0% found this document useful (0 votes)
12 views11 pages

Load Balancing

Download as pdf or txt
Download as pdf or txt
Download as pdf or txt
You are on page 1/ 11

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 1

Load Balancing in Low-voltage Distribution Network via Phase


Reconfiguration: An Efficient Sensitivity-based Approach
Bin Liu, Member, IEEE, Ke Meng, Senior Member, IEEE, Zhao Yang Dong, Fellow, IEEE, Peter K.C. Wong,
Senior Member, IEEE, and Xuejun Li

Abstract—Operational performance in the low-voltage distri- || Absolute value for a real number or magnitude
bution network (LVDN) can be undermined by its inherent un- for a complex number.
balances, which may become worse as the penetration of rooftop * Conjugate of a complex number.
solar continuously increases. To address this issue, load balancing
via phase-reconfiguration devices (PRDs), which can change Parameters
phase positions of residential customers as required, provides a λ The number of allowed PRDs to be switched
cost-efficient option. However, most reported approaches to con- each time.
trol PRDs require that demands of all residential customers are |Vn | The nominal voltage of the network, which is
available, which are not viable for many LVDNs without smart
meters or advanced metering infrastructure (AMI) installed. To 1.0 p.u. in per unit system.
bridging the gap in this field, this paper proposes a novel method Ci The set of customers connected to node i, and
to control PRDs purely based on measurable data from PRDs and Ci = Fi ∪ Xi , Fi ∩ Xi = ∅.
its controller. Based on limited information, sensitivity analysis in Fi The set of customers without PRD installed at
the network with PRDs is studied, followed by the optimization node i, also denoted as fixed customers.
model that comprehensively considers operational requirements
in the network. Moreover, slack variables are introduced to the R The set of measurable nodes.
model and penalized in the objective function to assure either a Xi The set of customers with PRD installed at node
strategy that is secure or with minimized violations can always be i, also denoted as adjustable customers.
provided. The model is a challenging mixed-integer non-convex µψ,k,j A parameter indicating the initial phase con-
programming (MINCP) problem, which is reformulated as an nection of a customer. µψ,k,j = 1 means the
efficient solvable mixed-integer second-order cone programming
(MISOCP) based on exact reformulations or accurate linear ap- j th customer of node k is initially connected to
proximations. Simulations based on two modified IEEE systems phase ψ.
and a real system in Australia demonstrate that an efficient ν−, ν0 Coefficients used for constraining the magni-
strategy can be provided to mitigate unbalances in the network. tudes of Vi− and Vi0 , respectively.
φ, ψ Subscripts to represent the phase in the network,
Index Terms—Distribution network, load balancing, phase where φ, ψ ∈ {a, b, c}.
reconfiguration, sensitivity analysis, voltage unbalance. σ Penalty coefficient for slack variables in the
model.
j Imaginary symbol of a complex number.
N OMENCLATURE 0
θφ,i The VA of phase φ at node i.
Abbreviations 0
θφψ,i The VA difference between phase φ and ψ at
AMI Advanced metering infrastructure node i.
BS/AS-PRDs Before/After switching PRDs Dφi,ψk Sensitivity of demand current of the j th cus-
DT Distribution transformer tomer at node k to voltage at phase φ of node
LP Linear programming i.
LVDN Low-voltage distribution network 0
Iφ,xy The current of phase φ in the DT, which is a
MILP Mixed-integer linear programming known parameter.
MINCP Mixed-integer non-convex programming max
Iψ,xy The upper magnitude limit for Iψ,xy .
MISOCP Mixed-integer second-order cone programming 0
Pφ,xy , Q0φ,xy Known active and reactive powers running out
PRD Phase-reconfiguration device
SOC Second-order cone from the secondary side of the DT.
n
UB/LB Upper/Lower bound Pk,j , Qnk,j The net active and reactive demands for the j th
UTOPF Unbalanced three-phase optimal power flow customer of node k.
0
UTPF Unbalanced three-phase power flow Vφ,i Known voltage of node i in phase φ.
0
VM/VA Voltage magnitude/angle Vφψ,i The phase-to-phase voltage (between phase φ
ZSC/NSC Zero-/Negative sequence current and phase ψ) at node i, which is a known
ZSV/NSV Zero-/Negative sequence voltage parameter.
min max
Operators Vφ,i , Vφ,i The lower and upper magnitude limits for Vφ,i,t .
∪/∩ Union/Intersection operator. x, y, xy Node indices of the primary and secondary
sides of the DT, and the index of line where
Manuscript received XX; revised XX. (Corresponding author: Bin Liu). The the DT locates.
authors would like to acknowledge that this work is funded by the Australian Zij The impedance or impedance matrix of line ij.
Renewable Energy Agency (ARENA) via the project “Demonstration of three-
phase dynamic grid-side technologies for increasing distribution network DER Variables
hosting capacity” and by UNSW Digital Grid Futures Institute. αψ,k,j A binary variable indicating the phase connec-
Bin Liu, Ke Meng and Zhao Yang Dong are with School of Electrical Engi-
neering and Telecommunications, The University of New South Wales, Syd- tion of a customer. αψ,k,j = 1 means the j th
ney 2052, Australia. (e-mail: bin.liu@unsw.edu.au, ke.meng@unsw.edu.au, customer of node k is connected to phase ψ.
zydong@ieee.org) ∆Iψ,k,j The variation of Iψ,k,j after switching PRDs.
Peter K.C. Wong is with Jemena Electricity Networks (Vic) Ltd, Melbourne
3000, Australia. (e-mail: peter.wong@jemena.com.au) ∆Vφ,i The variation of Vφ,i after switching PRDs.
Xuejun Li is with Jemena Ltd, Sydney, 2060, Australia. (e-mail: es- ωi− , ωi0 Slack variables for NSV and ZSV limits of node
epco@126.com)

