138 IEEE TRANSACTIONS ON SMART GRID, VOL. 11, NO.

1, JANUARY 2020

Tracing Power With Circuit Theory

Yu Christine Chen , Member, IEEE, and Sairaj V. Dhople , Member, IEEE

Abstract—Power tracing is the task of disaggregating the

g
SG , SL Complex-power injections at generator bus g ∈ G
power injection of a generator (or a load) into a sum of con- and load bus  ∈ L.
stituent components that can unambiguously be attributed to YN Network admittance matrix.
loads (generators) and losses. Applications of power tracing
γg , λ Portion of the generator g current that is allocated
g
range the broad spectrum of: transmission services pricing,
loss allocation in distribution networks, fixed-cost allocation, to the load  current and fraction of the load 
modelling bilateral transactions, and financial storage rights. current sourced from the generator g current.
μ , δg Coefficients capturing portion of generator g
This paper develops an analytical approach to power trac- g
ing leveraging elementary circuit laws. The method is rigorous complex-power injection that is consumed by load
from a system-theoretic vantage point, and it yields unam-
biguous results that are consistent with constitutive principles  and portion of load  complex-power injection
that describe the steady-state behaviour of power networks. served by generator g.
Moreover, it can be implemented with limited computational L Complex-power loss.
burden, applies to networks with arbitrary topologies, and pre- g
ωG , ωL  Coefficients capturing contributions of generator g
serves the coupling between active- and reactive-power injections. and load  to complex-power loss.
Numerical experiments indicate that given a solved power-flow
solution, disaggregations can be computed for a test system
with 2383 buses, 327 generators, and 2056 loads in 4.34 s on
a personal computer, hence establishing computational scalabil- I. I NTRODUCTION
ity. Furthermore, applications are demonstrated in distribution HIS paper presents a circuit-theoretic solution to power
and transmission networks with case studies focused on quan-
tifying the impact of distributed generation on loss allocation
T tracing: the problem of disaggregating power injections
from a subset of nodes in a power network into a sum of con-
and extracting nodal contributions to bilateral transactions,
respectively. stituent parts that are attributable to other nodes and allocable
to losses. It has been hypothesized that an agreeable solution
Index Terms—Downstream tracing, Kron reduction, loss allo-
cation, power flow, power tracing, upstream tracing. to this problem would pave the way for establishing an optimal
regulatory and economic environment for the transparent and
efficient operation of power networks [1]–[3]. Furthermore, a
N OMENCLATURE universally accepted and analytically justifiable approach to
G, L, N Sets collecting generator, load, and all buses. power tracing may potentially also be useful in modelling,
VG , VL Vectors of voltage phasors at generator and load analysis, and validation of contemporary notions in power
buses. systems economics such as transactive energy [4], peer-to-
g
VG , VL  Voltage phasors at generator bus g ∈ G and load peer energy trading [5], and blockchain electricity trading [6].
bus  ∈ L. However, as with most modelling and analysis tasks in power
IG , IL Vectors of current phasors at generator and load networks, nettling nonlinearities pose daunting analytical chal-
buses. lenges to power tracing. The majority of previously proposed
g  tracing methods have consequently revolved around numerical
IG , IL Current phasors at generator bus g ∈ G and load
bus  ∈ L. approaches, unverifiable assumptions, or sensitivity analysis.
SG , SL Vectors of complex-power injections at generator Therefore, the merits in a tracing method that is grounded
and load buses. in and conforms to circuit laws that underlie the steady-state
behaviour of power networks cannot be overstated.
Manuscript received October 31, 2018; revised February 8, 2019 and March We consider two types of disaggregations for complex-
28, 2019; accepted May 12, 2019. Date of publication May 22, 2019; date power injections in the network. In downstream tracing, the
of current version December 23, 2019. The work of Y. C. Chen was sup-
ported by the Natural Sciences and Engineering Research Council of Canada complex-power injection of a generator is decomposed into
under Grant RGPIN-2016-04271. The work of S. V. Dhople was supported a sum of parts that are attributed to loads and losses in the
in part by the National Science Foundation Sustainability Research Network network. Similarly, in upstream tracing, the complex power
under Grant 1444745. Paper no. TSG-01690-2018. (Corresponding author:
Yu Christine Chen.) drawn by a load is decomposed into a sum of parts that can
Y. C. Chen is with the Department of Electrical and Computer Engineering, be attributed to generators and allocated to the losses. (See
University of British Columbia, Vancouver, BC V6T 1Z4, Canada (e-mail: Fig. 1 for an illustration.) These are not merely semantic
[email protected]).
S. V. Dhople is with the Department of Electrical and Computer distinctions, and we demonstrate that they are contextually
Engineering, University of Minnesota, Minneapolis, MN 55455 USA (e-mail: useful depending on the application at hand. Furthermore,
[email protected]). drawing from desirable attributes discussed in [3], we seek a
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. tracing result that is consistent and unified. A consistent down-
Digital Object Identifier 10.1109/TSG.2019.2918258 stream tracing result is one where loads can be demonstrably
1949-3053 c 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 05:07:40 UTC from IEEE Xplore. Restrictions apply.
 CHEN AND DHOPLE: TRACING POWER WITH CIRCUIT THEORY 139

