1866 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 14, NO.

3, JULY 2023

Fairness-Aware Distributed Energy Coordination for

Voltage Regulation in Power Distribution Systems
Shiva Poudel , Member, IEEE, Monish Mukherjee , Member, IEEE, Rabayet Sadnan , Student Member, IEEE,
and Andrew P. Reiman , Member, IEEE

Abstract—The accelerating deployment of solar photovoltaics SDA Switch Delimited Areas.

into low-voltage distribution networks can cause reverse power UA/DA Upstream Area / Downstream Area.
flow and overvoltage problems. However, if coordinated properly, B. Sets
the real and reactive power flexibility of these resources enables
distribution operators to manage their networks more efficiently. E/V Set of edges/buses in a graph, G.
Existing literature is rich in droop-based control (Volt-Watt and VT Set of buses with split phase transformers.
Volt-VAr) and optimization-based distributed energy coordination K Set of areas in a distribution network.
for four-quadrant control of photovoltaics to prevent overvoltage
issues. While optimal coordination can effectively mitigate over-
H Set of buses with DERs in a network.
voltage, it treats resources at sensitive parts of the grid unfairly. To ES Set of switches in a distribution network.
address this concern, we propose a distributed optimal power flow x Set of network variables {Pij , Qij , vj }.
formulation that incorporates fairness in curtailing photovoltaic y Set of DERs operational variables {ppv , q pv }.
generation and utilizes the reactive power capability of smart Ek /Vk Set of edges/buses in an area k.
inverters. Fair curtailment of photovoltaic systems is studied with
aggregation at each of two layers in a distribution network: 1)
Hk Set of buses with DERs in an area k.
area-level fairness and 2) feeder-level fairness to demonstrate the Φj Set of phases of a bus j.
flexibility of the proposed approach in resource aggregation. To ex- Φij Set of phases of a line ij.
plore the trade-off, the fairness-aware control actions are compared
against the performance of a centralized controller that aims to C. Variables
maximize the aggregate PV generation without incorporating fair- αi Curtailment factor for PV at bus i.
ness. Simulation results show that introducing area-level fairness αareak Fair curtailment factor of areak .
increased curtailment by 1.11 percentage points, and feeder-level
fairness increased curtailment by 4.7 percentage points compared Pijφ , Qφij Real and reactive power flow for phase φ ∈ Φij
to a fairness-agnostic control. in branch ij.
pv pv
pφj , qjφ Real and reactive power output of a PV inverter
Index Terms—Distributed algorithms, photovoltaic systems,
power distribution, quadratic programming, voltage control.
at bus j for phase φ ∈ Φ.
vjφ Voltage magnitude square for φ phase of bus j,
φ ∈ Φj .
NOMENCLATURE
D. Parameters
Load
A. Acronyms pφj Real demand for φ phase of bus j, φ ∈ Φj .
Load
ADMM Alternating Direction Method of Multipliers. qjφ Reactive demand for φ phase of bus j, φ ∈ Φj .
ADMS Advanced Distribution Management Systems. pj,avail , spv
pv
j Inverter rating and generation for PV at bus j.
DEC Distributed Energy Coordination. rψφ /x ψφ
Resistance/Reactance value between phase ψ and
ij ij
DER Distributed Energy Resources. phase φ of buses i and j respectively.
OPF Optimal Power Flow. v/v Minimum/maximum voltage magnitude square.
PV Photovoltaic.
QP Quadratic Programming. I. INTRODUCTION
A. Background
Manuscript received 1 October 2022; revised 27 January 2023; accepted 26
February 2023. Date of publication 7 March 2023; date of current version 21 N RECENT years, reduced prices, increasing environmental
June 2023. This work was supported by the U.S. Department of Energy, Office of
Electricity, Advanced Grid Research and Develop Program. Pacific Northwest
National Laboratory is operated for DOE by the Battelle Memorial Institute un-
I awareness, and financial incentives have led to unprece-
dented growth of rooftop PV [1]. Furthermore, with the devel-
der Grant DE-AC05-76RL01830. Paper no. TSTE-01006-2022. (Corresponding opment of new approaches for affordable electricity, policies
author: Shiva Poudel.)
The authors are with the Pacific Northwest National Laboratory, Richland, WA for community solar programs are being enabled [2]. These
99354 USA (e-mail: [email protected]; [email protected]; new PV resources, if coordinated properly, can help distribution
[email protected]; [email protected]). system operators improve network performance and offer a cost-
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TSTE.2023.3252944. effective alternative to network upgrades [3]. However, leaving
Digital Object Identifier 10.1109/TSTE.2023.3252944 these resources uncoordinated in the hands of different entities
1949-3029 © 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: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 POUDEL et al.: FAIRNESS-AWARE DISTRIBUTED ENERGY COORDINATION FOR VOLTAGE REGULATION 1867

without grid visibility can create significant issues like voltage apparent power ratings; therefore, the reactive power alone
violation, power quality degradation, and network congestion cannot always yield sufficient voltage regulations when the PV
due to reverse power flows during high PV generation and low power output is high [34]. Hence, some studies have adopted
consumption [4]. Among these, voltage rise due to reversed active power curtailment [35], or a combination of active power
power flow has often been a limiting factor for the growth of curtailment and reactive power utilization of DERs [11], [12].
DERs in distribution networks [5]. However, implementing active power curtailment without adopt-
ing any fair distribution of the corresponding curtailment burden
can effectively impose location-based penalization; the curtail-
B. Related Work ment tends to increase with electrical distance to the nearest
To mitigate this challenge, state-of-the-art studies have pro- substation [36]. Such an unequal share of the burden, though
posed several solutions. A straightforward approach is network optimal from the physics of power flow, may create unfair op-
upgrades (increase conductor size, reduce line impedances) [7]. portunities among customers, thus calling for research on fair PV
However, this requires expensive investments, especially for curtailment strategies. Furthermore, it is important to consider
underground feeders. Another solution involves using tap- fairness across different layers of the distribution network given
changing transformers and capacitor banks [16], [17]. The the flexibility in resource aggregation, such as an aggregator or
main limitation of this technique is that legacy devices cannot a distribution system operator [37].
respond quickly to the rapid fluctuation due to mechanical To incorporate fairness in curtailment, different approaches
limitations [18]. In addition, without proper coordination, the have been proposed, including droop-based curtailment [6], [7],
interaction between legacy devices and PV inverter operation [8], [15] and OPF-based formulations [10], [13], [14], [38].
may lead to an excessive operation of the regulating devices in A model-free distributed control is proposed in [6] where PV
the system [19]. A third approach focuses on utilizing the active inverters switch between different operational modes based on
and reactive power flexibility of DERs for voltage regulation centralized control signals to mitigate over-voltage. However,
by designing various control mechanisms. Although this is a the approach leads to unnecessary curtailments at times and
promising solution for the future smart grid, designing coordi- only applies to a certain lateral or feeder portion (i.e., no support
nation strategies remains challenging [20]. between different grid segments). Similarly, reference [7] uses
The coordinating strategies proposed in existing literature can different droop coefficients for inverters to share the curtailment;
be categorized as decentralized, centralized, or distributed based however, it requires a central controller to compute the voltage
on their control architecture. Local responsive resources such as sensitivity of the system. In [8], authors proposed a droop-based
storage devices and flexible demands can be used for voltage method that incorporates fairness, but the computational com-
control [21], [22], [23], [24], [25]. While energy storage devices plexity increases with the number of DER inverters, as every
and demand response schemes can help alleviate over-voltage controller has to be designed separately. An adaptive real cap-
problem, they have limited capacity [26], [27]. Additionally, the ping method using a consensus controller is proposed in [15] for
cost benefit ratio can be low if storage devices have to be sized fair PV curtailment; however, such methods are not applicable
to store the surplus energy from renewable resources [7], [25]. if resources are dispersed across different territory owned by
Furthermore, localized approaches fail to optimally utilize all separate entities [39]. Reference [38] proposed a multi-objective
available resources due to a lack of information exchange and optimization problem to minimize line power losses and unfair-
network observability. On the other hand, centralized schemes ness, however, it doesn’t strictly distribute the curtailment fairly.
such as [28], [29] focus on optimally allocating resources for This limitation is addressed in [10], where different OPF-based
voltage control. These approaches involves solving large com- schemes are studied; however, the proposed formulation is cen-
plex problems, limiting their scalability, and also require perfect tralized without coordination among DERs. Most of the existing
system information with direct access to all assets. This can fairness schemes are centralized in nature and lack a generalized
be challenging because of: 1) the limited and low-bandwidth approach of ensuring fairness. With the rapid proliferation of
communication infrastructure employed in existing distribution DERs in the distribution grid, it is challenging to manage such
networks [30] and 2) potential data unavailability due to privacy a multi-resource network by a single central controller [40].
matters or lack of metering infrastructure [31]. To address these Towards the goal of enabling fairness through collaborative
challenges, distributed coordination strategies have emerged decision-making, some recent studies have proposed distributed
as alternatives that distribute the coordination problem into approaches for fairness curtailment [13], [14]; however, they
computationally simpler subproblems and enable information are implemented on simplified use-cases (single-phase feeders
exchange among a limited set of interacting agents [32]. Such and single aggregator) and lack a generalized approach for
distributed decision-making algorithms are computationally ef- enabling fairness across different resource aggregation-levels.
ficient and pose the potential to respect the privacy of data, Table I summarizes the related literature on PV curtailment
measurements, and constraints. where the capabilities and limitations of different approaches
Existing work has proposed several distributed coordination are highlighted.
strategies for voltage regulation. Some studies focus on utiliz-
ing the reactive power flexibility of DERs during overvoltage
C. Contributions
periods [30], [33]. Although PVs equipped with smart invert-
ers can adjust their reactive power consumption for voltage This paper proposes a distributed coordination strategy that
regulation, their reactive power capability is bounded by incorporates fairness to solve the voltage regulation problem in
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 1868 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 14, NO. 3, JULY 2023

