IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO.

5, SEPTEMBER 2019 3605

Multi-Phase Losses Allocation Method for Active

Distribution Networks Based on Branch
Current Decomposition
Muhammad Usman , Member, IEEE, Massimiliano Coppo, Member, IEEE, Fabio Bignucolo ,
Roberto Turri , Senior Member, IEEE, and Alberto Cerretti

Abstract—Losses allocation, being a purely economical problem, MPLAP Multi-Phase LAP

demands a fair allocation procedure that can distribute the losses in NLAF Neutral Losses Allocation Factor
an impartial manner by avoiding cross-subsidies among end users. PLAF Phase Losses Allocation Factor
In view of this, a novel multi-phase losses allocation methodology is
proposed, which fairly segregates the losses associated with cross- QLP Quadratic Losses Partitioning
terms of phase-currents and also provides an explicit information RCLP Resistive Component based Losses Partitioning
about losses allocated to the neutral. The effect of phase loading on TAL Total Allocated Losses
another phase is comprehensively taken into consideration while TN Transmission Network
distributing losses among end users. Furthermore, allocation of VLC Variable Loss Coefficient
neutral losses to each phase of a node is avoided as these losses
exist in a system due to its inherent characteristics. Quadratic and
geometric losses partitioning approaches have been utilized to de-
I. INTRODUCTION
termine the share of losses between phase currents of same and
different nodes. The methodology is implemented on a modified OSSES Allocation can provide valuable information to the
IEEE 13-node test system in the presence of distributed generators
and results show that the proposed scheme leads to a more fair
penalization and reward of passive and active end users as com-
L DSOs regarding contribution of each end-user in the over-
all power system losses. Therefore, it is extremely important to
pared to the resistive component based losses allocation method. fairly allocate these losses to the end-users by avoiding cross-
In addition to this, concept of neutral losses allocation factor along subsidies among them. In order to have an impartial distribution
with their possible application in the management of distribution of losses, several LAPs have been proposed in the literature for
network has also been discussed briefly. both transmission and distribution networks; however, methods
Index Terms—Cross-terms redistribution, distribution system, developed for TNs cannot be applied directly to DNs because of
multi-phase losses allocation procedure, neutral losses allocation. their special characteristics and configuration. As a result, apart
from modifying the existing LAPs developed for the TNs to
ACRONYMS
make them applicable to the DNs, researchers are also propos-
BCDLA Branch Current Decomposition method for Losses
ing new circuit and tracing-based LAPs for the DNs.
Allocation
A derivate-based marginal loss coefficient and direct loss
CLP Classical Loss Partitioning
coefficient strategy is proposed in [1] which takes into account
DG Distributed Generator
temporal and spatial variations of end-users but involves
DN Distribution Network
complex calculations related to network Jacobian and Hes-
DSO Distribution System Operator
sian matrices. An improved Z-bus method is reported in [2],
GLP Geometric Losses Partitioning
which overcomes the shortcomings of Z-bus method [3], by
LA Losses Allocation
redistributing the losses allocated to the slack bus to all load
LAP Losses Allocation Procedure
buses based upon their energy consumption. A modified Y-bus
MPG Multiple Point Grounding
method which avoids reconciliation factors and minimizes
cross-subsidies is proposed in [4].
Manuscript received July 17, 2018; revised November 5, 2018 and January 9, In [5], total network losses are allocated using impedance
2019; accepted March 23, 2019. Date of publication March 29, 2019; date of cur- of each network branch which leads to a paradox where less
rent version August 22, 2019. Paper no. TPWRS-01098-2018. (Corresponding
author: Muhammad Usman.) reactive power consuming end-users are allocated more losses if
M. Usman, M. Coppo, F. Bignucolo, and R. Turri are with Depart- transmission angle of the branch becomes greater than the trans-
ment of Industrial Engineering, University of Padova, Padua 35131, Italy mission angle of its connected load. This paradox is removed in
(e-mail:, [email protected]; [email protected];
[email protected]; [email protected]). [6] by a resistance-based LAP which establishes a direct relation
A. Cerretti is with E-Distribuzione S.p.A., Rome 00198, Italy (e-mail:, between each branch current and net injected downstream nodal
[email protected]). currents. As a result, a single forward/backward sweep of a
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. network allocates each branch losses to its downstream nodes.
Digital Object Identifier 10.1109/TPWRS.2019.2908075 However, this method fails to address properly the issue of
0885-8950 © 2019 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: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.
 3606 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 5, SEPTEMBER 2019