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 2

i. The controllable switches communicate wirelessly with the


ρψ,xy Slack variable for DT capacity limit in phase controller to switch their phase positions regularly or when
ψ. monitored parameters exceed predefined thresholds. Reported
− +
τφ,i , τφ,i Slack variables for lower and upper VM limits methods to implement the scheme in practice are mostly based
of phase φ at node i. on heuristic algorithm supported by independent unbalanced
I −, I 0 NSC and ZSC running through the DT in phase three-phase power flow (UTPF) programs [4], [7], [13]–[18].
a. This type of method can always provide a feasible solution
Iψ,k,j Demand current change at phase ψ of node k after a reasonable time. However, the computation efficiency
contributed by the j th customer at this node. may become an issue with a large number of introduced
Vi− , Vi0 NSV and ZSV of node i in phase a. integer variables, and the optimality of the solution may not
Vφ,i Voltage of node i in phase φ. be guaranteed. Other linear-based [3] or mixed-integer linear
z−, z0 Introduced variables to limit NSC and ZSC in programming (MILP)-based algorithms [19], [20] either treat
the DT. all customers as constant current loads (complex values) or
do not take the network formulation into account, thus may
lead to inaccurate results. The comprehensive formulation
I. I NTRODUCTION of the problem, which is based on the unbalanced three-
With encouraging polices and incentives from government phase optimal power flow (UTOPF), is presented in [22].
agencies, the low-voltage distribution network (LVDN) is The method considers relevant operational constraints in the
experiencing a remarkable pace in developing rooftop solar decision-making process and can guarantee the security of a
generations, which, on the one hand, accelerates the transition network if controlled by the optimal strategy.
of the current power grid to a more sustainable one while, However, there is one hidden assumption in the discussed
on the other hand, brings more challenges to the network methods above. It is that demand information, i.e. net active
operation. Among all challenges, addressing the unbalance and reactive demands of all customers, are available in real-
issue becomes more urgent, particularly in the daytime when time operation, which makes effectively running power flows
high irradiance occurs around midday or during night peak in heuristic algorithms or doing optimizations in determinis-
hours [1], [2]. The unbalance issue can cause lots of opera- tic algorithms possible. For networks that can provide such
tional problems, e.g., increased power loss and stressed/light operational data through smart meters or advanced metering
phases in the network that may worsen the under/over-voltage infrastructure (AMI), the above methods are viable and effi-
issue [3]–[5]. The challenge is particularly significant in cient. However, for networks without such capability, installing
Australia as it is experiencing rapid development in rooftop or upgrading to a measurement system with such functionality
solar generations according to the report released by the merely for load-balancing purposes may become uneconom-
Australian Energy Council [6]. Like the LVDN operator in ical for some networks. Therefore, weak observability of the
European countries, power utilities in Australia usually run operational state of the network due to insufficient commu-
extensive three-phase four-wire grid along streetscape and nication or control system may become a barrier of applying
many residential customers, the number of which ranges from the above algorithms, particularly in some rural areas with
dozens to over one hundred, are connected to the nearest pole more significant unbalance issue. Lacking efficient algorithm
via bundled cables [1], [7]. With more and more customers when complete information is unavailable motivates us to
enrolling in PV generations in the future, effective measures develop practical yet accurate enough method to control PRDs
should be taken to address the unbalance issue in the network. in network operation optimally. The idea is to do the opti-
To mitigate unbalances, existing researches reveal that op- mization after analytically bridging future operational states
timally controlling PV inverters, electric vehicle charging, of measurable parameters, which are available in real-time
and batteries of residential customers in LVDN could bring operation from PRD controller and installed PRDs, with their
potential benefits [8]–[12]. However, before network operators current values based on sensitivity analysis in the unbalanced
could control such devices that are private assets of residents network.
via future implemented control/communication infrastructure Compared with existing work, contributions of this paper
and agreements, an alternative cost-efficient option for the are summarized as follows.
network operator is introducing the phase-reconfiguration de- 1) The data available from the PRDs and its control sys-
vices (PRDs) to dynamically reconfigure customers’ connected tem is comprehensively discussed. Notably, an efficient
phases [3], [4], [7], [13]–[20]. This measure is also currently method based on cosine law is employed to determine
under trial in Australia by network operators [21]. the angle differences between any two phase voltages1
The scheme of controlling PRDs in an LVDN is illustrated if only magnitudes of phase voltages and phase-to-phase
by Fig. 1, where the PRD controller is usually installed at voltages are available.
the secondary side of the distribution transformer (DT) and 2) Sensitivities in the LVDN with PRDs are presented and
controllable switches, each of which lies between the LVDN analyzed. Different from existing researches, sensitivity
and a customer, are placed at various locations. analysis in this paper takes the PRDs into account
and bridges their phase position changes with nodal
PRD
Controller
Receive: Measured data
Send: Control strategy voltage variations and current variations in the DT. In
Measured data other words, the formulation analytically depicts how the
values of measurable parameters in the network change
Distribution
Transformer PRD
Receive: Control strategy
Send: Measured data along with positions of all PRDs.
3) A novel sensitivity-based approach to control PRDs in
Wireless Communication &
LVDN is presented. The proposed method considers
Control
the network’s operational requirements based on data
PRD PRD
collected from the PRD controller and PRDs, which

1 Phase voltage throughout this paper refers to phase-to-ground voltage


Fig. 1. Illustration of controlling PRDs in an LVDN. unless specified.

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 3

is particularly useful for a network without smart me- phase. Moreover, the voltage at node x is assumed to be
ters/AMI installed. Moreover, as slack variables are constant, which can be estimated based on measured data at
introduced to the model and penalized in the objective node y, i.e.
function, the proposed method could guarantee that the Vx0 = Vy0 + Zxy Ixy
0
(2a)
required operational requirements for monitored nodes
are always satisfied or with minimized violations. The
0
Iφ,xy = 0
(Pφ,xy − jQ0φ,xy )/(Vφ,y
0
)∗ (2b)
problem is formulated as a mixed-integer non-convex With all available information, which is summarized in
programming (MINCP) problem and reformulated as an Table I, the operational state of the LVDN can be roughly
efficient solvable mixed-integer second-order cone pro- estimated via the measurable parameters, which is the basis
gramming (MISOCP) problem based on several reason- for the optimal control of PRDs.
able assumptions and reformulation techniques, making
it more convenient for practical applications. TABLE I
The remainder of the paper is outlined as follows. The AVAILABLE DATA IN LVDN WITH PRD S
mathematical formulation of the problem is presented in
Part II, where available information from PRD controller Equipment Available Data (Complex Numbers)
and PRDs, sensitivity analysis in the unbalanced network, 0 , V 0 , I0
DT Vφ,x φ,y φ,xy ∀φ
and developed mathematical model are discussed separately. 0 ∀φ, ∀i ∈ R
Measurable Nodes Vφ,i
Solution techniques are presented in Part III, and case studies
µφ,i,j ∀φ, ∀i, ∀j
based on the two modified IEEE systems and a real system Adjustable Customers n = P n + jQn ∀i, ∀j
Si,j i,j i,j
in Australia are performed in Part IV. The paper is concluded
with some discussions in Part V.

II. P ROBLEM F ORMULATION B. Sensitivity Analysis in LVDN with PRDs


A. Analysis on Available Information With the available data of the current period, the sensitivity
As discussed previously, parameters in a real network may in the LVDN can be explained based on the single-phase
be partially available via PRD controller and PRDs. These illustrative example, as shown in Fig. 2.
measurable data are listed below.
3
1) Data from PRD controller. The PRD controller can
provide the voltage magnitudes (VMs) in three phases I23
at the secondary side of the DT (denoted as |Vφ,y 0
| for I3
1 I12 2
phase φ), and the active and reactive powers flowing 4
0
out of the DT (denoted as Pφ,xy and Q0φ,xy for phase φ,
respectively). DT I24
2) Data from PRDs. For each adjustable customer, say the I1 I2 I4
j th customer at node i, the VMs of node i (denoted as
0 Fig. 2. Illustration network with 4 buses.
|Vφ,i | for phase φ), its phase position (denoted as µφ,i,j )
n
and net active/reactive demand (denoted as Pi,j /Qni,j ) For the network in Fig. 2, we have
can be measured and sent to the PRD controller. " # " #" #
V1 − V1 0 0 0 0 I1
It is noteworthy that all measured data are in the field of V1 − V2 0 Z12 Z12 Z12 I2
real numbers, while the formulation in LVDN is based on V1 − V3 = 0 Z12 Z12 + Z23 Z12 I3 (3)
V1 − V4 0 Z12 Z12 Z12 + Z24 I4
complex numbers. To bridge them, another two parameters
are required: the angle differences of any two phase voltages, As V1 is constant, the above expression illustrates how nodal
0 0 0 0 0 0
i.e., θab,i = θb,i −θa,i , θac,i = θc,i −θa,i , and the voltage angle voltages in the network respond to varying demand currents.
0 In a generalized unbalanced three-phase LVDN, the variations
(VA) of at least one phase, say θφ,i , for each node i.
0
θab,i 0
and θac,i are used to determine the relative positions of of node voltages ∆Vφ,i and demand currents ∆Iψ,k,j can be
0 0
Va,i , Vb,i and Vc,i 0
for node i in the field of complex number. bridged as
XX X
In this paper, they are assumed to be available through PRD 0
Vφ,i − Vφ,i = ∆Vφ,i = − Dφi,ψk ∆Iψ,k,j
controller and PRDs. Otherwise, their phase-to-phase voltage ψ k j∈Ck
0
magnitudes for ab and ac at each node (denoted as |Vab,i | and
0 ∀φ, ∀i (4)
|Vac,i | for node i) are assumed to be available. For the latter
0
case, θab,i 0
, θac,i can be calculated based on the cosine law as where D is the constructed matrix as in (3) for a generalized
follows, and the efficiency of which will be demonstrated later network.
in case studies. The expression (4) has been widely used in solving power
0 2 0 2 0 flow equations in unbalanced power systems [28]. However, to
0 |Va,i | + |Vψ,i | − |Vaψ,i |2 make it applicable in our case, several steps need to be taken.
θaψ,i = arccos 0 0
∀ψ ∈ {b, c} (1)
2|Va,i ||Vψ,i | Firstly, as voltages at non-measurable nodes and currents of
0 fixed customers cannot be measured, only measurable nodes
The parameter θφ,i (∀i) is used to illustrate the relative during the decision-making process are focused and we further
0 0
positions of Vφ,1 , · · · , Vφ,i , · · · for phase φ in the field of assume that variations of demand currents at fixed customers
complex number. Assuming the least stressed phase, i.e., the are neglectable, yielding
one with least total net demand, of the whole network is XX X
phase φ, we assume VAs of all nodes in the downstream ∆Vφ,i ≈ − Dφi,ψk ∆Iψ,k,j ∀φ, ∀i ∈ R (5)
0
of node y are identical and equal to θφ,y . This assumption ψ k j∈Xk
arises from the fact that VAs of all nodes in the same phase
are sufficiently small, as demonstrated in [22]–[27] and the Another rationality of this assumption is that the current
difference is expected to be smallest for the least stressed variations of fixed customers are deemed to be small because