TABLE I
SUMMARY OF LITERATURE REVIEW ON PV CURTAILMENT FOR OVERVOLTAGE MITIGATION

a collaborative decision-making environment. This paper aims modified IEEE 123-bus test feeder in Section VI. A few realistic
to bridge the existing gaps in energy coordination literature scenarios for implementation are simulated and discussed in
by developing a scalable OPF formulation for implementing Section VII. Finally, Section VIII concludes the paper.
fairness in a distributed manner. The specific contributions of
this paper are summarized below. II. FAIRNESS-AWARE DISTRIBUTED ENERGY COORDINATION
1) OPF-Based Formulation: A novel distributed OPF for- (FA-DEC)
mulation is proposed for solving the fairness-aware distributed
energy coordination problem. The proposed approach is a Distribution feeders and segments that are built to traditional
generic framework that enables the implementation of fairness specifications tend to have distributed energy resource (DER)
schemes at varying levels of resource aggregation; For example, hosting capacities that are less than the total amount of customer
fairness could be considered among resources: i) per distribution load. Unmitigated, this physical reality results in the inherently-
transformer, ii) within an SDA,1 iii) within a feeder, etc.). unfair situation where not all customers can be guaranteed
A three-phase unbalanced power distribution test feeder with equal access to the benefits of deploying DERs. Policies and
secondary models (service transformer and triplex nodes) is used controls designed to mitigate the problems associated with high
to demonstrate the proposed formulation. concentrations of DERs tend to exacerbate inequities unless fair-
2) Flexible Resource Aggregation: Two different resource ness is considered explicitly. For example, policies that require
aggregation levels are presented to demonstrate the capability customers to mitigate the marginal impacts of new DERs create
of the proposed framework in scalable resource aggregation. advantages for well-resourced early adopters (especially if early
Fairness is implemented across i) a switch-delimited area where adoption is or ever was incentivized); and optimal power flow
resources within the area are curtailed fairly and ii) the feeder algorithms that are permitted to dispatch customer resources will
where area agents share their burden to achieve fairness among tend to dispatch resources at the ends of long lines more than
all resources within the feeder. resources along the robust trunk line.
3) Trade-off Analysis: A detailed performance comparison Fair and equitable access to DER benefits can be pursued
is carried out to highlight the cost of catering for fairness in PV using a combination of policies and controls. In this work,
curtailment. A fairness scheme studied at different aggregation we focus on fairness in controls, specifically optimal energy
levels is compared against a fairness-agnostic control (FAC) dispatch. Different stakeholders may have different perspectives
approach that maximizes aggregate PV generation. on fairness. A non-exhaustive list of ways to apportioning the
burden of mitigating the impacts of DER to increase hosting
capacity includes 1) evenly, 2) proportionally, or 3) according
D. Paper Organization
to degree of responsibility for the impacts. Each of these can
The rest of the paper is organized as follows: Section II be measured and expressed as an objective function for opti-
discusses the background and motivation for fairness-aware mization purposes. In this work, we study the second method
distributed energy coordination (FA-DEC). Section III describes (proportional burden sharing).
the appropriate architecture for solving the FA-DEC problem. Achieving fairness according to any of the definitions listed
Section IV formulates the proposed distributed approach for comes at a cost of efficiency. In Section VI, we compare fairness
solving a fairness-aware coordination problem. The problem and efficiency between fairness-aware and fairness-agnostic
decomposition for distributed energy coordination is detailed controls. We also examine the effects of ensuring fairness at dif-
in Section V. The proposed approach is demonstrated using a ferent layers in the distribution system. Specifically, we consider
fairness at the feeder-section level as it might be implemented
1 A switch-delimited area represents a section of the distribution network that
at the neighborhood or community level (e.g., by a non-utility
can be isolated by opening a set of switches [41]. Such an area can represent a aggregator) as well as fairness at the feeder level as it might be
community or neighborhood coordinating via a distributed agent. implemented by a distribution system operator.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 POUDEL et al.: FAIRNESS-AWARE DISTRIBUTED ENERGY COORDINATION FOR VOLTAGE REGULATION 1869

of such actions and updates the UA/DA relation among layer-2

agents accordingly. Establishing such a relationship is crucial to
the system, as the convergence of the distributed coordination is
fundamentally dependent on this UA/DA relation.
2) Layer-2: Distributed Agents: The layer-2 agents, i.e., dis-
tributed computing agents, are physically located on the feeder
(e.g., at a smart switch), and each layer-2 agent is responsi-
ble for the control and optimal operations of their respective
neighborhood area. These layer-2 agents also communicate
with other layer-2 agents using peer-to-peer communication to
carry out the distributed energy coordination algorithm. Specif-
ically, layer-2 agents solve the local problem by leveraging the
Fig. 1. Distributed Architecture for FA-DEC. area-specific network model as obtained from central ADMS
and local-resource information (PVs, loads etc.) within the
given area. These distributed agents have access to the detailed
III. DISTRIBUTED ARCHITECTURE FOR FA-DEC regional/switch-level network model with individual customer
The proliferation of controllable grid-edge resources such information located within the given neighborhood and the
as DGs, smart meters, and controllable demand-side assets coupling variables of the shared boundary from the neighboring
adds to the complexity of distribution systems operations [42]. layer-2 agents (Fig. 1). the layer-2 distributed agents use this
Expensive communication and data processing infrastructure is information and communicate with neighboring layer-2 agents
needed to gather and process all the information at the cen- to obtain optimal control actions for PVs within their respective
tralized control center required by any model-based algorithm. areas.
To avoid such complexities, instead, local federated informa- 3) Layer-3: Device Agents: The Level-3 agents are the
tion can be shared among distributed agents at the grid edge device-level controllers located at each controllable bus and
to complement and enhance operations. This can be achieved communicate only with their respective Level-2 agents to imple-
through a layered-decomposition based architecture – capable ment the device-level control actions. They do not have access to
of hosting distributed optimization and controlling physical any network model. For this application, Level-2 agents decide
infrastructure systems as detailed in [43]. Such frameworks can the PV curtailment factor and reactive power absorption which
provide (i) extensibility, (ii) boundary deference, (iii) structural are communicated to individual device-level controllers that
scalability, and (iv) securability, and enables a power distribution execute the device-level control actions.
network to host a general distributed optimal energy coordina-
tion algorithm. This work leverages a layered architecture for B. FA-DEC Using Layered Architecture
decision-support that synergistically combines distributed, and The layered control architecture for the distribution system,
edge-control paradigms for voltage regulation augmented with illustrated in Fig. 1, enables the hosting of the proposed FA-DEC
the fairness dimensions. method. Specific to the problem, the central ADMS at layer-1
first directs the interconnection relation among the SDAs us-
A. Three-Layer Architecture ing the current operation topology of the distribution system.
This establishes the message passing protocols, i.e., sharing
The layered architecture combines a centralized ADMS, dis-
appropriate coupling variables with appropriate neighbors of
tributed decision-making agents, and local controllers to coordi-
the layer-2 agents. In FA-DEC, the overall coordination problem
nate PVs. Specifically, the architecture has three layers of agents
is decomposed into several sub-problems associated with each
that can compute and coordinate PVs’ flexibility to achieve
SDA which solved by the corresponding layer-2 agents using
network-level objectives.
local information and shared coupling variables (see Fig. 1). The
1) Layer-1: ADMS: The first layer of the distributed layered
coupling variables include physics-based network variables (like
architecture is a central ADMS that (i) maintains an accurate
power flow and voltage) at the shared buses and fairness param-
and updated network model and (ii) provides the relevant model
eters shared based on the level of enabling fairness. All agents
information such as lines, transformers, etc. to agents in layer-2
iteratively solves their sub-problems until consensus among the
(see Fig. 1). It also defines the relationships between different
shared-network variables and fairness parameters is reached and
layer-2 agents (i.e., distributed agents), dictating the message’s
the resulting set-points are then dispatched by the device agents
content and the direction of the message passing during the dis-
to execute the optimal operation for the distribution system.
tributed coordination. Specifically, the layer-1 agent establishes
the interconnections between layer-2 agents of upstream and
downstream areas (UA & DA). Note that the ADMS controller IV. FAIRNESS-BASED DER COORDINATION PROBLEM
doesn’t actively participate in solving the coordination problem The DER coordination approach presented in this work is
in an operational environment. However, if the distribution formulated as an OPF problem. The coordination strategy aims
network is reconfigured (changing the open/closed status of to minimize the curtailment of DERs across the fleet by co-
sectionalizing and tie switches), the layer-1 agent keeps track ordinating the active and reactive power availability of DERs

