This article has been accepted for publication in IEEE Transactions on Smart Grid.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 1

Sequential Clearing of Network-aware Local Energy

and Flexibility Markets in Community-based Grids
Saber Talari , Member, IEEE, Sascha Birk , Student Member, IEEE, Wolfgang Ketter , and Thorsten
Schneiders

Abstract—In this paper, network-aware clearing algorithms for B Set of buses in the network (LFM)./Set of blocks in
local energy markets (LEMs) and local flexibility markets (LFM) Distflow linearization (LEM).
are proposed to be sequentially run and coordinate assets and Di Set of consumers at bus i.
flexible resources of energy communities (ECs) in distribution
networks. In the proposed LEM clearing algorithm, EC managers Fi Set of flexibility options at bus i.
run a two-stage stochastic programming while considering ran- G Set of all generators.
dom events by scenario generation and network constraints using L Set of branches (ij) in the network with i as the
linearized DistFlow. As one of outcomes, maximum available up- ”from” bus.
and down-regulations provided by ECs are estimated in LEM and Ni Set of neighboring buses to bus i.
communicated to LFM. In the distributed LFM clearing algo-
rithm, an iterative auction is designed using a dual-decomposition Parameters and constants
technique (Augmented Lagrangian) which is solved by consensus ∆θij Upper limit for voltage angle difference between buses
alternating direction method of multipliers. The LFM algorithm i and j.
efficiently dispatches the flexibility provided by ECs in operating ∆θij Lower limit for voltage angle difference between
time while considering flexibility local marginal price as pricing buses i and j.
method. Network constraints are included in the algorithm with
an AC distribution optimal power flow for dynamic network ϵabs Absolute feasibility tolerance.
topology in which branches and buses are decomposed to solve ϵrel Relative feasibility tolerance.
the problem in distributed fashion. The designed LFM algorithm I l(ij) Maximum possible current to be flowed in line l
can respond to exogenous and endogenous signals for flexibility connect bus i to j.
requests. The simulation results in a test case display effectiveness ϕpf Power factor.
of two proposed LEM and LFM algorithms for an efficient
provision of flexibility. πω Probability of scenarios generated by ECMs.
ρ Penalty parameter in CADMM.
Index Terms—Dual-decomposition, energy communities, flex- pd Demand of consumer d ∈ Di .
ibility provision, iterative auction, market design, stochastic
programming. Variables
λ Vector of dual variables for augmented Lagrangian.
N OMENCLATURE θi Voltage angle at bus i.
p
Indices and Superscripts fij Active power flexibility flow from bus i to j in
ω Scenario generated by ECMs for real time. distribution network.
q
b Block index for linearization of DistFlow. fij Reactive power flexibility flow from bus i to j in
ch Charging energy storage. distribution network.
d Demand in a EC. Iij Complex current phasor from bus i to j in ec.
dc Discharging energy storage. iij Squared absolute value of current phasor from bus i
dg Distributed generation. to j in ec.
dr Demand response. Pj Net active power at bus j of ec.
ec Energy community. Pjk Active power flow from bus j to k in ec.
ess Energy storage system. Qj Net reactive power at bus j of ec.
i, (j) Index for buses. Vi Complex phasor voltage at bus i.
k Index for number of iteration in LFM algorithm. Also vi Squared absolute value of voltage phasor at bus i.
index for bus in (1)-(2).
pcc Point of common coupling. I. I NTRODUCTION
pv Photovoltaic. A. Motivation
t
wg
Sets
Time step.
Wind generation. E MERGING technologies in distribution energy networks
such as distributed energy resources (DERs), smart me-
ters and controllable loads have brought significant potentials
L̂ Set of branches (ji) in the network with j as the ”to” to provide flexibility in the power system as whole [1].
bus. For instance, transmission system operators (TSOs) can call
B Set of buses in LEM. flexibility from distribution systems for security issues in
Bl Set of buses connected by line l in LEM. transmission networks [2]. This helps to avoid major blackouts
T Set of time-steps. such as the Texas power rotating outages in 2021 [3]. Thus, a
Ω Set of scenarios in LEM. market mechanism to trade this type of flexibility is required

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 2

to coordinate flexibility resources efficiently and avoid any problems iteratively using ADMM, though the stochasticity
problems in distribution networks due to providing flexibility of load behavior is trained in the market using reinforcement
to the transmission network [4]. learning (RL). However, potential flexibility of end-users was
Having more installed photovoltaic (PV), electrical storage not defined in this work.
systems (ESS) and using controllable loads among customers, In [13], a microgrid (MG) operator manages the end-
raises distribution system operators’ (DSO) concern on grid users’ energy using two-stage programming while the available
congestion and maintaining the voltage of the network within flexible resources (i.e. ESS and DR) are also scheduled using
the admissible threshold. Thus, flexibility for DSO at distri- scenarios represented as random events in real-time. Various
bution network is another major driver of a flexibility market other studies such as [14], [15] propose flexibility evaluation
[5], [6]. One of the obstacles for most of DSOs is having methods to be applied in an LFM, though there is no network
no direct control of distributed energy resources (DERs) due constraint in those studies. The MG operator in [16] considers
to unbundling laws of market deregulation [7]. Thus, the the network restrictions to manage the end-users by a social
concept of customers aggregation within a community as welfare maximization. However, in this work, there is no
energy communities (ECs) becomes necessary. According to market mechanism to coordinate flexible resources.
Directive (EU) 2019/944, Article 16(2), ECs can own a part There are several works in which flexibility provision in
of a distribution network and directly manage it with(out) distribution networks follows no market-based approaches,
cooperation of a relevant DSO. Moreover, Article 16(3) states rather financial incentives [17] or technical methods such as
that ECs might be balance responsible parties to compensate grid reconfiguration [18] are the main basis. In the latter, the
any imbalances and provide flexibility, if needed [8]. Hence, transmission network is relieved by selecting an operational
network-aware management of communities is lawfully possi- scheme for the connected distribution grid and curtailing its
ble which plays a vital role to coordinate the DERs and more loads while also reconfiguring the network topology. EU clean
importantly to ensure distribution network integrity [9]. This energy package [19] states, that flexibility provision should be
can be addressed via a local energy market (LEM) in which market-based and accessible for all end-users.
customers can actively participate in and freely decide to buy In [20], prosumers with ESS are aggregated in EU NVADE
and sell electricity internally (within EC) or from upstream project to distribute their flexible potential, when requested. To
network aligned with Directive (EU) 2019/944. this end, aggregators and prosumers solve their own problem
As the Article 32, Directive (EU) 2019/944 states, DSOs in a distributed fashion using a Proximal Jacobian (PJ) ADMM
should provide market-based incentives for flexibility procure- to reach an equilibrium and meet the requested flexibility
ment, therefore, the challenge is to design the appropriate by DSO. A joint flexibility market between DSO and TSO
market mechanism to encourage flexible resources for an is proposed in [21] (as an EU project named CoordiNet)
efficient participation. To this end, the precise estimation of based on a cooperative game between them and some pric-
available flexibility from communities is necessary. To this ing mechanisms e.g. pay-as-bid and nodal pricing are also
end, LEM schedules EC to define the set-points of the local evaluated. Within the joint market, flexibility resources from
generation and demand as well as the flexibility capacity in different DSOs can be readily used by TSO and the market is
advance. Knowing this capacity, the operator of a Local Flex- cleared by an optimization problem formalized by Lagrangian
ibility Market (LFM) can dispatch the resources based on the Relaxation (LR). Researchers in Enera project, which is one of
flexibility demand from upstream network or the requirements the pioneering LFM platforms in Germany [22], apply a merit-
within the distribution network in operating time. Meanwhile, order curve as clearing algorithm for an LFM run by DSO to
the value of flexibility varies along market participants based provide flexibility to TSO and propose pay-as-bid as pricing
on the technology, the location and privacy issues. Thus, algorithm. A blockchain-based LFM is proposed in [23] in
suitable pricing mechanism and clearing algorithm in LFM which smart contracts would be made between sellers and
play an important role. buyers of flexibility and a greedy heuristic method is employed
B. Literature Review to match buyers and sellers for the EU project called Bright.
Recently, to manage energy in a community level (i.e. a Nevertheless, none of above-mentioned research works took
particular geographical area), LEM has been widely proposed care of network constraints.
which is able to coordinate end-users in a decentralized Researchers in [24] develop a central clearing algorithm
market-based fashion [10]. LEMs can provide energy and for LFM run by DSOs using a two-stage stochatic model
capacity services. For instance, authors in [11] employ Nash while considering network constraints with a linearized AC
bargaining to model energy sharing payments between PV OPF. Reference [25] proposes a continuous flexibility mar-
prosumers and community ESS in an EC. An EC manager ket for EU FlexGrid project considering network constraints
(ECM) in [9] operates an LEM in a way that stochastic nature while using asymmetric block bids for clearing the market
of available PV prosumers and ESS are modeled in a two-stage with a first-come, first-served principle. Within the Swiss
programming. The model is decomposed to run the market project DiGRiFlex, a non-linear distributional robust chance-
in decentralized way using alternating direction method of constrained optimization is applied in [26] to dispatch the
multipliers (ADMM). Thus, the ECM and prosumers solve flexible resources in an active distribution network - ADN
their problem iteratively to reach an equilibrium. In [12], (e.g. MG) while stochastic variables are realized and network
within the EU project (FlexGrid), ECM also decomposes constraints are considered. Flexible resources are aggregated
the problem and let the smart buildings solve their own by clustered queuing systems in [27] and DSO optimizes the