cross-subsidies among end-users. A power summation based

LAP, which in principle is identical to the previously mentioned
method, is proposed in [7] which develops a direct relation
between line losses and net injected complex nodal power, and
allocates cross-terms quadratically among end-users. However,
this method improperly allocates losses to the same power
consuming end-users which are present in a network at different Fig. 1. Example of a simple DN with series impedance and transversal
locations and distances with respect to the root node. The same admittances.
authors developed an energy summation method [8] which
takes into account the stochastic nature of load and generation
curves and can be successfully used for planning purposes.
However, from computational point of view, it fails to handle
large data sets.
A method based on upstream and downstream versions of pro-
portional sharing algorithm is reported in [9] which determines
the share of passive end-users (which are not equipped with dis-
tributed generators) and active end-users (which are equipped
with distributed generators) in each branch. A sequential shapely
value algorithm based LAP is proposed in [10] which is com-
putationally more efficient than conventional shapely algorithm
based LAP. Furthermore, it avoids power flow solution data if Fig. 2. Multi-phase branch model showing mutual coupling among phases
and ground.
metered data is available. A participation matrix based LAP is
discussed in [11] which avoids spatial cross-subsidies, recovers
total losses and is applicable in pricing mechanism of a radial
DN. However, it employs contractual power data rather than utilized by DSOs in planning and management of DNs by taking
real-time data in the allocation of cross-terms. specific actions such as the identification of highly unbalanced
All the above-mentioned methods have been developed us- nodes in a network and consequently, implementation of a suit-
ing single-phase equivalent representation of DNs by assuming able unbalance compensation scheme at those nodes as reported
symmetrical network configuration and balanced loading condi- in [13]. Furthermore, a fair financial policy/plan can be built on
tions. However, as DNs are inherently unbalanced, single-phase the top of LA results, in order to penalize/reward the end-users
based LAPs do not provide explicit information about losses al- for their negative/positive impact on the network losses as well
located to each phase of a node including neutral. As per authors as to charge them for their network usage.
knowledge, the only three-phase LAP is reported in [12]; how- The paper is organized as follows. Section II discusses the
ever, this methodology imposes same weighting factors to the shortcomings of three-phase LAP [12], whereas a novel MPLAP
cross-terms of different phase-currents. As a result, high losses is introduced in Section III. In section IV, comparative analysis
can be allocated to lightly loaded phase(s) due to the flow of high between method reported in [12] and proposed MPLAP is car-
current in heavily loaded phase(s). Furthermore, it allocates neu- ried out on a modified IEEE-13 node test system. The paper is
tral losses to each phase in proportion to the latter current which concluded in Section V.
makes this scheme unfair as neutral losses are not dependent
upon the energy consumption of end-users but are related to the II. THREE-PHASE BRANCH CURRENT DECOMPOSITION BASED
degree of unbalance in a network. Furthermore, no LAP dis- METHOD FOR LOSSES ALLOCATION
cusses the possible applications of the results of LAPs in better
The three-phase BCDLA is briefly recalled here, as it sets
management of DNs.
the basis for the development of MPLAP, with focus on the as-
This paper fills these gaps by introducing a novel MPLAP,
sumptions implied in this method. Consider a simple radial DN
which introduces appropriate weighting factors related to the
as shown in Fig. 1 where each branch is modelled as a multi-
cross-terms by taking into account the real-time value of phases
conductor branch as shown in Fig. 2. According to RCLP ap-
and neutral current, and also develops a mathematical formula-
proach which is the foundation of three-phase BCDLA method,
tion for explicitly representing the losses allocated to the neutral.
active power losses in a three-phase branch b can be computed
Moreover, a valuable discussion on neutral losses allocation fac-
as follows
tors is also presented.
  
The proposed methodology can be successfully applied to any (b)
pl =  i∗(b) ⊗ Rabc · i(b) (1)
radial distribution network (high, medium or low-voltage DN),
equipped with or without neutral conductor. In the absence of (b) (b) (b) (b)
the neutral conductor, the proposed scheme would allocate the where pl = [pla plb plc ]T represents branch active power
(b) (b) (b)
phase losses associated with the cross-terms in a fairly manner, losses, i(b) = [ia ib ic ]T represents branch current, Rabc =
whereas in the presence of the neutral conductor, additional in-  {Zabc } where Zabc is the Kron reduced form of the original
formation about NLAFs would also be available. From practical multi-conductor impedance matrix Zp and ⊗ is component wise
point of view, the information obtained from LA results can be multiplication.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.
 USMAN et al.: MULTI-PHASE LOSSES ALLOCATION METHOD FOR ACTIVE DNS BASED ON BRANCH CURRENT DECOMPOSITION 3607

The total active power losses in such a branch can then be

determined as
(b) (b) (b) (b)
pltot = pla + plb + plc (2)
Since Rabc is symmetrical, (2) can be expanded by invoking
Rxy = Ryx = Rx/y where x, y ∈ {a, b, c}
     
(b)
pltot = i∗a · Raa · ia + i∗b · Rbb · ib + i∗c · Rcc · ic Fig. 3. Two phase generic radial DN for calculation of active power losses in
terminal and non-terminal branches.
   
+ Ra/b i∗aib + i∗bia + Rb/c i∗bic + i∗cib
 
lightly loaded phase(s) will be held accountable for the losses for
+ Rc/a i∗aic + i∗cia (3)
which they are not responsible. On the other hand, end-users of
which can further be written as heavily loaded phase(s) will be penalized less and consequently,
      losses will not be distributed fairly among phase(s).
(b)
pltot = i∗a · Raa · ia + i∗b · Rbb · ib + i∗c · Rcc · ic Furthermore, BCDLA reduces a n × m original branch
      impedance matrix Zp to a 3 × 3 phase impedance matrix Zabc
+ Ra/b 2 i∗a · ib +Rb/c 2 i∗b · ic under the assumption that neutral conductor is solidly grounded
   i.e., earthing resistance Rgr ≈ 0, at both of its ends and as a re-
+ Rc/a 2 i∗c · ia (4) sult, voltage is equal at all points on a conductor. However, this
is generally not the case in real DNs where neutral is grounded
Finally, (4) can be split into each phase losses components at a single or multiple points through a small earthing resis-
    
(b)
pltot = i∗a · Raa · ia + Rab  i∗a · ib tance (Rgr ≈ 5–15 Ω) and consequently, voltage profile along
a neutral conductor follows a descending (single point ground-
    
ing) or U-shaped (MPG) pattern. Under such scenarios, applying
+Rac  i∗c · ia + i∗b · Rbb · ib
Kron-reduction on Zp would lead to the over-approximation of
      the results. Although, in the case of solidly-grounded MPG sys-
+Rba  i∗a · ib + Rbc  i∗b · ic
tems, the assumption implied in Kron reduction technique can be
     practically realized. But, it is unrealistic to impose neutral losses
+ i∗c · Rcc · ic + Rca  i∗c · ia on each phase as these losses exist in a system due to uneven
   distribution of single-phase end-users which leads to inherent
+Rcb  i∗b · ic (5) unbalancing of the grid. Therefore, in the proposed methodol-
Since ogy, explicit information about the losses allocated to the neutral
    is provided.
 i∗x · iy =  i∗y · ix (6) The equal distribution of cross-terms among phases and al-
location of neutral losses to phases give rise to cross-subsidies
(5) can then be written as and ultimately make BCDLA-based methodology unfair.
    