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 1870 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 14, NO. 3, JULY 2023

whenever a voltage violation is observed or anticipated. The

objective function of our problem is given by (1).

min F = f (x, y) + αj (x, y); αj ∈ [0, 1] (1)
x,y
j∈H

where x is the set of network variables (active and reactive

power flows, voltages), y is the set of operational variables for
DERs (active and reactive power outputs). As the study aims
to implement proportional burden sharing among customer, a Fig. 2. Modeling service transformer serving loads via a triplex secondary.
curtailment factor variable (αj ) is introduced that indicates the
actual curtailment of the PV system connected to bus j, as
a proportion of its available output (ppv j,avail ) and the actual voltage drop in a line segment can be represented as:
dispatch of the DER (ppv
j ) is given by (2). Alongside the fairness
objective, other system objectives such as minimizing loss, viφ = vjφ − Hij
P φ Q φ
Pij − Hij Qij , ∀(ij) ∈ E, ∀φ ∈ Φij (5)
operational cost can be included via f (x, y). Since, this study
focuses on enabling fairness, the network objective has been where viφ = |Viφ |2 is the square of voltage magnitude at bus i for
simply modeled to ensure feasible operation (f (x, y) = 0). Q Q
phase φ and Hij and Hij are the matrices for active and reactive
ppv
j,curt. power flows, respectively on the line ij.
ppv pv
j = (1 − αj ) pj,avail where αj = ∀j ∈ H (2)
ppv
j,avail ⎡ √ ab √ ⎤
−2rij
aa
rij
ab
− 3xij rij
ac
+ 3xac
ij
⎢ √ √ ⎥
A. Distribution System Model Hij
P
=⎢
⎣rij + 3xij
ba ba
−2rij
bb
rij
bc
− 3xbc ⎥
ij ⎦ (6a)
√ √
Power distribution networks are unbalanced due to unequal rij
ca
− 3xcaij rij
cb
+ 3xcbij −2rij
cc
single-phase loads and non-equilateral spacing of three-phase ⎡ √ √ ac ⎤
overhead and underground lines [44]. Therefore it is imperative −2xaa
ij xab
ij + 3rij
ab
xac
ij − 3rij
⎢ √ ba √ bc ⎥
to model the coupling effects across multiple phases. The power Q
Hij =⎢ x
⎣ ij
ba
− 3rij −2xbb
ij xij + 3rij ⎥
bc
⎦ (6b)
flow model used to describe unbalanced three-phase systems √ √ cb
distribution systems is presented as follows. xca
ij + 3rij
ca
xij − 3rij
cb
−2xcc
ij
1) Power Flow Constraints: Let a radial distribution network
be represented by a directed graph, where G = (V, E); the set ψφ
where rij and xψφij are the resistance and reactance values
V represents buses, and the set E, represents edges associated between phase ψ of bus i and phase φ of bus j of a branch (ij).
with the branches. For a multi-phase distribution network, the The system voltages are to be maintained within the acceptable
set of available phases at bus i and branch ij are represented ANSI limits as shown in (7).
within the sets Φj ∈ {a, b, c} and Φij ∈ {a, b, c}, respectively.
By ignoring losses, and voltage angle approximation [45], the v ≤ vjφ ≤, v ∀j ∈ V, ∀φ ∈ Φj . (7)
branch flow model is decomposed into real and imaginary terms
of phase quantities and can be represented by the linear equations The linearized power flow equations are a function of vjφ ; there-
as shown in (3) and (5).
fore, v and v are set to 0.952 and 1.052 , respectively. Further-
 φ  φ
Pij = pφj + Pjk , ∀j ∈ V, ∀φ ∈ Φj (3a) more, the equipment models such as voltage-dependent loads,
i:i→j k:j→k voltage regulators, and capacitor banks need to be parameterized
  based on respective control variables and the vjφ [46].
Qφij = qjφ + Qφjk , ∀j ∈ V, ∀φ ∈ Φj (3b) 2) Modeling Service Transformer: The standard method of
i:i→j k:j→k providing service to a customer is from a center-tapped single-
Load pv Load pv phase transformer. Customers’ load and PV power injected at
pφj = pφj − pφj , qjφ = qjφ − qjφ (4) their point of interconnection cause power flow across a ser-
vice transformer. Hence, it is important to formulate the power
where Pijφ and Qφij denote the real and reactive power flow in
flow and voltage constraints for the service transformer in the
a branch ij, incoming at bus j, pφj and qjφ represents the net optimization problem. Such modeling capability can accurately
real and reactive power injections at j as given by (4). Note account for the per-phase contribution of customer loads and the
Load Load
that pφj + jqjφ represents the static load model where PV systems. The model of a center-tapped transformer is shown
the power does not vary with changes in voltage. However, the in Fig. 2. The impedances Z0 , Z1 , and Z2 represent the individual
voltage-dependent load models, which are normally modeled winding impedances and can be determined using the per-unit
by the composite of constant impedance (Z), constant current impedance based on transformer rating. Suppose rw represents
(I), and constant power (P and Q) can be incorporated into the the per unit resistance for winding w, and xwv represents the per
formulation. The impact of such model is discussed in Section unit reactance (leakage impedance) between winding w and v;
VII-B using a linearized ZIP model discussed in Appendix. The the winding impedance can be derived using the following set

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 POUDEL et al.: FAIRNESS-AWARE DISTRIBUTED ENERGY COORDINATION FOR VOLTAGE REGULATION 1871

of equations,
Z0 = r1 + 0.5 (x12 + x13 − x23 )

Z1 = r2 + 0.5 (x12 + x23 − x13 )

Z2 = r3 + 0.5 (x13 + x23 − x12 ) (8)

The following equations represent the power flow in a service
transformer.
 ψ
Pijφ = pψ
j + Pjk , ∀i ∈ VT , ∀φ ∈ Φj , ∀ψ ∈ Φs (9a)
k:j→k Fig. 3. Representation of polynomial expression for quadratic constraints
 relating to the apparent power limits of PV inverters.
Qφij = qjψ + ψ
qjk , ∀i ∈ VT , ∀φ ∈ Φj , ∀ψ ∈ Φs (9b)
k:j→k

where, Φs ∈ {s1 , s2 } represents a set of phases in triplex nodes. inequality can alone provide the reactive power capability of
The impedance from (8) and flows from (9) are used to calculate the inverter.
the voltage drop across the transformer.
C. Enabling Fairness
B. PV Models
The proposed coordination approach to ensure feasible oper-
The proposed DER coordination approach is demonstrated
ation for minimal DER curtailments is given below.
using solar PVs (both residential & community-scale) as select
DERs. The available PV output (ppv 
j,avail ) is modeled through min F = αj (x, y) ∀j ∈ H
a typical solar profile and the actual dispatch (ppv j ) to maintain x∈{Pij
φ
ij ,vj ,αj },y∈{pj ,qj }
,Qφ φ pv pv
j
feasible operation is governed by αj as given in (2).
PVs with new generation smart inverters can contribute to subject to: (2)–(5), (7), (9), (11) (12)
voltage regulation (over-voltage) by utilizing their capability of
absorbing reactive power. However, the inverter is not a limitless The formulation in (12) aims at minimizing the net α’s for all
sink of reactive power; its reactive power is bounded by its PVs, making it agnostic to customer fairness. Hence directly
fixed apparent power rating. Let ppv pv
j and qj denote the real
implementing it would lead to higher curtailment for down-
and reactive power output of a smart inverter connected at bus j, stream customers, thereby depriving them of equal opportu-
then (10) defines the relation between active and reactive power nities to leverage incentives [36]. This case is referred to as
outputs with its inverter rating (spv
j ).
the fairness-agnostic control (FAC) scheme. To enable fairness
among customers, some additional fairness-based constraints
(ppv 2 pv 2 pv
j ) + (qj ) ≤ sj , ∀j ∈ H (10) are introduced as below:
1) Area-Level Fairness: To ensure fairness among customers
Note that φ is omitted here to describe the relation between in a particular neighborhood (areak ), the curtailment factors are
active and reactive power outputs with the apparent power rating; made equal among customers in an areak , given by (13).
the actual representation of real and reactive power of the PV
inverter in the formulation is shown in (4). Since both ppv j
αareak = α1k = α2k = · · · = αnk , where nk = |Hk | (13)
and qjpv are controllable, direct use of (10) is a non-convex
constraint not possible in our formulation. We adopt a polygonal 2) Feeder-Level Fairness: A system-wide fairness can be
relaxation [47] where the quadratic constraints in (10) can be achieved through a consensus of α for the PVs throughout the
replaced by the following set of linear inequalities. feeder, which can be modeled using (13) and (14).
√ √
− 3 (ppv pv pv pv
j + sj ) ≤ qj ≤ − 3 (pj − sj ),
pv
αarea1 = αarea2 · · · = αareak , where k = |K| (14)
√ √
− 3/2 spv pv
j ≤ qj ≤ 3/2 spv
j , These constraints are included in the problem formulation
√ √ when solving the fairness-aware optimization problem. Note that
3 (ppv pv pv
j − s j ) ≤ qj ≤ 3 (ppv pv
j + sj ), ∀j ∈ H. (11)
the curtailment factor is augmented with the available PV power
As the inverter is responsible for absorbing reactive power during to determine the PV system output as shown in (2). Area-level
overvoltage, only the right-hand side of these inequalities can be fairness is directly included as an equality constraint ensuring
formulated in the optimization problem, i.e., restricting the PV fair curtailment in a given area. On the other hand, feeder-level
system to act only as an inductive generator. Fig. 3 represents fairness is incorporated as a penalty in the objective function
how the linear inequalities can approximate the actual inverter of each area thus minimizing the difference in curtailment and
reactive power capability (black curve). Since the over-voltage ensuring fairness across the network. This is explained in detail
is usually experienced during peak PV generation, the first in Section V while decomposing the problem.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 1872 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 14, NO. 3, JULY 2023

