets is required incurring costs [6] that may seem redundant consumers’ power profiles would deviate in real-time from the
as peak load generally appears only for a few hours. To avoid day-ahead load schedule impacting congestion in the network.
or postpone such costs, this paper utilizes the grid’s demand Hence, it is essential that the sources of load uncertainty are
side management (DSM) feature to alleviate congestion in the identified and the propagation of uncertainty, reflected as the

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
3570 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

TABLE I
SUMMARY OF LITERATURE REVIEW OF CONGESTION MANAGEMENT METHODS IN DISTRIBUTION NETWORK

change in power flowing in the distribution network due to un- II. DETERMINISTIC MULTI-OBJECTIVE CONGESTION
certain variables, is studied. To the authors’ knowledge, [3], [25] MANAGEMENT SCHEME
is the only work that has incorporated uncertainty in modeling a To schedule flexible loads, DSM methods, using either load
dynamic tariff-based congestion management model. However, shifting or valley filling, have been utilized in the literature.
the proposed model only considered EVs as stochastic flexible
These methods have their advantages and disadvantages and
loads and is not distributed in nature, i.e. it relies on the network can be compared in terms of electricity price paid by customers
operator to forecast consumer flexibility and load requirements. and peak-to-average (PAR) of the power flowing in the network
This paper presents a robust multi-objective congestion man-
feeder as follows:
agement framework to alleviate congestions in the distribution r Electricity Price: Price-based optimization is based on
network. The contributions of this paper can be summarized as
determining the optimal schedule that minimizes the con-
follows:
r A multi-objective congestion management model is de- sumers’ electricity cost. Hence, price-based optimization
results in the most economic schedule. On the other hand,
veloped using valley filling to flatten the load curve to
valley-filling schemes aim to flatten the load curve resulting
an optimal level that alleviates congestion in the network
in higher electricity costs for the consumer.
while maintaining minimum consumer electricity costs. r PAR: Price-based optimization’s advantage in terms of
This would avoid the need for any extra congestion tariffs
minimum electricity cost consumption can result in re-
to be imposed to manage network congestions. Since the bound peaks, and hence a higher PAR. Flexible load is
success of the proposed congestion management scheme scheduled in the early morning and late-night hours when
in the real-time market depends on the accuracy of the
the price of electricity is the lowest resulting in peaks in
consumer load consumption model in the day-ahead mar- demand. On the other hand, flattening the load curve as a
ket, the robustness of the proposed multi-objective conges- result of valley-filling optimization results in a lower PAR
tion management model to uncertainties in consumer load
compared to price-based optimization.
scheduling is investigated. An iterative scheme is adopted In this paper, to gain the advantage of minimizing consumers’
to determine the multi-objective Pareto weight that ensures bills and flattening the load curve to alleviate congestion, a multi-
the robustness of the line loading despite uncertainties in
objective model is proposed.
consumer flexible load.
r The proposed method is tested using an IEEE 33 bus system
and the results are compared to that of the DDT congestion A. Modeling Assumptions
management scheme. Prior to formulating the problem, the necessary assumptions
The rest of the paper is organized as follows: Section II adopted are listed and justified as follows:
explains the proposed deterministic multi-objective congestion r Forecasted day-ahead prices, pt , are determined by the
management scheme. Sources of uncertainty in the proposed market operator and forwarded to the grid operators
model are identified and a robust solution is developed in Sec- who in turn sends them to the aggregators in the
tion III. In Section IV, the proposed multi-objective flexible load network [19].
scheduling is tested in an IEEE 33 bus system. The obtained r EVs and HPs are chosen as flexible loads. Not only do they
results are compared with price-based congestion management represent the heaviest residential loads, but they represent
method prior to concluding the paper. shiftable and non-shiftable flexible loads [9].

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ROBUST MULTI-OBJECTIVE CONGESTION MANAGEMENT IN DISTRIBUTION NETWORK 3571

r Aggregators are defined as entities that are not only respon- following constraints
sible for adjusting consumers’ demand to minimize their 1
costs but are also the retailers of the consumers [26]. Th,t
i
− Th,t−1
i
= (P hp Ch + Rhis (Th,t
s
− Th,t
i
)
r Aggregators are assumed to be motivated to participate in Chi h,t
the day-ahead market as they provide financially binding + Rhie (Tte − Th,t
i
)) (1c)
prices protecting consumers from unexpectedly high real-
1
time prices [27]. Th,t
s
− Th,t−1
s
= (Ris (T i − Th,t
s
) + Rhse (Tte − Th,t
s
))
r Aggregators have the option to not participate in the day- Chs h h,t
ahead market to schedule their flexible loads. The network (1d)
operator, hence, utilizes its load forecasting tools to com- hp
0 ≤ Ph,t ≤ Pmax
hp
(1e)
plete the missing aggregator’s load information.
r Aggregators have no incentive to lie. Thmin ≤ Th,t
i
≤ Thmax (1f)
where Chi and Chs represent the internal and surface heat capacity
B. Problem Formulation of house, h, respectively. R , R , and R represent thermal
is ie se

The Distribution Network Operator (DNO) is the entity re- resistance between the interior and surface, interior and external,
sponsible for ensuring the alleviation of any congestions in the and surface and exterior respectively. Ch symbolizes the coef-
distribution network. Price-based load shifting and valley-filling ficient of performance of the HP. Pmax hp
is the maximum heat
objective functions are combined into a single multi-objective output limit. Th,t is the internal temperature of house h at time,
i

optimization problem to schedule consumers’ flexible loads. t, and Thmax and Thmin are the maximum and minimum indoor
Since both objective functions are convex, the weighted-sum temperature preferences of the consumers. Tte is the external
method is adopted as follows: temperature and Th,t s
is the surface temperature of house h at
time t.
 