(b)
pltot = i∗a · Raa · ia + Rab  i∗a · ib
III. MULTI-PHASE BRANCH CURRENT DECOMPOSITION BASED
    
LOSSES ALLOCATION PROCEDURE
+Rac  i∗a · ic + i∗b · Rbb · ib
      The shortcomings of three-phase BCDLA method are ad-
+Rba  i∗b · ia + Rbc  i∗b · ic dressed in this section by proposing a novel MPLAP which
     treats the neutral conductor like any other phase conductor and
+ i∗c · Rcc · ic + Rca  i∗c · ia assigns appropriate weighting factors to the cross-terms in order
   to avoid equal distribution of losses associated with these terms
+Rcb  i∗c · ib (7) among end-users. In order to derive a mathematical formulation
of the proposed approach, a generic two-phase (r and s) radial
The terms in square brackets represent losses allocated to DN composed of two branches as shown in Fig. 3 is considered.
each phase of a branch. The first term within these brackets The active power losses of these branches are first calculated
(i∗x · Rxx · ix ) shows losses in each phase due to its self cur- as reported in the below-mentioned subsections which form the
rent whereas the remaining two entries, defined as cross-terms basis of proposed MPLAP. For the sake of clarity, a two-phase
{Rxy · (i∗x · iy )}, represent losses in a particular phase due branch model is considered, however, the proposed approach
to the current in other phase. It can be seen in (4) and (5), as can be extended to m number of conductors. It should be kept
proposed in [12], that cross-terms are equally divided among in mind that losses formulae presented in the following subsec-
phases which leads to a paradox in case of an unbalanced sys- tions should be used only for LA purpose. The standard formulae
tem as heavily loaded phase(s) will induce more losses in lightly for the calculation of active power losses in a multi-conductor
loaded phase(s) and, as a result, end-users connected to such branch can be found in [14].

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.
 3608 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 5, SEPTEMBER 2019

A. Active Power Losses of Terminal Branch induced losses can be achieved among end-users. In this paper,
these coefficients are determined by QLP and GLP approaches
Starting from terminal branch j, active power losses can be
computed as follows as explained in the Subsection III-C.
   Once these factors are calculated, branch active power losses
(j)
pl =  i∗(j) ⊗ R(j) rs ∗ i (j)
(8) can be written in the following form
  (j) (j)

(j)
where pl = [
(j) (j)
plr pls ]T represents branch active power losses (j) Ir∗i ii
αrr ii
∗ Rrr αrs ∗ Rrs I ri
(j) (j)
pl = ⊗ ∗ (16)
and i(j) = [ir is ]T represents branch currents. Since the ter- Is∗i ii
αsr ∗
(j)
Rsr ii
αss ∗
(j)
Rss I si
minal branch current is equal to its receiving node i current, (8)
can be written as Eq. (16) is used for the allocation of losses in a terminal branch
   (j) (j)  to the phases of its downstream node.
(j)
(j) pr Ir∗i Rrr Rrs I ri
pl = (j) =  ∗
⊗ (j) (j)
· (9)
ps I si Rsr Rss I si
B. Active Power Losses of Non-Terminal Branch
The total active power losses in branch j can then be computed The branch current for a non-terminal branch can be written
as as a sum of net injected nodal currents
(j)
ptot = p(j)
r + ps
(j)
(10) 
    i(b) = Iq (17)
(j)
ptot = Ir∗i · Rrr
(j)
· Iri + Ir∗i · Rrs
(j)
· I si q∈Q
   
+ Is∗i · Rsr (j)
· Iri + Is∗i · Rss(j)
· I si (11) where Q is a set of downstream nodes being supplied by a non-
terminal branch. Consequently, active power losses for a non-
(j) terminal branch k in Fig. 3 can be computed as
As Rrs is symmetric, (11) can be re-written by substituting
(j) (j)
Rrs = Rsr = R(j)   (k) (k) 
    (k) Ir∗h + Ir∗i Rrr Rrs I rh + I ri
(j) pl = ⊗ ∗
ptot = Ir∗i · Rrr
(j)
· Iri + Is∗i · Rss
(j)
· I si ∗ ∗
I sh + I si (k) (k)
Rsr Rss I sh + I si
(18)
+ R(j) · Ir∗i · Isi + Is∗i · Iri (12)
Since Utilizing (13) and (15), (18) can be expanded as shown in (19)
at the bottom of this page, which can be re-written in compact
Ir∗i · Isi + Is∗i · Iri = (Ir∗i · Isi ) + (Ir∗i · Isi )∗ form as in (20)
= 2 Ir∗i · Isi (13) ⎧  ∗   (k) (k) ⎫
⎪
⎪ I R R ⎪
⎪
 ⎪
⎨ r
∗
w
⊗ rr rs ⎪
⎬
(12) can be further written as (k) I sw R
(k)
R
(k)
pl =  sr ss
     wq  (20)
(j) ⎪  wq
αrr ∗ Irq αrs ∗ Isq ⎪
ptot = Ir∗i · Rrr w∈Q ⎪ ⎪
(j)
· Iri + Is∗i · Rss
(j)
· I si ⎪
⎩ ⊗ q∈Q wq ⎪
⎭
wq
αsr ∗ Irq αss ∗ Isq
+ R(j) · 2 Ir∗i · Isi (14)
Equations (16) and (20) calculate active power losses in terminal
The cross-terms 2 Ir∗i · Isi can then be split into two com- and non-terminal branches respectively by introducing VLCs
ponents as follows which depend on the current in each phase of a node.
For a generic multi-phase branch having three-phase (a, b, c)
2 Ir∗i · Isi = αrs
ii
·  Ir∗i · Isi + αsr
ii
·  Ir∗i · Isi
and m neutral conductors, (16) and (20) can be generalized as
(15)
ii ii follows
where αrs and αsr are named as VLCs for phase r and s of node
i respectively. It is clear from (15) that equal division of cross-
pTl r =  I∗p ⊗ ŘTp r ∗ Ip (21)
terms among phases is avoided by introducing VLCs and, as a
result, the goal of LA problem then becomes to determine these  
factors which allow to take into account the loading effect of one pnT
l
r
= I∗wp ⊗ RnT
p
r
⊗ Ǐwqp (22)
phase on another phase, so that a fair segregation of cross-terms w∈Q q∈Q