Fig. 4 demonstrates how the proposed problem can be dis-

tributed between two coupled areas by decomposing the con-
straints associated with the central coordination problem, given
in (12), for each area based on the area-specific set of variables
(xk , yk ). Some of the variables are coupled and would appear
in the variable set for multiple areas. The general approach
for ADMM-based solutions involves duplicating these variables
by assigning one copy for each area and enabling a consensus
between these copies as given by (15).
Piφk jl = Piφk jl , ∀ik : ik → jl , {ik ∈ Vk , jl ∈ Vl } (15a)

Qφik jl = Qφik jl , ∀ik : ik → jl , {ik ∈ Vk , jl ∈ Vl } (15b)

viφk = vjφm , ∀ik : jm → ik , {ik ∈ Vk , jm ∈ Vm } (15c)

Fig. 4. Area Decomposition of a network with coupling variables.
α areal
=α areal
, ∀areal ∈ K & areal = areak (15d)

where the left-hand terms {Piφk jl , Qφik jl , viφk , αareal } are copies
for the sub-problem for a particular areak and the right hand
V. DISTRIBUTED COORDINATION METHOD
side terms are corresponding values from its neighbouring areas
The proposed DER coordination problem (12) and along with (aream , areal ), such that aream is UA and areal is DA of
the constraint (13), (14) enables a fair sharing of curtailments areak . In order to enable feeder-level fairness, (15d) ensures
among customers on top of the reactive power availability. How- consensus among the curtailment factors for all the areas. This
ever, solving the problem through a centralized controller may presents a decomposable structure of the problem allowing us
cause several complications like (i) violation of customer privacy to re-formulate (12) as:
by sharing their energy consumption details with the central
coordinator, (ii) requirements of direct access to all customer min F + min F . . . + min F, where k = |K| (16a)
x1 ,y1 x2 ,y2 xk ,yk
assets along with complete system information, (iii) the central-
ized problem is computationally challenging as its comprises of subject to: (15) (16b)
a large number of variable requiring convex relaxations thereby where {xk , yk } are the set of network variables and DER oper-
limiting its scalability, and (iv) the problem is also prone to ational variables belonging to the areak . This decouples the set
single-point failures. The proposed coordinated scheme can be of common variables and enables solving (16a) in a distributed
distributed into subproblems to alleviate the requirement for a fashion when the coupling constraints in (16b) are relaxed. This
centralized controller. is achieved by formulating the Augmented (partial) Lagrange
function of the problem as given by (17).
L := F (x1 , y1 ) + F (x2 , y2 ). . . + F (xk , yk )
A. Problem Decomposition
|K|
The DER coordination problem, given by (12), can be solved   ρv φ
+ (v − vjφm )2 + λvk (viφk − vjφm )
in a distributed fashion by decomposing it into area-specific 2 ik
k ik :jm →ik
sub-problems and decoupling the common variables between
|K|
the areas. This work adopts the alternating direction method of   ρP φ
multipliers (ADMM), a popular augmented Lagrangian-based + (P −Piφk jl )2 +λP φ φ
k (Pik jl −Pik jl )
2 ik jl
approach to decomposing an optimization problem into several k ik :ik →jl

smaller sub-problems. In ADMM, the solutions to small local ρQ φ

sub-problems are coordinated to find a solution to a large global + (Qik jl − Qφik jl )2 + λQ φ φ
k (Qik jl − Qik jl )
2
problem. This technique is useful for a large-scale distribution
|K| |K|
grid where distributed agents can aggregate the resources and   ρα
utilize locally available information generated from distributed + (αareal −αareal )2 +λα
k (α
areal
−αareal )
2
sensing and monitoring for coordinated control. The distributed k l,l=k
algorithm with ADMM offers better convergence character- |K| |K|
 
istics [48] and solves large-scale optimization problems in a L := F (xk , yk ) + (LP Q
k + Lk + Lk + Lk )
v α
distribution grid to coordinate flexible resources to provide grid k k
services [49], [50]. In our work, a distributed agent defined at (17)
a switch-delimited area is agnostic to the parameters/models where ρx are the penalty parameter associated with each type
of the participating resources in other areas. Each distributed of coupling variables {Pijφ , Qφij , vjφ , αj }, and λxk are the dual
agent solves the local problem and communicates the coupling variables (Lagrangian multiplier) associated with the coupling
variables to meet the convergence condition. constraint for each areak . The ADMM approach utilizes this

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 POUDEL et al.: FAIRNESS-AWARE DISTRIBUTED ENERGY COORDINATION FOR VOLTAGE REGULATION 1873

Augmented Lagrangian, which presents a completely decom-

posable problem, and performs alternating minimizations over
the variable sets for each area. At each iteration γ, areak calcu-
lates the next set of variables denoted by xγ+1
k .

B. Oscillations Damping
The Augmented Lagrangian in (17) is a complex problem to
solve and if ill-conditioned may lead to oscillations between
iterations [13]. Hence, we introduce a damping parameter for
α that penalizes the inter-iteration oscillations as given by (18).
Since the DERs injections are governed by αareak , damping α
would also damp the oscillations in the coupling Pijφ , Qφij .

G(αareak ) = β ∗ (αareak − αareak ,γ ) (18)

C. ADMM Algorithm Details Fig. 5. Modified IEEE 123-bus test case with secondary models (service
transformer and triplex configuration) divided into five switch delimited areas
As the Lagrangian in (17) can be decomposed into area- representing five neighborhoods. A distributed agent controls each area and
specific subproblems, at a given iteration γ, the sub-problem accesses the area-specific network model and resources within the boundary.
for primal variables, given by (19), are solved individually for
each area to compute the next set of iterates γ + 1. TABLE II
SUMMARY OF THE MODIFIED IEEE 123-BUS TEST FEEDER
Q
min F + G + LP
k + Lk + Lk + Lk
v α
(19)
xk ,yk

The next set of iterates (γ + 1) for the coupling variables are

exchanged between areas and the corresponding values are up-
dated. Finally the dual variables for each area are updated based
on (20). The coordination steps are described in Algorithm 1
and the algorithm terminates when the residuals for the coupling
variables become really small.

λP,γ+1
k,ik := λ P,γ
k,ik + ρP
P φ,γ+1
ik jl − P φ,γ+1
ik jl (20a) Algorithm 1: Fairness-Based Distributed Coordination.

λQ,γ+1 := λQ,γ Qφ,γ+1 − Qφ,γ+1 1: Initialize: γ = 1, ρx,1 = ρ0 , λx,1 φ φ
k = 0, vik = vjm = 1.04
k,ik + ρ
Q
k,ik ik jl ik jl (20b)
2: while Stopping criteria is not met do
P/Q 3: Compute the primal variables for each areak
where λk,ik are the dual variables for active and reactive
(xγ+1 ) by solving (19) for each areak .
power flows coupled with an area k and are updated ∀ik : ik → k
4: Each areak agent communicates the network
jl , {ik ∈ Vk , jl ∈ Vl },
 coupling variables (Piφ,γ+1
k jl
, Qφ,γ+1 φ,γ+1
ik jl , vjm ) to
λv,γ+1
k,ik := λv,γ
k,ik + ρ
v
viφ,γ+1
k
− v φ,γ+1
j m
(20c) neighbouring areas (area , area ), and αareak ,γ+1
m l

to all areas.
where λvk,ik are the dual variables for voltages coupled with an 5: Each areak updates the next iterates of the
area k and are updated ∀ik : jm → ik , {ik ∈ Vk , jm ∈ Vm }, coupling variables.
 area ,γ+1  6: Each area updates its dual variables using (20).
λα,γ+1
k,l := λα l ,γ
k,l + ρ
α
α l
− αareal ,γ+1 (20d)
7: Increase γ by 1
where λαk,l are the dual variables for α’s coupled with areak and 8: end while
are updated ∀areal ∈ K & areal = areak .