min α μh + (1 − α) Ph,t
ev T
βPh,t
ev Similarly, EV charging is modeled by:
ev ,P hp
μh ,Ph,t 
h h,t
Phreq ≤ Ph,t
h,t ev
(1g)

hp Thp hp t
+ Ph,t βPh,t + pt (Ph,t
ev
+ Ph,t ) (1a)
0≤ Ph,t
ev
≤ ah,t Phevmax (1h)
where weight, α ∈ [0, 1] is used to scalarize the objective func- where Phreq is the total required battery charge for EV h, energy
tions proportional to the valley-filling level desired. A higher at time t, Phevmax is the maximum charging power, and ah,t is
α represents a higher priority to flattening the load curve over how the availability of EV h at time t for charging.
minimizing the electricity cost for the consumers and vice versa.
hp
Ph,t
ev
and Ph,t represents the EV power and HP power for each C. Determining Optimal Pareto Solution
house, h, at time t respectively. It should be noted that h ∈ Nh Solving the multi-objective optimization problem (1) results
and t ∈ NT , where Nh and NT are the total number of houses in a set of optimal solutions referred to as the Pareto optimal
and hours in a day respectively. β ∈ RNT ×NT represent the price front [29]. Neither point on the Pareto front is considered better
sensitivity coefficient matrix. μh , is referred to as the water level than the other except in terms of the decision-makers objective.
of consumer h and it represents the summation of its flexible, To determine the Pareto front, the two objective functions of
f nf f
Ph,t , and non-flexible, Ph,t , load. The flexible load, Ph,t , is the (1) have to be initially normalized as they are not of the same
hp
summation of the existing EV load, Ph,t ev
and HP load Ph,t for type. The ideal point, z ∗ , and the nadir point, z nadir , defined
house h at time t. The valley-filling part of the multi-objective respectively as the lower and upper bound of the Pareto optimal
optimization problem, hence, aims to minimize the water level set [30] are utilized to normalize the objective functions (1a) as:
of the consumers and maximize the feeder’s capacity using the
fi − zi∗
following equality constraint: fin = , for i = 1, 2 (2)
zi
nadir − zi∗
hp nf +
s.t. Ph,t
ev
+ Ph,t = (μh − Ph,t ) , ∀h ∈ Nh (1b) where,

Heat pumps (HPs) utilize electrical energy to extract heat from f1 = μh
a low-temperature region (referred to as a source) and transfer it h
to a warmer area (referred to as a sink). Some of the commonly  T hp hp T
hp
f2 = Ph,t
ev
βPh,t
ev
+ Ph,t βPh,t + pt (Ph,t
ev
+ Ph,t ) (3)
used sources of HPs are air and ground-source HPs. Air-source
h,t
HPs extract heat from the external air for heating and pump heat
outside for residential cooling. On the other hand, ground-source fin is the ith normalized objective function, fi is the ith objective
HPs use groundwater and/or the earth to extract heat for heating function, zi∗ is the ith objective function ideal point, and zinadir
or pump heat into it for residential cooling. In this paper, air- is the ith objective function nadir point.
source HPs are utilized as flexible loads as they dominate global The multi-objective function (1) symbolizes a trade-off be-
HP sales [28]. The thermal dynamics of HPs are modeled by the tween total electricity price and feeder congestion level, as the

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
3572 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

weight, α, varies. An increase in the weight, α, increases, the


importance of valley filling load scheduling. This results in a
flexible load schedule that is more costly compared to that of
lower weights where price-based scheduling is dominant. How-
ever, a higher weight indicates lower congestion levels. Hence,
given the Pareto front, to determine the desired optimal Pareto
solution, an interactive method is utilized in which the decision-
maker is allowed to search for the optimal solution that achieves
its desired objection [31]. In this paper, the decision-maker is
the DNO and the goal of clearing congestions in the day-ahead
market would determine the desired optimal Pareto for solving
(1). In the absence of congestion, the weight α is chosen to be 0 to
schedule the flexible loads using price-based optimization. This
results in the most economical flexible load schedule resulting
in satisfied consumers. However, as the flexible load increases,
the scheduling of the soft load is shifted from price-based to
valley-filling optimization. This would result in flattening the
curve to the level needed to alleviate the existing congestion.
Using this method, congestion tariff is not imposed and hence
the price paid by the consumers is lower compared to existing
price-based congestion management schemes.
Initially, each aggregator schedules its flexible load using
(1) and α1 , where 1 represents the k iteration number, is set
Fig. 1. Flowchart for determining multi-objective Pareto weight and optimal
to 0. Hence, the flexible loads are initially scheduled using consumer load schedule.
economical price-based optimization to determine if there is
congestion in the network or not. Aggregator i consumers’ total
load requirement, Pi,t , is calculated as: Pareto weight of the multi-objective optimization problem and
i,nf i,f
the resulting consumers’ optimal load schedules.
Pi,t = E(Ph,t + Ph,t ), ∀i ∈ Na , ∀t ∈ NT (4) As an additional feature, to ensure the absence of manipu-
lation of day-ahead load schedules by the aggregators in the
Given that Na and Nd are the set of aggregators and load buses
distribution network, an imbalance penalty factor (IPF) [32]
in the network, E is the customer-to-bus mapping matrix where
is imposed by the utility operator in the real-time market.
E ∈ RNd ×Nh . Ph,t
i,nf i,f
and Ph,t are the non-flexible and flexible
i,nf i,f
This penalty is initially zero for all aggregators in the network
loads of customer, h, of aggregator i, where {Ph,t , Ph,t }∈ assuming they are all truthful. However, proportional to the
R . Before day-ahead market closure, aggregators forward
Nh
deviation between their day-ahead load schedule and real-time
their consumers’ load requirement, Pi,t , securely to the DNO. consumption, IPF for each aggregator is adjusted accordingly.
The network operator is responsible for determining the optimal Once the IPF exceeds a certain threshold the aggregator would
Pareto weight that resolves congestion and minimizes con- not be allowed to take part in the day-ahead market.
sumers’ costs. Hence, the DNO performs power flow analysis
to determine if the power flowing in the branches, Ftb , where
branch b ∈ Nb , exceeds the branches maximum limits, Fmax b
. III. DEVELOPING A ROBUST SOLUTION FOR THE PROPOSED
The Pareto weight, α , is then updated in proportion to the
k CONGESTION MANAGEMENT SCHEME
congestion level using, The proposed congestion management scheme relies on the
accuracy of the consumer load profiles submitted by the aggre-
αk+1 = αk + Dθ min(b, Ftb − Fmax
b
) (5) gators in the network, Pi,t , to the DNO. Despite the fact that
congestion management is executed in the day-ahead market, if
where θ represents the step size and D represents the power
the proposed scheme is highly sensitive to changes in demand,
transfer distribution factor. αk+1 is updated in proportion to the
the network can experience congestion in real-time. Sensitivity
most congested feeder in the network. As feeders that are not
is measured as the change in the line loading, ΔFtb , of a distri-
congested would result in negative αk+1 ,
bution network that reflects network congestion to changes in
αk+1 = (αk+1 )+ (6) network parameters. The line loading, Ftb , is highly sensitive
if the power flowing in the network feeder changes drastically
all negative αk+1 are replaced by zeros. This iteration continues due to changes in input variables. Hence, to consider a solution
between the DNO and the aggregators until the network is not of the multi-objective optimization problem robust, the network
congested, and the difference between the Pareto weights in two line loading, Ftb , should not be very sensitive to uncertainties
consecutive iterations is less than a small constant, τ . Fig. 1 in consumer load parameters, U . The stochastic-based robust
illustrates the flowchart of steps involved in determining the metric, R, is hence defined in this paper as the probability,