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 4

the number of PRDs to be switched each time is usually s.t. |I − | ≤ z − , |I 0 | ≤ z 0 (9b)


constrained (see the formulation (9) in Part II-C), and voltage −
√ √
6I = 2Ia,xy − (1 + j 3)Ib,xy − (1 − j 3)Ic,xy (9c)
levels are always around 1.0 p.u. The current variations at fixed
customers will be further investigated later in case studies. 3I 0 = Ia,xy + Ib,xy + Ic,xy (9d)
Secondly, as ∆Iψ,k,j is an intermediate variable, it should be (5), (7) and (8)

√ √
further expressed as a function of the decision variable αψ,k,j , 6Vi = 2Va,i − (1 + j 3)Vb,i − (1 − j 3)Vc,i ∀i (9e)
which can be derived based on the following expression of 3Vi0 = Va,i + Vb,i + Vc,i ∀i (9f)
Iψ,k,j .
n ∗
|Vi | ≤ ν |Vn | + ωi− , |Vi0 | ≤ ν 0 |Vn | + ωi0 ∀i
− −
(9g)
Iψ,k,j = αψ,k,j (Pk,j − jQnk,j )/Vφ,i ∀ψ, ∀k, ∀j (6) αψ,k,j ∈ {0, 1} ∀ψ, ∀k ∈ R, ∀j ∈ Xk (9h)
X
Then, we have αψ,k,j = 1 ∀k ∈ R, ∀j ∈ Xk (9i)
ψ
∂Iψ,k,j ∂Iψ,k,j ∗ min − 0
∆Iψ,k,j = ∆αψ,k,j + ∗
∆(1/Vψ,k ) (7a) Vφ,i − τφ,i ≤ |Vφ,i | = |Vφ,i + ∆Vφ,i |
∂αψ,k,j ∂(1/∆Vψ,k )
max +
n
(αψ,k,j − µψ,k,j )(Pk,j − jQn ≤ Vφ,i + τφ,i ∀φ, ∀i ∈ R (9j)
k,j )
ζψ,k,j = 0
(7b) |Iψ,xy | = + 0
|Iψ,xy max
∆Iψ,xy | ≤ Iψ,xy + ρψ,xy ∀ψ (9k)
(Vψ,k )∗ X X X
n ∗
γψ,k,j = αψ,k,j (Pk,j − jQn
k,j )∆(1/Vψ,k ) (7c) ψ k j∈Xk
|αψ,k,j − µψ,k,j | ≤ 2λ (9l)
where ∆Iψ,k,j represents the variation of Iψ,k,j ; ζψ,k,j = ωi− ≥ 0, ωi0 ≥ 0 ∀i (9m)
∂Iψ,k,j ∂Iψ,k,j − +
∂αψ,k,j and γψ,k,j = ∂(1/∆Vψ,k ∗ ) are the sensitivities of τφ,i ≥ 0, τφ,i ≥ 0 ∀φ, ∀i (9n)
∗ ρψ,xy ≥ 0 ∀ψ
αψ,k,j and 1/Vψ,k to Iψ,k,j , respectively, and ∆αψ,k,j and (9o)

∆(1/Vψ,k ) represent their variations.
The objective of SNTM is to minimize the sum of the
In (7), ∆Iψ,k,j , on the one hand, is affected by the change negative sequence current (NSC) and zero-sequence current
of customer’s phase position, i.e., αψ,k,j − µψ,k,j , and, on (ZSC) running through the DT, which are defined by (9b)-(9d),
the other hand, by the variation of the customer’s terminal plus the sum of penalized slack variables. Noting that positive
voltage Vψ,k . ζψ,k,j is in mixed-integer linear (MIL) form and values of slack variables imply that constraint violation occurs
can be easily implemented and dealt with in an optimization and σ is a large number, SNTM would seek a solution with

problem. However, ∆(1/Vψ,k ), which is strongly non-convex, the priority of ensuring network security. When such a secure
should be further reformulated to make it tractable, which will solution exists, the objective is equivalent to minimizing the
be discussed in the next section. current unbalance in the DT. Otherwise, SNTM will provide a
Similar to voltage variations, currents running through the solution that minimizes the current unbalance with minimized
DT will vary when PRD switching occurs, leading to constraint violations. Apart from (5),(7) and (8) that define the
0
Iψ,xy − Iψ,xy = ∆Iψ,xy ≈
X X
∆Iψ,k,j ∀ψ (8) sensitivities in LVDN, (9e)-(9g) require the voltage unbalance
are within allowed levels plus slack variables. (9h) and (9i)
k∈R j∈Xk
simulate the operation of PRDs and make sure each customer
Summarizing the analysis in this subsection, (5) illustrates can only be connected to one phase. Relaxed VM limits for all
the sensitivity of ∆Iψ,k,j to ∆Vφ,i . As ∆Iψ,k,j is not con- measurable nodes and relaxed current magnitude (CM) limits
trollable, (7a)-(7c) further depict how it will be affected by for the DT are specified by (9j) and (9k), respectively. To
αψ,k,j . Moreover, voltage variations will in turn affect ∆Iψ,k,j , limit the impact of switching PRDs on residential customers,
which is depicted by (7a) and (7c). Similarly, the impacts the number of PRDs to be switched is restricted, as expressed
of switching PRDs and voltage variations in the network on in (9l).
currents running through the DT are illustrated by (7)-(8). Several remarks on the formulation are given below.
1) To verify the accuracy of the sensitivity analysis, the
C. Mathematical Model proposed method will be compared with the results
Based on the sensitivity analysis, the mathematical model, refined by exact power flow calculations. In other words,
which is denoted as SNTM throughout the context, of opti- UTPF in [28] will be used to get the true operational
mally controlling PRDs in LVDN can be formulated as (9) state of the LVDN after fixing PRDs at their optimized
based on Figure.3. phase positions.
2) For comparison purposes, the proposed method will be
compared with the one that is implemented in practice
Node n Node k Node j to minimize CM unbalance in the DT [29]–[31]. This
Ia,nk Va,k Ia,kj method, which is denoted as SIPM throughout the con-
Ib,nk Ib,kj
text, is based on measured CMs at the secondary side of
a Vb,k the DT and all adjustable customers. Specifically, SIPM
b Ic,nk Vc,k Ic,kj can be formulated as the following linear programming
c
(LP) problem.
a,k,j b,k,j c,k,j
PRD for j-th
customer min F = z d (10a)
z z
s.t. ||Iφ,xy | − |Iψ,xy || ≤ z d ∀φ, ∀ψ (10b)
Pkn, j + jQkn, j X X
z 0
|Iφ,xy | = |Iφ,xy | + |ζφ,k,j | ∀φ (10c)
Fig. 3. Illustration of controlling PRD for an adjustable customer. k∈R j∈Xk

The objective of SIPM is to minimize the maximum


X X CM difference in the DT. However, as angles of phase
min F = z − + z 0 + σ (ωi− + ωi0 ) + σ ρψ,xy voltages are not considered, it may lead to ineffective
i ψ
X X − solutions.
+
+σ (τφ,i + τφ,i ) (9a) 3) In this paper, the scheme in Fig. 4 is presented to
φ i

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 5

optimally control PRDs in real-time operation, where Specifically, assuming the range of VA in phase φ for any
SNTM and SIPM are executed at each period with node is [δφmin , δφmax ], where δφmin = δφ −∆δ, δφmax = δφ +∆δ and
different available information, leading to various strate- both δφ and ∆δ are given parameters, linearly approximating
gies. The strategy is kept unchanged before they are (9j)ll leads to
updated in the next period. Moreover, to provide the min −
required information for making strategies, the operation Xφ,i cos δφ + Yφ,i sin δφ ≥ Vφ,i − τφ,i (11)
of LVDN is simulated by periodically solving the UTPF Linearizing another NCX constraint (7c) is based on the
in each period with updated information. ∗
approximation of 1/Vψ,k by the following expression [23].