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 3

TABLE I: Taxonomy table on existing local markets

Ref.
LEM LFM
Pricing
Network are in charge of running LEM in their own network1 through
Clearing Clearing constraints
[11] ADMM — Nash bargaining — an adaptive two-stage stochastic programming. The optimum
[12] ADMM
Finite Markov
— Iterative auction
Mid-market
— energy and available flexibility are decided in the first stage
[15] — —
Game rate as day-ahead decisions, while random events’ realization in
Max. social
[16]
welfare
— Nash bargaining AC OPF the second stage makes the real-time decisions on expected
-Not market-based
[18] —
(grid reconfig.)
— Distflow security actions. Indeed, random events are the stochasticity
[20] — -PJ-ADMM Iterative auction — of renewable energy sources (RESs) modeled by scenario
-Lagrangian pay-as-bid &
[21] — —
Relaxation LMP generation using Monte-Carlo simulation (MCS) method. A
[22] — -Merit order Pay-as-bid —
[23] —
-Matching by
Smart contracts —
linearized DistFlow [31] is also employed to make sure the
greedy heuristic
-Central two-stage Pool-based Linearied network constraints are satisfied. As a result, active and
[24] —
stochastic model pricing AC OPF
-Max. social Incentives by
reactive power exchange and maximum available flexibility
[27] — Branch Flow
welfare DSO at point of common coupling (PCC) of ECs are determined
-Central Continuous First-come,
[25] — DC OPF
clearing fist-served day ahead.
-Central Two-sided
[26] —
chance constraint
Set by DSO AR-OPF Aligned with Article 32, Directive (EU) 2019/944, the
[28] —
-Central Max. social
welfare
uni-sided VCG
auction
PF
calculation LFM operator2 receives this data and runs a network-aware
[29] —
-Min. flex.
cost
Regulated by
DSO
AC OPF decentralized market clearing algorithm in which the flexi-
[30]
Central SOCP -Central SOCP
FLMP AC OPF bility is dispatched to meet the flexibility requests3 without
joint clearing joint clearing
This paper
Stochastic model -Distributed clearing
FLMP
Component-based jeopardizing the distribution network constraints. There are
with DistFlow with CADMM AC OPF
multiple distributed/decentralized approaches reviewed in [32],
flexibility provision to the upstream market by maximizing [33] such as Optimality conditions decomposition (OCD)
her profit with offering incentives to aggregators without methods, analytical target cascading (ATC) method, and aux-
jeopardizing the network constraints. Authors in [28] examine iliary problem principle (APP). Yet, according to [34] and
different clearing algorithm for an LFM of the EU project [35], one of the fastest distributed optimization methods for
called EcoGrid 2.0, and propose the uni-sided VCG market as a non-convex AC OPF in a hierarchical structure where the
most beneficial one for both aggregators and the DSO due to a market operator plays the role of a coordinator is component-
good increase in social welfare. A power flow (PF) calculation based dual-decomposition technique. Therefore, we employ
is also applied to check the validity of contracts in terms of an adaptive component-based decomposition technique - Aug-
physical delivery. A mechanism is designed in [29] to clear mented Lagrangian method - to form a consensus problem
LFM while considering AC OPF with a reward function to that distributes the computation across components of the
regulate a incentive compatible pricing scheme for flexibility network (i.e. buses and branches) where the LFM is cleared
aggregators. However, all aforementioned projects have either while considering a distributed OPF (DOPF). The algorithm
proposed a LFM or LEM and none designed a framework that helps to harness the potential flexibility of ECs in a market-
sequentially or simultaneously clear both markets. based and distributed fashion. The problem is solved by
One work [30] is proposed a simultaneous clearing of consensus-ADMM (CADMM) which combines a coordination
energy and flexibility in both transmission and distribution step (i.e. LFM operator) among agents (i.e. ECs) interpreted as
levels using bi-level optimization model. In the upper-level tâtonnement process in the Walrasian auction [36]. The market
problem, TSO clears the joint energy and flexibility markets follows nodal pricing by computing the FLMP for ECs at each
with DC OPF as a linear programming (LP), whereas in the node of the grid. Moreover, the privacy of ECs are preserved
lower-level problem joint markets are cleared from DSOs’ through minimizing the private information communication as
perspective with AC OPF as second-order cone programming all ECs solve their own optimization problem, locally.
(SOCP) and flexibility local marginal price (FLMP) as pricing
rule. However the market structure is formed in a centralized D. Contribution
deterministic fashion without preserving participants’ privacy. The main contributions are listed as follows:
To the best of our knowledge, as summarized in Table. I, • A novel distributed sequential local energy and flexibility
the gap in this area is the lack of a market platform in which market structure is proposed in which the flexibility
ECs are treated as autonomous agents who can manage their capacity of a distribution network is defined through
communities and procure flexibility while satisfying their own LEM in a day-ahead time and demanded flexibility is
goals and grid constraints. In other words, there is no network- dispatched efficiently in LFM in the operating time.
aware local market design for sequential clearing of energy and
flexibility in distribution networks in the sense that flexibility 1 This is a low- or medium-voltage (LV/MV) grid with radial topology
evaluation is performed in LEMs to be communicated to a consists of customers with ESS, PV, and flexible loads as well as gas-fired
flexibility market for meeting the flexibility requests through small distributed generators (DGs).
2 To follow the liberalization of power systems and aligned with defined
dispatching flexibility in a distributed fashion.
tasks for DSOs in Article 31, Directive (EU) 2019/944, to have access to the
C. Solution approach distribution network data, we assume that the DSO has technical and financial
In this paper, a sequential clearing of local energy and flex- divisions in which the latter is in charge of running the LFM. We call both
ibility markets is proposed in the distribution networks com- divisions in this paper as DSO.
3 The model can respond to both security signals from TSO and DSO,
posed by several ECs which follows Directive (EU) 2019/944. though prioritized for TSO flexibility requests and we analyze, in this paper,
According to Article 16(2) of Directive (EU) 2019/944 ECMs the LFM response to TSO flexibility signals.

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 4

• A network-aware, stochastic LEM clearing algorithm is Local flexibility market

Data flow
proposed for ECs to individually act for their commu- Local Energy Market
Power flow

nity’s energy management so that each and any ECM ICT facilities

runs an adaptive two-stage stochastic programming to ECM

decide on optimum active and reactive power set-points of

their assets (i.e. flexible demands, ESS, DGs) as well as Customer 𝑛 Customer 2

exchange power with DSO at PCC. In addition, scenario Customer 1

generation using MCS method models random events (i.e.

uncertainty of PV and wind generation) of an EC in the ECn
TSO

second stage of the problem, in which maximum available DSO

Customer 3
flexibility delivered to the LFM in real-time is defined.
EC1
• An adaptive component-based dual-decomposition EC2

method called Augmented Lagrangian Relaxation (ALR)

is proposed as distributed clearing algorithm for LFM so Fig. 1. General structure of designed market
that flexibility provided by ECs are coordinated by LFM and linked into one or several branches with one connection
operator in an iterative auction to meet the flexibility point (i.e. PCC) to the public grid, forms an EC which is the
requests. foundation of the LEM. ECMs are in direct connection with
• LFM mechanism is not centrally run rather the market a DSO in the respective distribution network. All customers
operator (DSO) plays the role of a coordinator and the can communicate with the respective ECM via ICT facilities.
computation load is distributed over the agents (ECMs) The ECMs play the role of a mediator among customers and
which reduces the computational complexity for a single DSOs. An example of EC is a microgrid (MG) in which a
entity and preserves the privacy of actors. MG operator plays the role of ECM.
• Network constraints in both LEM and LFM are modeled Having smart systems e.g. Home Energy Management Sys-
in a sense that a linearized DistFlow is proposed for LEM tem (HEMS), customers are able to predict the demand and
algorithm with radial grids and an DOPF is included in self-production. Thus, they communicate the information such
LFM algorithm for dynamic network topology in which as production capacity and potential demand for a given time
the branches and buses are decomposed to solve the period to the ECM. Knowing this data and running an energy
problem in a distributed fashion. management algorithm, ECM makes decisions on the optimum
• A nodal pricing is proposed in the LFM to compute the set points of all productions/consumption, 24 hours before
FLMP as the final flexibility price to be traded between operating time and sends them back to relevant customers.
ECs and market operator within the iterative auction In other words, the ECM decides how much energy will be
process. supplied in the community with self-production and how much
The rest of the paper is organized as follows. The overall more needs to be imported from upstream network and what
structure of proposed LEM and LFM are explained in Section will be the potential regulation capacity. In the LEM, it is
II, the technical details of the clearing algorithms for both assumed that ECMs have no interaction among themselves,
LFM and LEM are presented in Section III, case studies and rather any potential energy deficit/exceed within communities
numerical results are given in Section IV, and finally Section is covered from the upstream network.
V concludes the paper. On the one hand, customers in the edge of the network
are usually formed in a radial topology and power flow is bi-
II. M ARKET S TRUCTURE
In this section, the structure of designed LEM and LFM directional which means that in addition to regular up-down
is presented. As depicted in Fig. 1, several ECs form a local current flow from upstream network, there is a bottom-up
energy market (LEM) in which ECMs are in charge of en- flow caused by the DERs on the demand side. Thus, by it is
ergy scheduling for prosumers/consumers in their community. important to design a network-aware energy management for
Maximum potential flexibility available in the community is ECM to consider bi-directional power flow in a radial network.
also determined in LEM by ECMs and communicated to To this end, a linearized DistFlow formulation with some
the respective DSO. In the real-time, the operator of LFM modifications to add bi-directional flow is offered in this paper
(e.g. DSO), calls the ECMs to provide desired flexibility to compute the active/reactive power flow while considering
based on signals received from TSO or the technical issues voltage drop from the PCC. A distribution substation is located
e.g. congestion detected in the distribution network. Thus, in the PCC used as the border between the community and the
LEM and LFM are operated sequentially to coordinate flexible rest of the network and taken as slack bus to each node of the
resources in distribution networks. The detailed description of feeder. In this sense, time varying export and import power
each market is expressed as follows. at PCC is scheduled by the ECM. Considering the existing
flexible resources in the EC, the up-/down-regulation to be
A. Local Energy Market Structure traded from PCC is decided by ECM which will be given to
It is assumed that customers are equipped with adequate DSO as the capacity of flexibility in the EC.
smart meters, and ICT infrastructure (e.g. like studied in On the other hand, due to the existing stochasticity caused
[37]) to log the data and communicate it. The collection of by weather-dependant RESs, demand or market price in the
customers, that are topologically located in the same grid area distribution network, ECM is encountered these random vari-