VI. DEMONSTRATION are added across the network to create over-voltage. The test
feeder is divided into five topological areas, each representing
The proposed approach is demonstrated using the modified
a community or a neighborhood. The areas are separated by a
IEEE 123-bus test case [51]. The original IEEE 123-bus is
switch delimited section of the feeder, as shown in Fig. 5. Table II
modified to include the service transformers and triplex sec-
summarizes the resource allocation across each area in the test
ondaries. All the single-phase nominal loads are replaced with
feeder.
a secondary model, and houses are populated for each triplex
node (See Fig. 5). Rooftop PVs are randomly populated across
A. Simulation Setup
the residential loads. The base case deployment of rooftop PVs
is 75%, where 75% of residents have a PV system. Addition- The proposed fairness-aware distributed control is intended
ally, community-scale PVs ranging from 120 kW to 400 kW to be implemented as an online application that can run in a

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 1874 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 14, NO. 3, JULY 2023

curtailment. The fourth metric is defined as:


Rtareak
U areak
= ∀t∈Tcurr (21)
|Tcurr |

where Rtareak is the range of curtailment at a specific time

interval t for an areak . Tcurr is the set of time intervals where
curtailment is necessary.
1) Case 1–Feeder-Level Fairness: In this case, we aim to
achieve fairness across the feeder network, i.e., all PV curtail-
ments in proportion. The area agents impose fairness constraints
on the local problem and, in addition, also share their curtail-
ment with other agents for achieving feeder-level fairness. The
formulation in (12) is implemented with area-level fairness con-
straint as represented in (13). During problem decomposition,
in addition to the network variables (P , Q, and v), the feeder-
level fairness constraint (14) is incorporated as a penalty in the
Fig. 6. Simulation set up for the test case. Load and PV profile for a day (top) objective function to ensure consensus among the curtailment
and voltage profile of the feeder (down). The red dashed line in the voltage factors. Fig. 7 shows the simulation results when feeder-level
profile represents the upper ANSI limit. fairness is implemented. The agreement in curtailment factor
among all the areas is shown in Fig. 7(a) for a time step (t = 48).
At each interval, the proportional distribution spreads the PV
smart grid environment and mitigate overvoltage in a scenario curtailment across the resources in all areas in proportion to the
with a high PV penetration. For the demonstration purpose, a exported power (See Fig. 7(b)). The cumulative PV curtailments
15-min time interval is considered as a control cycle. The load across each area are shown by the difference in dashed and solid
and PV profile for a day and the corresponding voltage profile are lines in Fig. 7(c). Observe that the curtailment is proportionally
shown in Fig. 6. Clouds and shading were not modeled and all distributed across each area.
PV systems followed the same daily profile. Loads, which were 2) Case 2–Area-Level Fairness: In this case, proportional
uncontrolled, were also modeled with the same daily profile curtailment is ensured within each area. However, the burden
for simplicity. Fig. 6 shows the voltage profile of the feeder from one area is not shared with other areas. The formulation
for a day. Due to the coincidence of peak PV generation and in (12) is implemented only with area-level fairness constraint
lower consumption, the uncontrolled voltage rises beyond the (13). During problem decomposition, only the network vari-
ANSI limit. At the start of every control cycle, the distributed ables (P , Q, and v) are communicated across participating
agents solve the distributed OPF to determine the necessary PQ areas. Fig. 8 shows the implementation of area-level fairness.
dispatch for the PV inverters. Once calculated, the setpoints It is observed that each area converges to a different value
are dispatched to all PV inverters and maintained throughout of curtailment factor, as shown in Fig. 8(a) for a particular
the control cycle. The agents solve the distributed OPF using interval (t = 55). At each interval, the proportional distribution
updated measurements (ploadj + jqjload and ppv
j,avail ) to calculate
in area-level fairness spreads the PV curtailment across the
the new setpoints for the next control cycle. In this way, the PV resources in all areas in proportion to the exported power (See
setpoints are dispatched based on real-time measurements. If no Fig. 8(b)). While resources within the area share the burden
overvoltage is observed in the system, the distributed OPF will proportionally, the curtailment is distributed unfairly across
be a normal power flow without any control signal. The proposed the areas. For example, at t = 55, resources within area 5 are
quadratic program is solved using CVXPY [52], where we have curtailed by a factor of 0.4, whereas resources within area 1
used Python to implement the desired model. are curtailed by a factor of 0.02. The cumulative PV export
by each area is shown in Fig. 8(c). The unfair curtailment
across different areas can be observed with dashed and solid line
B. Simulation Results
differences.
This section presents simulation results for the operating sce- When implementing area-level fairness, there is a disparity
nario generated using the setup described earlier. Two different in curtailment across different grid segments. This is because
case studies are conducted to demonstrate the implementation of the electrically distant areas are penalized in a typical radial
the fairness scheme in different layers of a distribution grid, and distribution feeder. Therefore, area-level fairness is expected to
the results are compared against the fairness-agnostic control reduce the volume of PV curtailment compared to feeder-level
scheme. To evaluate the trade-offs, four relevant metrics are fairness, as the area close to the substation/feeder head does not
calculated: i) total energy curtailed in kWh (kWh↓), ii) the per- share the burden of distant areas. For example, area-1 and area-3
centage of energy curtailed (%↓), iii) the percentage reduction in experience very little curtailment as they are electrically closer
net revenue from the PV systems (Net $↓), and iv) the average to the substation. Therefore, area-level fairness is not fair from
range of curtailment (U areak ), a measure of unfairness in the a system perspective; however, it can ensure localized fairness

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 POUDEL et al.: FAIRNESS-AWARE DISTRIBUTED ENERGY COORDINATION FOR VOLTAGE REGULATION 1875

Fig. 7. Implementation of feeder-level fairness for PV curtailment. (a) Convergence of curtailment factors for five areas at t = 48, (b) Curtailment factors for five
areas across different time intervals, and (c) Total PV export after curtailment for five areas (with Q support).

Fig. 8. Implementation of area-level fairness for PV curtailment with Q support. (a) Convergence of curtailment factors for five areas at an interval, t = 55,
(b) Curtailment factors for five areas across different time intervals, and (c) Total PV export after curtailment for five areas.

TABLE III
COMPARISON OF ENERGY CURTAILMENT BETWEEN FAIRNESS-AGNOSTIC AND FAIRNESS-AWARE CURTAILMENT

through sharing the burden among resources within a sensitive as shown in Fig. 8(b). With this, there is increased flexibil-
area. ity to curtail more PV from areas with higher sensitivity to
3) Performance Comparison: In this section, the perfor- over-voltage. Therefore, the total energy curtailment percentage
mance of feeder-level and area-level fairness are discussed and decreases to 5.77%, i.e., 1247.8 kWh lower than feeder-level
compared against the fairness-agnostic control (FAC). fairness.
In Case 1, the fairness-based curtailment at the feeder-level FAC refers to solving the OPF-based scheme summarized in
aggregation leads to the average curtailment of 9.37% for each (12) without incorporating fairness constraints into the formu-
area. Since all PVs in the feeder are curtailed with the same lation. This scheme maximizes the aggregate PV generation;
proportion, the average range of curtailment (U areak ) is zero therefore, customers/areas close to the substation are expected
for this case. The reduction in net revenue is equal (7.8%) to be favored. Table III shows the performance comparison of
across different segments of the network, thus implying a proportional burden sharing across different levels of aggrega-
fair distribution and equal access to economic incentives. On tion with the FAC, where ↓ represents the reduction level due to
the other hand, in case 2, the energy curtailment percentage curtailments. In fairness-agnostic control, the OPF formulation
varies between 16.76% and 0.30% among five areas. Simi- aims to minimize the total volume of curtailment across the
larly, the net reduction in revenue varies between 13.59% and feeder without considering fairness. Since no fairness is incor-
0.25%. While fairness is maintained among PVs within the porated, there is increased flexibility in curtailment. Therefore,
same area, curtailment factors could differ in different areas, this control favors PV systems that are less sensitive to voltage

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 1876 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 14, NO. 3, JULY 2023

Fig. 9. Contour plots showing curtailments in different locations for a specific interval. (a) feeder-level fairness, (b) area-level fairness, (c) maximum utilization.
The magenta-colored box on the left represents a source (i.e., a substation).

different proportions (from 0% to 100%). Therefore, resources

less sensitive to over-voltage (e.g., electrically closer to substa-
tion) are favored to minimize the cumulative curtailment. This
effect can also be realized in the time-series voltage profile in
Fig. 10, where the proportion of buses with voltages close to
1.05 p.u. is much higher for the FAC. It can better exploit the
available voltage headroom; hence, the proportion of customers
with nodal voltages close to the ANSI limit (1.05 p.u.) is much
higher with unfair curtailment. The curtailment factor for a given
Fig. 10. Time-series voltage profile for 738 residential customers after voltage
regulation using: Maximum utilization (left), and fairness-aware control (right). interval is studied with a contour plot to visualize the effect
across the feeder. Fig. 9 shows the curtailment factor throughout
the feeder for a specific time interval (t = 55). The feeder-level
fairness can be observed in Fig. 9(a), where all PVs are curtailed
with the same proportion. Similarly, for the area-level fairness,
it is observed that the curtailment factor is higher in area-5,
whereas area-1 and area-3 observe the least curtailment (See
Fig. 9(b)). For the maximum utilization case, the curtailment
occurs at some specific location within the feeder. These loca-
tions are electrically far from the source and are most sensitive
to over-voltage. For this specific scenario, three sensitive zones
are located in area-2, area-4, and area-5, as shown in Fig. 9(c).

