A Simple Method to Estimate Power Losses in Distribution Networks

Nassim A. IQTEIT , Ayşen BASA ARSOY , and Bekir ÇAKIR

Dept. of Electrical Engineering, Kocaeli University, 41380, İzmit /KOCAELİ, Turkey

[email protected], [email protected], [email protected]

Abstract but they do not have load profile functions with time (i.e. daily,
monthly or yearly load profiles) to calculate the average power
This paper presents a simple approximated formulas to losses in electrical networks. This paper presents simple
estimate active and reactive power losses in distribution approximation power loss formulas to estimate the active and
networks. The developed formulas are derived from Elgerd’s reactive power losses, taking load profiles into account. The full
power loss formulas considering load profiles. A simple model derivation of the proposed power loss formulas is discussed in
is also proposed to estimate the load profile based on several this manuscript.
parameters, such as average of load demand and load The contents of this paper are: simple model of load profiles,
limitations. Derivation of the load model and loss formulas active power loss formulas based on load demands, modified
are given along with their application on a sample power flow equations based on power profiles, an algorithm to
distribution system. The results show that the loss formulas find active and reactive power loss profiles, application on a 13
depending on the proposed load profile can estimate the bus distribution test system and its simulation results; finally,
power losses in distribution networks with high precision. summarizes the contributions and conclusions of this work.

1. Introduction 2. Simple Approximated Model of Load Profiles

The load profiles in conventional networks can be estimated The profiles of residential, commercial and industrial load
from the database of the distribution system, while in smart demands were modeled through field measurements and
networks can be obtained directly on time from the smart systems. statistical study [2]. The daily load profiles of P (t) or Q (t) at
Estimated load profiles in conventional distribution networks specific probability can be obtained by using a normal
helps the electrical companies in many fields; such as, covering distribution, where the Eq. (1) is the general daily load profile.
the energy demands of consumers, taking the economic and
management decisions, improving the performance of ( ) = ( )+ ( ) (1)
networks…etc. Many researches were carried out to determine
the model of load profiles; analytical model for determining where, XG (t): general daily load profile. μ(t): the mean curve of
uncertainty in distribution loads presented in [1], depends on the certain consumer. σ(t): the standard deviation curve of a certain
calculation of mean and variance of loading profile in each consumer based on daily load curves. κ1: the Gaussian
distribution transformer. Daily load profiles for all load types distribution value at specific probability Pr(%). For example,
carried out in [2] are modeled depending on the mean, standard κ1=1.3 at Pr =90% and κ1 = 1.65, Pr = 95%. The following
deviation and the normal distribution of load profiles. The daily proposed procedures present a simple and approximated model of
load profile in different seasons were also studied in [3]. Previous daily load profile depends upon a small amount of load data:
studies on load profile need to find mean and standard deviation
in constructing load profile model. The proposed model of load - According to Eq. (1), if the σ(t)=0, the daily load curve equal
profile in this paper consists of a little of load data; type of to the mean curve. Under this condition, the daily load
demands and their average values, maximum and minimum profile can be defined as XG (t)= μ(t)= μpu (t)×average{μ(t)}.
limitations, and the probability of load distribution between the At zero standard deviation, we can assume μpu (t) is the
given limitations. general behavior of residential, commercial or industrial
Estimation of active and reactive power losses in distribution load profile in per unit system. Where,
systems is one of the main goals for any electric utility company.
For example, reduction of real power losses increases the ( )
performances of distribution networks and decreases the financial = ( ) = (2)
×
costs. Also, reduction of reactive power losses reduces the voltage
drops and reactive power consumption and ameliorating - The per-unit mean profiles of residential, commercial and
loadability of networks [4]. For these reasons, many equations industrial were obtained through the results of [2], where the
depending on generated or consumed powers were proposed to area under each curve is equal to one. Fig.1 shows the per
calculate power losses in transmission and distribution systems: unit mean profiles of residential, commercial and industrial
Such as Kron's loss equation [5], Branch power loss equation [6], load demands.
and Elgerd’s loss equation [7]. The equations were used in the
calculation of power losses in the electrical system, economic - Approximated load profile can be derived which are based
dispatch analysis [5, 8] and determining the optimal size and on the values of average and minimum- maximum limits of
location of distributed generators or capacitors in primary load demands, as shown below:
distribution networks [4, 6, 9-12]. These equations have related
with single values of power (load, generation or injections power) ( )= ( ) + ( ) (3)
 Xr(t)
where, ( )= ([ ( )] ) − ( )
Δt
( ) = ( )√ − 1 = ( ) √ −1 (4) xmax − pu (t )
where: μ pu (t )
∑ [ ( )]
= (5) x min − pu ( t ) x max − pu
∑ ( )