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 5

ables to predict and decide accurately. In other words, making market-based and network-aware flexibility dispatch are pre-
decisions for flexibility capacity needs a promising insight of sented. The clearing algorithm for community management is
the operating time. To overcome this challenge, it is proposed based on a pre-emptive market run one day ahead, whereas the
to employ a two-stage optimization method by ECM in which flexibility dispatch algorithm follows a Walrasian tâtonnement
having the historical data from uncertain variables and their process via CADMM in real time.
prediction, energy decisions would be made while considering A. Local Energy Market algorithm
a possible deviation of stochastic variables in the operating
ECs are usually in low-voltage level of the grid with radial
time. The detailed model is discussed in section III-A.
topology where a bi-directional DistFlow can be employed for
B. Local Flexibility Market Structure AC OPF model. To this end, the line active (reactive) flow Pij
Within the designed LFM, the DSO (flexibility operator (Qij ), as depicted in Fig.2, are split up into two items Pij+ , Pij−
section) is in charge of clearing the market 15 minutes before (Q+ −
ij , Qij ) where each indicates the different direction of the
the operating time. As discussed, ECMs send the flexibility ca- power. Thus, bi-directional DistFlow is formulated as follows:
pacity they can procure at 12:00 PM the day before operating
time to the respective DSO. The flexibility value (thereafter
X
Pj = Pij − rij iij − Pjk , (1)
called flexibility initial price), which is the initial price of k∈Nj
flexibility provision for ECs, is kept by the ECMs for solving X
their individual problems in the LFM. Qj = Qij − xij iij − Qjk , (2)
The distribution network is requested for flexibility provi- k∈Nj
sion based on two signals: one is called by TSO asking for 2
reserve provision due to any security reason in the upstream vj = vi − 2(rij Pij + xij Qij ) + (rij + x2ij )iij , (3)
network (high voltage level), another signal is based on any Pij2 + Q2ij
security issues within the operating distribution network such iij = , (4)
vj
as congestion problems or voltage distortion. The cumulative
of both signals is the total flexibility demand to be procured where Pij = Pij+ − Pij− , and Qij = Q+ − 2
i − Qj ; iij = Iij ,
by all ECs within the studied distribution network. Noted that 2 2
vj = Vj , and vi = Vi . Therefore, different variables including
we consider the TSO signal for flexibility with higher priority P, Q, i, v are calculated in an optimization problem while the
and in this work the reaction of proposed LFM to TSO signal (1)-(4) are considered as constraints. Nevertheless, there are
is analyzed, though the process TSO signal is generated is sources of non-linearity in these equations, including Pij2 and
beyond the scope of this work. Q2ij . In case, the optimization problem intends to be solved
An iterative auction mechanism is offered in this paper to in linear manner, these variables are required to be linearized.
clear the LFM in a way that flexibility operator section of DSO To this end, a piece-wise linearization approach is employed
checks potential convergence of the algorithm, iteratively, [38]. The procedure is formulated in (5)-(9) and depicted in
based on updated variables by agents. In this market, agents Fig. 3.
are ECMs and network security section of DSO. ECs are B
X
looking for their own objective which is minimizing the cost Pij2 = hsijb Λpijb , (5)
of flexibility provision. Meanwhile, network security section b=1

of the DSO runs a distributed AC DOPF, iteratively, to satisfy (∆max

ijb )
2
− (∆max
ijb−1 )
2
the network constraints. hsijb = = (2b − 1)∆max
ij1 , (6)
∆max
ijb − ∆max
ijb−1
To this end, a distributed clearing algorithm is formulated
using ALR to be solved by CADMM. Within this algorithm, it B
X
is assumed that each bus is connected to an EC and all buses Λpijb = Pij+ + Pij− , (7)
and branches of the distribution grid are decomposed. ECMs b=1
calculates and updates their own variables such as optimum VjN Iij
max
flexibility while network security section of DSO analyzes ∆max
ijb = b , (8)
B
if the calculated flexibility from ECMs satisfies network
constraints. The iterations continue to yield an equilibrium 0 ≤ Λpijb ≤ ∆max
ijb , (9)
which is checked by flexibility operator section of DSO at
The function, Pij2 , is divided into B blocks in which the
each iteration. The LFM iterative process is run 15 minutes
upper bound of each block ∆max ijb , achieved by division of
prior to operating time.
maximum capacity of the line ij into the total number of
Noted that DSO evaluates the requested flexibility before
desired blocks B (see (8)). b is an index indicating each
running the LFM and only executes the market when the
individual block and treated as a given value, hsijb is the slope
request is less than 85% of the maximum available flexibility,
otherwise the request would be declined. 𝑖 𝑗 𝑘
From slack bus 𝑃!"# , 𝑄!"
#
𝑃!"$ , 𝑄!"
$ #
𝑃"% #
, 𝑄"% $
𝑃"% $
, 𝑄"% To the end bus
III. A LLOCATION AND C LEARING M ECHANISM FOR
D ESIGNED L OCAL M ARKETS 𝑟!" , 𝑥!" 𝑟"# , 𝑥"#
In this section, the mathematical modeling of community 𝑉! , 𝑃! , 𝑄! 𝑉" , 𝑃" , 𝑄" 𝑉% , 𝑃% , 𝑄%
operation management in LEM as well as the algorithm for Fig. 2. Bi-directional DistFlow.

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 6

X X
𝑃!"% Pj = pzt,jec , Qj = zec
qt,j , (17)
!
𝐵 ( ∆%&'
!" zec ∈Zec zec ∈Z
zec
Zec : {dc, ch, pv, wg, d, dr, dg}, qt,j = pzt,jec tan(cos−1 ϕpf )
$
ℎ!"( = (2𝐵 − 1)∆%&'
!"* and ϕpf is power factor of the network.
2
iij ≤ I l(ij) , ∀i, j ∈ Bl , (18)
2
!
V 2j ≤ vj ≤ V j , ∀j ∈ B, (19)
16∆%&'
!"
$
ℎ!"( = 7∆%&' 2 2
!"+
Pij+ + Pij− ≤ I l(ij) V j ϕpf , ∀i, j ∈ Bl , (20)
!
9∆%&' $
!" ℎ!"( = 5∆%&'
!")
− 2 2
−1
! Q+
ij + Qij ≤ I l(ij) V j sin(cos ϕpf ), ∀i, j ∈ Bl , (21)
4∆%&' $
!" ℎ!"( = 3∆%&'
!"(
%&' !
∆!" $
ℎ!"# = ∆%&' where (18) - (21) are the current, voltage and active, reactive
!"#

power flow constraints in the day-ahead decisions, respec-

𝑃!"
𝑏=1 𝑏=2 𝑏=3 𝑏=4 𝑏=𝐵 tively. The ESS constraints for the first-stage decisions are
$ $ $ $ $
Λ !"# Λ !"% Λ !"& Λ !"' Λ !"( as follows:
Fig. 3. Linearization of DistFlow. SOCt,j =SOCt−1,j + pch t,j ηch −
(22)
of the function in block b, and Λpijb is the value of Pij in block dc
pt,j /ηdc , ∀t ∈ T , ∀j ∈ Bess ,
b in which its summation over blocks equals to the summation where ηdc and ηch are the discharging and charging efficiency
of power flow in line ij in both directions i.e. Pij+ and Pij− of ESS, respectively. SOC at the beginning of time horizon
(see (7)). According to (9), Λpijb at each block b can be zero t = 1 should be equal to the one at the end of time horizon
or any positive value less or equal than the upper bound of t = T , SOC1,j = SOCT ,j .
the block. Eventually, (5) would turn into (10) as follows: The second-stage function is formulated in (23), where the
B
+ −
X scenarios, ω, are generated using MCS method (similar to
Pij2 = ∆max
ij1 (P ij + P ij ) (2b − 1), (10) [39]) in a sense that solar radiance is modeled using Beta
b=1
distribution function and wind speed modeled using Rayleigh
The same approach is applied for Q2ij and eventually (1) -
distribution function. Many equal-probability scenarios are
(4) turn into (11) - (14): X generated and reduced to the desired number Ω using forward
Pj = (Pij+ − Pij− ) − rij iij − +
(Pjk −
− Pjk ), (11) method [40] with particular probabilities πω .
k∈Nj
X G(yec , ω) = min πω cy yec (23)
Qj = (Q+
jk − Q−
jk ) − xij iij − (Q+ −
jk − Qjk ), (12) yec

k∈Nj
where Ω is the number of scenarios, yec : {pdr dc c
t,j,ω , pt,j,ω , pt,j,ω ,
p(q),up p(q),dn
Rt,ω , Rt,ω , Pij,ω , Qij,ω , iij,ω , vj,ω } and cy is the price
vj = vi − 2(rij [Pij+ − Pij− ] + xij [Q+
− Q− 2 2
ij ]) + (rij + xij )iij ,
ij of relevant yec unit.
PB PB (13) The respective constraints for the second stage are differ-
s p s q
b=1 hijb Λijb + b=1 hijb Λijb entiate active and reactive nodal balances between day-ahead
iij = , (14)
vj and real-time problems, second-stage voltage balance, second-
This work proposes a two-stage programming as LEM stage current flow and limitation of power flow for scenarios.
clearing algorithm for ECMs in which total operation cost Other constraints that link second- and first-stage decisions are
and expected security cost are minimized. According to the as follows:
(15), the day-ahead decisions for EC (xec ) are made in the pch ch
t,j,ω ≤ pt,j , ∀t ∈ T , ∀j ∈ Bess , ∀ω ∈ Ω, (24)
first stage as here-and-now decisions and real-time variables pdc dc
t,j,ω ≤ pt,j , ∀t ∈ T , ∀j ∈ Bess , ∀ω ∈ Ω, (25)
(yec ) are defined in the second stage within security cost
function (G) as wait-and-see decisions. Within this multiperiod SOCt,j,ω ≤ SOCt,j ∀t ∈ T , ∀j ∈ Bess , ∀ω ∈ Ω, (26)
market clearing, all xec includes DERs set points (e.g. DG,
ESS, and DR) together with exchange active (reactive) power pdr dr
t,j,ω ≤ pt,j ∀t ∈ T , ∀j ∈ Bdr , ∀ω ∈ Ω, (27)
with upstream network at PCC (i.e. p(q)pcc t ) and maximum
available flexibility at PCC which is active (reactive) up- p(q),dn
Rt,ω ≤ Rt
p(q),dn p(q),up
, Rt,ω ≤ Rt
p(q),up
, ∀t ∈ T , ∀ω ∈ Ω,
p(q),up p(q),dn
and down-regulation (i.e. Rt , Rt ) are optimized for (28)
horizon time T which
XX is 24 hours.
min (cx xec + Eω [G(yec , ω)]) (15) where (24)-(26) are the link between day-ahead and real-
xec ,yec time decisions for ESSs, (27) limits the security action of DR
t∈T j∈B
where xec : {pdg dr dc ch pcc p(q),up p(q),dn into DR schedule in day-ahead stage, and the same limitation
t,j , pt,j , pt,j , pt,j , p(q)t , Rt , Rt , iij ,
vj , Pij , Qij }, cx is the price of relevant xec unit, and Eω [.] is applies in (28) for up- and down-regulation of active power
the expected value operator over ω. (p) and reactive power (q). Indeed, the ECM defines maximum
p(q),up p(q),dn
s.t. up- and down-regulation of active power, Rt , Rt , at
(11) − (14) (16)
certain time period t based on the available flexible resources
where i.e. ESS, DR after considering potential risks (i.e. deviation

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 7

s
0:00 12:00 0:00 t-15min t yij is the series admittance and bch ij indicate the charging
shif t

Operating time
susceptance. Tij = tij eθi represents the complex tap ratio
of the transformer, while tij is the tap ratio magnitude in which
LEM clearing process LFM clearing process for lines, tij = 1. Tap ratio is not dynamically modified while
$ % !"/$%
𝑝!"# , 𝑞!"# , 𝑅!
𝑥( 𝑥( , 𝜆 running the LFM, hence, the tap ratio is treated as a parameter.
ECM1 𝑧( ECM1 𝑟, 𝑠
ECM2
$ % !"/$%
𝑝!"# , 𝑞!"# , 𝑅!
𝑥( DSO
The consensus variables and constraints in (29) - (32) make
DSO 𝑣& DSO 𝑥( , 𝜆
. 𝜃&
"#
𝑧( ECM2 𝑟, 𝑠
Flexibility
the decomposable problem and ensure the consensus between
. Energy 𝑝&' Network
managemetn 𝑞"# security . operator variables of decomposed buses and branches.
. &'
𝑥( .
section section 𝑥( , 𝜆 section p p
ECMi
$ % !"/$%
𝑝&"# , 𝑞&"# , 𝑅& 𝑧( ECMi 𝑟, 𝑠
fij = fij (i)
, (λfijp ) (i, j) ∈ L ∪ L̂ (29)
q q
fij = fij(i) , (λfijq ) (i, j) ∈ L ∪ L̂ (30)
Fig. 4. Data flow in proposed sequential LFM & LEM.
of RESs generation from forecast) that may encounter in real- vi(ij) = vi , (λvij ) (i, j) ∈ L ∪ L̂ (31)
time operation. θi(ij) = θi , (λθij ) (i, j) ∈ L ∪ L̂ (32)
If the certain EC is located in bus i of a distribution where buses’ decomposition include the vector variable xf :
p q
network, then the notation of available flexibility at a certain xf := [(fij (i)
, fij )
(i) (i,j)∈L∪L̂
, (vi , θi )i∈B ] (33)
time step to use in LFM is turned into Rip,up , Rip,dn , Riq,up , and branches’ decomposition involve vector variable zf :
Riq,dn representing maximum active up- and down-regulation, zf := [(fij p q
, fij , vi(ij) , θi(ij) )(i,j)∈L∪L̂ ] (34)
and maximum reactive up- and down-regulation, respectively. p q p q
where fij(i) , fij(i) , vi , and θi are local variables and fij , fij ,
Likewise, the notation of active and reactive power exchange vi(ij) , and θi(ij) are sharing variables. Accordingly, the single-
of an EC at PCC, p(q)pcc ec
t , is turned into p(q)i in LFM which period decomposable problem is formulated as follows:
4
indicates power exchange of EC at bus i. .
X X
min Fi (fpi ) = c2,f fp2i + c1,f fpi + c0 (35)
B. Local Flexibility Market algorithm fpi ,fqi ,
xf ,zf i∈B i∈B
After closing the LEM, the resulting flexibility range
p(q),up p(q),dn subject to Rip,dn ≤ fpi ≤ Rip,up ,
Ri and Ri as well as the active pec
i and reactive i∈B (36)
ec
qi power of ECs at the PCC connected to bus i ∈ B of Riq,dn ≤ fqi ≤ Riq,up , i∈B (37)
the distribution network are communicated to the LFM. The q
p q
following steps, as depicted also in Fig. 4, are taken to clear (fij + pec 2 ec 2
ij ) + (fij + qij ) ≤ sij , (i, j) ∈ L ∪ L̂ (38)
the LFM.
1) Estimate distribution network usage: Before running the |V i | ≤ vi(ij) ≤ |V i |, (i, j) ∈ L ∪ L̂ (39)
LFM algorithm, DSO estimates the capacity of the network ∆θij ≤ θi(ij) − θj(ji) ≤ ∆θij , (i, j) ∈ L (40)
which is supposed to be occupied by ECs’ interaction with X p X X
upstream network. Thus, a power flow (PF) is run using the fij(i) + pec
ij(i) = fpup
i
+ pec
i
output of LEM i.e. pec ec ec
i and qi and the results pij and qij
ec
j∈Bi j∈Bi g∈Gi
(41)
with (i, j) ∈ L ∪ L̂ shows the line load share and potential
X X
− fpdn
i
− vi2 gish , i∈B
congestion in the network. d∈Di
The DSO sends back the results of PF to the ECs so X q
X X
ec
that each EC realizes the line loads connected to its bus (i), fij i
+ qij = fqup
i
+ qiec
as well as the voltage magnitude v and the voltage angle j∈Bi j∈Bi g∈Gi
X X (42)
θ. The line usage is considered as given data in the LFM − fqdn
i
+ vi2 bsh
i , i∈B
algorithm, whereas voltage angle and magnitude will be used d∈Di
for initialization of the iterative process. p
fij + pec
ij = gij vi2(ij) +
2) Forming LFM algorithm using ALR: The component-
c
based decomposition of a π-model of distribution networks gij (vi(ij) vj(ji) cos(θi(ij) − θj(ji) ))+ (43)
on buses and branches can be performed as shown in Fig. 5 bcij (vi(ij) vj(ji) sin(θi(ij) − θj(ji) )), (i, j) ∈ L ∪ L̂
which is inspired by [35]. q ec
fij + qij = −bij vi2(ij) −
4 ECsare operated in low-voltage and medium-voltage level e.g. 400V and bcij (vi(ij) vj(ji) cos(θi(ij) − θj(ji) ))+ (44)
11kV and at their PCC are connected to bus i of distribution network with a
c
transformer operated in medium-voltage level e.g. 20kV or 110kV. gij (vi(ij) vj(ji) sin(θi(ij) − θj(ji) )), (i, j) ∈ L ∪ L̂
𝑖𝑖 𝑗𝑗 where (36) and (37) indicate the limit of the flexibility provi-
𝑝𝑝𝑖𝑖𝑖𝑖(𝑖𝑖) + 𝑖𝑖𝑞𝑞𝑖𝑖𝑖𝑖(𝑖𝑖) 𝑝𝑝𝑖𝑖𝑖𝑖 + 𝑖𝑖𝑞𝑞𝑖𝑖𝑖𝑖 𝑝𝑝𝑗𝑗𝑖𝑖 + 𝑖𝑖𝑞𝑞𝑗𝑗𝑗𝑗 𝑝𝑝𝑗𝑗𝑖𝑖(𝑗𝑗) + 𝑖𝑖𝑞𝑞𝑗𝑗𝑖𝑖(𝑗𝑗) sion of each EC using outputs of LEM. In (38) the line usage
𝑦𝑦𝑠𝑠
𝑝𝑝𝑖𝑖𝑗𝑗(𝑖𝑖) 𝑝𝑝𝑖𝑖𝑖𝑖 𝑇𝑇𝑖𝑖𝑖𝑖 𝑝𝑝𝑗𝑗𝑗𝑗 𝑝𝑝𝑗𝑗𝑖𝑖(𝑗𝑗) due to the energy allocation of the LEM and the flexibility
𝑞𝑞𝑖𝑖𝑗𝑗(𝑖𝑖) 𝑞𝑞𝑖𝑖𝑖𝑖 𝑐𝑐𝑐
𝑏𝑏𝑖𝑖𝑖𝑖 𝑏𝑏𝑗𝑗𝑗𝑗𝑐𝑐𝑐 𝑞𝑞𝑗𝑗𝑗𝑗 𝑞𝑞𝑗𝑗𝑖𝑖(𝑗𝑗) allocation of the LFM is limited by the overall line capacity
𝜈𝜈𝑖𝑖(𝑖𝑖𝑖𝑖) 𝑗𝑗 𝑗𝑗 𝜈𝜈𝑗𝑗(𝑗𝑗𝑗𝑗) 𝜈𝜈𝑗𝑗 sij , (ij) ∈ L ∪ L̂. (39) and (40) limit the voltage magnitude at
𝜈𝜈𝑖𝑖 2 2
𝜃𝜃𝑖𝑖 𝜃𝜃𝑖𝑖(𝑖𝑖𝑖𝑖) 𝜃𝜃𝑗𝑗(𝑗𝑗𝑗𝑗) 𝜃𝜃𝑗𝑗 each bus, as well as the voltage angle difference between two
buses. Constraints (41) and (42) present the active and reactive
Fig. 5. Decomposed π-Model of the buses and branches. nodal balance of bus i connected to ECi in which pec ec
i /qi are