network admittance matrix while incorporating equivalent-

admittance representations of generators and loads. Then, by
leveraging fundamental circuit-theoretic definitions, the disag-
gregation of currents is straightforwardly extended to obtain
those of complex-power injections. In this way, the seemingly
(a) (b) regressive and obviously unglamorous task of tracing currents
emerges as the perfect foil to the more involved problem
Fig. 1. The proposed tracing approach uncovers (a) downstream
and (b) upstream disaggregations that are consistent (fractional components
of tracing power. Since we deal with complex-valued cur-
sum up to load and generator values) and unified (allocations include contri- rent and voltage phasors at every step, the resulting upstream
butions to losses, marked above as red ellipses). Note that the coloured links and downstream complex-power tracing results preserve the
capture power disaggregations and not the physical topology of the network.
nontrivial couplings between active and reactive powers. For
example, our approach quantifies how much of the active
recovered by summing up fractional decompositions across power drawn by a particular load derives from both the active-
all generators (and vice versa for the upstream case). A uni- and reactive-power injections of generators. To better contex-
fied disaggregation is one where the decomposition of nodal tualize the elementary algebraic operations performed on the
injections innately embeds allocations to system loss. Namely, admittance matrix to arrive at the power tracing results, we
losses emerge as integral constituents of the upstream and establish correspondences with a number of circuit-theoretic
downstream tracing results, and they are not stitched into the notions, including (but not limited to): equivalent admittance
formulations as an afterthought. representations of nonlinear shunt elements, Kron reduction of
Given the fundamental nature of the tracing problem and complex electrical networks, super-node circuit analysis, and
its latent promise in refashioning power system operations conservation of power.
and control tasks, it has attracted extensive attention from We summarize salient features of our approach and highlight
several directions. We focus our review of relevant litera- its contributions over prior art discussed above. First and fore-
ture squarely on power tracing methods, but also reserve a most, the proposed disaggregations of currents (and powers)
few comments on applications. The seminal work in [1] and are grounded in and can be verified to satisfy circuit laws that
follow-up efforts in [7]–[10] are grounded in the so-called govern the sinusoidal steady-state behaviour of AC networks.
proportional sharing rule. This is an assumption that at any Second, our approach can be applied to both transmission
given bus, outgoing active-power flows on the lines are com- and distribution networks of arbitrary size and topological
posed of a sum of parts, which are proportional to the inflows. and constitutional complexity. Given a solved power flow,
While these efforts are formative, the proportional sharing we trace complex power with minimal computational bur-
method is assumptive by design [8]. Another body of work den (predominantly attributable to the computation of inverses
leverages a variety of graph-theoretic notions to address the of admittance and admittance-like matrices). Furthermore,
tracing problem [2], [11], but these methods lead to iterative the proposed decompositions preserve and reflect the cou-
computations. Optimization problems with varying degrees of plings between active- and reactive-power injections. Finally,
computational complexity offer an appealing numerical alter- the downstream and upstream power-tracing results are con-
native [3], [12], [13], but these are inherently algorithmic, and sistent and acknowledge losses in a unified manner; these
demonstrate tenuous links to the underlying circuit laws that characteristics are conceptually illustrated in Fig. 1.
dictate power flows. Previous attempts at leveraging circuit- The remainder of this manuscript is organized as follows.
theoretic notions for power tracing replace constant-power Mathematical notation and the power-system model utilized
elements with admittance equivalents in an attempt to counter in the paper are established in Section II. In Section III,
their nonlinearity [14], [15]. While our approach shares this we address the downstream disaggregation of generator cur-
as a common motif, these efforts neither result in consistent rents into load contributions (and the corresponding upstream
upstream and downstream disaggregations, nor do they incor- dual). This forms the basis for our main results in Section IV,
porate losses in a unified manner. Related efforts that leverage where we decompose generator power contributions into com-
circuit-theoretic notions for the problem of network-usage ponents that satisfy loads and can be attributed to losses
allocation include [16]–[18]. Finally, we bring attention to lit- (and the corresponding dual upstream problem). Next, in
erature that has explored applications of power tracing in trans- Section V, we provide two numerical case studies that demon-
mission services pricing [19], loss allocation in distribution strate applications of the approach in: i) distribution network
networks [20], fixed-cost allocation [3], modelling bilateral loss allocation, and ii) analysis of bilateral transactions in
transactions [21], [22], and financial storage rights [23]. transmission networks. Numerical results from our approach
Unlike most previously proposed approaches that begin are compared to those generated from the proportional sharing
with a solved power flow and attempt to directly tackle the method in [1]. Concluding remarks and directions for future
problem of tracing power, we embark on a more circuitous work are offered in Section VI.
path, with the intention of ensuring that the method yields
consistent and unified results. First, we formulate and pro-
pose a solution to the more fundamental problem of tracing II. P RELIMINARIES AND P OWER S YSTEM M ODEL
currents. Downstream and upstream current tracing can be Below, we introduce relevant notation and describe the
accomplished with elementary algebraic manipulations of the power system model used in the remainder of the paper.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:07:40 UTC from IEEE Xplore. Restrictions apply.
 140 IEEE TRANSACTIONS ON SMART GRID, VOL. 11, NO. 1, JANUARY 2020

A. Notation
The transpose of a vector or matrix is denoted by
(·)T , complex conjugate by (·)∗ , real and imaginary parts
of a complex number by Re{·} and Im{·}, respectively,√
magnitude of a complex scalar by |·|, and j := −1.
A diagonal matrix formed with entries of the vector X
stacked along the main diagonal is denoted by diag(X).
The spaces of N-dimensional real- and complex-valued vec- Fig. 2. One-line diagram for a 4-bus network illustrating adopted notation.
tors are denoted by RN and CN , respectively; the spaces of The network is composed of generators at buses in G = {1, 2} and loads at
M × N real- and complex-valued matrices are denoted by buses in L = {3, 4}.
RM×N and CM×N , respectively. The (m, n) entry, i.e., the TABLE I
entry in the m-th row and n-th column, of the matrix X is P OWER - FLOW S OLUTION FOR 4-B US S YSTEM IN F IG . 2.
denoted by [X]mn . A LL Q UANTITIES A RE IN p.u.