μ pu = 1
E(Xr) = μ is the mean value of Xr , Xr is the random variable in x min − pu
the set {x1,x2,…,xNd}. The subset {xmax(t), μ(t), xmin(t)} represents
the maximum deviation in the set of Xr. This subset can be used t = mΔt
to find out the approximate value of the λ ratio. The approximate
value of λ is shown in Eq. (6). Fig.2: Maximum, minimum, and mean functions of the general
load demand based on random subset of Xr(m).
( ) ( ) ( )
≈ (6) Table 1: Examples compare between exact and approximate
[ ( ) ( ) ( )] values of load profiles at specific time.
Profile of Subset λ ratio The value of
Fig.2 clarifies the relation between maximum, minimum, and set Xr at of Xr profile XG
mean functions of general load demands and their average values. time m Exact Appx Exact Appx
Replacing the average values of these functions instead of the {0.9,1.075,1. {1.1,0.9, 1.00512 1.00675 1.080 1.094
general load functions is the second approximation that can be 1,0.95,0.9,1, 0.988}
used to simplify the calculations of λ ratio. Where the symbol ̅ 0.97,0.94,1.0 Range ≈
is the average of x(t). Eq. (7) represents the approximate value of 85,0.96} ±10%
{0.8,1.25,0.7 {1.3,0.7, 1.04304 1.05962 1.282 1.331
λ, based on the range of load limits.
5,0.7,1.3,1,1. 1.01}
1,1.2,1.15, Range ≈
̅ ̅ ̅ ̅ 0.85} ±30%
≈ = (7)
̅ ̅ ̅ ̅ {1.4,1.3,0.55 {1.49,0.5 1.11992 1.15944 1.459 1.528
,0.7,1.49,1,0. 1,1.006}
51,0.86,1.39, Range ≈
Now Eq. (1) can be reformulated based on the approximate value 0.86} ±50%
of λ as shown in Eq. (8). Table 1 displays three numerical
examples and the comparison between exact and approximated 3. Power Losses
values of λ ratio, and profile values of XG.
3.1 Exact Power Loss Formulas
( )= ( ) 1+ √ −1 (8)
The total active and reactive power losses in distribution
network with N buses can be calculated by Elgerd’s Eq. (9) and
3
Residential Eq. (10), respectively [4]. The equations show the power losses
depending on the active and reactive power injections.
Load profiles

Commercial
2
Industrial

1 = + + − (9)

0
2 4 6 8 10 12 14 16 18 20 22 24
= + + − (10)
(a)
where,
3
Load profiles (pu)

2 =[ … … ], =[ … … ],
α = [αij], β = [βij], γ = [γij] and ξ = [ξij].
1 While the size of matrices α, β, γ and ξ is N×N. The coefficients
0
αij, βij, γij and ξij are functions with magnitude and angle of phasor
2 4 6 8 10 12 14 16 18 20 22 24
voltage ∠ . Also, αij and βij depend on resistive entries rij of
(b) impedance matrix [Zij], while γij and ξij depend on reactance
Time (Hours) entries xij. Pi and Pj are the active power injections at buses i and
Fig.1.The per unit mean load profiles based on a) the j, respectively; Qi and Qj are the reactive power injections at buses
measurement data [2] , b) 5th order polynomial function of curves i and j, respectively. Where the power injection is the difference
in (a) between the power generations and load demands on the same
bus.
3.2 Power Loss Formulas Based on Load Profiles - The length of the regular interval m is ∆ = and the

Fig. 3 displays the change of the power injections on the nodes endings of this interval are ( − 1) and , where m= 1,
of a distribution network at each moment of time. The distribution 2,…, M.
system was assumed perfectly balanced in the three-phase
system. The load profiles and the line segments were supposed to - If X(t) is assumed a linear function along interval m , then
be perfectly balanced. The line impedance of each segment was the values of Xm_avi and can be approximated by Eq.
constant with time, but the state variables (Vi (t) and δi (t)) were (11).
not constant because they depend on the power demands.
_ = ( ) ≅ −
∆ ( )
- (11)
Vi (t)∠δi (t) V j (t )∠δ j (t ) = ( ) ≅ −
( )
zij = rij + jxij
= z ij ∠ θ ij - The average integral of X(t) through the period T is
Pi (t),Qi (t) Pj (t ), Q j (t ) approximated in Eq. (12)
Pi (t ) Qi (t ) Pj (t) Qj (t)
( )= ( ) =∑ ( ) =∑ ⇒
( )
t t t t

Fig. 3. Load profiles on distribution network ( ) ≅ ∑ ≅ ∑ − (12)

3.2.1 Linearization Principle and Average Integral

of Power Profiles 3.2.2 Approximated Power Loss Formulas Based
on Load Profiles
Linearization technique and additive property of integral were
used to derive a simple power loss equation based on power According to Eq. (11), over interval m the average integral of
profiles. Profiles of active and reactive load demands, power power injection profile (Pi (t) , Qi (t) , Pj (t) or Qj (t) ) can be
injections and power losses are nonlinear functions with time represented by single point and constant value at time =
variable. Also, active and reactive power loss relations are − . Dependent on approximated values of power
nonlinear with profiles of active and reactive power injections. injection profiles, the active power loss profile for each interval
Fig. 4 shows a general nonlinear function X(t), where X(t) could m was derived in the following equation.
represent the power profiles of load demands (PDi (t), QDi (t)),
power generations (PGi (t), QGi (t)), power injections (Pi (t), Qi (t))
= ( )= ( ), ( ), ( ), ( )
at node i , or power losses (PLT (t), QLT (t)).
Through one period T the function X(t) is divided by M regular = , , ,
intervals with length of Δt for each one. The length of the regular
interval (∆ = ) depends on the power measurement devices = + + − (13)
(smart, digital or conventional meters) or accuracy of analysis.
Similarly, the reactive power profile for each interval m can be
X(t) was considered approximate linear function through each
calculated by using Eq. (13) only by replacing γm and ξm instead
interval m. Where Xm(t) is the linear approximation of X(t) over
of αm and βm, respectively.
interval m. Xm_avi is the average integral value of Xm(t) through
According to Eq. (12) and Eq. (13), the average approximation
time Δt . is the weight of the average integral value of X(t) in
of active and reactive losses based on load profiles are obtained
interval m.
using Eq. (14) and Eq. (15), respectively.
X(t)
X m (t ) : Linear +
= ∑ (14)
Xm_avi Same area + −

X(t) +
= ∑ (15)
+ −

Xm
t t
... t
... t
where,
= − =[ … … ]
t, m
0 1 2 m-1 m M-1 M 1
= − =[ … … ]
T 2
= − and = −
Fig. 4. Linearization principle of general nonlinear function X(t)
= − = , = − = ,
The following relations have been obtained by using fig. 4.
= − = , = − = .
While the size of matrices αm, βm, γm and ξm is N×N. The
coefficients αmij , βmij , γmij and ξmij are calculated for each interval Table 3: Load data of a 13 bus radial distribution network
m. BUS P (pu) Q (pu) Range Probability Type of
where, (± R%) (Pr%) load
1 0 0 - - -
= − = cos( − ) (16a)
2 0 0 - - -
3 0.3045 0.1828 5 90 R
4 0.4059 0.2435 30 90 R
= − = sin( − ) (16b) 5 3.0351 2.2519 15 90 80%R+20%I
6 0.7060 0.4353 10 90 C
7 1.0509 0.7006 20 90 I
=γ − = cos( − ) (16c) 8 1.0631 0.6571 10 90 R
9 0.4578 0.2861 13 90 C
10 0.8299 0.5532 15 90 R
=ξ − = sin( − ) (16d) 11 0.4450 0.2966 50 90 75%R+25%C
12 1.0580 0.6172 20 90 R
where, 13 0.3046 0.1521 5 90 R
= , = , = , =− , = R: residential C : Commercial I: Industrial
, =− .

3.3 Algorithm of Power Losses Estimation

The coefficients αmij , βmij , γmij and ξmij are calculated for each
interval m. Power flow equations in reference [5] were reformed
to include the state variables and power injections at each value
of m. where the state variables Vmi and δmi can be calculated by
solving the modified power flow equations (Eq. (17) and Eq. Pmi = PGmi− PDmi = −PDmi
(18)). Numerical methods such as Newton-Raphson method is
PDi =[PD1i PD2i ...PDmi...PDMi] QDi =[QD1i QD2i ...QDmi...QDMi]
used to solve Eqs. (17) and (18). The power injections Pmi and Qmi
are calculated by Eq. (11).  1 T 
PDmi= PDim−   QDmi = QDi m− 1  T 
 2 M   2  M 
=∑ cos( − + ) (17)

= −∑ sin( − + ) (18)

where, Ymij is the entry of admittance matrix at values of m, i and

j, Fig. 5 represents an algorithm to find the profiles and the total
average of active and reactive power losses. m≤ M

4. Applied Example
V , δ , P and Q

The following applied example was formed through

reconfiguring the distribution test system given in [13] and its
load profiles were estimated depending on the proposed simple
approximation model of load profiles.
18MVA, 12.5kV, 13 bus distribution network, is shown in
fig.6. Nodes 1 and 2 are unloaded. Nodes 3, 4, 8, 10, 12, and 13
have only residential loads. Nodes 6 and 9 have only commercial
loads. Node 7 is connected only with industrial load. Buses 5 and
11 have a mixed loads. The line and the load data are shown in
table 2 and 3, respectively. The base values of this network are 10
MVA and 12.5kV.

Table 2: Line data of a 13 bus radial distribution network.

BUS(ns) BUS(nr) R (pu) X(pu) ½ B (pu)
1 2 0.0043 0.0121 0.000035
2 3 0.0060 0.0170 0.000050
3 4 0.0031 0.0090 0.000026
4 5 0.0090 0.0250 0.000073
2 7 0.0030 0.0085 0.000025
4 6 0.0170 0.0221 0.000043
2 8 0.0290 0.0380 0.000074 Fig.5. Algorithm to find active and reactive power losses by
8 9 0.0230 0.0290 0.000056 using simple formulas.
9 10 0.0400 0.0500 0.000010
10 11 0.0290 0.0380 0.000074
9 12 0.0370 0.0460 0.000010
12 13 0.0220 0.0270 0.000060
 Algorithm for power loss calculations is achieved using
MATLAB. Table 5 includes the results of average power losses
of 13 node radial distribution network. The results were estimated
by using the load profile and their average and maximum values.
Also, fig. 8 shows the active and reactive power loss profiles
using the proposed equations of power losses. Table 5 confirms
that the results of power loss equations are very close to the results
of power losses when they are directly calculated by power flow
programs. The power loss values are different when using single
values average/maximum demand and power profiles because the
Fig.6. 13 bus radial distribution network. power losses of the network are a nonlinear function of load
profiles. For example, the active power loss based on the load
5. Simulation Results and Discussion profiles equals to 0.0407pu, but based on the average values of
these load profiles equals to 0.03453pu. The active power loss
The proposed model of load profile has been used to estimate the based on maximum values of the load profiles is equal to
load profiles at each node of 13 bus system. Type of load, 0.1466pu. The calculation of active power loss by using Eq. (14)
maximum and minimum limitations and the probability percent is more accurate and effective than using Eq. (9) and the same
of load distribution were considered besides the load mean value way calculating reactive power loss by using Eq. (15) is more
to estimate each load profiles at each node of the system. The accurate and effective than using Eq. (10).
consideration details of our case study given in table 3, where the Fig. 8 shows that the maximum active power loss is equal to
probability percent of all distributed loads are assumed 90%. 0.11pu, while the maximum reactive loss is equal to 0.19pu. Fig.8
Fig.7 shows the estimated load profiles of 13 bus system at each also shows that the maximum active and reactive power losses
node of the network. Table 4 shows the average values of active happened at time 20:00.
and reactive load profiles that shown in fig.7. The effects of
limitations on load profiles can be noted by comparing the values Table 5: Power losses in 13 bus radial distribution network
of P and Q in table 3 and 4. The estimated load profiles increasing Calculations based Active/ losses Power flow
by increasing mean of load demand, a range of load limitations, on Reactive formula calculation
and probability factor κ1. Maximum load P(pu) 0.14660 0.14660
demands Q(pu) 0.26761 0.26762
Load 1&2 Load 3 Load 4
1 1 2 Average load P(pu) 0.03452 0.03453
M W (o r) M v a r

M W (o r) M v a r

M W (o r) M v a r

P [bar]

0.5 Q [bar]
P [line]
0.5 1 demands Q(pu) 0.06502 0.06561
zero load
0
0 5 10 15
Q [line]

20 25
0
0 5 10 15 20 25
0
0 5 10 15 20 25
Load power profiles P(pu) 0.04070 0.04070
Hour
Load 5
Hour
Load 6
Hour Q(pu) 0.07622 0.07615
Load 7
2 4
M W (o r) M v a r

M W (o r) M v a r

M W (o r) M v a r

5
1 2

0 0 0
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0.2
Hour Hour Hour P [bar]
Load 8 Load 9 Load 10 Q [bar]
0.18
4 2 4
M W (o r) M v a r

M W (o r) M v a r

M W (o r) M v a r

P [line]
2 1 2 0.16 Q [line]
Plosses & Qlosses (pu)

0 0 0 0.14
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Hour Hour Hour 0.12
Load 11 Load 12 Load 13
2 4 1
M W (o r) M v a r

M W (o r) M v a r

M W (o r) M v a r

0.1
1 2 0.5
0.08
0 0 0
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
0.06
Hour Hour Hour

0.04

Fig.7. load profiles of a 13 bus radial distribution network 0.02

0
Table 4: Average values of load profile based proposed model 0 5 10
Hour
15 20 25

bus P proposed model (pu) Q proposed model (pu)

1 0 0 Fig.8. Active and reactive power losses profiles of 13 bus radial
2 0 0 distribution network
3 0.3207 0.1925
4 0.5346 0.3207 Fig.9 shows the real power loss profiles at two type of load
5 3.5194 2.6112 models: 1) discrete load model as shown in fig.7. 2) 5th order
6 0.7815 0.4818
polynomial functions load model. The average active power loss
7 1.2750 0.8500
in discrete mode is 0.04070pu, while in the polynomial model is
8 1.1761 0.7270
9 0.5210 0.3256 0.040707pu.
10 0.9623 0.6415 The proposed load and loss models can be used in the
11 0.6812 0.4541 estimation of power losses in networks, economic analysis and to
12 1.2830 0.7484 determine the optimal distribution generators that can be inserted
13 0.3207 0.1601 in distribution networks.
 [8] L. Wang and C. Singh, "Environmental/economic power
0.12 dispatch using a fuzzified multi-objective particle swarm
Ploss (Discrete load model) optimization algorithm", Electric Power Systems Research,
0.1 Ploss (5th poly. load model) vol. 77, no. 12, pp. 1654-1664, 2007.
Power losses (pu)