Updated Residential 1/Vψ,k ≈ kψX Xψ,k + kψY Yψ,k + bX X
ψ + j(hψ Xψ,k +
Demands

Unbalanced Three-phase hYψ Yψ,k + bYψ ) ∀ψ, ∀k (12)


Power Flow

Network
operational state where kψX , kψY , bX X Y Y
ψ and hψ , hψ , bψ are parameters to be fitted
Operation Simulation
LVDN @ Period t
Data Process & Making Strategies
for phase ψ.
From SNTM based on: Until
The general idea to fit the parameters is based on the
previous period Updated PRD
Phase Positions
(1) Voltages (Measurable nodes)
(2) Currents in the DT
Updated PRD
Phase Positions
next period
least-square method with a sufficient number of sampled
(3) P&Q of adjustable customers points in the feasible region of Vψ,k . This method has been
demonstrated with high accuracy, and more details can be
From Until
found in [23], [32], which are omitted here for simplicity.
SIPM based on: ∗
previous period Updated PRD
Phase Positions
(1) CMs of adjustable customers
Updated PRD
Phase Positions
next period
With (12), ∆(1/Vψ,k ) can be expressed as
(2) CMs in the DT

∆(1/Vψ,k ) = kψX ∆Xψ,k + kψY ∆Yψ,k + j(hX
ψ ∆Xψ,k +
Fig. 4. Scheme for real-time control of PRDs in LVDN. hYψ ∆Yψ,k ) ∀ψ, ∀k (13)

4) SNTM is a MINCP problem due to the introduced Combining (7c) and (13) leads to
integer variables and non-convex parts in (7c) and (9j). n
ζψ,k,j = (Pk,j − jQnk,j )[(kψX + jhX
ψ )αψ,k,j ∆Xψ,k +
Solution techniques to make the formulated problem
efficiently solvable will be discussed in the next section. (kψY + jhYψ )αψ,k,j ∆Yψ,k ] ∀ψ, ∀k, ∀j (14)
Obviously, non-convex parts in (14) now are κX ψ,k,j =
III. S OLUTION T ECHNIQUES αψ,k,j ∆Xψ,k and κYψ,k,j = αψ,k,j ∆Yψ,k . Noting that they
A. Reformulating Constraints are in bilinear form and each of them can be expressed as the
To make solution techniques clearer, constrains in (9) are product of a continuous variable and a binary variable, they
can be exactly reformulated as a set of MIL constraints based
categorized as shown in Table II, followed by the reformula- on McCormick envelope [33], yielding
tion or linearization of each type of constraint. n
ξψ,k,j = (Pk,j − jQn X X X Y
k,j )[(kψ + jhψ )κψ,k,j + (kψ +

TABLE II jhYψ )κYψ,k,j ] ∀ψ, ∀k, ∀j (15a)


C ONSTRAINT CATEGORIES (NLC: NON - LINEAR BUT CONVEX ; NCX: min
≤ κX max
αψ,k,j ∆Xψ,k ψ,k,j ≤ αψ,k,j ∆Xψ,k (15b)
NON - CONVEX ; LIN: LINEAR ; (9j)ul /(9j)ll : U PPER / LOWER LIMIT IN (9j))
(αψ,k,j − max
1)∆Xψ,k ≤ κX min
ψ,k,j − ∆Xψ,k ≤ (αψ,k,j − 1)∆Xψ,k (15c)
Category NLC NCX LIN min
αψ,k,j ∆Yψ,k ≤ κYψ,k,j ≤ αψ,k,j ∆Yψ,k
max
(15d)
Constraints (9b),(9g),(9j)ul ,(9k) (7c),(9j)ll Others
(αψ,k,j − max
1)∆Yψ,k ≤ κYψ,k,j − ∆Yψ,k ≤ (αψ,k,j − 1)∆Xψ,k
min
(15e)
To make the mathematical formulation and solution tech-
√ NLC constraints in (9) can be generally expressed as niques more clear, a schematic diagram describing the math-
m2 + n2 ≤ s + s0 with m, n, s being variables and s0
a constant. This inequality is a second-order cone (SOC) ematical formulation and solution techniques is also provided
constraint, which is convex and can be easily dealt with by in Fig.6.
commercial solvers.
STEP 1: STEP 2:
For NCX constrains, the lower limit in (9j) can be approx- Mathematical Formulation Solution Techniques
imately linearized according to Fig. 5 based on the fact that Linear
approximation Bilinear non-
(7c)
the VA difference of all nodes in the same phase is sufficiently Sensitivity analysis in
General non-convex
convex terms
Exact
LVDN with PRDs: Linear
small [22]–[27]. (5), (7)-(8)
terms reformulation
Lower limits approximation MIL/Linear
in (9j) terms

Mathematical Quadratic terms:


Objective function: SOC MISOCP Commercial
(9a) formulation (9b), (9g), upper limits
terms formulation solvers
(MINCP) in (9j), (9k)

Operational
constraints: (9b)-(9d), Other linear terms
(9e)-(9o)

Fig. 6. Schematic diagram for the mathematical formulation and solution


techniques.
A

B. Algorithm
Based on the above reformulation techniques, (9) is re-
Fig. 5. Illustration of dealing with voltage constraints. formulated as an efficient solvable MISOCP problem. The

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 6

formulation is implemented in Matlab with YALMIP [34] and B. Case I


solved by CPLEX 12.9.0 [35] on a desktop PC with Intel i9- In this case, 100 three-phase voltages (complex numbers)
9900 3.1 GHz CPU, 32 GB memory. will be randomly generated based on |Vφ0 | ∈ [0.8, 1.2],
IV. C ASE S TUDY θa0 = 0◦ , θb0 ∈ [−140◦ , −100◦ ] and θc0 ∈ [100◦ , 140◦ ].
A. Case setup Based on the randomly generated scenarios, only |Vφ0 |(∀φ)
0
Several cases based on the modified IEEE-13 bus system and |Vaψ |(∀ψ ∈ {b, c}) for each scenario will be input to (1)
(IEEE-13 LVDN), a practical system in Australia (Australian as known parameters. The solutions of (1) are then compared
LVDN), and the modified IEEE European LVDN (European with the originally scenarios.
LVDN) will be studied in this section. The topology of the
modified IEEE-13 bus system is presented in Fig. 7. 1 Random Scenario
Recovered Scenario

Imaginary Part (p.u.)