B. Power System Model

Consider an AC network with N buses, collected in the
set N , operating in sinusoidal steady state. We partition the
N buses in the network into G generator buses collected in
the set G = {1, . . . , G} ⊆ N ; and L load buses collected into generator and load buses by
in the set L = N \ G = {G + 1, . . . , N}. Aligned with  T
standard practices in steady-state power-system modelling, IG = [I1 , . . . , IG ]T =: IG1 , . . . , IGG ∈ CG ,
we adopt constant power load (CPL) models which imply  T
T G+1
that at these buses, a fixed amount of active and reactive IL = IG+1 , . . . , IN =: IL , . . . , IL
N
∈ CL , (5)
power is drawn. On the other hand, generators are mod-
elled as constant power sources (CPSs). This is reasonable, respectively, where Ii ∈ C denotes the phasor of the net current
since we assume that a solved power flow is available, and injected into bus i from the ground node. This includes the cur-
consequently, both the active- and reactive-power injections rent through any shunt elements connected to bus i; see Fig. 2
can be computed for the generators. The set of transmis- for an illustration. Kirchhoff’s current law (KCL) applied at all
sion lines is represented by E := {(m, n)} ⊆ N × N . Each buses can be compactly represented in matrix-vector form as
line is modelled using the -model with series admittance Y YGL
IG VG
ymn = gmn + jbmn ∈ C and shunt admittance ysh mn ∈ C. Denote = GGT , (6)
IL YGL YLL VL
the vector of shunt admittances by Ysh = [y1 , . . . , yN ]T ∈ CN ,
where where YGG ∈ CG×G , YGL ∈ CG×L , and YLL ∈ CL×L are
 (dimensionally consistent) decompositions of the admittance
ym := ymm + mk = gm + jbm
ysh (1) matrix Y. Finally, mirroring the notation adopted in (4)–(5), the
k∈Nm
nodal complex-power injections into the generator bus g ∈ G
is the total shunt admittance connected to bus m, Nm ⊆ N and load bus  ∈ L are expressed as
denotes the set of neighbours of bus m, and ymm ∈ C captures  ∗
g g g ∗   
passive shunt elements connected to bus m. With the above SG = VG IG , SL = VL IL . (7)
model in place, the network admittance matrix, denoted by The complex-power injections above are decomposed into real
YN , can be expressed as g g g
and imaginary parts as follows: SG = PG + jQG and SL  =
 
PL + jQL .
YN = Y + diag(Ysh ), (2)
The notation introduced so far is illustrated with a simple
where entries of Y are given by example next. We will periodically revisit this example in the
⎧ paper to demonstrate key concepts numerically.
⎨ (m,k)∈E ymk , if m = n, Example 1: Consider the 4-bus system with the one-line
[Y]mn := −ymn , if (m, n) ∈ E, (3)
⎩ diagram shown in Fig. 2. Generators are connected at buses
0, otherwise. in the set G = {1, 2}, and loads at buses in the set L =
Nodal voltages of generators and loads are denoted by {3, 4}. Voltage magnitudes at buses 1 and 2 are regulated
 T to be |VG1 | = 1.04 p.u. and |VG2 | = 1.02 p.u., respectively.
VG = [V1 , . . . , VG ]T =: VG1 , . . . , VGG ∈ CG , Transmission lines are modelled with lumped parameters, with
 T y12 = 0.553−j10.5, ysh 12 = j0.0880, y13 = 0.771−j10.5, y13 =
sh
T
VL = VG+1 , . . . , VN =: VL G+1
, . . . , VL
N
∈ CL , (4) j0.0790, y14 = −j6.90, y14 = j0.0330, y24 = 2.00 − j14.0,
sh

24 = j0.0430, y34 = −j11.8, and y34 = j0.152, all in p.u.

ysh sh

respectively, where Vi = |Vi |∠θi ∈ C represents the voltage The power-flow solution is reported in Table I (computed with
phasor at bus i. Also, denote the vectors of current injections bus 1 set to be the slack bus).

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:07:40 UTC from IEEE Xplore. Restrictions apply.
 CHEN AND DHOPLE: TRACING POWER WITH CIRCUIT THEORY 141

III. T RACING C URRENTS

In this section, we address the problem of decomposing
a particular generator current injection into constituent parts
that identifiably serve loads, as well as the dual problem of
extracting generator contributions that serve a particular load.
To begin, we establish some terminology:
1) Downstream current tracing: Currents injected by gen- (a) (b)
erators are disaggregated into components that are
attributed to loads. Specifically, we decompose the Fig. 3. (a) Downstream and (b) upstream current tracing for the 4-bus
g network in Fig. 2. Our approach uncovers the coefficients (γ14 , γ23 , γ13 , γ24 )
current injected by the g generator, IG , as a linear
combination of entries of IL as follows: and (λ14 , λ23 , λ13 , λ24 ) in closed form. We demonstrate that γ13 + γ23 = −1,
γ14 + γ24 = −1; and λ13 + λ14 = −1, λ23 + λ24 = −1. (Only grey-coloured edges

γg IL

g correspond to physical interconnections.)
IG = , ∀ g ∈ G. (8)
∈L

For the 4-bus system in Fig. 2, the notion of downstream Proof: Substituting (13) into (6) yields
current tracing in (8) is illustrated in Fig. 3a.  
2) Upstream current tracing: Currents injected by loads 0G YGG − diag ϒG YGL VG
= T , (14)
are disaggregated into components that are identifiably IL YGL YLL VL
sourced from generators. Specifically, we decompose the
 , as a linear combination where 0G is a G-length vector of all zeros. Elementary alge-
current injected by the  load, IL
braic manipulations of (14) yield the following expressions:
of entries of IG as follows:
  −1 
 g g IL = YLL − YGL T
YGG − diag ϒG YGL VL =: YL VL ,