   (k) (k)  hh 
(k) Ir∗h Rrr Rrs (αrr ∗ Irh + αrrhi hh
∗ Iri ) (αrs hi
∗ Ish + αrs ∗ I si )
pl = ⊗ ⊗
Is∗h (k)
Rsr Rss
(k) hh
(αsr hi
∗ Irh + αsr hh
∗ Iri ) (αss hi
∗ Ish + αss ∗ I si )
 ∗   ii  (19)
(k) (k) ih ii ih
I ri Rrr Rrs (αrr ∗ Iri + αrr ∗ Irh ) (αrs ∗ Isi + αrs ∗ I sh )
+ ⊗ ⊗
Is∗i (k)
Rsr Rss
(k) ii
(αsr ih
∗ Iri + αsr ii
∗ Irh ) (αss ih
∗ Isi + αss ∗ I sh )

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.
 USMAN et al.: MULTI-PHASE LOSSES ALLOCATION METHOD FOR ACTIVE DNS BASED ON BRANCH CURRENT DECOMPOSITION 3609

where Tr and nTr stand for terminal and non-terminal branch phase a current vector ia position is chosen as a reference, how-
respectively, and ŘTp r and Ǐwqp are defined as ever, in the absence of load connected to phase a, phase b current
⎡
(T r) (T r)
⎤ vector ib position becomes the reference.
ii ii
αaa ∗ Raa · · · αan ∗ Ranm
⎢ m
⎥
ŘTp r = ⎢
⎣ :
..
. :
⎥
⎦ D. Multi-Phase Losses Allocation Procedure
(T r) (T r)
αniim a ∗ Rnm a · · · αniim nm ∗ Rnm nm Once active power losses in terminal and non-terminal
⎡ wq wq
⎤ branches are determined using (21) and (22), the next step is
αaa ∗ Iwq · · · αan m
∗ Inm q to allocate these losses to the respective nodes which trace the
⎢ .. ⎥
⎢ : . : ⎥ path to the root node through these branches.
Ǐwqp =⎢ ⎥
⎣αnwq a ∗ Iaq · · · αnm nm ∗ Inm q ⎦
wq For all terminal branches, losses calculated using (21) are
m
allocated to the three-phases and m neutral conductors of their
terminal node. For a non-terminal branch b, allocated losses are
As it can be seen in (21) and (22), Rp contains the resistance determined by first expanding (22) as follows
of all conductors including neutral of a branch. Furthermore,
cross-terms are redistributed among phases on the basis of phase- (b)
 
pl =  I∗1 ⊗ Rp ⊗ Ǐ1q + · · · + I∗N ⊗ Rp ⊗ ǏN q
currents magnitude and as a result, end-users connected to these
q∈Q q∈Q
phases will be penalized/rewarded fairly. (28)
where N is the last node in the set Q.
C. Calculation of Variable Loss Coefficients Eq. (28) shows the contribution of each node in the over-
(b)
In order to determine VLCs, QLP and GLP approaches are all losses of branch b. Let β f denotes the losses caused by a
utilized as reported below. generic node f in a branch b, (28) can be written as follows
1) Quadratic Losses Partitioning Approach: Since power
(b)
 (b)
losses and branch phase-currents have a quadratic relation be- pl = βf (29)
tween them, the following constraint is imposed in order to de- f ∈Q(b)
pq qp
termine αxy and αyx
where
pq qp
αxy αyx 
= (23) (b) I∗f ⊗ (R ∗ q∈Q Ǐfq ) f ∈ Q(b)
Ix2p Iy2q βf = (30)
0 f∈/ Q(b)
Furthermore, from (15)
Eq. (30) represents the losses allocated to node f due to the flow
pq qp
αxy + αyx =2 (24) of current If in branch b. In this way, sweeping a network either
in a backward or forward manner, losses in all branches are
Combining (23) and (24) results in
allocated to the respective nodes as shown in (31)
2 ∗ Ix2p 2 ∗ Iy2q B
pq
αxy = qp
; αyx = (25)  (b)
Ix2p + Iy2q Iy2q + Ix2p βf = βf (31)
b=1
2) Geometric Losses Partitioning Approach: As cross-term
is a product of two phase-currents, geometric mean can also be where B is the set of branches which connect node f to the root
used to determine VLCs by imposing the following constraint node.
Eqs. (30)–(31) are the generic mathematical formulae of the
pq qp
αxy − logIxp = αyx − logIyq (26) proposed MPLAP which are based on the modified active power
pq qp losses expressions, (21) and (22), for terminal and non-terminal
Combining (24) with (26), coefficients αxy and αyx can be cal-
branches.
culated as
pq 1 Ix qp 1 Iy IV. CASE STUDY
αxy = 1 + log p ; αyx = 1 + log q (27)
2 I yq 2 I xp
The MPLAP outlined in Section III is applied on IEEE 13-
In both schemes, {x, y} represents the generic phase/neutral node test system whose multi-wire configuration is shown in
conductor and {p, q} represents generic nodes in a system. These Fig. 4 and results of the proposed scheme are compared with CLP
coefficients as can be seen from these equations relate the phase- [15] and RCLP-based BCDLA approaches. The distributed load
currents of a same node (self VLCs) and phase-currents of dif- connected between nodes 632 and 671 is replaced by two lumped
ferent nodes (mutual VLCs), and consequently the influence of loads as reported in [15] by connecting 2/3 of the total load at
phase-currents of all the nodes are properly taken into account newly introduced node 670 and 1/3 of the load at node 671. In
while allocating losses to a particular phase. order to show the applicability of the method for active DNs, a
Furthermore, in order to determine a factor among different single-phase DG is added at node 646 (100kW, 50kVar) whereas
phases, current of one phase has to be rotated in the direction a three-phase DG of type Y-PQ (200kW, 100kVar) is con-
of other phase-currents. In this work, for a multi-phase system, nected at node 680. Both DGs are considered as negative loads.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.
 3610 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 5, SEPTEMBER 2019