1 0.5

Distribution
Transformer 0
6 8
1

4 3 2 5 6 7 -0.5

5 4 3 2 17
-1

10 9 -0.5 0 0.5 1
7 12 13 15
Real Part (p.u.)
11 9
16
12 11 13
8 10 14
Fig. 8. The comparison of solutions from cosine law (1) and initially
generated random scenarios.
Fig. 7. Topology of the IEEE-13 LVDN (Orange: phase a; Green: phase b;
Purple: phase c). As shown in Fig. 8, (1) is accurate in providing VA
information at each node and is readily to be applied in
The IEEE-13 LVDN is with 17 single-phase powered cus- practice.
tomers, where five customers are with PV panels, and the
capacity for each of them is 5 kW. There are 5 PRDs installed
at customer 3,7,9,11 and 17 for the IEEE-13 LVDN. C. Case II
Moreover, ν − is set as 2%, which is based on the opera- 1) Operational performance at specific periods: The op-
tional requirements in Australian distribution networks [36], timal control of PRDs in the IEEE-13 LVDN is studied in
and ν 0 is set as 4.5%. σ is set as 1000 and the voltage this case based on the scheme presented in Fig. 4. Before
of the root node is a known parameter, which is set as introducing PRDs, i.e., for NPRD, phase positions of 5 ad-
2π 2π
Vx0 = [1.0ej0 , 1.0e−j 3 , 1.0ej 3 ]T for all periods throughout justable customers are c, b, a, c and c, respectively. Simulation
the day for clearer illustration. Other parameters and topology results are presented in Fig. 9-Fig. 15. Specifically, simulation
information of the system can be found in [37]. results of two periods (Period 8 and 18 corresponding to 8:00
The Australian LVDN is with 26 nodes in the main feeder, and 18:00) are investigated with more details, followed by the
and 69 customers, each of which is powered by single-phase or network performance analysis throughout the whole day2 .
three phases. Among the 69 customers, 16 of them are with PV The initial phase positions of 5 PRDs at period 8 are
panels, and 14 of them are equipped with PRDs. Other opera- a, b, c, a, c, which are the optimized results from the previous
tional parameters, including the topology, residential demands period. After 0.07 second, SNTM reports the optimized posi-
and PRD locations, can be found in [37]. Different from the tions of all PRDs as a, a, a, a, c. The maximum zero-sequence
modified IEEE-13 LVDN, voltages of the root node for this voltage (ZSV) and negative sequence voltage (NSV) levels of
system throughout the day are set according to historical data, all measurable nodes are 1.33% and 2.36% initially and are
as discussed later. 0.74% and 1.40% after switching PRDs to the updated posi-
The European LVDN is a much larger system with 907 tions. To verify the accuracy of the proposed method, UTPF
buses and 55 single-phase powered residential customers. is employed to get the exact operational state of the whole
Based on original data from [38], we further assume there network after fixing PRDs at their optimized phase positions3 .
are ten residential customers with PV installed and each of The calculated maximum ZSV and NSV by SNTM+PF are
the installations is 5 kW. Moreover, PV is assumed to be 0.76% and 1.39%, respectively, which are very close to the
generating at their capacity values at noon. Other data for the results reported by SNTM. The overall current unbalance level
European LVDN, including the PRDs location information can is reduced from 9.04 A to 4.95 A reported by SNTM, and
also be found in [37]. to 5.05 A by SNTM+PF. The result again shows that the
Cases to be studied are outlined as follows. simulation results of SNTM and SNTM+PF are close to each
1) Case I: This case is used to demonstrate the efficiency other.
of the employed method to determine the VAs of any VMs of all nodes at period 8 are presented in Fig. 9, where
two phases, as discussed in Section II-A. they vary significantly after switching PRDs to optimized
2) Case II: This case is based on the IEEE-13 LVDN. In phase positions. However, VMs all fall into the secure region
this case, the efficiency and optimality of SNTM will be for both the initial state and the updated state. Moreover, the
studied, followed by its comparisons with both SIPM errors for measurable nodes for SNTM and SNTM+PF are
and the case when no PRD is introduced (denoted as
NPRD). 2 For
the IEEE-13 LVDN, the time resolution of controlling PRDs is 1-hour.
3) Case III: This case is based on the Australian LVDN 3 “SNTM/SIPM+PF” is used to represent the refined simulation results by
and the European LVDN to show the effectiveness and UTPF after fixing PRD at optimized phase positions provided by SNTM/SIPM
practicality of the proposed method. throughout the context.

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 7

1.02
Phase a (Initial)
Phase b (Initial)
Phase a (SNTM+PF)
Phase b (SNTM+PF)
Phase a (SNTM)
Phase b (SNTM)
However, as PRDs are usually dispersedly installed in the
Phase c (Initial) Phase c (SNTM+PF) Phase c (SNTM) network, the operational requirements of most non-measurable
Voltage Magnitude (p.u.)

1.01 nodes are expected to be satisfied once they are reached at all
measurable nodes.
1 2) Accuracy analysis: To verify the rationality of neglect-
ing variations of demand currents at fixed customers, their
0.99 values before and after switching PRDs at periods 8 and 18
are presented in Fig. 11. As shown in the figure, the variations
0.98 are minimal, which further demonstrates approximating (4) by
(5) is accurate enough for the developed model.
0.97
2 4 6 8 10 12
2 Real Part (BS-PRDs) 4
Node Real Part (AS-PRDs)
Imaginary Part (BS-PRDs)
1.5 Imaginary Part (AS-PRDs) 3
Fig. 9. Voltage magnitudes at period 8 for the IEEE-13 LVDN.
1 2

Current (p.u.)
0.5
1
sufficiently small, which, along with other simulation results 0
at this period, demonstrates the accuracy of the sensitivity- 0
-0.5
based method. -1
The simulation results of period 18 are presented in Fig. 10 -1
and Table III. -1.5
-2

5 10 15 5 10 15
TABLE III
Customer Index Customer Index
S IMULATION RESULTS OF SNTM, SNTM+PF, SIPM+PF AT PERIOD 18.
Fig. 11. Demand currents of fixed customers before/after switching PRDs
SNTM SIPM (BS/AS-PRDs) at period 8 (left) and 18 (right).
Indicator/Method
Initial SNTM+PF Initial SIPM+PF
z− + z0 3.15 2.50 3.78 1.91 3) Optimality analysis: In this section, the optimality of the
zd 5.09 3.50 6.14 3.09 solution provided by SNTM will be investigated based on the
max NSV(i ∈ R) 0.32% 0.54% 1.185% 1.12% following steps for period 8 in the IEEE-13 LVDN.
max ZSV(i ∈ R) 1.45% 0.74% 3.11% 2.80%
PRD 1) Enumerate all possible strategies to control PRDs. As
c, c, c, a, b c, c, a, c, b a, b, c, a, c b, b, c, a, c there are 5 PRDs in the network, the total number of
phase positions
possible strategies is 35 = 243.
2) For each possible strategy, fix positions of all PRDs
accordingly and run the UTPF for the network.
1.02
Phase a (Initial)
Phase b (Initial)
Phase a (SNTM+PF)
Phase b (SNTM+PF)
Phase a (SNTM)
Phase b (SNTM)
Phase a (SIPM+PF)
Phase b (SIPM+PF)
VM Lower Limit 3) Analyze the simulation results of all possible strategies,
Phase c (Initial) Phase c (SNTM+PF) Phase c (SNTM) Phase c (SIPM+PF)
find the one with minimal current unbalance in the DT
Voltage Magnitude (p.u.)

1
and without any constraint operational violations.
The simulation results, including the upper bound (UB) and
0.98
lower bound (LB) of VMs, UB and LB of voltage unbalances,
number of PRDs switched, and the current unbalance level for
0.96
each possible strategy are presented in Fig.124 .
0.94 8 NSV (All Strategies) NSV (Strategy from SNTM)
Voltage Magnitude (p.u.)

Voltage Unbalance (%)

ZSV (All Strategies) ZSV (Strategy from SNTM)


1 6
0.92
2 4 6 8 10 12
4
Node 0.95

2
UB (All Strategies) UB (Strategy from SNTM)
LB (All Strategies) LB (Strategy from SNTM)
Fig. 10. Voltage magnitudes at period 18 for the IEEE-13 LVDN. 0.9
5
All Strategies Strategy from SNTM All Strategies Strategy from SNTM
Number of Switched PRDs

Current Unbalance (A)

4 20
In Fig. 10, VMs of phase a at node 9-11 are below the lower
3
limit initially. However, this violation is adequately addressed 15

by ameliorating the power flow conditions by switching two 2


10
PRDs to other phases. As shown in Table III, current unbal- 1

ance, whether measured as the sum of NSC and ZSC or as 0


50 100 150 200 50 100 150 200
Strategy Index Strategy Index
its magnitude unbalance, has been effectively mitigated by
both SNTM and SIPM, and SIPM even performs better than
Fig. 12. Analysis of simulation results for all possible PRD control strategies
SNTM. Moreover, voltage unbalances of all nodes fall under (Constraints for VM range, NSV, ZSV and allowed number of PRDs to be
the required level. However, if we look at the VM levels across switched are [0.94,1.1], 2%, 4.5% and 3, respectively.
the network as shown in Fig. 10, it is evident that the strategy
provided by SIPM leads to VM violations at node 7-13 in Simulation results in Fig.12 show that the PRD control
phase c, which implies that making a strategy by SIPM that strategy provided by SNTM achieves the minimal current
neglects the network constraints may worsen the network’s unbalance for this period of all possible strategies. Moreover,
operational performance. Post-analysis also indicates that no operational violation is
It is noteworthy that not all nodes can be guaranteed to
satisfy the requirements on the VM or voltage unbalance level 4 As all CMs in the DT are under its upper limit for all strategies, the results
because only part of nodes in the network can be monitored. are not presented here.