IL = λ IG , ∀  ∈ L. (9)
  −1
g∈G VG = − YGG − diag ϒG YGL VL . (15)
For the 4-bus system in Fig. 2, the notion of upstream Isolating IG from (6), we get
current tracing in (9) is illustrated in Fig. 3b.
The coefficients γg and λ in (8) and (9) can be determined
g
IG = YGG VG + YGL VL . (16)
∀ g ∈ G,  ∈ L given the topology of the network and the
solved power flow. We discuss this next. Substituting VG = −(YGG − diag(ϒG ))−1 YGL VL and VL =
−1
YL IL from (15) into (16), we get
A. Downstream Current Tracing   −1 
−1
IG = YGL − YGG YGG − diag ϒG YGL YL IL
In this section, we derive the coefficients γg in analytical
closed form. We present this in the form of a lemma next, =: IL . (17)
following which, several remarks are provided to explain the
result from a circuit-theoretic perspective. Extracting the g-th entry of IG in (17), we arrive at (10).
Lemma 1 (Downstream Current Tracing): The current 1) Circuit-theoretic Interpretation: We provide a few
g remarks that yield a circuit-theoretic interpretation to two
injected by generator g ∈ G, IG , can be uniquely disaggre-
gated into a linear combination of currents injected into each key terms in Lemma 1: the vector ϒG and the matrix YL .
load bus, as follows: First, with regard to ϒG , note that the disaggregation in (10)
 would be algebraically consistent with any complex-valued
γg IL
 G × G matrix, say ϒ, which can satisfy the power-flow solu-
g
IG = , ∀ g ∈ G. (10)
∈L tion IG = ϒVG . However, only a diagonal matrix: i) can be
uniquely determined given IG and VG , and ii) preserves the
In (10), γg ∈ C is the (g, ) entry of the G × L matrix topology of the network. This establishes the uniqueness of
  −1  the disaggregation, and lends an appealing circuit-theoretic
−1 interpretation to the entries of ϒG . Particularly, given the
= YGL − YGG YGG − diag ϒG YGL YL , (11)
steady-state power-flow solution in (7), entries of the vector
where YL ∈ CL×L is given by ϒG are (almost surely not realizable) equivalent-admittance
representations of the generators. As an example, with ϒG1 =
  −1
YL = YLL − YGL
T
YGG − diag ϒG YGL , (12) IG1 /VG1 and ϒG2 = IG2 /VG2 , the circuit in Fig. 4a is equivalent
to the one in Fig. 2 in sinusoidal steady state. With regard
and ϒG ∈ CG satisfies the following relationship, to YL , note that it corresponds to the admittance matrix of
which captures the power-flow solution at the G generator the Kron-reduced network where all generator buses (mod-
buses: elled with admittances ϒG ) are eliminated. As an example,
  the network in Fig. 4a reduces to the one in Fig. 5a through
IG = diag ϒG VG . (13) this process.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:07:40 UTC from IEEE Xplore. Restrictions apply.
 142 IEEE TRANSACTIONS ON SMART GRID, VOL. 11, NO. 1, JANUARY 2020

TABLE II
C URRENT T RACING FOR 4-B US S YSTEM IN F IG . 2. A LL Q UANTITIES A RE IN p.u. (a) D OWNSTREAM . (b) U PSTREAM

(a) (b)

rows expectedly sum to generator current injections, indicat-

ing the validity of the decomposition. It is also consistent,
because in Table II–(a), columns 2 and 3 sum up as follows:
3 + γ 3 I 3 = −I 3 and γ 4 I 4 + γ 4 I 4 = −I 4 . Thus, as indi-
γ13 IL 2 L L 1 L 2 L L
cated in (18), it is indeed the case that γ13 + γ23 = −1 and
γ14 + γ24 = −1.
(a) (b)
B. Upstream Current Tracing
Fig. 4. Equivalent-admittance models for generators and loads, highlighted
in red, are used in (a) downstream, and (b) upstream tracing, for the network Next, mirroring the result in Lemma 1, we derive in closed
g
in Fig. 2. form the coefficients λ that allow the disaggregation of load
currents into contributions from generators.
Corollary 1 (Upstream Current Tracing): The current
injected by load  ∈ L, IL  , can be uniquely disaggre-

gated into a linear combination of currents injected into each

(a) (b) generator bus,
 g g

Fig. 5. Kron-reduced networks from Fig. 4 are highlighted in red. (a) From IL = λ IG . (21)
Fig. 4(a) with admittance matrix YL , i.e., IL = YL VL . (b) From Fig. 4(b)
with admittance matrix YG , i.e., IG = YG VG . g∈G
g
In (21), λ ∈ C is the (, g) entry of the L × G matrix

2) Consistency: Notice that γg precisely represents the
= YGLT
− YLL (YLL − diag(ϒL ))−1 YGL T
YG−1 , (22)
fractional contribution of the load  current to the generator g
current. Indeed, the values of γg are such that where YG ∈ CG×G is given by

γg = −1, ∀  ∈ L, (18) YG = YGG − YGL (YLL − diag(ϒL ))−1 YGL
T
, (23)
g∈G
 = −   and ϒL ∈ CL satisfies the following relationship, which
which implies that we can express IL g∈G γg IL . This captures the power-flow solution at the L load buses:
conclusively establishes the consistency of the disaggregation
in (10), since it implies that the -th load current can be recov- IL = diag(ϒL )VL . (24)
ered by summing up its contributions to all generators in the
system. The identity in (18) can be proved as follows. Sum Proof: The result above can be derived in an analogous
up both sides of (10) over all generators to get fashion to Lemma 1 and the derivation is not included.
 g   Offering due respect to brevity, we refrain from repeat-
IG = γg IL

= γg IL

, (19) ing detailed remarks on the circuit-theoretic interpretation to
g∈G g∈G ∈L ∈L g∈G entries of ϒL , and the fact that YG represents the Kron-
where the second equality above follows by simply switch- reduced admittance matrix of the network where all load
ing the order of summation. Also, treating the network as a buses (modelled with admittances ϒL ) are eliminated. As
supernode, we get from KCL that an illustration of these ideas for the upstream tracing case,
 g  equivalent-admittance models for the CPLs in Fig. 2 are

IG = − IL . (20) computed to give rise to the circuit in Fig. 4b, which
g∈G ∈L reduces to the one in Fig. 5b through Kron reduction of
Since (19) and (20) hold ∀ IL , IG , we see that (18) is true. all load buses (modelled with admittances ϒL ). Furthermore,
Example 2: Here, we illustrate the downstream current- in this case,
tracing ideas introduced above using the 4-bus system  g
λ = −1, ∀ g ∈ G. (25)
described in Example 1. See Fig. 3a for an illustration. In
∈L
Table II–(a), given the steady-state power-flow solution, we 
g g g
report the disaggregation of generator current injections into This implies that we can express IG = − ∈L λ IG , which
load contributions as described in (10). We note that the establishes the consistency of the disaggregation in (21).

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:07:40 UTC from IEEE Xplore. Restrictions apply.
 CHEN AND DHOPLE: TRACING POWER WITH CIRCUIT THEORY 143

Example 3: Revisiting the 4-bus system from Example 1,

we illustrate the upstream current-tracing concepts introduced
above. See Fig. 3b for an illustration. Table II–(b) reports
pertinent quantities obtained by disaggregating load-current
injections into generator contributions following (21). We first
observe that the decomposition is valid as the rows sum
to load current injections. Also, consistency is demonstrated (a) (b)
by the fact that λ13 + λ14 = −1 and λ23 + λ24 = −1, as
indicated in columns 2 and 3 of Table II–(b), which show Fig. 6. (a) Downstream and (b) upstream complex-power tracing in
λ13 IG1 + λ14 IG1 = −IG1 and λ23 IG2 + λ24 IG2 = −IG2 . the 4-bus network. Our approach uncovers in closed form the coefficients
(μ13 , μ14 , μ23 , μ24 ) and (δ13 , δ23 , δ14 , δ24 ), as well as (ωG
1 , ω2 ) and (ω3 , ω4 ).
G L L
Coefficients (ωG 1 , ω2 ) and (ω3 , ω4 ) enable allocation of complex-power
G L L
IV. T RACING C OMPLEX P OWER loss, L (marked as a red ellipse), to generator and load complex-power
injections, respectively. (Only grey-coloured edges correspond to physical
In this section, we address the problem of disaggregating interconnections.)
the generator complex-power outputs into contributions to the
loads and system losses (and the dual problem). To aid the
discussion, we define—while mildly abusing terminology and With the choice
notation1 —the complex-power loss in the system as V   g ∗  g
μ = − Lg λ , ωG = 1 −
g g
μ , (28)
 g  VG
 ∈L
L= SG + SL . (26)
V
g
g∈G ∈L where L and VG denote the nodal voltages of the  load and
g
g generator buses, respectively, and λ = [ ]g (see (22)), we
System loss is typically defined as the real part of (26),
have the following:
which includes line losses [24]. We generalize this defini-
(1) The fractional component of generator g output that is
tion to its complex-valued counterpart to ensure consistency in g g
consumed by load  is given by μ SG , i.e.,
subsequent developments, where we uncover contributions of
 g g
generator and load buses to system loss. As such, the resulting 
SL =− μ SG , ∀  ∈ L. (29)
disaggregations in downstream and upstream tracing naturally g∈G
incorporate allocations of line losses to the generators and
loads in the network, respectively. (2) The contribution of generator g to the complex-power
g g
With the definition of system loss in place, we now loss, L, is given by ωG SG , i.e.,
introduce notions of downstream and upstream tracing:  g g
L= ωG SG . (30)
1) Downstream power tracing: The complex-power out-
g∈G
put of generators are disaggregated into contributions to
system losses and components that are consumed by the Before delving into the proof to (29) and (30), a few remarks
g g g
loads in the network. For the 4-bus system in Fig. 2, the are in order. The decomposition of SG into the terms ωG SG and
g g
notion of downstream tracing is illustrated in Fig. 6a. {μ SG }∈L is not profound. Particularly, for any set of complex
g g  g
2) Upstream power tracing: The complex power con- variables, {μ }∈L , and with the choice ωG = 1 − ∈L μ ,
sumed by each load is disaggregated into its contribution the expression in (27) is algebraically consistent, i.e., the right
g
to system losses and components that are identifiably hand side sums to SG . What is to be emphasized is that with the
g g g g g
sourced from generators. For the 4-bus system in Fig. 2, particular choice of μ in (28), the terms μ SG and ωG SG rep-
the notion of upstream tracing is illustrated in Fig. 6b. resent the contribution of the g-th generator output to the -th
Both downstream and upstream complex-power tracing draw load and to the system losses, respectively. This is established
from and build upon the current tracing results in Section III. in (29) and (30), which we prove next.
Proof: To show (29), express the complex-power injection
A. Downstream Power Tracing into load bus  ∈ L as follows:
∗
  
Here, we state one of the main results of this paper, which SL = VL IL . (31)
pertains to the decomposition of generator complex-power
 from (21) into the above to get
Substitute for IL
injections into constituent parts that are: i) consumed by each
load in the network, and ii) allocated to system loss. ⎛ ⎞∗
 g g  
Theorem 1 (Downstream Power Tracing): Express the ⎝ g g ∗
SL
= VL λ IG ⎠ = 
VL λ  IG
complex-power injection at generator bus g as follows:
g∈G g∈G
 g g   g g
g g g
SG = ωG SG + μ SG . (27) LV  g ∗ g
= g λ SG =: − μ SG , (32)
∈L VG
g∈G g∈G

1 The variable L was previously used to denote the number of load buses. where the third equality above is obtained by substituting
(IG )∗ = SG /VG .
g g g
Subsequent usage, however, should be contextually obvious.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:07:40 UTC from IEEE Xplore. Restrictions apply.
 144 IEEE TRANSACTIONS ON SMART GRID, VOL. 11, NO. 1, JANUARY 2020

TABLE III
C OMPLEX - POWER T RACING FOR 4-B US S YSTEM IN F IG . 2. A LL Q UANTITIES A RE IN p.u. (a) D OWNSTREAM . (b) U PSTREAM

(a) (b)

TABLE IV
C OMPARISON OF (a) D OWNSTREAM AND (b) U PSTREAM T RACING Proof: The proof proceeds analogously to that for
R ESULTS FOR 4-B US S YSTEM IN E XAMPLE 1 O BTAINED VIA Theorem 1 and is omitted in the interest of brevity.
THE P ROPOSED M ETHOD AND THE O NE IN [1].
The power-tracing results highlighted in (27) and (34) are
A LL Q UANTITIES A RE IN p.u.
consistent and unified. In particular, notice that (29) establishes
consistency in downstream tracing since the fractional decom-
positions across all generators demonstrably sum up to loads.
Analogously, (36) establishes consistency in upstream tracing.
Furthermore, notice from (30) and (37) that the results are
unified since the compositional decomposition of generators
and loads innately include allocations to system loss.
Example 4: Direct application of (27) to the 4-bus system
from Example 1 yields disaggregation of generator complex-
 from (29) into (26)
Next, to show (30), substitute for SL power injections, pertinent values from which are reported
and consider the following steps: in Table III–(a). For instance, the complex-power injection at
 g  g g  g  g g
L= SG − μ SG = SG − μ SG bus 1 is 3.97 + j1.37 p.u., out of which 2.36 + 0.403 p.u. is
consumed by the load at bus 3, 1.58 + 0.153 p.u. is consumed
g∈G ∈L g∈G g∈G g∈G ∈L
  by the load at bus 4, and 0.0274 + 0.813 p.u. is—for lack of
  