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ROBUST MULTI-OBJECTIVE CONGESTION MANAGEMENT IN DISTRIBUTION NETWORK 3573

TABLE II
ARRIVAL AND DEPARTURE TIME PDF PROPERTIES

P, of the change in the line loading, ΔFtb , as a result of load


uncertainties being less than factor δ of the branch’s maximum
limit, Fmax
b
. The proposed model should have robustness, R,
less than η as follows,
R = P(ΔFtb ≤ δFmax
b
)≤η (7)
To achieve the desired robustness, R, the classical uncertainty
propagation theorem [3] is adopted. This is explained in the
following subsections after identifying the uncertain variables,
U , in the network and characterizing their stochastic nature using
probability density functions (PDF).

A. Identification of Uncertain Variables


To schedule their consumers’ flexible load with higher accu-
racy, aggregators in the distribution network have to factor in any
source of uncertainty in their optimal load scheduling. Assuming
accurate day-ahead price forecast by the market operator, pt , ag-
gregators have to incorporate flexible load-related uncertainties,
U . For example, the amount of charge EV consumers require on Fig. 2. PDF of EV (a) arrival, and (b) departure times.
returning home, Phreq , depends on their distance traveled during TABLE III
the day. In the occurrence of extra traffic or a road blockage, an CASE STUDY PARAMETERS
EV driver would require to travel a larger distance and hence
require more battery charge, Phreq , than what was forecasted in
the day-ahead schedule. These uncertainties are hence due to
the stochastic nature of such flexible loads and it has to be iden-
tified, represented using PDFs, and included in the congestion
management model. The sources of uncertainty, U , considered
in this paper and their respective modeling are as follows:
r Uncertainty in modeling EVs: The stochastic nature of EVs
is due to the random behavior of each EV owner. Schedul-
ing EV charging depends on three uncertain parameters:
time of arrival to the residence level charger, Δtar , the state
of charge of the EV battery on arrival, ΔP req , and the time
of departure the next day with a full EV battery, Δtdep . To
simulate EV availability in residential parking lots, data was used. RMSE was determined to be 0.00845 for de-
obtained from the US department of transportation [33] parture time, and 0.005336 for arrival time demonstrating
is used to model arrival and departure times of cars to the function’s fit for prediction. Hence, the availability of
residences. A two-term Gaussian model, represented by EV h, represented as ah,t (1h), depends on the arrival
(8), is fit as illustrated in Fig. 2 respectively. and departure times obtained probabilistically from the
gaussian PDF. The initial state of charge (SOC) of each EV
2 2
P(t) = a1 e−((t−b1 )/c1 ) + a2 e−((t−b2 )/c2 ) , ∀t ∈ NT is randomized to be normally distributed with a mean of
(8) 30% and a standard deviation of 4%. Hence, the required
The probability of departure/arrival time, t, where t is power to charge EV h battery, Phreq , can be determined
between 1 and 24, is represented by P(t). a1 and a2 from the stochastic SOC as it is assumed that the car should
represents the scaling factors, b1 and b2 represents the be completely charged on departure.
mean, and c1 and c2 represents the standard deviations r Uncertainty in modeling HPs: The operation of HPs relies
of the PDF. Table II demonstrates the properties of the on the ambient temperature, Tte . In the event of colder
arrival and departure time density functions. To evaluate the weather in winter (or higher temperatures in summer), the
hp
goodness of the fit, the Root Mean Squared Error (RMSE) heat pump would require more electrical power, Ph,t , to