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 8

observed, which, together with the accuracy analysis in the last NPRD and SIPM can be brought down to within the upper
subsection, demonstrates that the provided strategy is likely to limits effectively by SNTM, as shown in Fig. 15.
achieve global optimality. In summary, the simulation results of Case II demonstrate
4) Operational performance during the whole day: The that the proposed sensitivity-based method is accurate enough
operational performances of the network throughout the whole and can provide a satisfactory strategy that effectively im-
day are presented in Fig. 13-Fig. 15, where for NPRD, 5 proves the network’s operational performance, i.e., mitigating
adjustable customers at all periods are fixed at their original current unbalance in the DT while ameliorating both VM and
phase position c, b, a, c, c respectively as shown in Fig. 7. As voltage unbalance levels.
shown in Fig. 13, SNTM always provides a strategy that better
minimizes the current unbalance (z − + z 0 ) than NPRD, while D. Case III
SIPM may lead to a worse result, particularly around the
midday. It is noteworthy that although SIPM is to minimize the 1) Australian LVDN: For the Australian LVDN, the VMs
CM unbalance (z d ), the true operational state of the network at the primary side of the DT, i.e., node x, throughout the
reveals that this cannot be guaranteed due to neglecting the whole day are presented in the middle subgraph of Fig. 165 ,
phase angles of relevant terms as shown in Fig. 13. where voltage volatility and unbalance can be observed.
Simulation results for this case are presented in the top and
bottom subgraphs of Fig. 16 and Fig. 17.
NPRD
40 SNTM+PF
SIPM+PF
z-+z0 (A)

30
50 NPRD
20 SNTM+PF

z-+z0 (A)
40
10
30
20
10
NPRD Time (hour)
30 SNTM+PF

VM@Node x (p.u.)
25 SIPM+PF
1.09 Phase a
20 Phase b
zd (A)

1.08
Phase c
15 1.07
10 1.06
5 1.05
2 4 6 8 10 12 14 16 18 20 22 24
Time (hour) VM Level (p.u.) 1.1
1.08
1.06
Fig. 13. Current unbalance throughout the whole day for the IEEE-13 LVDN. 1.04 NPRD Average (NPRD) VM Upper Limit
1.02 SNTM+PF Average (SNTM+PF)
2 4 6 8 10 12 14 16 18 20 22 24
Time (hour)

1.06

1.04
Fig. 16. Current unbalance and voltage magnitude levels throughout the whole
1.02
day for the Australian LVDN.
VM Level (p.u.)

0.98
2
0.96 NPRD Average (NPRD) NSV Limit
SNTM+PF Average (SNTM+PF)
NSV Level (%)

0.94 1.5

0.92
1
0.9 SIPM+PF Average (SIPM+PF) VM Limits
NPRD Average (NPRD)
0.88 SNTM+PF Average (SNTM+PF) 0.5
2 4 6 8 10 12 14 16 18 20 22 24
Time (hour)
4 Time (hour)
ZSV Limit
ZSV Level (%)

3
Fig. 14. Voltage magnitude levels throughout the whole day for the IEEE-13
LVDN. 2

SIPM+PF Average (SIPM+PF) NSV Limit 2 4 6 8 10 12 14 16 18 20 22 24


NPRD Average (NPRD) Time (hour)
NSV Level (%)

3 SNTM+PF Average (SNTM+PF)

2
Fig. 17. Voltage unbalance levels throughout the whole day for the Australian
1
LVDN.
ZSV Limit Time (hour)
10
As shown in the figure, current unbalances in the DT can be
ZSV Level (%)

8
6
effectively mitigated, leading to a 67.10% reduction averagely
4 of the whole day. On operational security, the network experi-
2
ences violations of upper VM limits from around period 15 to
2 4 6 8 10 12 14
Time (hour)
16 18 20 22 24
period 16, which can be adequately addressed after controlling
PRDs by the proposed method, as shown in Fig. 16.
Fig. 15. Voltage unbalance levels throughout the whole day for the IEEE-13 Moreover, voltage unbalance level exceeds the upper limit
LVDN. slightly around period 6:00 and around 21:00, and significantly
from around 15:00 to 19:00 in the network. However, this
Based on Fig. 14, VMs for SNTM at most periods are issue can also be adequately addressed by optimally switch-
within the secure region except slight violations at period 20 ing PRDs, as shown in Fig. 17. Regarding the computation
and 21. However, it is found that the violations occur only efficiency, solving SNTM once only takes approximately 0.27
at the non-measurable node 10, and VMs of all measurable second, which implies the proposed method is readily applied
nodes are within the secure region for all periods. By contrast, in practical systems.
violations of VM limits frequently occur throughout the day
for the other two methods. Similar results are observed for 5 Different from the IEEE-13 LVDN, time resolutions for the Australian
voltage unbalance levels, where the significant violations for LVDN and the European LVDN are 5 minutes and 1 minute, respectively.

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 9

150
The current unbalances in the DT, NSV, ZSV and VM levels NPRD

z-+z0 (A)
100 SNTM+PF
at node 11 in the Australian LVDN before and after switching
50
PRDs are present in 18.

VM Level (p.u.)
1.08 NPRD Average (NPRD)
1.06
SNTM+PF Average (SNTM+PF)
BS-PRDs 1.04
z-+z0 (A)

30 AS-PRDs 1.02
20 1
0.98
10 0.96
2 4 6 8 10 12 14 16 18 20 22 24
2 6
NPRD Average (NPRD) NSV Limit ZSV Limit
NSV/ZSV(%)

NSV (BS-PRDs) ZSV (BS-PRDs) NSV Limit SNTM+PF Average (SNTM+PF) 5

NSV Level (%)


4

ZSV Level (%)


NSV (AS-PRDs) ZSV (AS-PRDs) ZSV Limit 1.5
3 4
2 1 3
1
2
0.5
2 4 6 8 10 12 14 16 18 20 22 24 1
1.11
1.1
VM@Node 11 (p.u.)

VM@Node 11 (p.u.)

1.1 2 4 6 8 10 12 14 16 18 20 22 24 2 4 6 8 10 12 14 16 18 20 22 24
1.08
Time (hour) Time (hour)
1.06 1.09
1.04
Phase a (BS-PRDs) Phase a (AS-PRDs) VM Limits 1.08
1.02 Phase b (BS-PRDs) Phase b (AS-PRDs)

1
Phase c (BS-PRDs) Phase c (AS-PRDs)
1.07
Fig. 19. Current unbalance (in the DT), VM levels and NSV/ZSV for the
2 4 6 8 10 12 14 16 18 20 22 24
Time (hour)
13 14
Time (hour)
15 16 European LVDN throughout the day.

Active Power 60
40
Reactive Power
Fig. 18. Current unbalance (in the DT), NSV/ZSV and VM levels of node
20 50

Net Demand (kW/kVA)


11 before and after switching PRDs (BS-PRDs/AS-PRDs) for the Australian

Computation Time (s)


LVDN throughout the day. 0 40

As shown in Fig. 18, current unbalances in the DT are -20 30


significantly mitigated after optimizing positions of all PRDs.
-40
Moreover, before switching PRDs, violations for both voltage 20