= 1−
g
μ SG =:
g g g
ωG SG . (33) a better word—dissipated as loss. The contributions to loss
g∈G ∈L g∈G
from the two generators sum to 0.0664 + j0.970 p.u., which
indeed coincides with the total system loss as defined in (26)
This completes the proof. and computed using the power-flow solution in Table I. See
Fig. 6a for an illustration.
B. Upstream Power Tracing On the other hand, the disaggregation of load complex-
Mirroring the result in Theorem 1, below, we address the power injections in the 4-bus system from Example 1 is
problem of disaggregating the complex-power consumed by obtained by applying (34) and is reported in Table III–(b).
loads into constituent parts that are: i) identifiably sourced See Fig. 6b for an illustration. As an example, the load at
from each generator, and ii) allocated to system loss. bus 3 sources 2.35 + j0.986 p.u. from the generator at bus 1
Corollary 2 (Upstream Power Tracing): Express the and 0.668 + j0.0582 p.u. from the generator at bus 2, out of
complex-power injection of load bus  as follows: which 3.00 + j0.379 p.u. is consumed by the load itself while
 0.0194 + j0.665 p.u. is allocated to system loss.
  
SL = ωL SL + δg SL

. (34)
For comparison, in Table IV, we report results obtained
g∈G
using the active-power tracing method in [1]. Differences
With the choice may conceivably be attributed to significant couplings between
V ∗
g
 active- and reactive-power injections, withdrawals, and flows
δg = − G γg , 
ωL =1− δg , (35) that are dealt with by [1] in a disjoint fashion.
VL g∈G

where VL and V denote the nodal voltages of the  load and
g V. N UMERICAL C ASE S TUDIES
G
g generator buses, respectively, and γg = [ ]g (see (11)), we In this section, we first present results from a numeri-
have the following: cal case study that focuses on the computational cost of the
(1) The fractional component of load  that is served by proposed method. Next, we discuss two applications for the
generator g is given by δg SL , i.e.,
power tracing method developed in Section IV: one focuses
 on a distribution network with radial topology and the other
δg SL

g
SG = − , ∀ g ∈ G. (36) on a transmission network with meshed topology.
∈L
(2) The contribution of load  to the complex-power loss, A. Computational Cost for Different Networks
 S , i.e.,
L, is given by ωL L Computationally intensive operations in the proposed

 
L= ωL SL . (37) method predominantly include algebraic manipulations of
∈L network-admittance-like matrices. In Table V, we report the

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:07:40 UTC from IEEE Xplore. Restrictions apply.
 CHEN AND DHOPLE: TRACING POWER WITH CIRCUIT THEORY 145

Fig. 7. Downstream active-power tracing in a 5-bus distribution feeder. Fig. 8. Upstream active-power tracing in the 5-bus distribution feeder. System
System loss allocated to feeder head. Case (a): before DG at bus 5; case loss allocated to each load. Case (a): before DG at bus 5; case (b): after DG
(b): after DG at bus 5. at bus 5.

TABLE V TABLE VI
C OMPUTATION T IMES [S EC ] R EQUIRED TO O BTAIN D ISAGGREGATIONS C OMPARISON OF U PSTREAM T RACING R ESULTS FOR S YSTEM IN F IG . 8
IN D OWNSTREAM AND U PSTREAM C OMPLEX - POWER T RACING FOR O BTAINED VIA THE P ROPOSED M ETHOD AND THE O NE IN [1].
39-B US N EW E NGLAND , 118-B US IEEE, AND 2383-B US A LL Q UANTITIES A RE IN p.u.
P OLISH S YSTEMS

computation times (given the power-flow solution) required

to obtain downstream and upstream complex-power disag-
gregations for the 39-bus New England, 118-bus IEEE, and nodes are load buses. With regard to the adopted notation,
2383-bus Polish test systems on a conventional laptop (Intel for the system of interest, G = {1} and L = {2, 3, 4, 5}.
Core i5 processor at 2.6 GHz with 16 GB 1600 MHz DDR3 We consider two cases: (a) the active-power injections at
memory). The results indicate that downstream power trac- load buses are set as P2L = −0.800, P3L = −0.500,
ing is more computationally expensive compared to upstream P4L = −0.900, and P5L = −1.00, all in p.u.; (b) dis-
power tracing. This is because there are more load buses tributed generation (DG) at bus 5 serves some of the local
than generator buses in all three networks. Consequently, alge- load, and consequently the new net injection at bus 5 is
braic operations in downstream tracing have to be performed 
P5L = −0.500 p.u. (Injections at other load buses are the same
with matrices of dimension 29 × 29, 64 × 64, 2056 × 2056 as in case (a).) With this setup, the power-flow solution for
for the 39-, 118-, 2383-bus systems, respectively. In contrast, case (a) reveals that the feeder head injects P1G = 3.34 p.u.,
the dimensions of the largest matrices for which we per- with system loss Re{L} = 0.138 p.u. In case (b), the feeder
form algebraic operations in upstream tracing are 10 × 10, head injection is P1G = 2.79 p.u., with system loss Re{ L} =
54 × 54, 327 × 327 for the 39-, 118-, 2383-bus systems, 0.0917 p.u. As expected, the addition of DG reduces the
respectively. total system loss because less power is required from the
feeder head.
B. Loss Allocation in Distribution Networks 1) Downstream Tracing: Applying the downstream tracing
Transmission-loss allocation has long been recognized as method and extracting the real part of (27), we obtain the
a challenging problem due to the nonlinear nature of the disaggregation of the feeder-head active-power injection into
power-flow equations [24], [25]. Here, we apply the proposed contributions to each load and to the system loss, values from
downstream and upstream power-tracing methods to a rep- which are reported in the table in Fig. 7. As illustrated by both
resentative distribution network, and discuss two ways in the table and the one-line diagram in Fig. 7, loss incurred in
which the total feeder loss can be allocated. In the interest the feeder (marked in red) is allocated to bus 1 (the feeder
of presentation clarity, we focus our attention on only the real head). Such an allocation scheme would be useful in operating
part of the complex-valued loss defined in (26). a vertically integrated power system in which the utility bears
Consider the 5-bus distribution-system feeder depicted in the cost of transmission losses. Since this system contains only
Fig. 7. Bus 1 represents the feeder head (the secondary of one generator, there is no ambiguity in allocating generator
the step-down transformer that connects the feeder to the contributions to loads and loss. Thus, the method in [1] yields
bulk system) and is modelled as a generator bus. All other the same numerical results.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:07:40 UTC from IEEE Xplore. Restrictions apply.
 146 IEEE TRANSACTIONS ON SMART GRID, VOL. 11, NO. 1, JANUARY 2020