the division of losses differs significantly from the behaviour of

both RCLP and MPLAP due to the fact that the former method
employs impedance whereas the latter methods use resistance
of a branch. Furthermore, in CLP approach, negative losses are
calculated for lightly loaded phase b of branches 670 and 671
which is counter-intuitive as losses in a branch cannot be neg-
ative. This is due to the presence of paradox in CLP method
which is reported comprehensively in [6] and is successfully
eliminated in RCLP and MPLAP approaches.

A. Bifurcation of Cross Terms Based on Variable

Loss Coefficients
Table II shows the losses allocated to each node of the net-
work through CLP, RCLP and MPLAP procedures. The CLP
approach, on the one hand, allocates negative losses to the pas-
sive users connected at node 645 and 680 and on the other hand,
Fig. 4. Modified IEEE 13-bus test system with DGs connected at node 646 allocates positive losses to the shunt capacitances connected at
and 680. node 611 and 675 due to the presence of paradox in its formu-
lation. The effect of allocating cross-terms by utilizing VLCs,
calculated through QLP approach, is evident from the results as
losses allocated to each phase of all nodes by RCLP and MPLAP
are completely different from one another. RCLP based method
results in more allocation of losses to the passive and active end-
users as compared to MPLAP. The Total Allocated Losses (TAL)
with RCLP are 111.02 kW and −19.83 kW for passive and active
end-users respectively, whereas these values become 96.31 kW
and −6.01 kW with MPLAP. These results show that although
TAL to all the nodes are the same with RCLP and MPLAP,
however, the latter allows a more fair distribution of the losses
by properly taken into account the impact of cross-term related
losses as shown in Fig. 6.
Fig. 5. TAL over each time instant for a real Italian low-voltage 37-Bus DN.
B. Impact of Cross-Term Redistribution Schemes on
Losses Partitioning
The voltage at three-phase voltage regulator (RG60) is imposed The losses partitioning among phases of same/different nodes
to 1.0 pu. Load flow analysis is carried out using correction strongly depends upon the scheme chosen for the segregation of
current injection method reported in [16]. cross-terms. Table III shows the results of losses allocated to
Before discussing the main results of the article, a comparison each phase of a node based upon GLP and QLP schemes and
has been made between MPLAP and Z-bus [3] LA methodolo- it is evident from the results that TAL to each node and losses
gies, which clearly demonstrates the reasons of not considering partitioning among phases vary significantly from one scheme
the Z-bus scheme to compare its results with MPLAP in the to another.
subsequent sections. Fig. 5 shows the TAL through Z-bus and Comparison of the results of Table III with the results reported
MPLAP and it can be observed that Z-bus over/under allocates in Table II reveals that TAL to each node and losses partitioning
the total system losses and therefore, requires a reconciliation among phases through GLP-based MPLAP are more close to
factor in order to redistribute the losses among the end-users TAL through RCLP-based LAP as compared to the results ob-
to make the TAL equal to the total system losses. Furthermore, tained through QLP-based MPLAP. The reason of this behaviour
Z-bus method allocates losses to the slack bus which have to be can be understood by referring to Fig. 7 which shows the division
reallocated to the system nodes as it is conceptually wrong to of losses associated with cross-terms between two phases of a
allocate losses to the slack bus in a DN. This aspect is clearly same/different node(s) depending upon their currents ratio. It is
reported in [17]. evident that if ratio between Ixi /Iyj is close to 1, the division of
Referring to Table I, it can be seen that the total active power losses does not vary significantly between phases when achieved
losses in each branch, calculated using losses partitioning ex- through GLP-based MPLAP. Since, as discussed before, cross-
pressions of CLP, RCLP and MPLAP methods, are same irre- terms are equally divided between phases irrespective of their
spective of the method used for calculating them. However, a loading in RCLP based approach, it becomes clear that LA pat-
significant difference in partitioning of losses among phases can tern in both methods (GLP and RCLP) is close to each other
be observed among these methods. The behaviour of CLP in when a less diversified currents ratio exists between phases.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.
 USMAN et al.: MULTI-PHASE LOSSES ALLOCATION METHOD FOR ACTIVE DNS BASED ON BRANCH CURRENT DECOMPOSITION 3611