unbalance and VM occur at around 19:00 and from 13:00 to -60


10
16:00, respectively. The violations are adequately addressed
after rearranging the positions of all PRDs. The simulation -80
results further demonstrate that the efficiency of the proposed 2 4 6 8 10 12 14 16 18 20 22 24 10 20
Number of PRDs
30 40 50
Time (hour)
method in balancing the network while ameliorating the net-
work’s operational performance.
Fig. 20. Net demands (left) and computation efficiency under various number
2) European LVDN: Simulations results, including the cur- of PRDs (right) for the European LVDN.
rent unbalances in the DT, VM levels, and voltage unbalance
of all nodes through the day, is presented in Fig. 19, which,
similar to the Australian LVDN, show that current unbalance
in the DT are mitigated based on the proposed method for all number of PRDs is over 30. Therefore, if the algorithm is
periods compared with NPRD, i.e., the case when no PRD is going to be executed every 1 minute, which is the case for
installed in the network. It is noteworthy that, voltage unbal- the studied system, we can be sure that the proposed method
ance violation still exists after optimizing PRD positions, and is efficient when the number of PRDs is under 30. Moreover,
there is significant voltage rise around the noon. However, this even if the system is with more PRDs, the proposed method
is reasonable noting that: 1) Merely switching PRDs cannot can still be applicable either by imposing a time limit for the
guarantee that the network meets all operational requirements, solver or lower the frequency of executing the algorithm, say
as discussed previously; 2) There is a significant export of every 10 minutes.
active power in the midday due to high PV generations, as
shown on the left side of Fig. 20. Nonetheless, compared V. C ONCLUSIONS AND D ISCUSSIONS
with NPRD, the voltage unbalance violations are avoided or
reduced for almost all periods where the violation occurs, and This paper studies the optimal control of PRDs in the low-
VMs throughout the day are within the required range. voltage distribution network when only partial information is
To further address the unbalance issue while guaranteeing available. Major conclusions are summarized as follows.
the security of the network, coordinate control of PRDs with 1) To collect the angle differences of nodal phase voltages,
other controllable equipment, either from the network operator which are critical in formulating the optimization model
or residential customers, is needed, which falls into our future to control PRDs, an efficient and accurate optimization-
research interest. based method is presented based on measured magni-
Regarding the computation efficiency, the solver takes 0.13 tudes of phase voltages and phase-to-phase voltages.
second each time it is executed. Noting that the solver takes 2) Sensitivities of an LVDN with PRDs are systematically
less time than that for the Australian LVDN although the analyzed, which provides an efficient method to bridge
European LVDN is larger in size, the efficiency of the pro- future operational states of measurable parameters after
posed method is further tested for the period corresponding switching PRDs with their current values.
to 12:00 for the European LVDN. Specifically, PRD locations 3) Optimally controlling PRDs via the proposed method
will be randomly selected with the total number ranging from can effectively mitigate current unbalances in the DT
1 to 55, and each scenario constructed will be solved with its while ensuring the network’s operational requirements at
computation time recorded. most periods. By contrast, inappropriate control of PRDs
The simulation results under various numbers of PRDs are may worsen the network’s operational performance.
presented on the right side of Fig. 20, which shows that the It is also noteworthy that PRDs are controlled regularly
solver generally takes more time as the number of PRDs throughout the day in this paper. To further limit the impacts of
increases, and the increment is more significant when the the switching process on residential customers, the proposed

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 10

method can be executed to update the control strategy only [11] S. Karagiannopoulos, P. Aristidou, and G. Hug, “A centralised control
when monitored unbalance, either current unbalance in the method for tackling unbalances in active distribution grids,” in 2018
Power Systems Computation Conference (PSCC), Dublin, Ireland, 2018.
DT or voltage unbalance of measurable nodes, exceeds a [12] F. Geth, J. Tant, R. Belmans, and J. Driesen, “Balanced and unbalanced
predefined level, or VM violation is detected for monitored inverter strategies in battery storage systems for low-voltage grid sup-
nodes. However, as substantial voltage unbalance may pass port,” IET Generation, Transmission Distribution, vol. 9, no. 10, pp.
929–936, 2015.
from the upstream medium-voltage distribution network to [13] T.-H. Chen and J.-T. Cherng, “Optimal phase arrangement of distribu-
LVDN, and the target of balancing the network may be tion transformers connected to a primary feeder for system unbalance
unachievable merely by PRDs under some extreme operational improvement and loss reduction using a genetic algorithm,” IEEE Trans.
Power Syst., vol. 15, no. 3, pp. 994–1000, 2000.
conditions, other measures may be required to further tackle [14] C. H. Lin, C. S. Chen, H. J. Chuang, and C. Y. Ho, “Heuristic rule-based
the unbalance issue. Therefore, exploring the formulation phase balancing of distribution systems by considering customer load
when other controllable devices, e.g. PV inverter, battery patterns,” IEEE Trans. Power Syst., vol. 20, no. 2, pp. 709–716, 2005.
[15] C. H. Lin, C. S. Chen, M. Y. Huang, H. J. Chuang, M. S. Kang,
energy storage system, tap position of the DT, are incorporated C. Y. Ho, and C. W. Huang, “Optimal phase arrangement of distribution
in the network and developing efficient algorithms fall in feeders using immune algorithm,” in Proc. International Conference on
our future research interests. Moreover, how to consider the Intelligent Systems Applications to Power Systems, Toki Messe, Niigata,
Japan, 2007.
generation or demand uncertainties when updating the control [16] L. Chia-Hung, C. Chao-Shun, C. Hui-Jen, H. Ming-Yang, and H. Chia-
strategies is also worthy of more effort. Wen, “An expert system for three-phase balancing of distribution feed-
Besides, switching PRDs could have some impacts on ers,” IEEE Trans. Power Syst., vol. 23, no. 3, pp. 1488–1496, 2008.
[17] R. A. Hooshmand and S. Soltani, “Fuzzy optimal phase balancing of
appliances of residential customers. Nevertheless, to the au- radial and meshed distribution networks using bf-pso algorithm,” IEEE
thors’ experience, the switching process is not likely to cause Trans. Power Syst., vol. 27, no. 1, pp. 47–57, 2012.
significant impacts or damages on common residential ap- [18] K. Wang, S. Skiena, and T. G. Robertazzi, “Phase balancing algorithms,”
Electric Power Systems Research, vol. 96, pp. 218–224, 2013.
pliances. Moreover, when installing PRDs in a real power [19] J. Horta, D. Kofman, D. Menga, and M. Caujolle, “Augmenting der
system, customers with life-support equipment or PV panels hosting capacity of distribution grids through local energy markets and
are excluded to minimize the risk of incurring outage or dynamic phase switching,” in Proc. the Ninth International Conference
on Future Energy Systems - e-Energy ’18, Karlsruhe, Germany, 2018.
temporary shutdown of such devices. [20] X. Geng, S. Gupta, and L. Xie, “Robust look-ahead three-phase bal-
On placing PRDs, there are experience-based and ancing of uncertain distribution loads,” in The Hawaii International
optimization-based methods. In the former one, based on Conference on System Sciences (HICSS), Maui, Hawaii, 2018.
which the PRD locations are selected for the European LVDN, [21] ARENA, “Demonstration of three dynamic grid-side technologies in
australia,” https://arena.gov.au/projects/demonstration-of-three-dynamic
customers with higher demand levels are usually selected -grid-side-technologies/.
because they could balance the network more effectively, [22] B. Liu, K. Meng, Z. Y. Dong, P. Wong, and T. Ting, “Unbalance
particularly when the penetration of PV is high. In the latter, mitigation via phase-switching device and static var compensator in low-
voltage distribution network,” IEEE Trans. Power Syst., 2020.
placing PRDs could be formulated as an optimization problem [23] H. Ahmadi, J. R. Marti, and A. von Meier, “A linear power flow
considering both installation and operational constraints based formulation for three-phase distribution systems,” IEEE Trans. Power
on historical data, which is the one used in this paper for Syst., vol. 31, no. 6, pp. 5012–5021, 2016.
[24] L. Gan and S. H. Low, “Convex relaxations and linear approximation for
the IEEE-13 LVDN and the Australian LVDN. However, opf in multiphase radial networks,” in Proc. Power Systems Computation
placing PRDs may also be impacted by specific PRD control Conference, Dublin, Ireland, 2018.
algorithms. How to install PRDs considering the specific [25] B. A. Robbins and A. D. Dominguez-Garcia, “Optimal reactive power
dispatch for voltage regulation in unbalanced distribution systems,” IEEE
control algorithm, and to achieve a compromise between the Trans. Power Syst., vol. 31, no. 4, pp. 2903–2913, 2016.
observability and controllability in the network is beyond the [26] J. A. Castrillon, J. S. Giraldo, and C. A. Castro, “MILP for optimal reac-
scope of this paper and also needs more efforts in our future tive compensation and voltage control of distribution power systems,” in
Proc. IEEE Power & Energy Society General Meeting, Chicago, USA,
work. 2017.
[27] J. C. Lopez, J. F. Franco, M. J. Rider, and R. Romero, “Optimal
R EFERENCES restoration/maintenance switching sequence of unbalanced three-phase
distribution systems,” IEEE Trans. Smart Grid, vol. 9, no. 6, pp. 6058–
[1] P. Wong, A. Kalam, and R. Barr, “Modelling and analysis of practical 6068, 2018.
options to improve the hosting capacity of low voltage networks for [28] T. Jen-Hao, “A direct approach for distribution system load flow solu-
embedded photo-voltaic generation,” IET Renew. Power Gener., vol. 11, tions,” IEEE Trans. Power Del., vol. 18, no. 3, pp. 882–887, 2003.
no. 5, pp. 625–632, 2017. [29] H. Fang, W. Sheng, J. Wang, Y. Liang, J. Wang, and S. Wang, “Research
[2] M. M. Haque and P. Wolfs, “A review of high PV penetrations in LV on the method for real-time online control of three-phase unbalanced
distribution networks: Present status, impacts and mitigation measures,” load in distribution area,” Proceedings of the CSEE, vol. 35, no. 9,
Renewable and Sustainable Energy Reviews, vol. 62, pp. 1195–1208, 2015.
2016. [30] G. Grigoras, M. Gavrilas, B. Neagu, O. Ivanov, I. Tris, tiu, and C. Bulac,
[3] J. Zhu, M.-Y. Chow, and F. Zhang, “Phase balancing using mixed-integer “An efficient method to optimal phase load balancing in low voltage
programming,” IEEE Trans. Power Syst., vol. 13, no. 4, pp. 1487–1492, distribution network,” in 2019 International Conference on Energy and
1998. Environment (CIEM), Timisoara, Romania, 2019.
[4] J. Zhu, G. Bilbro, and M.-Y. Chow, “Phase balancing using simulated [31] S. Liu, R. Jin, H. Qiu, X. Cui, Z. Lin, Z. Lian, Z. Lin, F. Wen, Y. Ding,
annealing,” IEEE Trans. Power Syst., vol. 14, no. 4, pp. 1508–1513, Q. Wang, and L. Yang, “Practical method for mitigating three-phase
1999. unbalance based on data-driven user phase identification,” IEEE Trans.
[5] A. Kharrazi, V. Sreeram, and Y. Mishra, “Assessment techniques of Power Syst., vol. 35, no. 2, pp. 1653–1656, 2020.
the impact of grid-tied rooftop photovoltaic generation on the power [32] B. Liu, K. Meng, P. Wong, Z. Y. Dong, C. Zhang, B. Wang, T. Ting,
quality of low voltage distribution network - A review,” Renewable and and Q. Qu, “Improving operational feasibility of low-voltage distribution
Sustainable Energy Reviews, vol. 120, 2020. network by phase switching devices,” in Proc. 8th IET International
[6] “Solar report (January 2019),” Australian Energy Council, Report, Conference on Renewable Power Generation, Shanghai, China, 2019.
2019. [Online]. Available: https://www.energycouncil.com.au/media [33] A. Gupte, S. Ahmed, M. Cheon, and S. Dey, “Solving mixed integer
/15358/australian-energy-council-solar-report -january-2019.pdf bilinear problems using MILP formulations,” SIAM Journal on Opti-
[7] F. Shahnia, P. J. Wolfs, and A. Ghosh, “Voltage unbalance reduction mization, vol. 23, no. 2, pp. 721–744, 2013.
in low voltage feeders by dynamic switching of residential customers [34] J. Lofberg, “Yalmip : a toolbox for modeling and optimization in
among three phases,” IEEE Trans. Smart Grid, vol. 5, no. 3, pp. 1318– Matlab,” in Proc. IEEE International Symposium on Computer Aided
1327, 2014. Control Systems Design, New Orleans, USA, 2004.
[8] “Distributed voltage control strategy for LV networks with inverter- [35] I. IBM, “Cplex user’s manual,” 2015. [Online]. Available: https:
interfaced generators,” Electric Power Systems Research, vol. 107, pp. //www.ibm.com/analytics/cplex-optimizer
85 – 92, 2014. [36] “Electricity distribution code (version 9),” Essential Services Commis-
[9] S. Weckx and J. Driesen, “Load balancing with EV chargers and pv sion, Report, 2015. [Online]. Available: https://www.esc.vic.gov.au/sit
inverters in unbalanced distribution grids,” IEEE Trans. Sustainable es/default/files/documents/Electricity-Distribution-Code-Version-9.pdf
Energy, vol. 6, no. 2, pp. 635–643, 2015. [37] “Simulation data for LVDNs,” https://drive.google.com/open?id=1SxS
[10] P. Fortenbacher, M. Zellner, and G. Andersson, “Optimal sizing and iQR05nDituIQDmZnDQU o1YWy22zU.
placement of distributed storage in low voltage networks,” in 2016 Power [38] “Simulation data for the IEEE European LVDN,” https://site.ieee.org/
Systems Computation Conference (PSCC), Genoa, Italy, 2016. pes-testfeeders/resources/.