Fig. 10. Contributions of (a) loads and (b) generators for all bilateral trans-
actions of magnitude 1.00 p.u. in the WECC system. (x → y represents a
Fig. 9. Tracing active power in the 9-bus network for the base case. transaction where generator at bus x increases output by 1.00 p.u. and load at
(a) Downstream tracing of P1G = 0.720 p.u., P2G = 1.63 p.u., and P3G = bus y increases consumption by 1.00 p.u.).
0.850 p.u. (b) Upstream tracing of P5L = −0.900 p.u., P7L = −1.00 p.u., and
P9L = −1.25 p.u.

2) Upstream Tracing: The upstream tracing method

enables unbundling transmission losses so they can be allo-
cated to each consumer in the distribution network. Extracting
the real part of (34), we get the disaggregation of the
active-power injection of each load into components that are:
i) sourced from the feeder head at bus 1, and ii) allocated to
system loss. Particularly, components of each load allocated to
the system loss are reported in the table in Fig. 8 for cases (a) Fig. 11. (a) Downstream and (b) upstream active-power tracing in the WECC
and (b). We note that the contribution of a load to loss is system after the 1 → 5 bilateral transaction of 1.00 p.u.
affected by two factors: the amount of power demanded by
that load and its proximity to the feeder head. As a trend,
loads that are located further down the feeder contribute more assume the largest fraction of the transactions in each case.
to loss, see, e.g., contributions from buses 4 and 5 as compared Closer inspection reveals a slew of nonlinear effects. For
with those from buses 2 and 3 in case (a). On the other hand, instance, the remainder of the buses do not appear to contribute
in case (b), the load at bus 5 demands less power from the in proportion to their electrical distances from each transac-
feeder head, and so it contributes less to the system loss. We tion consistently. More importantly, these results conclusively
compare upstream tracing results obtained via the proposed demonstrate that all nodes in the network contribute to all
method with the one in [1], and the results are summarized in bilateral transactions. While this is intuitive, our approach
Table VI. In contrast to Example 4, for this case, we see that allows one to numerically quantify contributions of nodal
the results closely match those obtained from the approach injections and withdrawals to transactions.
in [1]. This is presumably due to the fact that reactive-power
components of loads have little effect on line active-power VI. C ONCLUDING R EMARKS AND F UTURE W ORK
flows.
We developed a circuit-theoretic method to trace complex-
power injections from generators to loads (and vice versa).
C. Bilateral Transaction Allocation The proposed method leveraged a suite of circuit-theoretic
Bilateral transactions are power-trade agreements between constructs to arrive at the disaggregation of generator (load)
suppliers and consumers of electricity. While the financial complex-power injections into constituent parts, each of which
agreement is settled on a node-to-node basis, the physical are attributable to loads (generators) and losses. We presented
paths taken by the transacted power depend on the network applications of power tracing to loss allocation in distribution
topology and parameters. Applying the proposed downstream networks and bilateral transaction allocation in transmission
and upstream tracing methods to the Western Electricity networks. With numerical case studies, we demonstrated that
Coordinating Council (WECC) 3-machine 9-bus system, we the proposed method can be implemented with limited compu-
illustrate how all generators and loads contribute to a bilateral tational burden, applies to networks with arbitrary topologies,
transaction. As a base case, extracting the real part of the gen- and reflects the coupling between active- and reactive-power
erator disaggregation in (27), Fig. 9a shows the downstream injections.
active-power tracing from generators to loads. Conversely, As part of future work, applications of power tracing to
upstream active-power tracing is obtained from the real part fixed-cost allocation, transmission-services pricing, and vali-
of (34) and is shown in Fig. 9b. dating bilateral transactions for distribution-network markets
In Figs. 11a–11b, we plot downstream and upstream dis- could be developed. From a theoretical perspective, decou-
aggregations resulting from all possible bilateral transactions pling assumptions that are common to power systems analysis
between generator-load pairs of magnitude 1.00 p.u. in the could be leveraged to not only obtain insights on dependence
WECC network. Notice that the origin and destination buses of tracing coefficients on network attributes but also potentially

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:07:40 UTC from IEEE Xplore. Restrictions apply.
 CHEN AND DHOPLE: TRACING POWER WITH CIRCUIT THEORY 147