TABLE I
BRANCH LOSSES IN THE MODIFIED IEEE 13-NODE TEST SYSTEM BASED UPON CLP, RCLP AND MPLAP

TABLE II
ALLOCATED LOSSES TO NODES IN THE MODIFIED IEEE 13-NODE SYSTEM BASED UPON CLP, RCLP AND MPLAP

reduction on the original impedance matrix of each branch in a

network. However, as it is evident from Fig. 8, this assumption is
no longer valid in realistic scenarios where neutral is grounded
at a single/multiple points through finite grounding resistance.
The figure shows that it is very unlikely to have same voltage at
both ends of a neutral conductor. Furthermore, it can be noted
in Table IV that TAL to end-users for solidly grounded neutral
case are lower than the TAL in multiple-grounded and single-
grounded network by 15.0% and 16.3% respectively, which
suggests that losses allocated under solidly grounded-neutral
Fig. 6. Impact of redistribution of losses related to cross-terms by RCLP and
assumption do not reflect the true system losses. Moreover, al-
MPLAP. location of neutral losses to each phase connected end-users
is conceptually wrong as explained in detail in E. The results
clearly justify the development of a MPLAP in which neutral
can be treated as other phases of a network.
C. Impact of Neutral-Ground Configuration
In [12], neutral losses are allocated to the respective phases
D. Discussion on Losses Allocated to DG Units
of each node under the assumption that voltage at both ends of a
neutral conductor is the same (which is true only in an unrealistic In the analysed network, RCLP allocates negative losses to
case of solidly grounded neutral) which justifies applying Kron Distributed Generators (DGs) connected at node 680 and 646,

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.
 3612 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 5, SEPTEMBER 2019

TABLE III
ALLOCATED LOSSES TO NODES IN THE MODIFIED IEEE 13-NODE SYSTEM BASED ON GEOMETRIC AND QUADRATIC LOSSES PARTITIONING APPROACHES

whereas MPLAP allocates positive losses to DG connected at

node 646. The idea of allocating negative losses to DGs stems
from the fact that local provision of energy from DGs fulfill-
ing local demand reduces the branch current, which ultimately
lowers down the overall network losses. This is true for the case
of the DG connected at node 680 as it supplies active and re-
active power to the loads connected in its vicinity and reduces
the overall losses from 129.9 kW (no DG case) to 86.9 kW.
Consequently, negative losses are allocated to this DG.
However, the DG connected at phase a of node 646 causes a
reverse power flow as no load is connected to the same phase of
Fig. 7. Fraction of losses allocated to Ix through QLP and GLP approaches. the nodes 645 and 646, eventually leading to increase in losses
with respect to the passive case. Moreover, installation of this
DG also causes additional losses in other phase(s) and neutral
of these branches due to mutual coupling among conductors.
Consequently, total network losses increase in this case from
86.9 kW (DG at node 680 is connected only) to 91.183 kW (both
DGs are present). Therefore, it is rightly justified to allocate
positive losses to this node.
The shortcomings of RCLP-based LAP can be further ob-
served by referring to Fig. 10, which shows the losses allocated,
by both RCLP and proposed MPLAP methodologies, to the DG
connected at node 646 by incorporating real load profiles, as
shown in Fig. 9, for IEEE-13 bus loads. A careful observation
reveals that during the first half of time interval 12–13, losses
Fig. 8. Neutral voltage for single and multiple point grounding configuration. allocated to node 646 by both RCLP and MPLAP methodolo-
gies have opposite signs. During this interval, total system losses
are increased by a small amount in the presence of connected
TABLE IV DG. Consequently, it is required to allocate positive losses to
TOTAL ALLOCATED LOSSES FOR DIFFERENT NEUTRAL-GROUND DG connected node during this interval. However, due to in-
CONFIGURATION
appropriate division of cross-terms among phases of same and
different nodes by RCLP method, it allocates negative losses to
this DG during this time interval as can be clearly observed in
Fig. 11. Therefore, it can be deduced that as cross-terms are
equally divided among phases in RCLP method, the benefi-
cial/detrimental effect of any DG/load on system losses can be

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.
 USMAN et al.: MULTI-PHASE LOSSES ALLOCATION METHOD FOR ACTIVE DNS BASED ON BRANCH CURRENT DECOMPOSITION 3613

Fig. 9. Active power profiles of DG and three of the end-users.

Fig. 12. Losses allocated to the passive nodes by RCLP and MPLAP
methodologies.

E. Discussion on Losses Allocation to Neutral

The losses allocated to the neutral arise several interesting
questions such as what is the meaning of NLAFs? Do they con-
Fig. 10. Total system losses before and after connection of DG at node 646
vey the same information as PLAF? What does negative sign of
and losses allocated to the DG at node 646. NLAFs mean? and more importantly, can NLAFs be exploited
and play an active role in the management of DNs? The answers
to these questions are briefly presented below.
PLAFs represent losses allocated to each phase of a node
due to the net power injection/absorption of its connected loads,
which would not occur, otherwise, in a system in the absence of
these users, whereas NLAFs are an indication of degree of unbal-
ance in a network, which in turn depends upon the non-uniform
distribution of end-users and their variable distinct energy con-
sumption patterns. For a perfectly balanced system, there would
be no flow of current in the neutral and as a result, no losses would
be assigned to it. On the other hand, for an unbalanced system,
specifically low voltage DNs, it is unreasonable to allocate share
of neutral losses to the end-users, particularly single-phase end-
users, as their connection leads to inherent unbalancing of the
Fig. 11. TAL to the DG at node 646 during time interval 12–13. grid.
Furthermore, both PLAFs and NLAFs can have positive and
negative signs as can be seen in Table II and III results, how-
masked due to over/under estimation of positive/negative con- ever, factors with same signs have no correlation between them.
tribution of connected end-users, which ultimately results in an Passive end-users are allocated positive PLAFs as they consume
unfair allocation of positive/negative losses to the connected energy and, therefore, are responsible for an increase of losses
end-users. in a system. On the other hand, active end-users can be allocated
Apart from the negative LA to DG at 646, RCLP-based both positive (in case of over-production which leads to reverse
method wrongly allocates negative losses to passive loads dur- power flow in a branch) and negative (by partially/completely
ing few time instants as shown in Fig. 12, even though these fulfilling the local demand and therefore reducing the branch
loads are not controllable, and are the cause of the losses in the currents) LA factors depending upon their impact on losses.
system. On the other hand, the proposed MPLAP methodology Negative NLAFs are allocated to those nodes which facili-
avoids such negative losses allocation to the passive end-users. tate the reduction of the net neutral current in their upstream
This shows the superior performance of the proposed scheme branches and therefore, reduce the degree of unbalance in a
over RCLP-based LAP. network.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.
 3614 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 5, SEPTEMBER 2019