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 8

given data from LEM, the line usage pec ec

ij /qij as output of
Algorithm 1 LFM iterative clearing algorithm using CADMM
PF in previous step, and the flexibility allocation fpi /fqi as 1: Initialize: ϵabs , ϵrel , τdecr , τincr , µdecr , µincr , σ, θ, v and
p q
well as flexibility flow fij /fij are decision variables. Energy consensus variables in (29)-(32)
and flexibility flow are computed using (43) and (44) when 2: while ||rk+1 ||2 ≤ ϵpri and||sk+1 ||2 ≤ ϵdual do (Checked
the flow direction is from i to j for both directions of the by LFM operator)
c
line L ∪ L̂. In addition, gij = ℜ(yii ), bcij = ℑ(yii ), gji
c
= 3: ECMs update xk+1 using (50)
c fi
ℜ(yjj ), bji = ℑ(yjj ), gij = ℜ(yji ), bij = ℑ(yji ), gji = 4: DSO (network security section) computes zfk+1 using
ℜ(yij ), bji = ℑ(yij ), where y is the component of admittance (51)
ij

bch
ij
s s
yij +j 2 −yij 5: ECMs update λ using 49
matrix Yij and defined as yii = t2ij
, yij = shif t ,
tij e−jθi 6: ECMs calculate primal and dual residual using (52)
s
−yij bch
yji = s
, and yjj = yij +j ij and (53)
jθ
shif t 2 . The compact form of
tij e i
7: DSO (flexibility operator section) updates ϵpri and
ALR for the decomposed problem can be written as follows: dual
ϵ using (54) and (55)
Lp (xf , zf , λ) =f (xf ) + g(zf ) + λT (Axf + Bzf )
(45) 8: if k == 1 then
ρ
+ ||Axf + Bzf ||2 9: Update ρ using (56)
2
where 10: else if k mod σ == 0 then
λ = [λfijp , λfijq , λvij , λθij ], (ij) ∈ L ∪ L̂, (46) 11: Update ρ using (56)
12: end if
and f (xf ), g(zf ) are the separation of function of bus vari-
13: end while
ables, xf (feasible set defined by (36), (37), (41), and (42))
and function of branch variables zf (feasible set defined by
subject to: (38), (39), (40), (43), and (44).
(38), (39), (40), (43), and (44)), respectively. Axf + Bzf = 0
represents the consensus constraints defined in (29) - (32) and ECMs compute the dual variables for the next iteration k +
ρ is a positive penalty factor. 1 using the dual variables in (49). The primal r and dual
3) Solving ALR problem of LFM algorithm using CADMM: residual s are calculated in (52) and (53), respectively. The
The ALR problem is separated using CADMM by the set of primal residual indicates the difference between the values of
xf in (47) and zf in (48) to obtain the decomposed problem local and sharing variables , while the dual residual indicates
for the buses and the branches, respectively. To this end, the change of the local and sharing variables, compared to the
the dual variables are fixed and updated in (49) using an previous iteration.
iteration process. Thereafter, to obtain the updated iterates for rk+1 =Axk+1f + Bzfk+1
k+1 k+1 k+1 k+1
the iteration k + 1 the consecutive problems are solved: p
=[(fij p
− fij q
), (fij q
− fij ), (52)
(i) (i)
xk+1
f = arg min Lρ (xf , zfk , λk ), (47)
xf ∈X (vik+1 − vik+1 ), (θik+1 − θik+1 )].
zfk+1 = arg min Lρ (xk+1
f , zf , λk ), (48) (ij) (ij)

λ k+1
z ∈Z
k f
= λ + ρ(Axf + Bzfk+1 ).
k+1
(49) sk+1 =ρAT B(zfk+1 − zfk )
k+1 k k+1 k
p p q q
Considering (35) and (47), the problem for agent ECi =[(fij − fij ), (fij − fij ), (53)
connected to bus i is formulated as (50) to compute the (vik+1
(ij)
− vik(ij) ), (θik+1
(ij)
− θik(ij) )].
flexibility share and other bus variables at iteration k + 1.
The CADMM algortihm of LFM converges when
xk+1
fi := min xfi
Fi (fpi ) ||rk+1 ||2 ≤ ϵpri and ||sk+1 ||2 ≤ ϵdual , where the feasibility
X p pk q qk tolerances ϵprip and ϵdual are defined as:
+ (λfijp (fij − fij ) + λfijq (fij − fij )
j∈Ni
(i) (i)
ϵ = Nλ ϵabs + ϵrel max||Axk ||, ||Bz k ||,
pri
(54)
p
+ λvij (vi − vik(ij) ) + λθij (θi − θik(ij) )) (50) ϵ dual abs rel
= Nz ϵ + ϵ ||A λ ||, T k
(55)
ρ p pk 2 q qk 2
and the absolute and relative tolerances ϵabs and ϵrel are
+ ((fij − fij ) + (fij − fij ) positive [41]. In order to improve convergence, an adaptive
2 (i) (i)

k 2 k
+ (vi − vi(ij) ) + (θi − θi(ij) ) ), 2 penalty factor ρ is implemented according to [35]. Each set of
subject to: (36)), (37), (41), (42). consensus variables has one corresponding ρ, which is updated
The network security section5 runs (51) to compute the as follows:
k
branch variables at iteration k + 1. ρ (1 + τincr ),
 if ||rk+1 || > µincr ||sk+1 ||,
pk qk ρk+1 = ρk (1 + τdecr )−1 , if ||sk+1 || > µdecr ||rk+1 ||,
X p q
zfk+1
ij
:= min
z
λfijp (fij(i)
− fij ) + λfijq (fij (i)
− fij )
fij

 k
(i,j)∈L∪L̂ ρ , otherwise,
(56)
+ λvij (vik − vi(ij) ) + λθij (θik − θi(ij) )
The parameters τincr /τdecr define the step size of the update,
ρ pk p 2 qk q 2
+ ((fij − fij ) + (fij − fij ) while µincr /µdecr keep the residuals within a range of each
2 (i) (i)
other. Depending on the network size, the update is delayed by
+ (vik − vi(ij) )2 + (θik − θi(ij) )2 ), a number of σ iterations. This prevents oscillations of ρ and
(51) further improves the convergence (see section IV for details).
5 This section is a part of DSO and in charge of distribution network security. Algorithm 1 summarizes the market clearing of the LFM.