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
3574 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

maintain the indoor temperatures, Th,t


i
, within the desired solving (8) deterministically. λ is the Lagrange multiplier of
min max
range, [Th , Th ]. However, there is uncertainty associ- the active constraints. For each aggregator, i, the associated
ated with the forecasted outdoor temperature, ΔT e , which Lagrangian function, L(x, λ), is
impacts the day-ahead load scheduling. Uncertainty in the L(x, λ) = xT β̄x + g T x + m − λ(Ax − b) (14)
ambient temperature, ΔT e , hence is modelled as a normal
density function with a standard deviation, σT e , of 1 °C. Minimizing the Lagrangian function over x results in,
x = 0.5β̄ −1 (AT λ − g) (15)
B. Sensitivity Analysis
Using the Lagrangian’s function’s differentiation over the La-
The robustness of the proposed congestion management
grangian multipliers, the results obtained are substituted in (15).
scheme is impacted by the uncertainty in scheduling the flexible
Optimal variable x∗ is hence defined as,
load, U , identified in the previous subsection. If consumers
deviate from their load requirements, the multi-objective Pareto x∗ = β̄ −1 AT (Aβ̄ −1 AT )−1 (b + 0.5Aβ̄ −1 g) − 0.5β̄ −1 g (16)
weight, α, determined in Section II, might not manage existing
The change in the optimal solution, x∗ , as a result of changes in
network congestion. To ensure the robustness of the proposed
the uncertain variables, U , reflected by changes in b, is,
multi-objective congestion management scheme, the uncertainty
propagation theorem [3] is adopted. Based on [3], the uncer- ∂x
= β̄ −1 AT (Aβ̄ −1 AT )−1 (17)
tainty in power flowing in a network branch, ΔFtb , is a func- ∂b
tion of uncertainties in EV and HP modeling parameters, i.e. Hence, to determine the impact on congestion in the distribution
U = {Δtar , Δtdep , ΔP req , and ΔT e }. Hence, propagation of network, change in the power flow, ΔFt , in the network at the
uncertainty from the input variables to the line loading can be time, t, as a result of a change in EV and HP loading due to their
determined as follows, uncertainties is determined as follows,
      
hp 2 ev )2  ∂x
ΔFtb = DEi (ΔPi,t ) + (ΔPi,t (9) ΔFt = DEi ΔPi,t + ΔPi,t =
ev hp
DEi Δb
b∈Nb i∈Nd i i
∂b
where, (18)
 2  ev 2 where D is the power transfer distribution factor and Ei is the
ev 2
∂Pi,t
ev
∂Pi,t
(ΔPi,t ) = Δtar + Δtdep customer to bus mapping matrix for aggregator i.
∂tar ∂tdep
  2
∂Pi,tev
C. Robustness Enhancement of Proposed Scheme Through
+ ΔP req (10) Uncertainty Propagation
∂P req

and, As a result of the uncertainty in scheduling consumers


 hp
2 load, the deterministic scheme explained in Section II may
hp 2 ∂Pi,t not clear congestions in the real-time market using the deter-
(ΔPi,t ) = ΔT e (11)
∂T e mined the Pareto weight, α. The modeled congestion man-
e
agement scheme identifies four sources of uncertainty, U =
For the rest of the paper, ζ ar , ζ dep , ζ P , and ζ T represents {Δtar , Δtdep , ΔP req , and ΔT e }. Using (17-18), the sensitiv-
hp
ity of the consumers’ load, and hence the line loading ΔFt ,
ev ev ev ∂Pi,t
∂Pi,t ∂Pi,t ∂Pi,t
∂tar , ∂tdep , , and
∂P req respectively. Hence, to ensure
∂T e
the effectiveness of the proposed model the sensitivity of the is determined for changes in each of the uncertain variables
optimal flexible loads, and consequently line loading, as a result independently. Respective rows of vector b corresponding to
of uncertain loading is studied. As each aggregator, i, is re- changes in EV power as a result of changes in arrival time,
sponsible for its own flexible load scheduling, the sensitivity of departure time, and EV charge required results in ζ ar , ζ dep ,
multi-objective congestion management problem (1) variables, and ζ P as sensitivity coefficients. Similarly, rows of vector b
represented by xit , where, corresponding to changes in HP power as a result of changes
e
in ambient temperature result in ζ T as sensitivity coefficient.
hp T
xit = [Pi,t
ev
Pi,t ] , ∀i ∈ Na , ∀t ∈ NT (12) Based on (9), the change in power flowing in the distribution
network, ΔFtb , depends on the sensitivity of the EV and HP
to the uncertainty of EV arrival time, departure time, SOC, and
power, ζ U , to changes in the uncertainty variables, U . The robust
ambient temperature is studied. Hence, rewriting (1) in standard
metric, R, is hence determined as the probability, P, of the
form for each aggregator, i, in the network (hence subscript i
change in line loading, ΔFtb , as a result of load scheduling
has been dropped),
uncertainties, U , as follows,
min xT β̄x + g T x + m  
x 
R=P ΔFt ≤ δFmax
b
s.t. Ax = b, (λ) (13) b∈Nb
 
where β̄ = (1 − α)β, constant m = αPtnf 1,
and gt = [α1 + 
(1 − α)(pt 1)T ]. A and b are the coefficients of the active con- =P DEi ζiU ΔU ≤ δFmax (19)
straints related to EV and HP modeling (1b-1f) determined after i∈Nd

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ROBUST MULTI-OBJECTIVE CONGESTION MANAGEMENT IN DISTRIBUTION NETWORK 3575

Algorithm 1: Robust Enhanced Multi-objective Congestion


Management Scheme.
Input: Consumers’ comfortable internal temperature range: Thmin
and Thmax , required EV charging power, Phev , EV availability
for charging, ah,t , and PDFs of uncertain parameters,
U = {Δtar , Δtdep , ΔP req , and ΔT e }.
Output: Multi-objective function Pareto weight, α, and optimal

load schedule, Pi,t , for aggregator, i.
0
Initialize: Set α = 0 for iteration k = 0.
1) while (αk+1 − αk > τ ) do
a) Using αk and mean of the random variables, U , solve
multi-objective optimization (1) for i ∈ NA to determine

consumers’ optimal load, Pi,t .
Fig. 3. Day-ahead electricity price. b) Each aggregator forwards their total optimal consumer

loads, Pi,t , ∀i ∈ NA and ∀t ∈ NT , to the DNO.
c) The network operator performs power flow analysis to
To measure the robustness of the proposed scheme, the robust determine the power flow in each branch in the network,
metric, R (19), is determined for the result of the Pareto optimal Ftb .
obtained in the deterministic case following Fig. 1. To enhance d) The DNO checks for congestion and determines the extent
the robustness of the proposed model, if R is greater than η, of congestion, Ctb , with respect to a branch’s maximum
limit.
then the Pareto is updated by a step size, γ, proportional to
Ctb = Ftb − Fmaxb
, ∀b ∈ Nb
the sensitivity of the line loading to the uncertainty in the load e) The DNO updates the Pareto weight, αk , such that,
schedule. Algorithm 1 illustrates the steps involved to determine αk+1 = αk + Dθ min(b, Ftb − Fmax b
)
the optimal Pareto weight, α, taking into consideration the f) if (α k+1
< 0) then
uncertainty of the consumers’ load. On executing Algorithm αk+1 = 0

1, the probability of the resulting optimal solution, Pi,t , causing 2) Using αk , solve (17-18) to determine the sensitivity, ζ U , of
congestion in the network due to randomness of EV arrival time the line loading to changes in the random variables, U .
or departure time, EV power required, and ambient temperature, 3) Determine the robustness, R, of the solution (19) to flexible
should be below η. load uncertainties, U .
4) whileR > η do
a) Set αk+1 = αk + γζiU , where γ is a step size. Since αk
IV. CASE STUDY, RESULTS AND DISCUSSION obtained from step 1 clears congestion in the network,
The IEEE 33 bus system [35], is adopted as a case study in αk+1 ≥ αk also clears congestion as the weight of
this paper. The distribution network consists of 32 residential flattening the load curve increases.
b) Using the updated Pareto weight, αk+1 , the multi-objective
load buses representing both flexible and non-flexible loads.
function (1) is solved for each aggregator i ∈ NA to
The default loading of the IEEE 33 bus system represents the ∗
determine consumers’ optimal load, Pi,t .
non-flexible base load profiles of the load buses. The flexible c) Repeat steps 2 and 3.
residential loads consist of EVs and HPs [18]. The case study
involves a total of 32 aggregators, one aggregator representing
consumers of each load bus. GAMS [36] and MATLAB R2021a
were used for executing the optimizations.
Solving the normalized multi-objective function (1), Fig. 5
illustrates the effect of varying the multi-objective function
A. Deterministic Multi-Objective Flexible Load Scheduling weights, α, on the power flowing, Ft , in the main feeder of the
Using the case study explained above, Fig. 3 illustrates the IEEE 33 bus system. An α of 0 represents price-based flexible
forecasted day-ahead price of electricity, pt . To develop the load scheduling. Hence, as the aggregator schedules the load by
Pareto front, the multi-objective function weight, α, is varied minimizing consumption costs, the load is concentrated at times
in (1) from 0 to 1, in increments of 0.01. The resulting Pareto in which the cost of electricity is low leading to congestion at
optimal front of the multi-objective optimization is illustrated times t = 20 − 23. However, as α increases, the dominance of
in Fig. 4. Neither point on the Pareto front is considered better the valley-filling load scheduling is observed to increase and the
than the other except in terms of the higher-level objective of a curve is flattened. An α of 1 represents valley filling flexible
user. An α closer to 0 increases the importance of price-based load scheduling. The impact of changing α is seen in the PAR of
scheduling resulting in a more economical consumer load sched- the load buses. An α of 0 results in a mean PAR of 1.3936 while
ule. This is illustrated in Fig. 4 as a decrease in the price objective an α of 1 returns a mean PAR of 1.0011. Hence, the PAR keeps
function at the cost of an increase in the valley filling objective improving as α increases. Fig. 6 illustrates the trade-off between
value. Similarly, an α closer to 1 increases the importance of electricity price and that of feeder congestion level as the weight,
valley-filling-based scheduling resulting in a more flattened load α, of the objective function (1) varies. The importance of valley
curve. filling load scheduling increases as the weight, α, increases.

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
3576 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

Fig. 4. Pareto optimal front, ideal and nadir points.


Fig. 7. Comparison using proposed multi-objective load scheduling versus
DDT congestion management scheme.

scheme [9]. Comparison is executed in terms of electricity price


paid by consumers and Peak-To-Average (PAR) of the power
flowing in the main feeder as follows.
r Electricity Price: DDT congestion management results in
scheduling the load such that the electricity cost of the
consumers is minimized. Hence, as illustrated in Fig. 7,
flexible loads are scheduled mainly in the early and late
hours of the night when the price of electricity is the lowest.
If the line was not congested, consumers had to pay a
total of 4853.8$, based on electricity prices illustrated in
Fig. 3. However, as a result of congestion at t = 21 − 22,
Fig. 5. Effect of varying multi-objective function weight, α, on power flowing
in main feeder.
a congestion tariff of 0.063, and 0.075$/kW was, hence,
imposed on the consumers at that time to alleviate the
congestion. This tariff is eventually paid by the consumers
as a total increase in electricity cost by 21.6%. On the
other hand, no tariffs are imposed using the proposed multi-
objective congestion management scheme. The proposed
algorithm aims to improve the welfare of the consumer in
terms of minimizing electricity price while maintaining
indoor temperatures within consumers preferences and
charging their EVs based on their availability. Moreover,
the proposed scheme improves the welfare of the network
operator, in terms of flattening the curve and reducing the
PAR. Consumers following the proposed multi-objective
optimization would end up paying a total of 4964.6, which
is 15.9% lower than price-based optimization.
r PAR: As DDT congestion management scheme aims to
Fig. 6. Percentage congestion in the network versus electricity cost as a
function of Pareto weight.
minimize electricity costs, consumer loads end up being
concentrated at times in which the electricity price is low
causing congestion. To alleviate congestion, tariffs need
This results in a flexible load schedule that is more costly com- to be imposed to clear the peak, as illustrated in Fig. 7.
pared to that of lower weights where price-based scheduling is The PAR of the resulting power flowing in the feeder is
dominant. However, a higher weight indicates lower congestion observed to be 1.362 which is caused as a result of the
levels. peak in demand at t = 21 − 22 causing congestion. On
1) Comparison of Results with Existing Distributed Dynamic the other hand, the proposed multi-objective optimization
Tariff Congestion Management Scheme and Discussion: To resulted in a congestion-free network with a PAR of 1.152
evaluate the performance of the proposed multi-objective con- which is sufficient enough to have a congestion-free feeder
gestion management scheme, the results obtained on simulating at minimum electricity price.
the IEEE 33 bus case study are compared with that of the exist- From the above discussion and comparison, the superiority of
ing Distributed Dynamic Tariff (DDT) congestion management the proposed multi-objective flexible load scheduling to manage

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ROBUST MULTI-OBJECTIVE CONGESTION MANAGEMENT IN DISTRIBUTION NETWORK 3577

Fig. 9. Comparison between the robust and deterministic schemes in terms of


Fig. 8. Sensitivity of the IEEE 33 bus system main feeder line loading to power flowing in the main feeder.
changes in EV power required.

in the network by 1% of the feeder limit was determined to


congestion in the network with minimal impact on electricity be less than 0.15%. Fig. 9 illustrates the difference in power
prices has been demonstrated. flowing in the main feeder as a result of the change in the Pareto
weight. The deterministic scheme having a lower Pareto weight
B. Robustness of the Proposed Multi-Objective Congestion of α = 0.4328 did not satisfy the robustness constraint (19). As
Management Scheme observed at times t = 21 − 22, power flowing in the main feeder
As a result of the uncertainties in the load scheduling de- is just below the feeder limits. Any slight increase in consumer
sign parameters, U , as described in Section III, the IEEE demand at those times would cause congestion in the network.
33 bus system is adopted to study the sensitivity of the Hence, the robust scheme was executed and it resulted in having
power flowing in its main feeder to uncertainties in the load, a larger Pareto which ensures that the power flowing in the main
ζU . Four uncertain parameters are considered in the design, feeder is below its congestion limit, as illustrated in Fig. 9.
where U = {Δtar , Δtdep , ΔP req , and ΔTe }. Using the results
obtained from the deterministic case, the sensitivity, ζU = V. CONCLUSION
{ζ ar , ζ dep , ζ P , ζ Te } = { ∂t
∂Ft
ar
, ∂t∂Ft
, ∂Ft , ∂T
dep ∂Preq
∂Ft
e
}, of the power In this paper, a multi-objective flexible load scheduling al-
flowing in the main feeder to the random design variables was gorithm was proposed to manage congestions in the distribu-
determined by executing (17-18). For example, as a result of tion network. The proposed model combines the advantages of
uncertainty in power required by an EV, the sensitivity factor, price-based and valley-filling demand-side management (DSM)
ζPreq , which represents the change in power flowing in the main and based on the congestion level in the network branches, a
feeder of the IEEE 33 bus system as a result of 1 kWh change weight is assigned to each objective. The higher the congestion,
in the consumer loads due to more EV charge required, was the higher is the weight given to the valley-filling portion of
observed to complement the power flowing in the main feeder. the multi-objective problem compared to that of the price-based
This is illustrated in Fig. 8. When the main feeder is lightly objective function. This results in determining an optimal Pareto
loaded, the sensitivity is observed to be higher compared to peak from the resulting Pareto front that is sufficient enough to relieve
times. This results in flattening the curve and ensuring that no congestion at the minimum electricity price. Moreover, the
congestion occurs. stochastic nature of flexible loads has also been incorporated
Using the sensitivities and the PDFs of the four uncertain into the congestion management scheme. A sensitivity analysis
parameters, U , described in Section III-A, Algorithm 1 was was executed to determine the propagation of uncertainty in
executed. Using the deterministic model, a Pareto weight, α = the proposed model ensuring that the probability of overloading
0.4328, resolves the congestion in the distribution network as the network is less than a specific level given the uncertainty
illustrated in Fig. 9. However, on evaluating the robustness (19), involved in scheduling the flexible loads. The proposed con-
R, of the resulting distribution due to the uncertainties in the gestion management scheme was compared with the existing
load, U , given γ = 1%, change in the line loading, ΔF b , in the Distributed Dynamic Tariff price-based congestion management
main feeder was determined to be 0.1207 MW. This violates the scheme to demonstrate the superiority of the technique. The
robustness constraint and the Pareto had to be updated following proposed multi-objective model resulted in the most economical
Algorithm 1 until a change in the line loading, ΔF b , in the solution in the presence or absence of congestion and an im-
main feeder was determined to be lesser than γ = 1%. A Pareto proved peak to average ratio compared to that of the price-based
weight, α = 0.4928, resulted in a change in line loading of method. Therefore, combining the advantages of price-based
the main feeder, ΔF b , equal to 0.0997 MW. Hence, given the and valley-filling DSM into one optimization problem resulted
normal density function of the random variables, the probability in a model that is economical for the consumer and also relieves
of the proposed consumers load profile to cause congestion congestion.