VII. DISCUSSION
Fig. 11. Effect of ZIP load models on curtailment factor.
A. Simulation Duration and Control Cycle
In this work, a single day for the worst-case scenario (i.e., the
rise (i.e., closer to the substation or feeder head) and penalizes coincidence of peak PV generation and low demand) is consid-
those in weaker areas. The total energy curtailment is the least for ered. Nonetheless, from the perspective of fairness, the behavior
the FAC, around 19% less than area-level fairness and almost of curtailment remains the same throughout the time of interest
half compared to feeder-level fairness. However, it introduces and will be more evident during those days when curtailment
unfairness, as shown by the average curtailment range among is required. We have used a 15-min time interval to include
PVs in each area calculated for a day. In fairness-agnostic the variations in PV generation and load demand. However, the
control, the average curtailment range (U areak ) is non-zero proposed approach could be implemented in a lower resolution
for areas 2, 4, and 5. For example, in area 2, on average, the (e.g., every minute). Such frequent control cycles are expected
difference between the maximum and minimum curtailment is to increase performance as it reduces the uncertainty faced
0.37. by longer control cycles [10]. The duration of a control cycle
Another interesting observation can be made upon comparing depends on: i) the availability of real-time measurements from
the time-series voltage profile of the FAC with the fairness-aware the feeder, ii) the computing capability of distributed agents,
solution. Observe that the bus voltages are below the ANSI and iii) the speed of communication infrastructure. If there are
limit, i.e., 1.05 p.u. Since FAC doesn’t fairly apply curtailment, fluctuations in load demand or PV generation before the end
it can curtail resources across the network with significantly of a control cycle, local controls such as droop-based [8] or

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 POUDEL et al.: FAIRNESS-AWARE DISTRIBUTED ENERGY COORDINATION FOR VOLTAGE REGULATION 1877

Fig. 13. Implementation of feeder-level fairness for PV curtailment. (a) Con-

vergence of curtailment factor for four areas at t = 48, (b) Curtailment factors
for four areas across different time intervals.

Fig. 12. EPRI K1 circuit divided into 4 switch-delimited areas [53]. TABLE IV
COMPARISON OF ENERGY CURTAILMENT BETWEEN FAIRNESS-AGNOSTIC AND
FAIRNESS-AWARE CURTAILMENT
search-based [9] methods could be implemented before moving
to the next control cycle.

B. Impact of Voltage-Dependent Load Models

This section describes the impact of voltage-dependent load
models on PV curtailment. The voltage-dependent load models
can be included in the OPF formulation, as shown in the Ap-
pendix. ZIP load models are incorporated in this work to model
the voltage dependency. While the load models vary with cus-
tomer class, season, and time of the day, we incorporate the same
ZIP coefficients for all loads (Z = 0.4, I = 0.4, & P = 0.2) at towards agreement in curtailment factor among all the areas
every simulation time step for simplicity. However, time-varying for a time step (t = 48) which demonstrates the convergence
load models could be easily incorporated within the proposed of the proposed approach. At each interval, the PV curtailment
formulation if available. gets distributed across the resources in all areas in proportion
We incorporated the voltage-dependent load models in our to the exported power (See Fig. 13(b)). This demonstrates the
problem formulation and investigated the change in load and PV applicability of the proposed algorithm for real distribution
curtailment. Fig. 11 shows the increase in feeder demand and networks.
the corresponding reduction in PV curtailment at different time Table IV presents the performance comparison of the fairness-
steps. During the over-voltage period, most nodal voltages are aware distributed control with the FAC case. The fairness-based
above the nominal voltage, i.e., 1.0 p.u. Thus, the overall demand curtailment at the feeder-level aggregation leads to the average
for the grid increases due to voltage-depended loads. As a result, curtailment of 9.19% for each area. Since all PVs in the feeder
there is less reverse flow, which lowers the PV curtailment. are curtailed with the same proportion, the average range of
curtailment (U areak ) is zero for the fairness-aware case. As the
C. Implementation in a Realistic Feeder FAC aims to minimize the net curtailment across the feeder with-
out considering fairness, it favors the PV systems that are less
To demonstrate the applicability to real distribution networks,
sensitive to over-voltage and heavily curtails those in the weaker
the proposed FA-DEC approach is implemented for a real feeder
areas. The total energy curtailment is significantly less for the
located in the Southeastern United States [53], referred to as the
FAC as compared to the feeder-level fairness case. However, this
EPRI K1 circuit, and is publicly available at [54]. Fig. 12 shows
introduces unfairness, as indicated by the average curtailment
the one-line diagram of the EPRI K1 feeder which has been
range (U areak ) among PVs in each area (see Table IV). For
sectionalized into four switch-delimited areas. To demonstrate
example, the average difference between the maximum and
an over-voltage scenario, the base case deployment of the PV
minimum curtailed PVs in Area 2 and Area 3 of the system are
systems considers 90% of residential customers with rooftop
91% and 87% respectively, which demonstrates that the FAC
PVs along with a set of commercial customers. The proposed
case curtails the PV systems in locations that are most sensitive
fairness-aware distributed control is implemented for the feeder-
to over-voltages.
level aggregation case (see Case 1 in Section VI-B) considering
the same simulation setup as described in Section VI-A and
its performance is compared with the fairness-agnostic control D. GridAPPS-D Implementation
(FAC). GridAPPS-D is an open-source, standards-based platform
Fig. 13 shows the simulation results when feeder-level fair- that enables the cost-effective development and deployment
ness is implemented. Fig. 13(a) shows the iterative evolution of advanced applications for distribution system planning and

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 1878 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 14, NO. 3, JULY 2023

operations [42]. While the conceptual design of the platform constant power (P and Q). For a given phase, the voltage-
enables standards-based data abstractions and data exchange dependent load demand pload
j + jqjload can be represented as,
mechanisms, the centralized architecture lacks the ability to
coordinate and control different DERs, smart devices, and intel-  
ligence deployed across the distribution grid. This calls for new aZ
p aIp
pload
j = pload
j,0 v j + N Vj + aP
p (22a)
architectural approaches that allow a large-scale integration of vj
N Vj
devices and intelligence at the grid edge to enhance flexibility  
in grid automation and implement control actions across tens of aZq aIq
qjload = qj,0
load
v j + Vj + aP
q (22b)
millions of endpoints. vjN VjN
Towards this goal, a distributed application architecture for
communication and control of distribution network assets is j,0 + jqj,0 is the load demand of load at node j at
where pload load

proposed in [55] that defines a structure of communication and the nominal voltage VjN ; aZ
p , ap , ap , aq , aq , aq are coefficients
I P Z I P