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 9

0.06
External grid 1700

1200 0.05
110 kV bar
700
0.04
Transformer 20 kV
1 200

€/kWh
11 kV

kWh
20 kV bar 2 0.03
9 3
Line 0

-300
10 4 11
0.02

Line 5
15 14 5 12
6 -800
Bus 2 7 8
13
Total active load Power exchange at PCC DG bus 2
Line 1

DG bus 3 DG bus 5 DG bus 6 0.01

EC 1 -1300
Total DR ESS charge ESS discharge
DR price (bus 9) DG price Retail market price
Bus 6 -1800 0
Bus 3

Line 4
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Line 2

EC 5 Time (h)
EC 2
Fig. 7. Results of LEM clearing for EC2 with respective prices.
Bus 5
Bus 4 TABLE II: Data communication from LEM to LFM at time slot 16.
EC 4 up dn up up
EC 3 pec
i qiec Rip Rip Riq Riq cRp
i i
Switch
(kWh) (kvar) (kWh) (kWh) (kvar) (kvar) (C/kWh)
Line 3 1 1326.2 356.9 0 0 162.8 127.2 0.046
2 1253.6 355.2 31.9 30.7 142.1 0 0.046
Fig. 6. MV dynamic network topology as test Case1 3 1250.5 13.1 30.2 30.7 495.9 0 0.048
IV. C ASE S TUDY & N UMERICAL R ESULTS 4 1269.2 355.7 47.2 30.7 119.1 0 0.048
To check the performance of the proposed sequential LEM 5 1269.2 355.7 47.2 30.7 119.1 0 0.05
and LFM, two cases are employed with following specifica- The figure shows the optimum allocation of energy in
tions: the community as optimum set-points of DGs, ESS charge
- Case1 is a modified 6-bus and 20-kV distribution network (discharge), flexible loads, and exchange active power at PCC
with dynamic topology as depicted in Fig. 6. The network (traded with DSO or retailers) considering the respective
can be operated in radial or ring topology. The reference bus is prices. In peak hours (t=12, 19 - 22), as shown, DGs are mainly
connected to a 110 kV network with a transformer and the rest scheduled to generate as well as ESS are discharged and
of buses i are connected to a 15-bus and 11-kV radial network DR are programmed to be called for supporting the internal
as ECi via transformers with nominal apparent power of 2.3 demand and sell the extra energy to DSO/retailers cause DG
MW. The ECs are similar, except the number and location of operation cost and DR price are lower than retail price.
DER in the network. The details on the network data for 5 In addition, the data to be communicated to LFM including
ECs are in [42]. As shown in Fig. 6, the MV network has a traded active and reactive power at PCC (p(q)ec i ), maximum
switch in line 3 to check the operation of the market in both p(q)up/dn
available up- and down-regulation (Ri ), plus regulation
radial and ring topology in order to prove the applicability of
price as flexibility initial price (cRip ) for all five ECs at time
the LFM market in dynamic network topology [42].
step 16 are summarized in Table. II. As mentioned, ECs
- Case2 is a modified radial 82-bus network which originates
differences are in the number and location of installed DERs,
from the SimBench dataset with network code “1-HV-urban–
thus, their energy management at each time slot follows a
0-sw” [43]. The reference bus is connected to a 220 kV
relatively similar pattern.
network, while the nodes of the 81 ECs are operated at 110
kV. The details of the modified grid can be found in [42]. B. LFM clearing results
Most analysis are conducted on Case1 for giving a better Given the data in Table II, LFM will be run to allocate
vision on the market performance. However the scalability of the flexibility and clear the market for any single time slot.
the market and its applicability on larger grids are shown using In this example, c0 and c2,f in (35) are supposed to be zero
Case2. CPLEX solver is used to solve and clear each and and c1,f = cRip . Before running LFM, a PF is run given pec i
any ECs and IPOPT is employed to run the LFM clearing and qiec to compute the distribution line flows and discover
algorithm. Using a 8GB-RAM, 1.1 GHz Quad-Core Core potential congestion. Due to limited space, we do not show
i5, LEM algorithm takes 1.17 seconds and the running time the PF results, yet the selected test case is resilient enough
of LFM algorithm depends on the selection of the initial without any congestion.
parameters which is discussed in this section. First, the sensitivity of CADMM to the initial parameter se-
A. LEM clearing results lection is analyzed. The experiment indicates that ϵabs = 10−6
To show the performance of the LEM clearing algorithm, and ϵrel = 2 ∗ 10−5 , are giving the most accurate results for
EC2 which is connected to bus 3, in Case1, is chosen for in the open switch (radial network), as well as the closed switch
detail analysis. EC2 has four gas-fired DGs with capacity 1.6 ring network in Case1. Regarding Case2, ϵabs = 10−6 and
MW (buses 4, 6, 11, 15), two WGs (buses 12, 8) and two PVs ϵrel = 8 ∗ 10−5 are chosen. Table III displays the dependency
(buses 5, 10), with capacity 100 kW each, two ESSs (buses of number of iterations to delay σ of ρ-update for Case1.
3, 11) with capacity of 299 kW, ηdc =0.95 and ηch =0.9, and Note that Gap in this table represents the difference between
10% of loads connected to buses 9, 15 are flexible for DR the result of the decomposed and centralized problem. Thus,
programs. Output of scenario generation and reduction is ten as seen in the closed-switch ring network, σ = 10 and in
scenarios for PVs and WG production for next-day 24 hours. the open (ring network) σ = 20 yield the fastest convergence
The results of LEM clearing for EC2 and next-day 24 hours (less iterations) with a reasonable amount of Gap. In both
are depicted in Fig. 7. topologies, when σ = 5, the algorithm does not converge. In

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 10

TABLE III: Sensitivities regarding the ρ-update delay σ in Case1

Up-regulation request - no congestion (t =16)
Open Ring Closed Ring

50 kWh 100 kWh 150 kWh

EC5 EC4 EC3 FLMP
σ Iter. Gap(%) Iter. Gap(%) EC2 EC1 Flex. Quantity
5 DNC — DNC — FLMP
10 191 -0.05 150 0.21 Flex. Quantity
15 187 -0.1 174 -0.09
FLMP
20 187 0.06 181 -0.23
Flex. Quantity
25 320 -0.17 166 -0.23
30 221 -0.19 325 0.1 -35 -25 -15 -5 5 15 25 35 45
FLMP (€/MWh), Flex. Quantity (kWh)
35 265 0.14 204 0.20 Up-regulation request - with congestion (t = 16)
40 259 0.21 249 0.18

50 kWh 100 kWh 150 kWh

EC5 EC4 EC3 FLMP
45 234 -0.17 224 0.11 EC2 EC1
Flex. Quantity
50 262 -0.24 250 0.17 FLMP
55 264 -0.17 249 -0.20
Flex. Quantity
60 273 -0.18 243 0.22
FLMP
Network with 5 ECs Network with 81 ECs
10 1 100 Flex. Quantity
||r||2 pri
-35 -25 -15 -5 5 15 25 35 45
FLMP (€/MWh), Flex. Quantity (kWh)
10 3 10 2
Down-regulation request - no congestion (t =15)
EC5 EC4 EC3

50 kWh 100 kWh 150 kWh

EC2 EC1 FLMP
0 25 50 75 100 125 150 0 250 500 750 1000 1250 Flex. Quantity
101
100 FLMP

10 3 10 2 Flex. Quantity
||s||2 dual
FLMP

0 25 50 75 100 125 150 0 250 500 750 1000 1250 Flex. Quantity

5.0
FLMP [ct/kWh]

FLMP [ct/kWh]

5.0 -40 -30 -20 -10 0 10 20 30 40 50 60

FLMP (€/MWh), Flex. Quantity (kWh)
2.5 2.5 Fig. 9. Flexibility quantity and FLMP for all ECs in different
0.0 0.0 flexibility demands.
0 25 50 75 100 125 150 0 250 500 750 1000 1250
Iterations Iterations the real world practice. In lower flexibility demand settings
Fig. 8. LFM Convergence rate in networks with 5 (Case1) and 81 (i.e. 50 kWh), some ECs e.g. EC5 sells down-regulation to
ECs (Case2). the market (as it is closer to the upstream network) even
Case2, σ = 20 showed the best performance. For the ρ-update when up-regulation is needed (see top graph of Fig. 9).
τdecr = 0.9, τincr = 1.15, µdecr = 1.15 and µincr = 1/µdecr This phenomena is also seen in down-regulation flexibility
proved to result in reasonable oscillation during the iterations request. This behaviour possibly arises from the freedom for
for both Cases. ECs to dispatch both up- and down-regulation under any
Fig. 8 compares the convergence of primal and dual resid- circumstance, they try to better off as much as possible and
uals as well as the FLMP between Case1 and Case2. On one might schedule for down-regulation even if up-regulation is
hand, Case1 is supposed to provide 100 kWh flexibility request requested, specially when the distribution lines have enough
for TSO (at time slot 16) in which the LFM is cleared in 150 capacity for this arbitrage. Indeed, as the flexibility request
iterations for the ring topology in 12.47 seconds (see left-hand increases, there would be no room for such behavior in the
side of Fig. 8) with a system of 2.4 GHz 16-core i9 and 64 market particularly when lines get closer to congestion. It is
GB RAM. Noted that this iteration takes 14.67 seconds for all the matter of market design and plausible to prevent such
the radial topology. On the other hand, Case2 procures 1.62 phenomena by additional constraints and regulations.
MW flexibility request in which takes the clearing algorithm The performance of the markets over the course of a day
1414 iterations in 1423.76 seconds to reach an equilibrium (see is depicted in Fig. 10. The top graph of Fig. 10 displays
right-hand side of Fig. 8) when solved on a single machine. maximum available flexibility for each EC as the output of
In application, within the algorithm the computation would be the LEM algorithm whereas the reaction of ECs to procure
distributed over the agents which reduces the computational 50 kWh up-regulation within LFM is shown in the bottom
complexity even in less powerful machines. graph. The LFM operator just accepts 12 time steps flexibility
Fig. 9 displays the flexibility market price i.e. FLMP and request, since the total available up-regulation in the rest of
flexibility quantity provided by each EC in Case1 to procure time steps, scheduled by LEM, is not sufficient to procure
50, 100, and 150 kWh flexibility requests in three different 50kWh. LEM outputs, as depicted in the top graph, depend on
conditions including up-regulation request with and without available flexible resources at each EC and as the specification
congestion at time slot 16, and down-regulation request with- of EC4 and EC5 are similar and slightly different from EC2
out congestion at time slot 15. As shown in the top and bottom and EC3 , the results are somewhat following the same pattern
graphs, when there is no congestion, FLMP for all ECs are still different from EC1 with less flexibility resources. It is
almost the same, whereas in the middle graph when there is interesting to notice that while available flexibility for both
congestion in the grid, the market is cleared with different EC4 and EC5 are similar with the same flexibility initial price,
prices for each EC. Noted that finding the optimum flexibility they react differently in the LFM because of their location
initial prices for each EC was out of the scope of this paper, in the grid (see bottom graph of Fig. 10). EC5 tends to
therefore, the flexibility initial prices are set relatively small sell unnecessary down-regulation to the LFM to better off
for ECs as given parameters yet comparable with retail market in some time slots taking the advantage of being closer to
prices and the EC domestic production prices. This assumption the upstream network, unlike EC4 . This behavior forces other
leads to also relatively small range of FLMPs compared with agents to procure more up-regulation which leads to higher