TABLE V influenced by the selection of losses partitioning approach as

COMPARISON BETWEEN COMPUTATIONAL PERFORMANCE
OF RCLP AND MPLAP
clearly indicated by the results.
It has also been demonstrated that voltage along neutral con-
ductor follows a unique profile based upon the type of neutral-
grounding topology, which has a direct influence on TAL, and as
result, it rules out the assumption of equal potential at all points
along the neutral which is necessary for Kron reduction. Un-
der this assumption, total allocated losses are relatively lower as
compared to the case when neutral is not solidly grounded and
therefore, it is impractical to apply Kron reduction technique on
a n × m original impedance matrix.
Since NLAFs are an indication of network balancing condi-
tion, they can be effectively utilized by DSO in the identification
REFERENCES
of critical nodes, which have been assigned positive NLAFs, in
order to implement suitable unbalance reduction management [1] J. Mutale, G. Strbac, S. Curcic, and N. Jenkins, “Allocation of losses
in distribution systems with embedded generation,” IEE Proc.—Gener.
schemes at these nodes. Transmiss. Distrib., vol. 147, no. 1, pp. 7–14, 2000.
[2] N. Zhao, Y. Song, Z. Bie, S. Takahashi, and Y. Sekine, “Improved Z-bus
loss allocation method through redistribution,” in Proc. 39th Int. Univ.
F. Computational Performance Power Eng. Conf., 2004, pp. 1101–1105.
[3] A. J. Conejo, F. D. Galiana, and I. Kockar, “Z-bus loss allocation,” IEEE
The comparison between computational performance of Trans. Power Syst., vol. 16, no. 1, pp. 105–110, Feb. 2001.
RCLP and MPLAP based LAPs for the test case and several [4] J. S. Daniel, R. S. Salgado, and M. R. Irving, “Transmission loss allocation
through a modified Ybus,” IEE Proc.–Gener. Transmiss. Distrib., vol. 152,
real-time low voltage DNs has been shown in Table V. The sim- no. 2, pp. 208–214, 2005.
ulations are carried out on MATLAB 2017b using DELL 64-bit [5] W. L. Fang and H. W. Ngan, “Succinct method for allocation of network
OS, core i7 with a processor speed of 2.80 GHz and 16 GB RAM. losses,” IEE Proc.–Gener. Transmiss. Distrib., vol. 149, no. 2, pp. 171–174,
2002.
It can be seen that the proposed methodology takes slightly more [6] E. Carpaneto, G. Chicco, and J. S. Akilimali, “Branch current de-
time in comparison to the existing RCLP methodology. This is composition method for loss allocation in radial distribution systems
due to the fact that GLP and QLP factors are calculated for each with distributed generation,” IEEE Trans. Power Syst., vol. 21, no. 3,
pp. 1170–1179, Aug. 2006.
branch in an iterative loop and sequential way during a backward [7] M. Atanasovski and R. Taleski, “Power summation method for loss allo-
feeder sweep. Since, the calculation of the factors of one branch cation in radial distribution networks with DG,” IEEE Trans. Power Syst.,
is totally independent from the calculation of the factors of the vol. 26, no. 4, pp. 2491–2499, Nov. 2011.
[8] M. Atanasovski and R. Taleski, “Energy summation method for loss allo-
remaining branches, a possible reduction in the computational cation in radial distribution networks with DG,” IEEE Trans. Power Syst.,
time of MPLAP can be achieved by determining these parame- vol. 27, no. 3, pp. 1433–1440, Aug. 2012.
ters in parallel by employing parallel computational techniques. [9] P. M. Costa and M. A. Matos, “Loss allocation in distribution networks with
embedded generation,” IEEE Trans. Power Syst., vol. 19, no. 1, pp. 384–
Furthermore, calculation of these factors in a vectorized way 389, Feb. 2004.
can further reduce the computational time as compared to the [10] S. Sharma and A. R. Abhyankar, “Loss allocation of radial distribution
iterative loop technique. system using Shapley value: A sequential approach,” Int. J. Elect. Power
Energy Syst., vol. 88, pp. 33–41, 2017.
[11] Z. Ghofrani-Jahromi, Z. Mahmoodzadeh, and M. Ehsan, “Distribution loss
V. CONCLUSIONS allocation for radial systems including DGs,” IEEE Trans. Power Del.,
vol. 29, no. 1, pp. 72–80, Feb. 2014.
Losses allocation methodologies can provide beneficial in- [12] E. Carpaneto, G. Chicco, and J. S. Akilimali, “Loss partitioning and loss
allocation in three-phase radial distribution systems with distributed gen-
formation for economical treatment of losses as well as the role eration,” IEEE Trans. Power Syst., vol. 23, no. 3, pp. 1039–1049, Aug.
of end-users in planning and management of DNs. Therefore, 2008.
these approaches should distribute network losses among end- [13] M. Usman, M. Vanzetto, M. Coppo, F. Bignucolo, and R. Turri, “Losses
allocation methods in planning and management of low voltage active
users in a fair manner by properly taking into account their con- distribution networks,” in Proc. 53rd Int. Univ. Power Eng. Conf., 2018,
tribution in the overall system losses. In order to achieve this pp. 1–6.
objective, a novel MPLAP is proposed which addresses the [14] W. H. Kersting, Distribution System Modeling and Analysis, 2nd ed. Boca
Raton, FL, USA: CRC Press, 2007.
shortcomings of solely available three-phase RCLP-based [15] W. Kersting, “The computation of neutral and dirt currents and power
BCDLA method by redistribution of losses, which are re- losses,” in Proc. IEEE PES Power Syst. Conf. Expo., 2004, pp. 213–218.
lated to the cross-terms, among end-users connected to the [16] K. Sunderland, M. Coppo, M. Conlon, and R. Turri, “A correction current
injection method for power flow analysis of unbalanced multiple-grounded
same/different phases of nodes. Neutral losses are not assigned 4-wire distribution networks,” Elect. Power Syst. Res., vol. 132, pp. 30–38,
to the respective phases due to the fact that these losses are part of 2016.
a system because of its inherent unbalanced nature and therefore, [17] M. Usman, M. Coppo, F. Bignucolo, and R. Turri, “Losses management
strategies in active distribution networks: A review,” Elect. Power Syst.
it is unrealistic to penalize end-users for these network losses. Res., vol. 163, pp. 116–132, 2018.
The results indicate that losses are allocated to the end-users in a [18] W. H. Kersting, “Radial distribution test feeders,” in Proc. IEEE Power
more fairly manner when partitioning of losses is achieved by the Eng. Soc. Winter Meeting, 2001, vol. 2, pp. 908–912.
[19] G. Celli et al., “Containment of power losses in LV networks with high
proposed MPLAP as compared to RCLP-based LAP. Addition- penetration of distributed generation,” CIRED—Open Access Proc. J.,
ally, segregation of share of cross-term-based losses is strongly vol. 2017, no. 1, pp. 2183–2187, 2017.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.
 USMAN et al.: MULTI-PHASE LOSSES ALLOCATION METHOD FOR ACTIVE DNS BASED ON BRANCH CURRENT DECOMPOSITION 3615