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRD.2020.3022061, IEEE
Transactions on Power Delivery
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. XX, NO. XX, AUG 2019 11

Bin Liu (M’19) received the Bachelor, Master and Xuejun Li received Master degree of electric power
Ph.D. degrees all in electrical engineering from system from Shandong University, Ph.D. degree
Wuhan University, Wuhan, China, in 2009, China from Beijing Jiaotong University and MBA degree
Electric Power Research Institute, Beijing, China, from The University of Sydney. Dr. Li is currently
in 2012 and Tsinghua University, Beijing, China, the Deputy Executive General Manager of Jemena
in 2015, respectively. He is currently a Research and looks after the gas and electricity T&D network
Associate at The University of New South Wales, assets. His research interest includes asset manage-
Sydney, NSW, Australia. He also worked as a Re- ment, information system management, and project
search Assistant at The Hong Kong Polytechnic management. He is a Charted Engineer and member
University, Hong Kong, in 2012 and as a power of IET (The Institute of Engineer and Technology).
system engineer in China from 2015 to 2017. His
research interests include optimization theory and its
application in power and multi-energy systems, renewable energy integration
and energy storage.

Ke Meng (M’10-SM’19) received the Ph.D. de-


gree in electrical engineering from the University
of Queensland, Brisbane, QLD, Australia in 2009.
He is currently a Senior Lecturer with the School
of Electrical Engineering and Telecommunications,
The University of New South Wales, Sydney, NSW,
Australia. He was previously a Lecturer with the
School of Electrical and Information Engineering,
The University of Sydney, Sydney, Australia. His
research interests include pattern recognition, power
system stability analysis, wind power and energy
storage.

Zhao Yang Dong (M’99-SM’06-F’17) received the


Ph.D. degree in electrical engineering from The
University of Sydney, Sydney, NSW, Australia, in
1999. He is currently a professor of energy systems
with The University of New South Wales, Sydney,
Australia, where he is also Director of UNSW Digi-
tal Grid Futures Institute. His previous roles include
Ausgrid Chair and Director of the Ausgrid Centre
for Intelligent Electricity Networks, University of
Newcastle, Australia. His research interest includes
power system planning, power system dynamics
and stability, smart grid/microgrid, renewable energy
systems, electricity market, and load modelling. Prof. Dong has been serving
as an Editor for several IEEE transactions and IET journals. He is a Fellow
of IEEE.

Peter K.C. Wong (SM’17) was born in Hong


Kong. He received his BSc (Eng) degree from the
University of Hong Kong in 1983, and a PhD degree
from Victoria University, Australia, in 2017, both in
electrical engineering. Peter is currently the Network
Technology & Measurement Manager in Jemena,
a multi Australian utility operating in electricity,
gas and water areas, and is responsible for new
technologies, future networks, Distributed Energy
Resources (DER) integration, smart meter strategies
and applications.
Peter is a Chartered Professional Engineer, a Fel-
low of the Institution of Engineers Australia, Senior Member of the Institution
of Electrical & Electronics Engineers, and a Corporate Member of the Institu-
tion of Engineering and Technology (U.K.). Peter has a distinguished career
as a professional engineer. He received the Energy Networks Association
Overseas Study Tour Award in 2011 which he used to study smart grid
developments in Asia, Europe and the U.S. He was awarded the John Madsen
medal for the best paper published in the Australian Journal of Electrical &
Electronic Engineering in 2015, and received the Lifetime Achievement in
Energy Award from the Electrical Energy Society of Australia in 2017.

0885-8977 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Auckland University of Technology. Downloaded on September 13,2020 at 08:34:56 UTC from IEEE Xplore. Restrictions apply.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy