You are on page 1of 6

SIMPLIFIED CALCULATION OF VOLTAGE AND LOSS SENSITIVITY FACTORS IN

DISTRIBUTION NETWORKS

Qiong Zhou Janusz W. Bialek


Institute for Energy Systems Institute for Energy Systems
University of Edinburgh, University of Edinburgh,
Edinburgh EH9 3JL, UK Edinburgh EH9 3JL, UK
jogrey2001@yahoo.co.uk Janusz.Bialek@ed.ac.uk

Abstract – Increased penetration by Distributed Gen- and DNOs (Distribution Network Operators) tend to
eration has caused an increased interest in sensitivity operate along "install and forget" principles whereby
analysis of distribution networks. However application of security of supply is achieved by maintaining wide secu-
Jacobian-matrix based sensitivity factors encounters seri- rity margins rather than active management typically
ous problems as generally it is difficult to perform load
seen at the transmission level. Consequently DNOs
flow calculations in distribution networks due to a lack of
on-line information. To avoid those problems, we have normally don't have accurate information regarding
proposed to use the Gauss-Seidel method and Z-bus matrix actual generation and demand profile required to per-
to derive voltage and loss sensitivity factors in closed form form a load-flow and calculate the Jacobian matrix.
as explicit functions of real and reactive power injections, With the lack of an up-dated Jacobian matrix, fixed
without having to solve the full ac load-flow problem. sensitivity factors derived for a base case may be highly
Comparison with the sensitivities derived using the vari- inaccurate.
able Jacobian matrix, made on a model of 71-bus network, To overcome those problems, this paper concentrates
has shown a very good agreement.
on calculating the values of sensitivities of voltages and
losses to power injections in distribution networks,
Keywords: power system analysis, distribution net-
works without using the Jacobian matrix. The calculated sensi-
tivities could be practically applied for loss allocation
[4] or for generation curtailment to manage voltage
1 INTRODUCTION constraints [8]. Readers are referred to those references
Sensitivity analysis at the transmission level is a rou- for further details of practical application.
tine task performed for outage analysis, congestion In this paper we apply the Gauss-Seidel method and
management, transmission pricing etc. Generally, its aim Z-bus matrix to calculate sensitivity factors in closed
is to establish how network flows, transmission losses or form as explicit functions of nodal injections. This
voltages change due to changed real and/or reactive makes it possible to update the sensitivities when operat-
power generation at a given node. Sensitivity factors are ing conditions change, without having to solve the full
normally calculated from the Jacobian matrix of a AC load-flow problem in order to calculate the Jacobian
solved load-flow case. matrix.
Examples of sensitivity analysis include [1] where a The proposed methodology has been illustrated using
network topology rescheduling methodology is pro- a model of 71-bus real distribution network. Comparison
posed to alleviate power system emergency conditions between sensitivity factors derived using the proposed
in transmission networks. In [2], sensitivity matrix de- methodology and those using the Jacobian matrix shows
rived from the Jacobian matrix is proposed, where the a very good agreement. Finally, the incremental loss
voltage magnitudes are only dependent on the reactive sensitivity coefficients developed in [4] are compared
power injections assuming small R/X ratios in lines. with the proposed factors showing again a good agree-
Paper [3] proposed a loss sensitivity analysis method ment.
based on the generalized generation shift distribution
factor (GGDF). Again the assumption of small R/X 2 JACOBIAN MATRIX AND SENSITIVITY
ratios in transmission networks was used. A very popu- FACTORS
lar use of sensitivity factors (called power transfer dis- If a load flow case has been solved using a Newton-
tribution factors) is in congestion management [15]. Raphson method, an obvious way to derive voltage and
Increased penetration of Distributed Generation (DG) loss sensitivity factors is to use the Jacobian matrix :
in distribution networks has caused an increased interest
in transmission-style sensitivity analysis. However  ∂P/∂θ ∂P/∂V  θ
J=  = F    (1)
application of Jacobian-matrix based sensitivity factors ∂Q/∂θ ∂Q/∂V   V  
encounters serious problems in distribution networks as
generally it is difficult to perform load flow calculations. where P and Q are the vectors of real and reactive
Distribution networks traditionally have been passive nodal injections and θ and V are the vectors of voltage

16th PSCC, Glasgow, Scotland, July 14-18, 2008 Page 1


angles and magnitudes. By inverting the Jacobian ma- are treated as complex current injections obtained from
trix, one can calculate the voltage sensitivity factors the solved load flow results [5] . In [6], the Z bus matrix
∂V ∂P and ∂V ∂Q and the loss sensitivities [13]. As was applied in the contingency analysis in order to de-
power injections are nonlinear functions of voltages, the termine line overloads. Paper [7] studied the voltage sag
Jacobian matrix changes with operating conditions. problem in transmission networks using Z bus matrix
Hence each time power injections change, a new load techniques.
flow has to be solved, based on which the updated Jaco- 3.1 Voltage sensitivity factors
bian matrix and sensitivities factors can be calculated.
The network nodal equation is
However running a full load-flow for every change of
operating conditions in distribution networks may not be & &  V
& or I N  = YNN
& Y & 
feasible due to a lack of on-line information about the I& = Y
&V Ns 

N
 (2)
&I &T Y&ss   V&s 
actual generation/demand profile in every node. An  s   Y Ns
alternative could be to use a fixed Jacobian matrix, and
therefore fixed sensitivity factors. where V & is the vector of all N voltages except the
N
The fixed Jacobian matrix is usually calculated at a slack bus, V s is the voltage at the slack bus, I& N and I&s
&
‘flat start’ when:
are the corresponding nodal current injections, and
• Bus voltages are maintained at around 1.0p.u
Y& & &
during normal operation. NN , YNs and Yss are the corresponding blocks of the
• The phase angle differences of bus voltages admittance matrix. A dot on top of a symbol signifies a
at two ends of the lines are small. complex number.
The fixed Jacobian matrix may not represent the true The vector of complex currents I& N can be expressed
function very well when the operating conditions as
change. Simulation results reported in Section IV con-
firmed that conclusion. I& N = YNN N + YNsVs
& V& & & (3)

3 SENSITIVITY ANALYSIS BASED ON & is


from which the vector of nodal voltages V N
GAUSS-SEIDEL METHOD AND Z BUS MATRIX
Gauss-Seidel method has been widely used in the past & =Y
V & −1 (I& − Y
& V& ) = Z& I& − V
& (4)
N NN
N Ns s N s
for load-flow calculations due to limited memory of
early computers. Gauss-Seidel method is known to be where & −1
Z& = Y is the impedance matrix and
less sensitive to the choice of the initial operating point NN

than the Newton-Raphson method and, while being & = Z& Y


V & V& is a vector reflecting voltage support
s Ns s
generally slower in terms of convergence, is more robust from the slack bus.
when applied in distribution networks [13, 14]. All these
A complex voltage at node i , V&i , can now be calcu-
features make the Gauss-Seidel method more convenient
to use by DNOs to derive sensitivity factors in distribu- lated as
tion networks although successful implementation of &*
Sj
variants of the Newton-Raphson method have also been V&i = ∑ Z& ij & * − V&si for i=1, 2, …, N (5)
reported [16]. j≠s Vj
The Gauss-Seidel load flow method can be applied
using either the bus admittance matrix (Y), or the bus where S& j = P j + jQ j is the complex power injection.
impedance matrix (Z) [9]. Z bus matrix gives an indica-
tion of how much power injection at a given bus affects Equation (5) can be used in Gauss-Seidel method of
voltages at other buses. Compared with the admittance solving load flow equations. In Gauss-Seidel iterations,
matrix Y, the Z bus matrix method has better conver- the voltage V&i on the left-hand-side would be the cur-
gence performance as the buses not directly connected rently calculated value while V& on the right-hand-side
j
also have impact on one another. The convergence per-
formance using Z bus matrix is comparable to the New- would be the last available value.
ton-Raphson approach if the only voltage-specified bus Let us calculate the derivative of a complex nodal
in the system is the slack bus [10]. A typical distribution voltage with respect to a real power injection ∂V&i ∂Pk :
network meets that requirement because embedded
generators normally run at fixed power factors [11] and  S& *j 
∂ 
therefore can be treated as P-Q loads. The main disad-  V& * 
∂Vi
& j  ∂V&si
vantage of using the Z-matrix is that it is dense while the = ∑ Z& ij  − for i=1, 2, …, N (6)
admittance matrix is sparse. ∂Pk j ≠ s ∂Pk ∂Pk
Obviously the main application of Z-bus matrix is for
the fault analysis. Other applications include the alloca-
tion of network losses where the generators and loads

16th PSCC, Glasgow, Scotland, July 14-18, 2008 Page 2


As V&si is a function of the constant slack bus voltage ∂ Vi 1  ∂V& 
= ReV&i* i 
 (12)
∂Qk Vi  ∂Qk 
∂V&si
and network parameters, = 0 and:
∂Pk ∂V&i
where is calculated from (11). The derived sen-
∂Qk
∂S& *j ∂V& j*
V& j* − S& *j sitivities are in closed form as explicit functions of nodal
∂V&i ∂Pk ∂Pk injections.
= ∑ Z& ij
( )
for i=1, 2, …, N (7)
∂Pk j ≠ s V& j*
2 Shunt compensators (reactors or capacitor banks) can
be modeled as a shunt branch admittance, so that the
changes of bus voltages due to the switching of shunt
∂S& *j compensators can be calculated by modifying the Z bus
Note that if j=k then = 1 , otherwise
∂Pk matrix.
∂S& *j 3.2 Loss Sensitivity factors
= 0 . Hence
∂Pk Loss sensitivity factors can be derived in an approach
similar to the method suggested in [5], where the actual
∂V&i − Z& ij S& *j ∂V& j* Z& ik V&k* losses (not the incremental loss) are expressed using the
= ∑ +
( ) ( )
for i=1, 2, …, N Z bus matrix and the nodal current injections.
∂Pk j ≠ s ,k V& *
2 ∂P 2
j
k V&k* Network losses can be calculated as:
(8)  N +1   N +1  N +1 
* & * 
Ploss = Re  ∑ V& j I&*j  = Re  ∑ V& j  ∑ Y& jk Vk 
Now the derivatives of the voltage magnitude with  j =1  
 j =1  k =1 

[ ] [ ] [ ]
respect to power injections can be derived the following
way. For the i-th bus: = Re V & *V
& TY & * = Re V & T GV& * − Re V
& T jBV&*
(13)
∂ Vi ∂V i* ∂V&
2
Vi = V&iV& i* => 2Vi = V&i + V& i* i
(9) where Y & = G + jB . Note that the voltages and the
∂Pk ∂Pk ∂Pk
admittance matrix now include the slack bus and that
The two terms on the right are conjugates of each
other. As the derivative of the magnitude is a real num-
[
& T jBV
Re V ]
& * = 0 - see [5].
In order to derive the sensitivity factors of network
ber, the imaginary components must cancel each other
losses to individual nodal injections, it is assumed that
out and:
an incremental change in the bus power injection is
 ∂V&  absorbed at the slack bus. This assumption can be justi-
∂ Vi
ReV&i* i 
1
= (10) fied as in distribution networks where the grid supply
∂Pk Vi  ∂Pk 
  point performs the function of the slack bus [4]. The
derivatives can be then calculated as:
Substituting the values of ∂V&i ∂Pk calculated from
(8) produces the required sensitivities of voltage magni-
∂Ploss ∂ Re V
=
[
& T GV
&* ∂ V
= Re 
]
& TG *
V
(
& TG ∂ V 
& +V )
&*  ( )
tudes with respect to real power injections. It is worth ∂Pk ∂Pk  ∂Pk ∂Pk 
emphasizing that the derived sensitivities are in closed (14)
form as explicit functions of nodal injections.
Now let us consider sensitivities of voltage magni- The two terms in the brackets are conjugate trans-
tudes with respect to reactive power injections. First it is poses of each other. Hence
necessary to derive the derivative ∂V&i ∂Qk . Following ∂Ploss ∂ V
= 2 Re 
&T
GV
( )