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
3578 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

REFERENCES [24] S. Deb, A. K. Goswami, R. L. Chetri, and R. Roy, “Congestion man-


agement considering plug-in electric vehicle charging coordination in
[1] T. Vo, A. Haque, P. Nguyen, I. Kamphuis, M. Eijgelaar, and I. Bouwman, distribution system,” in Proc. 4th Biennial Int. Conf. Nascent Technol.
“A study of congestion management in smart distribution networks based Eng., 2021, pp. 1–5.
on demand flexibility,” in Proc. IEEE Manchester PowerTech, 2017, [25] S. Huang, Q. Wu, Z. Liu, and H. Zhao, “Sensitivity analysis of dy-
pp. 1–6. namic tariff method for congestion management in distribution net-
[2] N. Damsgaard, “Study on the effective integration of distributed energy works,” in Proc. IEEE Power Energy Soc. Gen. Meeting, 2015,
resources for providing flexibility to the electricity system,” Rep. Eur. pp. 1–6.
Commission, 2015. [26] A. Ramos, “Aggregators and retailers in electricity markets: Roles and
[3] S. Huang, Q. Wu, L. Cheng, Z. Liu, and H. Zhao, “Uncertainty manage- conflicts,” in Deliverable 8.5 Energy Storage Deployment Handbook.
ment of dynamic tariff method for congestion management in distribu- STORY H2020 Project, 2020, ch. 3.
tion networks,” IEEE Trans. Power Syst., vol. 31, no. 6, pp. 4340–4347, [27] “Day-ahead market high level design,” Independent Electricity System
Nov. 2016. Operator, 2019.
[4] R. Verzijlbergh, L. Vries, and Z. Lukszo, “Renewable energy sources and [28] T. Abergel, “Heat pumps,” Int. Energy Agency (IEA), 2021.
responsive demand. Do we need congestion management in the distribu- [29] A. D. Belegundu and T. R. Chandrupatla, Optimization Concepts and
tion grid,” IEEE Trans. Power Syst., vol. 29, pp. 2119–2128, Sep. 2014. Applications in Engineering, 2nd ed. Cambridge, U.K.: Cambridge Univ.
[5] A. Haque, P. Nguyen, F. Bliek, and J. Slootweg, “Demand response for real- Press, 2011.
time congestion management incorporating dynamic thermal overloading [30] L. He, H. Ishibuchi, A. Trivedi, H. Wang, Y. Nan, and D. Srinivasan,
cost,” Sustain. Energy, Grids Netw., vol. 10, pp. 65–74, 2017. “A survey of normalization methods in multiobjective evolutionary algo-
[6] A. N. M. M. Haque, P. H. Nguyen, W. L. Kling, and F. W. Bliek, rithms,” IEEE Trans. Evol. Computation, vol. 25, no. 6, pp. 1028–1048,
“Congestion management in smart distribution network,” in Proc. 49th Dec. 2021.
Int. Universities Power Eng. Conf., 2014, pp. 1–6. [31] A. Ghane-Kanafi and E. Khorram, “A new scalarization method for finding
[7] A. A. Sallam and O. P. Malik, Demand-Side Management and Energy the efficient frontier in non-convex multi-objective problems,” Appl. Math.
Efficiency. Piscataway, NJ, USA: IEEE Press, 2019, pp. 429–463. Modelling, vol. 39, no. 23, pp. 7483–7498, 2015.
[8] S. Huang, Q. Wu, L. Cheng, and Z. Liu, “Optimal reconfiguration-based [32] M. R. Hamouda, M. E. Nassar, and M. M. A. Salama, “A novel energy
dynamic tariff for congestion management and line loss reduction in dis- trading framework using adapted blockchain technology,” IEEE Trans.
tribution networks,” IEEE Trans. Smart Grid, vol. 7, no. 3, pp. 1295–1303, Smart Grid, vol. 12, no. 3, pp. 2165–2175, May 2021.
May 2016. [33] Federal Highway Admin., “2017 national household travel survey,” U. S.
[9] S. Huang, Q. Wu, H. Zhao, and C. Li, “Distributed optimization based Dept. Transp., Washington, DC, USA, 2017.
dynamic tariff for congestion management in distribution networks,” IEEE [34] Joint Committee for Guides in Metrology, “JCGM 100: Evaluation of mea-
Trans. Smart Grid, vol. 10, no. 1, pp. 184–192, Jan. 2019. surement data: Guide to the expression of uncertainty in measurement,”
[10] S. Huang and Q. Wu, “Dynamic subsidy method for congestion manage- JCGM, 2008.
ment in distribution networks,” IEEE Trans. Smart Grid, vol. 9, no. 3, [35] M. Baran and F. Wu, “Network reconfiguration in distribution systems for
pp. 2140–2151, May 2018. loss reduction and load balancing,” IEEE Trans. Power Del., vol. 4, no. 2,
[11] M. R. Sarker, M. A. Ortega-Vazquez, and D. S. Kirschen, “Optimal pp. 1401–1407, Apr. 1989.
coordination and scheduling of demand response via monetary incentives,” [36] GAMS Development Corporation, “General algebraic modeling system
IEEE Trans. Smart Grid, vol. 6, no. 3, pp. 1341–1352, May 2015. (GAMS) release 24.2.1,” Washington, DC, USA, 2013.
[12] Z. Wang and S. Wang, “Grid power peak shaving and valley filling
using vehicle-to-grid systems,” IEEE Trans. Power Del., vol. 28, no. 3,
pp. 1822–1829, Jul. 2013.
[13] M. Shinwari, A. Youssef, and W. Hamouda, “A water-filling based schedul-
ing algorithm for the smart grid,” IEEE Trans. Smart Grid, vol. 3, no. 2,
pp. 710–719, Jun. 2012.
[14] K. Zhou and L. Cai, “A dynamic water-filling method for real-time hvac Omniyah Gul M Khan (Graduate Student Mem-
load control based on model predictive control,” IEEE Trans. Power Syst., ber, IEEE) received the B.Sc. and M.Sc. degrees in
vol. 30, no. 3, pp. 1405–1414, May 2015. electrical engineering from the American University
[15] Y. Mou, H. Xing, Z. Lin, and M. Fu, “Decentralized optimal demand-side of Sharjah, Sharjah, U.A.E, in 2008 and 2010, re-
management for phev charging in a smart grid,” IEEE Trans. Smart Grid, spectively. She is currently working toward the Ph.D.
vol. 6, no. 2, pp. 726–736, Mar. 2015. degree from the Electrical and Computer Engineering
[16] N. Chen, C. W. Tan, and T. Q. S. Quek, “Electric vehicle charging in smart Department, University of Waterloo, Waterloo, ON,
grid: Optimality and valley-filling algorithms,” IEEE J. Sel. Topics Signal Canada. Her research interests include demand side
Process., vol. 8, no. 6, pp. 1073–1083, Dec. 2014. management, cyber security of distribution systems,
[17] M. Liu, P. Phanivong, and D. Callaway, “Electric vehicle charging control and artificial intelligence.
in residential distribution network: A decentralized event-driven realiza-
tion,” in Proc. IEEE Conf. Decis. Control, 2017, pp. 214–219.
[18] W. Liu, Q. Wu, F. Wen, and J. Øtergaard, “Day-ahead congestion man-
agement in distribution systems through household demand response and
distribution congestion prices,” IEEE Trans. Smart Grid, vol. 5, no. 6,
pp. 2739–2747, Nov. 2014.
[19] S. Huang, Q. Wu, S. S. Oren, R. Li, and Z. Liu, “Distribution locational
marginal pricing through quadratic programming for congestion manage- Amr Youssef (Senior Member, IEEE) received the
ment in distribution networks,” IEEE Trans. Power Syst., vol. 30, no. 4, B.Sc. and M.Sc. degrees from Cairo University,
pp. 2170–2178, Jul. 2015. Cairo, Egypt, in 1990 and 1993, respectively, and
[20] S. Hanif, H. B. Gooi, T. Massier, T. Hamacher, and T. Reindl, “Distributed the Ph.D. degree from Queens University, Kingston,
congestion management of distribution grids under robust flexible build- ON, Canada, in 1997. He was with Nortel Networks,
ings operations,” IEEE Trans. Power Syst., vol. 32, no. 6, pp. 4600–4613, Ottawa, ON, Canada, the Center for Applied Crypto-
2017. graphic Research, University of Waterloo, IBM, and
[21] R. Ciavarella, M. Somma, G. Graditi, and M. Valenti, “Congestion man- also with Cairo University. He is currently a Professor
agement in distribution grid networks through active power control of with the Concordia Institute for Information Systems
flexible distributed energy resources,” in Proc. PowerTech, 2019, pp. 1–6. Engineering, Concordia University, Montreal, QC,
[22] M. Attar, A. Supponen, and S. Repo, “Distribution system congestion Canada. He has authored more than 240 referred
management through local flexibility market,” in Proc. CIRED Berlin journal and conference publications in areas related to his research interests. His
Workshop, 2020, pp. 769–772. research interests include information security, cyber-physical systems security,
[23] H. Khomami, R. Fonteijn, and D. Geelen, “Flexibility market design for and wireless communications. Dr. Youssef was with about 75 technical program
congestion management in smart distribution grids: The dutch demon- committees of cryptography and data security conferences. He was also the
stration of the interflex project,” in Proc. IEEE PES Innov. Smart Grid Co/Chair of Africacrypt 2010, Africacrypt 2020, conference selected areas in
Technol., 2020, pp. 1191–1195. cryptography, SAC 2014, SAC 2006, and SAC 2001.