coordination layers based on the laminar coordination frame- representing the percentages of constant impedance, current, and
work [43]. Such architecture is leveraged to develop a distributed power of active and reactive power and should satisfy:
app API within the GridAPPS-D platform. In the future, the
proposed distributed energy coordinator method will be imple- aZ
p + ap + ap = 1, aq + aq + aq = 1
I P Z I P
(23)
mented within the GridAPPS-D platform, where the distributed
The ZIP load model representation in (22) is a function of both
agents will be integrated with the distributed app API of the
Vj and vj = Vj2 . Directly including these equations in OPF
platform.
will make the formulation discussed in (4) non-linear. The term
√
Vj = vj can be linearized around vj = 1.0 based on its Taylor
VIII. CONCLUSION expansion [56]:
√
We proposed a distributed OPF formulation that incorporates vj ≈ 0.5 + 0.5vj . (24)
fairness for voltage regulation in power distribution systems.
An ADMM-inspired distributed approach is presented where Substituting (24) in (22), we get a linear representation of ZIP
agents in different areas of the distribution grid share the burden load models that can be easily incorporated into the linear power
for voltage regulation. It is shown that a generic formulation balance equations.
for adopting scalable resource aggregation enables flexibility    
in implementing fairness across the different sections of the aZp aIp aIp
pj = pj,0
load load
+ 0.5 N vj + 0.5 N + ap P
network. The simulation results on the modified IEEE 123-bus vjN Vj Vj
test feeder demonstrate that the proposed approach mitigates (25a)
voltage violations on the network and is effective in removing    
locational penalizations. The fairness scheme is studied across aZq aIq aIq
qjload = qj,0
load
+ 0.5 v j + 0.5 + aP
q
different layers of a distribution network, and the trade-offs of vjN VjN VjN
fairness-based controls are explored against maximum utiliza- (25b)
tion to observe the increase in energy curtailment and corre-
sponding unfairness among resources.
REFERENCES
In this work, we considered only the flexibility of PV systems
to mitigate the overvoltage. While other forms of flexibility (i.e., [1] K. S. Wolske, P. C. Stern, and T. Dietz, “Explaining interest in adopt-
ing residential solar photovoltaic systems in the United States: Toward
voltage regulators and OLTCs) and controllable resources (i.e., an integration of behavioral theories,” Energy Res. Social Sci., vol. 25,
flexible loads and storage devices) can be used in coordination pp. 134–151, 2017.
to lessen the amount of active power curtailment, they are not [2] National Community Solar Partnership. [Online]. Available: https://www.
energy.gov/communitysolar/community-solar
explored in this work because the emphasis is on employing [3] P. Scott, D. Gordon, E. Franklin, L. Jones, and S. Thiébaux, “Network-
a distributed approach to assess the impact of fairness in PV aware coordiof residential distributed energy resources,” IEEE Trans.
curtailment across different layers in the distribution system. Smart Grid, vol. 10, no. 6, pp. 6528–6537, Nov. 2019.
[4] B. Bletterie, S. Kadam, R. Bolgaryn, and A. Zegers, “Voltage control
Future research will examine the effects of additional fairness with PV inverters in low voltage networks–in depth analysis of different
models, including egalitarian and revenue-driven ones, and de- concepts and parameterization criteria,” IEEE Trans. Power Syst., vol. 32,
sign a coordinated control mechanism to include other flexible no. 1, pp. 177–185, Jan. 2017.
[5] B. Uzum, A. Onen, H. M. Hasanien, and S. Muyeen, “Rooftop solar PV
devices/resources in the system. penetration impacts on distribution network and further growth factors–A
comprehensive review,” Electronics, vol. 10, no. 1, 2020, Art. no. 55.
[6] F. Olivier, P. Aristidou, D. Ernst, and T. Van Cutsem, “Active management
APPENDIX of low-voltage networks for mitigating overvoltages due to photovoltaic
units,” IEEE Trans. Smart Grid, vol. 7, no. 2, pp. 926–936, Mar. 2016.
This appendix summarizes the derivation of voltage- [7] R. Tonkoski, L. A. C. Lopes, and T. H. M. El-Fouly, “Coordinated active
dependent load models that can be incorporated in the OPF power curtailment of grid connected PV inverters for overvoltage preven-
tion,” IEEE Trans. Sustain. Energy, vol. 2, no. 2, pp. 139–147, Apr. 2011.
formulation ((4) in Section IV-A). [8] M. G. Kashani, M. Mobarrez, and S. Bhattacharya, “Smart inverter volt-
For demonstration purposes, the loads are modeled by the watt control design in high PV-penetrated distribution systems,” IEEE
composite of constant impedance (Z), constant current (I), and Trans. Ind. Appl., vol. 55, no. 2, pp. 1147–1156, Mar./Apr. 2019.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 POUDEL et al.: FAIRNESS-AWARE DISTRIBUTED ENERGY COORDINATION FOR VOLTAGE REGULATION 1879