& *  = 2 Re V

& T *G ∂ V 
&

( ) (15)
the same steps as before, the equivalent of (8) is: ∂Pk  ∂Pk   ∂Pk 
∂V&i − Z& ij S& *j ∂V& j* Z& V& *
= ∑ − j ik k The vector of complex derivatives ∂V
& ∂P can be
( ) ( )
for i=1, 2, …, N k
∂Qk j ≠ s ,k V& *
2 ∂Q 2
j
k V&k* calculated from (8).
(11) 3.3 Discussion
∂V&i The derived voltage magnitude and loss sensitivity
Equation (11) can be solved for similarly as factors are explicit functions of the complex nodal volt-
∂Qk
ages V& and nodal injections S& = P + jQ . By com-
i j j j
before and the sensitivity of voltage magnitude with
respect to reactive power injection is: parison, the Jacobian-based sensitivity factors are the
explicit functions of nodal voltages only. This shows the
main advantage of using the proposed methodology. To

16th PSCC, Glasgow, Scotland, July 14-18, 2008 Page 3


calculate up-dated Jacobian-based sensitivity factors, we Jacobian matrix, and the result was then compared with
have to solve the full AC Newton-Raphson load flow the sensitivity factors obtained from the proposed
and invert the Jacobian matrix. This may cause prob- method.
lems in distribution networks as discussed in the Intro-
duction. On the other hand the proposed sensitivity
factors can be updated directly using actual power injec-
tions without having to solve the full AC load flow. If
the voltages used in the calculations are from a solved
load-flow, the sensitivities calculated using the proposed
methodology are identical to those calculated using the
Jacobian-matrix. If the values of voltages are not up-
dated from the base case, the values of sensitivity factors
are approximate. However those approximations are
closer to the true values than if a constant Jacobian ma-
trix was used. The reason is that although the voltages
are kept constant, actual values of nodal injections are
used so the sensitivities are at least partly modified. Our
simulations have shown that using nodal voltages up-
dated after just two Gauss-Seidel iterations provides
almost exact values of sensitivity factors.
The next issue is that of computational complexity.
For a distribution network with (N+1) nodes, the calcu-
lation of Jacobian-based sensitivity factors requires
inverting a (2N)x(2N) Jacobian matrix – assuming that
all embedded generators operate in a constant power
factor mode. For the proposed methodology, calculation
of ∂V&i ∂Pk using (8), and ∂V&i ∂Qk using (11), re-
quires solving two NxN linear equations. Note that solv-
ing linear equations is far less computationally intensive
than inverting a matrix of the same size. If the sensitivi-
ties of all nodal voltage magnitudes with respect to both
real and reactive power injections in all the nodes were
required, that would require solving (NxN) linear equa-
tions 2N times. However it is unlikely that all nodal
sensitivities with respect to all injections would be re-
Fig. 1 The 71-bus rural distribution network
quired – usually a DNO is interested only in a limited
number of important locations. Therefore it is likely that
the proposed methodology is far less computationally 4.1 Voltage sensitivities
intensive that inverting (2N)x(2N) Jacobian matrix. Fig. 2 shows four sets of voltage sensitivities: sensi-
tivities of the voltage at bus 32 and 33 to real and reac-
4 TEST EXAMPLES tive power injection at bus 32 and 33, all four as the
Fig. 1 shows a 71-bus, 33/11kV rural distribution function of real power generation at bus 33. The plots of
network in Scotland used to test the accuracy of the actual values of sensitivities are annotated V33_P33,
proposed method. Bus 1 is the slack bus as it is the grid V33_P32, V33_Q33 and V33_Q32. Those graphs were
connection point. The voltage of interest is at bus 33 produced using the inverted Jacobian matrix but the
(bottom left corner of the diagram). proposed methodology produced almost exactly the
Two embedded generators are connected to bus 32 same values after just two Gauss-Seidel iterations to
and 33 respectively. The output of the generator at bus calculate the nodal voltages used in (8)-(12). That
32 is kept constant at 1.5MW/0MVar. Generation at bus showed that the maths used in the derivations was cor-
33 is increased gradually from 0MW to 2.5MW at unity rect. Sensitivities calculated using a constant inverted
power factor. Further increases of generation would Jacobian matrix would correspond to constant sensitivi-
breach the voltage constraints. We have tested also ties represented by horizontal lines (not shown). The
generators operating at a different power factor than additional index _base on the graphs corresponds to the
unity and the results were very similar to those presented values of sensitivities calculated using the proposed
here. method when the actual injections are used but the volt-
In order to assess the accuracy of the proposed meth- ages used in (8)-(12) are kept constant from the base
odology, a load flow program was run for each operat- case.
ing point to calculate the sensitivity factors based on the Fig. 2 confirms very good performance of the pro-
posed methodology. Just two Gauss-Seidel iterations