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.
KHAN et al.: ROBUST MULTI-OBJECTIVE CONGESTION MANAGEMENT IN DISTRIBUTION NETWORK 3579

Magdy Salama (Life Fellow, IEEE) received the Ehab F. El-Saadany (Fellow, IEEE) was born in
B.Sc. and M.Sc. degrees in electrical engineering Cairo, Egypt, in 1964. He received the B.Sc. and
from Cairo University, Cairo, Egypt, in 1971 and M.Sc. degrees in electrical engineering from Ain
1973 respectively, and the Ph.D. degree from the Shams University, Cairo, Egypt, in 1986 and 1990,
University of Waterloo, Waterloo, ON, Canada, in respectively, and the Ph.D. degree in electrical engi-
1977. He is currently a Professor with the Department neering from the University of Waterloo, Waterloo
of Electrical and Computer Engineering, University ON, Canada, in 1998. He was a Professor with ECE
of Waterloo. He has consulted widely with govern- Department, University of Waterloo, till 2019. He is
mental agencies and the electrical industry. His re- currently a Professor with Electrical and Computer
search interests include the operation and control of Engineering Department and the Director of the Ad-
distribution systems, smart microgrids, power quality vanced Power and Energy Research Center, Khalifa
monitoring and mitigation, asset management, and electromagnetics. He is a University, Abu Dhabi, UAE. His research interests include smart-grid operation
Registered Professional Engineer with the Province of Ontario, Canada. and control, microgrids, self healing, power quality, distributed generation,
power electronics interfacing, and mechatronics. He is also a Registered Profes-
sional Engineer with the Province of Ontario, Canada.

Authorized licensed use limited to: Raghu Engineering College - Visakhapatnam. Downloaded on July 11,2023 at 09:58:38 UTC from IEEE Xplore. Restrictions apply.

You might also like