[9] S. Ghosh, S. Rahman, and M. Pipattanasomporn, “Distribution voltage [31] K. E. Antoniadou-Plytaria, I. N. Kouveliotis-Lysikatos, P. S. Georgi-
regulation through active power curtailment with PV inverters and so- lakis, and N. D. Hatziargyriou, “Distributed and decentralized volt-
lar generation forecasts,” IEEE Trans. Sustain. Energy, vol. 8, no. 1, age control of smart distribution networks: Models, methods, and fu-
pp. 13–22, Jan. 2017. ture research,” IEEE Trans. Smart Grid, vol. 8, no. 6, pp. 2999–3008,
[10] M. Z. Liu et al., “On the fairness of PV curtailment schemes in resi- Nov. 2017.
dential distribution networks,” IEEE Trans. Smart Grid, vol. 11, no. 5, [32] D. K. Molzahn et al., “A survey of distributed optimization and control
pp. 4502–4512, Sep. 2020. algorithms for electric power systems,” IEEE Trans. Smart Grid, vol. 8,
[11] S. Weckx, C. Gonzalez, and J. Driesen, “Combined central and local active no. 6, pp. 2941–2962, Nov. 2017.
and reactive power control of PV inverters,” IEEE Trans. Sustain. Energy, [33] W. Zheng, W. Wu, B. Zhang, H. Sun, and Y. Liu, “A fully distributed
vol. 5, no. 3, pp. 776–784, Jul. 2014. reactive power optimization and control method for active distribution net-
[12] X. Su, M. A. Masoum, and P. J. Wolfs, “Optimal PV inverter reactive power works,” IEEE Trans. Smart Grid, vol. 7, no. 2, pp. 1021–1033, Mar. 2016.
control and real power curtailment to improve performance of unbalanced [34] D.-L. Schultis, A. Ilo, and C. Schirmer, “Overall performance eval-
four-wire LV distribution networks,” IEEE Trans. Sustain. Energy, vol. 5, uation of reactive power control strategies in low voltage grids with
no. 3, pp. 967–977, Jul. 2014. high prosumer share,” Electric Power Syst. Res., vol. 168, pp. 336–349,
[13] D. Gebbran, S. Mhanna, Y. Ma, A. C. Chapman, and G. Verbič, “Fair 2019.
coordination of distributed energy resources with Volt-Var control and PV [35] M. Zeraati, M. E. H. Golshan, and J. M. Guerrero, “A consensus-based
curtailment,” Appl. Energy, vol. 286, 2021, Art. no. 116546. cooperative control of PEV battery and PV active power curtailment for
[14] S. Poudel, M. Mukherjee, and A. P. Reiman, “A fairness-based distributed voltage regulation in distribution networks,” IEEE Trans. Smart Grid,
energy coordination for voltage regulation in distribution systems,” in vol. 10, no. 1, pp. 670–680, Jan. 2019.
Proc. IEEE Green Technol. Conf., 2022, pp. 45–50. [36] S. Xu, Y. Xue, and L. Chang, “Review of power system support functions
[15] S. Alyami, Y. Wang, C. Wang, J. Zhao, and B. Zhao, “Adaptive real power for inverter-based distributed energy resources-standards, control algo-
capping method for fair overvoltage regulation of distribution networks rithms, and trends,” IEEE Open J. Power Electron., vol. 2, pp. 88–105,
with high penetration of PV systems,” IEEE Trans. Smart Grid, vol. 5, 2021.
no. 6, pp. 2729–2738, Nov. 2014. [37] M. H. Amini et al., “Demand response in future power networks: Panorama
[16] C. Masters, “Voltage rise: The big issue when connecting embedded and state-of-the-art,” in Sustainable Interdependent Networks II. Berlin,
generation to long 11 kV overhead lines,” Power Eng. J., vol. 16, no. 1, Germany:Springer, 2019, pp. 167–191.
pp. 5–12, 2002. [38] P. Lusis, L. L. H. Andrew, S. Chakraborty, A. Liebman, and G.
[17] S. Toma et al., “Optimal coordinated voltage control in distribution sys- Tack, “Reducing the unfairness of coordinated inverter dispatch in PV-
tem,” in Proc. IEEE Power Energy Soc. Gen. Meeting - Convers. Del. rich distribution networks,” in Proc. IEEE Milan PowerTech, 2019,
Elect. Energy 21st Century, 2008, pp. 1–7. pp. 1–6.
[18] L. Cheng, Y. Chang, and R. Huang, “Mitigating voltage problem in dis- [39] M. Mousavi and M. Wu, “A DSO framework for market participation
tribution system with distributed solar generation using electric vehicles,” of DER aggregators in unbalanced distribution networks,” IEEE Trans.
IEEE Trans. Sustain. energy, vol. 6, no. 4, pp. 1475–1484, Oct. 2015. Power Syst., vol. 37, no. 3, pp. 2247–2258, May 2022.
[19] K. M. Muttaqi, A. D. Le, M. Negnevitsky, and G. Ledwich, “A coordinated [40] M. Baggu, “Integration of microgrids, DER aggregators, and DERMS with
voltage control approach for coordination of OLTC, voltage regulator, and ADMS,” Nat. Renewable Energy Lab., Golden, CO, USA, Tech. Rep.,
DG to regulate voltage in a distribution feeder,” IEEE Trans. Ind. Appl., NREL/PR-5D00-82680, Apr. 2022.
vol. 51, no. 2, pp. 1239–1248, Mar./Apr. 2015. [41] R. Sadnan, S. Poudel, A. Dubey, and K. P. Schneider, “Layered co-
[20] H. Sun et al., “Review of challenges and research opportunities for ordination architecture for resilient restoration of power distribution
voltage control in smart grids,” IEEE Trans. Power Syst., vol. 34, no. 4, systems,” IEEE Trans. Ind. Informat., early access, May 24, 2022,
pp. 2790–2801, Jul. 2019. doi: 10.1109/TII.2022.3177464.
[21] L. Wang, A. Dubey, A. H. Gebremedhin, A.K. Srivastava, and N. Schulz, [42] R. B. Melton et al., “Leveraging standards to create an open platform
“MPC-based decentralized voltage control in power distribution systems for the development of advanced distribution applications,” IEEE Access,
with EV and PV coordination,” IEEE Trans. Smart Grid, vol. 13, no. 4, vol. 6, pp. 37361–37370, 2018.
pp. 2908–2919, Jul. 2022. [43] J. D. Taft, “Comparative architecture analysis: Using laminar structure to
[22] S. Fattaheian-Dehkordi, A. Abbaspour, H. Mazaheri, M. Fotuhi-Firuzabad, unify multiple grid architectures,” Nat. Lab., Richland, WA, USA, Pacific
and M. Lehtonen, “A new framework for mitigating voltage regulation is- Northwest, Tech. Rep. PNNL-26089, 2016.
sue in active distribution systems considering local responsive resources,” [44] W. H. Kersting, “Distribution system modeling and analysis,” in Elec-
IEEE Access, vol. 9, pp. 152585–152594, 2021. tric Power Generation, Transmission, and Distribution: The Electric
[23] Y. Wang, K. Tan, X. Y. Peng, and P. L. So, “Coordinated control of dis- Power Engineering Handbook. Boca Raton, FL, USA:CRC Press, 2018,
tributed energy-storage systems for voltage regulation in distribution net- pp. 26–1.
works,” IEEE Trans. Power Del., vol. 31, no. 3, pp. 1132–1141, Jun. 2016. [45] L. Gan and S. H. Low, “Convex relaxations and linear approximation for
[24] M. Zeraati, M. E. H. Golshan, and J. M. Guerrero, “Distributed control optimal power flow in multiphase radial networks,” in Proc. IEEE Power
of battery energy storage systems for voltage regulation in distribution Syst. Comput. Conf., 2014, pp. 1–9.
networks with high pv penetration,” IEEE Trans. Smart Grid, vol. 9, no. 4, [46] R. R. Jha, A. Dubey, C.-C. Liu, and K. P. Schneider, “Bi-level Volt-Var
pp. 3582–3593, Jul. 2018. optimization to coordinate smart inverters with voltage control devices,”
[25] Q. Xie et al., “Use of demand response for voltage regulation in power dis- IEEE Trans. Power Syst., vol. 34, no. 3, pp. 1801–1813, May 2019.
tribution systems with flexible resources,” IET Gener., Transmiss. Distrib., [47] H. Ahmadi and J. R. Martı, “Linear current flow equations with application
vol. 14, no. 5, pp. 883–892, 2020. to distribution systems reconfiguration,” IEEE Trans. Power Syst., vol. 30,
[26] R. Tonkoski and L. A. Lopes, “Voltage regulation in radial distribution no. 4, pp. 2073–2080, Jul. 2015.
feeders with high penetration of photovoltaic,” in Proc. IEEE Energy Conf., [48] S. Boyd et al., “Distributed optimization and statistical learning via the
2008, pp. 1–7. alternating direction method of multipliers,” Found. Trends Mach. Learn.,
[27] M. J. E. Alam, A. Somani, R. B. Melton, and T. E. McDermott, “Transac- vol. 3, no. 1, pp. 1–122, 2011.
tive approach for engaging distribution network assets for voltage manage- [49] R. Gupta, S. Fahmy, and M. Paolone, “Coordinated day-ahead dispatch
ment in southern california edison distribution feeders,” Nat. Lab., Pacific of multiple power distribution grids hosting stochastic resources: An
Northwest, Richland, WA, USA, Tech. Rep. PNNL-27650, Jun. 2018. ADMM-based framework,” Electric Power Syst. Res., vol. 212, 2022,
[28] G. Valverde and T. Van Cutsem, “Model predictive control of voltages Art. no. 108555.
in active distribution networks,” IEEE Trans. Smart Grid, vol. 4, no. 4, [50] Y. Liu, H. Beng Gooi, and H. Xin, “Distributed energy management for the
pp. 2152–2161, Dec. 2013. multi-microgrid system based on ADMM,” in Proc. IEEE Power Energy
[29] H. S. Bidgoli and T. Van Cutsem, “Combined local and centralized voltage Soc. Gen. Meeting, 2017, pp. 1–5.
control in active distribution networks,” IEEE Trans. Power Syst., vol. 33, [51] “IEEE PES test feeder, 123-bus Feeder,” [Online]. Available: https://cmte.
no. 2, pp. 1374–1384, Mar. 2018. ieee.org/pes-testfeeders/resources/
[30] Z. Tang, D. J. Hill, and T. Liu, “Distributed coordinated reactive power [52] S. Diamond and S. Boyd, “CVXPY: A Python-embedded modeling lan-
control for voltage regulation in distribution networks,” IEEE Trans. Smart guage for convex optimization,” J. Mach. Learn. Res., vol. 17, no. 1,
Grid, vol. 12, no. 1, pp. 312–323, Jan. 2021. pp. 2909–2913, 2016.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.
 1880 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 14, NO. 3, JULY 2023

[53] F. E. Postigo Marcos et al., “A review of power distribution test feeders Rabayet Sadnan (Student Member, IEEE) received
in the United States and the need for synthetic representative networks,” the B.Sc. and M.Sc. degrees in electrical and elec-
Energies, vol. 10, no. 11, 2017, Art. no. 1896. tronics engineering from the Bangladesh University
[54] “IEEE PES test feeder, EPRI test circuits,” [Online]. Available: https:// of Engineering And Technology, Dhaka, Bangladesh,
cmte.ieee.org/pes-testfeeders/resources/ in 2015 and 2017, respectively, and the second M.Sc.
[55] A. A. Anderson et al., “Distributed application architecture and linknet degree in mathematics in 2021 from Washington State
topology processor for distribution networks using the common informa- University, Pullman, WA, USA, where he is currently
tion model,” IEEE Access, vol. 10, pp. 120765–120780, 2022. working toward the Ph.D. degree. He is appointed
[56] B. Chen, C. Chen, J. Wang, and K. L. Butler-Purry, “Sequential service as a Power System Research Engineer with Pacific
restoration for unbalanced distribution systems and microgrids,” IEEE Northwest National Laboratory, Richland, WA. His
Trans. Power Syst., vol. 33, no. 2, pp. 1507–1520, Mar. 2018. research interests include renewable energy, opti-
mization, distribution system analysis, and distributed optimization.

Shiva Poudel (Member, IEEE) received the B.E.

degree from the Department of Electrical Engineer-
ing, Pulchowk Campus, Kathmandu, Nepal, in 2013,
the M.S. degree from the Electrical Engineering and
Computer Science Department, South Dakota State
University, Brookings, SD, USA, in 2016, and the Andrew P. Reiman (Member, IEEE) received the
Ph.D. degree in electrical engineering from Wash- B.S. degree in electrical engineering from the Univer-
ington State University, Pullman, WA, USA, in 2020. sity of Michigan, Ann Arbor, MI, USA, and the M.S.
He is currently a Power Systems Research Engineer and Ph.D. degrees in electrical engineering from the
with Pacific Northwest National Laboratory, Rich- University of Pittsburgh, Pittsburgh, PA, USA. He is
land, WA. His research interests include distribution currently a Senior Power Systems Research Engineer
system modeling and analysis, resilience assessment, and transactive energy with Pacific Northwest National Laboratory, Rich-
systems. land, WA, specializing advanced distribution system
operations, algorithms for grid-edge systems, and
distribution system modeling and simulation.

Monish Mukherjee (Member, IEEE) received the

B.E. degree from the Department of Electrical Engi-
neering, Jadavpur University, Kolkata, India, in 2016,
and the Ph.D. degree in electrical and computer engi-
neering from Washington State University, Pullman,
WA, USA. He is currently a Power Systems Research
Engineer with Pacific Northwest National Labora-
tory, Richland, WA. His research interests include
transactive energy systems, distribution system mod-
eling and simulation, grid resiliency, and condition
monitoring of high voltage power equipment.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO AMAZONAS. Downloaded on September 13,2024 at 02:09:50 UTC from IEEE Xplore. Restrictions apply.