16th PSCC, Glasgow, Scotland, July 14-18, 2008 Page 4


were enough to provide very good estimates of the sen- 0.06
sitivities. Using constant values of the voltages (i.e. base
0.04
case) produced reasonably good results for the sensitivi-

Loss sensitivity factor (MW/MW)


ties with respect to real power. Using constant values of 0.02 L_P33
L_P32
voltages when calculating the voltage sensitivities with 0
respect to reactive power produced nearly horizontal
-0.02
lines and therefore a similar result to keeping the Jaco-
bian matrix constant. -0.04

The results also confirmed that keeping the Jacobian -0.06


L_P32_base
matrix constant introduces significant errors. Fig. 2 -0.08 L_P33_base
shows that such a strategy would introduce large errors
-0.1
as the voltage sensitivities can change quite significantly
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5
with changing operating conditions. Generation at bus 33 (MW)

0.06 V32_Q33
V32_Q33_base Fig. 3 Loss sensitivities to real power at bus 33 as the
0.055 function of real power generation at bus 33.
V sensistivity factor (pu/MW, pu/MVAr)

V33_Q32_base
0.05

0.045
V33_Q32 0.06

Loss sensitivity factor (MW/MVAr)


0.04 0.04

0.035
0.02

V33_P33 V33_P33_base 0
0.03

0.025 V33_P32_base -0.02


V33_P32
0.02 -0.04
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 L_Q32
-0.06 L_Q32_base
Generation at bus 33 (MW)
-0.08
Fig. 2 Sensitivities of voltage at bus 33 to real power at bus 32
-0.1
and 33 as the function of real power generation at bus 33. 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5
L_Q33 Generation at bus 33 (MW) L_Q33_base
4.2 Loss sensitivities
Fig. 3 shows sensitivities of network losses to real Fig. 4 Loss sensitivities to reactive power at bus 33 as the
power generation at buses 32 and 33 as the function of function of real power generation at bus 33.
real generation at bus 33 while Fig. 4 shows sensitivities
of network losses to reactive power generation at buses
32 and 33 (using the same scale as in Fig. 3). The nam- 5 APPLICATION OF SENSITIVITY
ing convention is similar to that in Fig. 2. ANALYSIS IN LOSS CHARGING
Fig. 3 and Fig. 4 confirm very good performance of In the UK, the DNOs tend to use the loss adjustment
the proposed method. The graphs show that the loss factors (LAFs) method to allocate network losses to
sensitivities are not constant (due to the approximately individual users, and the LAF values are either deter-
square nature of network losses) and therefore do need mined by the substitution method (with/without) for
to be recalculated whenever operating conditions large users, or roughly assigned according to voltage
change. The graphs of actual values of sensitivities are levels for smaller users [4]. LAFs might not be consis-
shown as L_P33, L_P32, L_Q32 and L_Q33. Those tent and might result in cross-subsidies so two types of
graphs were produced using the inverted Jacobian ma- loss coefficients, Marginal Loss Coefficients (MLCs)