[9] D. Hung, N. Mithulananthan and R. Bansal, "Analytical

0.08
strategies for renewable distributed generation integration
0.06 considering energy loss minimization", Applied Energy, vol.
105, pp. 75-85, 2013.
0.04 [10] D. Hung, N. Mithulananthan and R. Bansal, "Analytical
Expressions for DG Allocation in Primary Distribution
0.02 Networks", IEEE Transactions on Energy Conversion, vol.
25, no. 3, pp. 814-820, 2010.
0
0 5 10 15 20 25 [11] D. Hung and N. Mithulananthan, "Multiple Distributed
Hours Generator Placement in Primary Distribution Networks for
Loss Reduction", IEEE Transactions on Industrial
Fig.9. Comparison between the power losses profiles at discrete Electronics, vol. 60, no. 4, pp. 1700-1708, 2013.
load model and 5th order polynomial model of load profiles. [12] M. Shahzad, I. Ahmad, W. Gawlik and P. Palensky, "Load
Concentration Factor Based Analytical Method for Optimal
6. Conclusions Placement of Multiple Distribution Generators for Loss
Minimization and Voltage Profile Improvement", Energies,
This paper presents a simple approximated formulas to vol. 9, no. 4, p. 287, 2016.
estimate the active and reactive power losses in distribution [13] W. Grady, M. Samotyj and A. Noyola, "Minimizing network
networks based on load profiles. Also, it presents a simple harmonic voltage distortion with an active power line
approximated model to estimate load demand profiles. The conditioner", IEEE Transactions on Power Delivery, vol. 6,
models of loads and losses are examined on a 13 bus distribution no. 4, pp. 1690-1697, 1991.
network. The following conclusions can be drawn from this
study: 1) type of load and its average value, the range of load
changes, and the normal distribution factor κ1 are the main
parameters that can effect the behavior of load profile. 2) The
developed loss formulas give to more precise results when
compared to the power loss formula based on average/maximum
load demand. 3) the proposed formulas of active and reactive
power losses is easy to apply, is capable of using discrete load
profiles, and has a small error when using an appropriate number
of time intervals (M). Because of these features, the simple
approximated formulas can be used not only in power loss
analysis but also economic dispatch and estimate the optimal
setting of distributed generators that can be inserted into
distribution networks.

7. References

[1] D. Nguyen, "Modeling Load Uncertainty in Distribution

Network Monitoring", IEEE Transactions on Power
Systems, vol. 30, no. 5, pp. 2321-2328, 2015.
[2] J. Jardini, C. Tahan, M. Gouvea, S. Ahn and F. Figueiredo,
"Daily load profiles for residential, commercial and
industrial low voltage consumers", IEEE Transactions on
Power Delivery, vol. 15, no. 1, pp. 375-380, 2000.
[3] E. BOBRIC, G. CARTINA and G. GRIGORAS, "Clustering
Techniques in Load Profile Analysis for Distribution
Stations", Advances in Electrical and Computer
Engineering, vol. 9, no. 1, pp. 63-66, 2009.
[4] D. Hung and N. Mithulananthan, "Loss reduction and
loadability enhancement with DG: A dual-index analytical
approach", Applied Energy, vol. 115, pp. 233-241, 2014.
[5] Saadat, “Power Systems Analysis”, McGraw-Hill, 2000.
[6] C. Chang, "Reconfiguration and Capacitor Placement for
Loss Reduction of Distribution Systems by Ant Colony
Search Algorithm", IEEE Transactions on Power Systems,
vol. 23, no. 4, pp. 1747-1755, 2008.
[7] O. Elgerd, “Electric Energy Systems Theory: An
Introduction”, New York: McGraw-Hill, 1971.