Muhammad Usman (M’19) received the B.Sc. Roberto Turri was born in Padova, Italy, in 1958. He
(Hons.) degree in electrical engineering from the Uni- received the Dr. Ing. degree in electrical engineering
versity of Engineering and Technology, Lahore, Pak- from the University of Padova, Padua, Italy, in 1984,
istan, in 2011, and the M.Sc. degree in electrical and the Ph.D. degree from the University of Wales,
drives, from the Technical University of Munich, Mu- Cardiff, U.K., in 1987. He was with the Physics De-
nich, Germany, in 2016. He is currently working to- partment, University College of Swansea, Swansea,
ward the Ph.D. degree in electric energy engineering U.K. In 1990, he joined the Electrical Engineering
with the University of Padova, Padua, Italy. His main Department, University of Padova, where he is cur-
research interests include modeling, control, and op- rently an Associate Professor in power systems. His
timization of power systems as well as the analysis of main research interests include power system analy-
the role of distributed energy resources in the man- sis and simulation, smart grids, and assessment and
agement of medium- and low-voltage distribution networks. mitigation of human exposure to low-frequency electromagnetic fields.

Massimiliano Coppo (M’16) received the Ph.D. de-

gree in electrical energy engineering from the Univer-
sity of Padova, Padua, Italy, in 2016. He is currently
a Research Associate with the University of Padova.
Alberto Cerretti received the M.Sc. degree in electri-
His main research interests include modeling and sim-
ulation of power systems for smart grid management cal engineering from the University of Padova, Padua,
Italy, in 1984. From 1986 to 1989, he was with the
and energy markets participation, network stability,
Assicurazioni Generali-Industrial Risk Department
and power quality analysis related to integration of
and CESI Network Study Division, Snamprogetti. In
distributed resources in electrical networks.
1989, he joined ENEL, Rome, Italy, where he is cur-
rently in charge of Network Efficiency and Technical
Losses, Operation and Maintenance Function, Infras-
tructure and Networks Italy Division. From 2001 to
2003, he headed the SCADA Systems, Communica-
Fabio Bignucolo is a Research Fellow with the De- tion, Protection and Metering Division in Enel Dis-
partment of Industrial Engineering, University of tribuzione and from 2006 to 2007, he was in charge of Network Planning and
Padova, Padua, Italy. He has authored or co-authored Operation Responsible in north east part of Italy. From 2008 to 2010, he was
more than 50 papers presented at national and inter- the Head of Engineering and Standardization Function with ENEL Distribution
national conferences or published on esteemed in- Companies, Romania, and from 2013 to 2014, he was in charge of Comitol-
ternational journals. His research interests include ogy and Standardization Responsible for ENEL Infrastructure and Networks
computer applications in electrical power engineer- Division. His main expertise includes protection systems, connection criteria of
ing, regulation of distribution networks hosting dis- end-users plants to the electric distribution networks, power quality, SCADA
persed generators, innovative control architectures, systems, operation of MV distribution networks, fault passage indicators, MV
and modeling of components and plants. Recently, he network automatic fault selection and supply restore systems, protection set-
is also working on network applications for electro- tings and regulations, smart grid system architecture including both automation
chemical storage units aiming at providing ancillary services to the transmission system and DER control systems, components/conformance tests, and standard-
and distribution grids. ization activities.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:28:38 UTC from IEEE Xplore. Restrictions apply.