facilitate computations. Finally, while we provide comparisons [19] M. Pantos and F. Gubina, “Ex-ante transmission-service pricing based
of numerical results with those obtained from the proportional on load-flow patterns,” IEEE Trans. Power Syst., vol. 19, no. 2,
pp. 796–801, May 2004.
sharing method in [1], exhaustive numerical case studies that [20] P. M. Costa and M. A. Matos, “Loss allocation in distribution networks
compare the present approach with a wider body of previous with embedded generation,” IEEE Trans. Power Syst., vol. 19, no. 1,
ones could be performed. pp. 384–389, Feb. 2004.
[21] A. R. Abhyankar, S. A. Soman, and S. A. Khaparde, “Tractability of
bilateral transactions considering multiplicity of solution space in real
R EFERENCES power tracing,” in Proc. IEEE Power India Conf., 2006, pp. 1–8.
[22] S. Cvijić and M. D. Ilić, “Part I: A new framework for modeling and
[1] J. Bialek, “Tracing the flow of electricity,” IEE Proc. Gener. Transm. tracing of bilateral transactions and the corresponding loop flows in
Distrib., vol. 143, no. 4, pp. 313–320, Jul. 1996. multi-control area power networks,” IEEE Trans. Power Syst., vol. 29,
[2] D. Kirschen, R. Allan, and G. Strbac, “Contributions of individual gen- no. 6, pp. 2706–2714, Nov. 2014.
erators to loads and flows,” IEEE Trans. Power Syst., vol. 12, no. 1, [23] J. A. Taylor, “Financial storage rights,” IEEE Trans. Power Syst., vol. 30,
pp. 52–60, Feb. 1997. no. 2, pp. 997–1005, Mar. 2015.
[3] A. R. Abhyankar, S. A. Soman, and S. A. Khaparde, “Optimization [24] A. J. Conejo, F. D. Galiana, and I. Kockar, “Z-bus loss allocation,” IEEE
approach to real power tracing: An application to transmission fixed Trans. Power Syst., vol. 16, no. 1, pp. 105–110, Feb. 2001.
cost allocation,” IEEE Trans. Power Syst., vol. 21, no. 3, pp. 1350–1361, [25] Q. Ding and A. Abur, “Transmission loss allocation based on a new
Aug. 2006. quadratic loss expression,” IEEE Trans. Power Syst., vol. 21, no. 3,
[4] K. M. Jhala, B. Natarajan, A. Pahwa, and H. Wu, “Stability of transac- pp. 1227–1233, Aug. 2006.
tive energy market-based power distribution system under data integrity
attack,” IEEE Trans. Ind. Informat., to be published.
[5] T. Morstyn, A. Teytelboym, and M. D. Mcculloch, “Bilateral contract
networks for peer-to-peer energy trading,” IEEE Trans. Smart Grid,
vol. 10, no. 2, pp. 2026–2035, Mar. 2019.
[6] M. T. Devine and P. Cuffe, “Blockchain electricity trading under
demurrage,” IEEE Trans. Smart Grid, vol. 10, no. 2, pp. 2323–2325,
Mar. 2019.
[7] J. Bialek, “Allocation of transmission supplementary charge to real and
reactive loads,” IEEE Trans. Power Syst., vol. 13, no. 3, pp. 749–754,
Aug. 1998.
[8] J. W. Bialek and P. A. Kattuman, “Proportional sharing assumption in Yu Christine Chen (M’15) received the B.A.Sc.
tracing methodology,” IEE Proc. Gener. Transm. Distrib., vol. 151, no. 4, degree in engineering science from the University of
pp. 526–532, Jul. 2004. Toronto, Toronto, ON, Canada, in 2009 and the M.S.
[9] F. F. Wu, Y. Ni, and P. Wei, “Power transfer allocation for open and Ph.D. degrees in electrical engineering from
access using graph theory-fundamentals and applications in systems the University of Illinois at Urbana–Champaign,
without loopflow,” IEEE Trans. Power Syst., vol. 15, no. 3, pp. 923–929, Urbana, IL, USA, in 2011 and 2014, respectively.
Aug. 2000. She is currently an Assistant Professor with the
[10] C. Achayuthakan, C. J. Dent, J. W. Bialek, and W. Ongsakul, “Electricity Department of Electrical and Computer Engineering,
tracing in systems with and without circulating flows: Physical insights University of British Columbia, Vancouver, BC,
and mathematical proofs,” IEEE Trans. Power Syst., vol. 25, no. 2, Canada, where she is affiliated with the Electric
pp. 1078–1087, May 2010. Power and Energy Systems Group. Her research
[11] D. Kirschen and G. Strbac, “Tracing active and reactive power between interest includes power system analysis, monitoring, and control.
generators and loads using real and imaginary currents,” IEEE Trans.
Power Syst., vol. 14, no. 4, pp. 1312–1319, Nov. 1999.
[12] A. R. Abhyankar, S. A. Soman, and S. A. Khaparde, “Min–max fairness
criteria for transmission fixed cost allocation,” IEEE Trans. Power Syst.,
vol. 22, no. 4, pp. 2094–2104, Nov. 2007.
[13] M. S. S. Rao, S. A. Soman, P. Chitkara, R. K. Gajbhiye,
N. Hemachandra, and B. L. Menezes, “Min–max fair power flow tracing
for transmission system usage cost allocation: A large system per-
spective,” IEEE Trans. Power Syst., vol. 25, no. 3, pp. 1457–1468,
Aug. 2010.
[14] J.-H. Teng, “Power flow and loss allocation for deregulated transmission Sairaj V. Dhople (M’13) received the B.S., M.S.,
systems,” Int. J. Elect. Power Energy Syst., vol. 27, no. 4, pp. 327–333, and Ph.D. degrees in electrical engineering from
May 2005. the University of Illinois at Urbana–Champaign,
[15] W.-M. Lin, T.-S. Zhan, and C.-H. Huang, “A circuit theory based load Urbana, IL, USA, in 2007, 2009, and 2012, respec-
flow tracing method considering counter-flow contribution,” in Proc. 5th tively. He is currently an Associate Professor
WSEAS Int. Conf. Instrum. Meas. Circuits Syst., Hangzhou, China, 2006, with the Department of Electrical and Computer
pp. 312–317. Engineering, University of Minnesota, Minneapolis,
[16] A. J. Conejo, J. Contreras, D. A. Lima, and A. Padilha-Feltrin, “Zbus MN, USA. His research interests include modeling,
transmission network cost allocation,” IEEE Trans. Power Syst., vol. 22, analysis, and control of power electronics and power
no. 1, pp. 342–349, Feb. 2007. systems with a focus on renewable integration. He
[17] S. M. Abdelkader, D. J. Morrow, and A. J. Conejo, “Network usage was a recipient of the National Science Foundation
determination using a transformer analogy,” IET Gener. Transm. Distrib., CAREER Award in 2015 and the Outstanding Young Engineer Award from
vol. 8, no. 1, pp. 81–90, Jan. 2014. the IEEE Power and Energy Society in 2019. He is an Associate Editor
[18] Y. C. Chen and S. V. Dhople, “Power divider,” IEEE Trans. Power Syst., of the IEEE T RANSACTIONS ON E NERGY C ONVERSION and the IEEE
vol. 31, no. 6, pp. 5135–5143, Nov. 2016. T RANSACTIONS ON P OWER S YSTEMS.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on June 28,2022 at 05:07:40 UTC from IEEE Xplore. Restrictions apply.