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 11

Maximum Available Flexibility

prices accordingly to adjust the available flexibility within
60
LEM.
40 • LFM convergence rate is reasonable in both small- and

20
larg-scale grids given the fact that this algorithm is not
supposed to run by one entity rather distributed among
kWh

0
agents which causes even faster computation process.
-20 • As expected, without congestion, FLMP would be equal
among agents of LFM, while having a congestion makes
-40 EC1 EC2 EC3
EC4 EC5 the flexibility market prices different among ECs based
-60 on the location and the contribution to the market.
1 2 3 4 5 9 10 11 14 15 16 17
Selected Time (h) • In some cases particularly when the flexibility request
80 70
Flex. quantity & price is low or there is no congestion, the market equilibrium
60 60 would happen by dispatching unnecessary flexibility (i.e.
40 50 up-regulation is requested while an EC delivers down-
regulation). This is the matter of the market design and
20 40

€/MWh
can be prevented by adding more regulations on different
kWh

0 30
circumstances.
-20 20 • Over the course of 24 hours, the designed markets are

-40
EC2 - Flex. quantity
EC4 - Flex. quantity
EC3 - Flex. quantity
EC5 - Flex. quantity 10
performed consistently and a strong bond between LEM
FLMP Flexibility initial price and LFM is trivial in a sense that the dispatched flexibility
LEM price
-60 0
1 2 3 4 5 9 10 11 14 15 16 17 and FLMP are determined in LFM based on flexibility
Selected Time (h)
initial prices and availability defined in LEM.
Fig. 10. Performance of the local markets in different hours of a day
with 50kW flexibility request. In addition, one of the capabilities of the proposed algorithm
FLMP compared with flexibility initial prices as shown in is reactive power reserve provision for voltage regulation,
the bottom graph. A regulator or policy maker might not be which can be analyzed in the future work. Finding the op-
interested in such behavior and need to modify the market timum flexibility initial prices for ECs is another potential
rules. LEM price at the bottom graph of 10 is indeed the future work. Comparing different market rules to improve the
retail price in LEM and depicts much lower value compared performance of the market under any circumstance, including
to FLMP. communication issues like data losses, is another suggestion
Eventually, it is worth noting that such designed LFM, can for the future work. Considering a Reinforcement Learning
be operated as the Intraday market for ECs, since the market algorithm for players e.g. ECs and DSOs to learn in the market
is not only meeting DSO/TSO security needs. It also covers environment for an efficient participation might be worthwhile
internal ECs security needs when the estimations in LEM vary extension.
closer to the real time.
ACKNOWLEDGMENTS
V. C ONCLUSION
We would like to thank the reviewers and the Editor for
To provide flexibility requests in distribution networks, local the constructive comments during the review process and also
markets have been designed in this paper in a way that Daniel Gebbran for sharing thoughts during the project.
LEM and LFM have been cleared, sequentially. ECMs run R EFERENCES
a network-aware two-stage programming to schedule their [1] W. Ketter, J. Collins, M. Saar-Tsechansky, and O. Marom, “Information
optimum available flexibility in the frame of LEM in a day- systems for a smart electricity grid: Emerging challenges and opportuni-
ahead scheme. LFM, on the other hand, has been designed to ties,” ACM Transactions on Management Information Systems (TMIS),
vol. 9, no. 3, pp. 1–22, 2018.
efficiently dispatch the flexibility of ECs in real time without [2] J. Villar, R. Bessa, and M. Matos, “Flexibility products and markets:
jeopardizing the network constraints. LFM aims to satisfy any Literature review,” Electric Power Systems Research, vol. 154, pp. 329–
flexibility requests via an iterative auction formulated by ALR 340, 2018.
[3] J. W. Busby, K. Baker, M. D. Bazilian, A. Q. Gilbert, E. Grubert,
and solved by CADMM. V. Rai, J. D. Rhodes, S. Shidore, C. A. Smith, and M. E. Webber,
A test case including an LV/MV grid operated by ECMs “Cascading risks: Understanding the 2021 winter blackout in texas,”
and MV grid owned by a DSO has been employed to assess Energy Research & Social Science, vol. 77, p. 102106, 2021.
[4] S. Talari, M. Khorasany, R. Razzaghi, W. Ketter, and A. S. Gazafroudi,
the efficiency of proposed local markets. A large-scale grid “Mechanism design for decentralized peer-to-peer energy trading con-
also used to prove the scalability of the proposed method. sidering heterogeneous preferences,” Sustainable Cities and Society,
The results have shown the followings: vol. 87, p. 104182, 2022.
[5] H. Liao and J. V. Milanović, “Flexibility exchange strategy to facilitate
• Based on the specifications of the ECs including available congestion and voltage profile management in power networks,” IEEE
source of flexibility, domestic demand, and also exoge- Transactions on Smart Grid, vol. 10, no. 5, pp. 4786–4794, 2018.
[6] X. Jin, Q. Wu, and H. Jia, “Local flexibility markets: Literature review
nous factors e.g. retail prices, the maximum flexibility on concepts, models and clearing methods,” Applied Energy, vol. 261,
decided by ECMs in LEM can differ. This has direct p. 114387, 2020.
impact on flexibility provisions in LFM, hence, it makes [7] A. Chapman, A. Fraser, L. Jones, H. Lovell, P. Scott, S. Thiebaux, and
G. Verbic, “Network congestion management: Experiences from bruny
sense for ECMs to better predict effective elements e.g. island using residential batteries,” IEEE Power and Energy Magazine,
upstream flexibility requests and set the flexibility initial vol. 19, no. 4, pp. 41–51, 2021.

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 12

[8] “Directive (eu) 2019/944 of the european parliament and of the council [29] G. Tsaousoglou, J. S. Giraldo, P. Pinson, and N. G. Paterakis, “Mech-
of 5 june 2019 on common rules for the internal market for elec- anism design for fair and efficient dso flexibility markets,” IEEE
tricity and amending directive 2012/27/eu,” https://eur-lex.europa.eu/ transactions on smart grid, vol. 12, no. 3, pp. 2249–2260, 2021.
legal-content/EN/TXT/?uri=CELEX%3A02019L0944-20220623, 2022. [30] T. Jiang, C. Wu, R. Zhang, X. Li, H. Chen, and G. Li, “Flexibility
[9] J. L. Crespo-Vazquez, T. AlSkaif, Á. M. González-Rueda, and clearing in joint energy and flexibility markets considering tso-dso
M. Gibescu, “A community-based energy market design using decentral- coordination,” IEEE Transactions on Smart Grid, 2022.
ized decision-making under uncertainty,” IEEE Transactions on Smart [31] Z. Guo, W. Wei, L. Chen, Z. Dong, and S. Mei, “Parametric distri-
Grid, vol. 12, no. 2, pp. 1782–1793, 2020. bution optimal power flow with variable renewable generation,” IEEE
[10] M. E. Honarmand, V. Hosseinnezhad, B. Hayes, and P. Siano, “Local Transactions on Power Systems, vol. 37, no. 3, pp. 1831–1841, 2021.
energy trading in future distribution systems,” Energies, vol. 14, no. 11, [32] A. Kargarian, J. Mohammadi, J. Guo, S. Chakrabarti, M. Barati, G. Hug,
p. 3110, 2021. S. Kar, and R. Baldick, “Toward distributed/decentralized dc optimal
[11] S. Cui, Y.-W. Wang, Y. Shi, and J.-W. Xiao, “Community energy power flow implementation in future electric power systems,” IEEE
cooperation with the presence of cheating behaviors,” IEEE Transactions Transactions on Smart Grid, vol. 9, no. 4, pp. 2574–2594, 2016.
on Smart Grid, vol. 12, no. 1, pp. 561–573, 2020. [33] D. K. Molzahn, F. Dörfler, H. Sandberg, S. H. Low, S. Chakrabarti,
[12] G. Tsaousoglou, N. Efthymiopoulos, P. Makris, and E. Varvarigos, R. Baldick, and J. Lavaei, “A survey of distributed optimization and
“Multistage energy management of coordinated smart buildings: A control algorithms for electric power systems,” IEEE Transactions on
multiagent markov decision process approach,” IEEE Transactions on Smart Grid, vol. 8, no. 6, pp. 2941–2962, 2017.
Smart Grid, 2022. [34] D. Gebbran, S. Mhanna, Y. Ma, A. C. Chapman, and G. Verbič, “Fair
[13] S. Talari, M. Yazdaninejad, and M.-R. Haghifam, “Stochastic-based coordination of distributed energy resources with volt-var control and
scheduling of the microgrid operation including wind turbines, photo- pv curtailment,” Applied Energy, vol. 286, p. 116546, 2021.
voltaic cells, energy storages and responsive loads,” IET Generation, [35] S. Mhanna, G. Verbič, and A. C. Chapman, “Adaptive admm for dis-
Transmission & Distribution, vol. 9, no. 12, pp. 1498–1509, 2015. tributed ac optimal power flow,” IEEE Transactions on Power Systems,
[14] G. De Zotti, S. A. Pourmousavi, J. M. Morales, H. Madsen, and N. K. vol. 34, no. 3, pp. 2025–2035, 2018.
Poulsen, “A control-based method to meet tso and dso ancillary services [36] J. Levin, “General equilibrium,” Microecon. Notes, 2006.
needs by flexible end-users,” IEEE Transactions on Power Systems, [37] N. R. Pérez, J. M. Domingo, G. L. López, J. P. C. Ávila, F. Bosco,
vol. 35, no. 3, pp. 1868–1880, 2019. V. Croce, K. Kukk, M. Uslar, C. Madina, and M. Santos-Mugica, “Ict
[15] Y. Ye, D. Papadaskalopoulos, Q. Yuan, Y. Tang, and G. Strbac, “Multi- architectures for tso-dso coordination and data exchange: a european
agent deep reinforcement learning for coordinated energy trading and perspective,” IEEE Transactions on Smart Grid, 2022.
flexibility services provision in local electricity markets,” IEEE Trans- [38] A. C. Rueda-Medina, J. F. Franco, M. J. Rider, A. Padilha-Feltrin, and
actions on Smart Grid, 2022. R. Romero, “A mixed-integer linear programming approach for optimal
[16] S. Schwarz, H. Flamme, G. Gürses-Tran, M. Cupelli, and A. Monti, type, size and allocation of distributed generation in radial distribution
“Agent-based power scheduling framework for interconnected local systems,” Electric power systems research, vol. 97, pp. 133–143, 2013.
energy communities incorporating dso objectives,” in IECON 2019-45th [39] S. Talari, M. Shafie-Khah, N. Mahmoudi, P. Siano, W. Wei, and
Annual Conference of the IEEE Industrial Electronics Society, vol. 1. J. P. Catalão, “Optimal management of demand response aggregators
IEEE, 2019, pp. 6642–6648. considering customers’ preferences within distribution networks,” IET
Generation, Transmission & Distribution, vol. 14, no. 23, pp. 5571–
[17] W. Alharbi and K. Bhattacharya, “Incentive design for flexibility pro-
5579, 2020.
visions from residential energy hubs in smart grid,” IEEE Transactions
[40] Y. Dvorkin, Y. Wang, H. Pandzic, and D. Kirschen, “Comparison of
on Smart Grid, vol. 12, no. 3, pp. 2113–2124, 2021.
scenario reduction techniques for the stochastic unit commitment,” in
[18] X. Jin, Y. Mu, H. Jia, Q. Wu, T. Jiang, M. Wang, X. Yu, and
2014 IEEE PES General Meeting— Conference & Exposition. IEEE,
Y. Lu, “Alleviation of overloads in transmission network: A multi-
2014, pp. 1–5.
level framework using the capability from active distribution network,”
[41] S. Boyd, N. Parikh, E. Chu, B. Peleato, J. Eckstein et al., “Distributed
International Journal of Electrical Power & Energy Systems, vol. 112,
optimization and statistical learning via the alternating direction method
pp. 232–251, 2019.
of multipliers,” Foundations and Trends® in Machine learning, vol. 3,
[19] A. Nouicer, A. M. Kehoe, J. Nysten, D. Fouquet, L. Hancher, and
no. 1, pp. 1–122, 2011.
L. Meeus, The EU clean energy package (ed. 2020). European
[42] S. Talari and S. Birk, “Mv-lv-grid-data and scripts for clearing algo-
University Institute, 2020.
rithm,” https://github.com/Pyosch/LEM LFM Paper, 2022.
[20] P. Olivella-Rosell, F. Rullan, P. Lloret-Gallego, E. Prieto-Araujo, [43] S. Meinecke, N. Bornhorst, L. P. Lauven, J. H. Menke, M. Braun,
R. Ferrer-San-José, S. Barja-Martinez, S. Bjarghov, V. Lakshmanan, and S. Drauz, “SimBench - Elektrische Benchmarknetzmodelle,”
A. Hentunen, J. Forsström et al., “Centralised and distributed optimiza- Universität Kassel, Fraunhofer IEE, RWTH Aachen University,
tion for aggregated flexibility services provision,” IEEE Transactions on Technische Universität Dortmund, Tech. Rep., 2020. [Online].
Smart Grid, vol. 11, no. 4, pp. 3257–3269, 2020. Available: https://simbench.de/wp-content/uploads/2021/09/simbench
[21] A. Sanjab, H. Le Cadre, and Y. Mou, “Tso-dsos stable cost allocation documentation de 1.1.0.pdf
for the joint procurement of flexibility: A cooperative game approach,”
IEEE Transactions on Smart Grid, pp. 1–1, 2022.
[22] E. Spot, “enera project: Ewe and epex spot to create local market
platform to relieve grid congestions,” Paris, Feb, 2018.
[23] C. Antal, T. Cioara, M. Antal, V. Mihailescu, D. Mitrea, I. Anghel,
I. Salomie, G. Raveduto, M. Bertoncini, V. Croce et al., “Blockchain
based decentralized local energy flexibility market,” Energy Reports, Saber Talari (S’17, M’20) is a team leader and
vol. 7, pp. 5269–5288, 2021. lecturer at the Faculty of Management, Economics,
[24] G. Kara, P. Pisciella, A. Tomasgard, H. Farahmand, and P. C. del and Social Sciences of University of Cologne (UoC),
Granado, “Stochastic local flexibility market design, bidding, and dis- Germany. He works on sustainable energy systems
patch for distribution grid operations,” Energy, vol. 253, p. 123989, with a focus on digitalization of energy sector, de-
2022. carbonization and algorithmic marketplaces for local
[25] A. Alarcón Cobacho, E. Prat, D. Vázquez Pombo, and S. Chatzi- energy trading. He holds a PhD in electrical and
vasileiadis, “Auction-based vs continuous clearing in local flexibility computer engineering from Portugal and has done
markets with block bids,” arXiv e-prints, pp. arXiv–2110, 2021. his PostDoc at UoC, Germany in energy economics.
[26] M. Rayati, M. Bozorg, R. Cherkaoui, and M. Carpita, “Distributionally He also gained industrial experience while working
robust chance constrained optimization for providing flexibility in an at Fraunhofer institute for energy economics and
active distribution network,” IEEE Transactions on Smart Grid, 2022. energy system technology as a project manager. He won Energies best paper
[27] K. Oikonomou, M. Parvania, and R. Khatami, “Deliverable energy flex- award in 2017 and currently serves as Associate Editor of IET, Generation,
ibility scheduling for active distribution networks,” IEEE Transactions Transmission, and Distribution.
on Smart Grid, vol. 11, no. 1, pp. 655–664, 2019.
[28] C. Heinrich, C. Ziras, T. V. Jensen, H. W. Bindner, and J. Kazempour,
“A local flexibility market mechanism with capacity limitation services,”
Energy Policy, vol. 156, p. 112335, 2021.

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Transactions on Smart Grid. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TSG.2023.3276024

IEEE TRANSACTIONS ON SMART GRID 13

Sascha Birk (S’22) received the B.Eng. degree in

electrical engineering from the Baden-Württemberg
Cooperative State University, Mannheim, in 2013
and the M.Sc. in renewable energy from the Univer-
sity of Applied Sciences Cologne, both in Germany.
He is currently working as a research assistant at the
University of Applied Sciences Cologne with the vir-
tual institute smart energy, while pursuing his Ph.D.
degree in information systems at the University of
Cologne, Germany. His research interest includes
the modeling of distributed energy resources, the
integration and optimization of such in distribution grids.

Wolfgang Ketter is Chaired Professor of Informa-

tion Systems for Sustainable Society at the Faculty
of Management, Economics, and Social Sciences at
the University of Cologne, Germany. In addition,
he is the coordinator of the Key Research Initia-
tive “Sustainable and Smart Energy and Mobility”
where research he leads focuses on how digital
transformation can create a faster and more stable
transition to sustainable energy and mobility. He also
is professor of Next Generation Information Systems
at the Department of Technology and Operations
Management, and Director of the Erasmus Centre for Future Energy Business
at the Rotterdam School of Management, Erasmus University. He has served
as Editor for Information Systems Research and MIS Quarterly. Since 2017
he is advisor to the German government on energy policy and since 2018 a
fellow of the World Economic Forum Global Future Council on Mobility. He
has won many prestigious awards, such as the AIS Impact Award 2020, the
INFORMS Wagner Prize Finalists Award 2020, the INFORMS ISR Best Paper
Award for 2020, the INFORMS Information Systems Society - Design Science
Award (2012 and 2021), the AIS Ciborra Award 2022, and the INFORMS
Information Systems Society - Practical Impacts Award 2022.

Torsten Schneiders is working as a professor for

energy storage and renewable energy systems at the
Cologne Institute for Renewable Energy. Since 2015,
he also is the Technical Director of the Virtual Insti-
tute for Smart Energy, an interdisciplinary research
platform for the digitization of the energy sector.

© 2023 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: NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY. Downloaded on May 16,2023 at 06:30:42 UTC from IEEE Xplore. Restrictions apply.