trix but the proposed methodology produced almost and Direct Loss Coefficients (DLCs), were proposed in
exactly the same values after just two Gauss-Seidel [4]. Those coefficients have been compared in this paper
iterations to calculate voltages used in (8)-(15). That with the coefficients calculated employing the proposed
showed that the maths used in the derivations was again method using a simple 4-bus example given in [4] and
correct. Using constant values of voltages to calculate shown in Fig. 5.
the sensitivities produced similar results to keeping the Table 1 compares three sets of loss coefficients:
sensitivities constant. MLCs and DLCs proposed in [4] with Z matrix Loss
Coefficients (ZLCs) proposed in this paper. The results
indicate that the proposed method can give consistent
loss allocation coefficients, and that they are very simi-
lar to those obtained from the Jacobian-based loss
analysis (MLCs and DLCs).

16th PSCC, Glasgow, Scotland, July 14-18, 2008 Page 5


The calculation in Table 1 is based on two scenarios exactly the same result after just two Gauss-Seidel
analyzed in [4] in which the generator at bus C produces iterations. Finally, the incremental loss sensitivity
400kW and 200kW respectively. The demand at bus A coefficients developed in [4] have been compared with
and B is fixed at 200kW. the proposed loss factors showing again a good
agreement.

7 REFERENCES
[1] N. Muller and V. H. Quintana, "Line and shunt switching
to alleviate overloads and voltage violations in power
networks," IEE Proceedings, vol. 136, pp. 246-253, 1989.
[2] B. Stott and O. Alsac, "Fast Decoupled Loadflow," IEEE
Fig. 5 Four-bus example Transactions on Power Apparatus and Systems, vol. PAS-
93, pp. 859-869, 1974.
[3] Y.-C. Chang, W.-T. Yang, and C.-C. Liu, "A New Method
Table 1 Z-bus Loss Coefficients (ZLCs) compared with MLC for Calculating Loss Coefficients," IEEE Transaction on
and DLC coefficients proposed in [4] Power Systems, vol. 9, pp. 1665-1671, 1994.
[4] J. Mutale, G. Strbac, S. Curcic, and N. Jenkins, "Alloca-
Case 1 Case 2 tion of Losses in Distribution Systems with Embedded
EG output (kW) 400 200 Generation," IEE Proc.-Gener. Transm. Distrib., vol. 147,
Total Loss (kW) 1.97 1.21 pp. 7-14, 2000.
Node MLCs -0.0001 -0.0041 [5] A. J. Conejo, F. D. Galiana, and I. Kockar, "Z-Bus Loss
A DLCs -0.0001 -0.004 Allocation," IEEE Transactions on Power Systems, vol.
ZLCs -0.0001 -0.004 16, pp. 105-110, 2001.
Node MLCs 0.0018 -0.0041 [6] E. B. Makram, K. P. Thornton, and H. E. Brown, "Selec-
B DLCs 0.0019 -0.0041
tion of Lines to be Switched to Eliminate Overloaded
ZLCs 0.0019 -0.004
Lines Using a Z-matrix Method," IEEE Transactions on
Node MLCs 0.0058 -0.0021
C DLCs 0.0059 -0.002
Power Systems, vol. 4, pp. 653-661, 1989.
ZLCs 0.0059 -0.002 [7] Y. S. Lim and G. Strbac, "Analytical Approach to Prob-
abilistic Prediction of Voltage Sags on Transmission Net-
works," IEE Proceedings - Generation Transmission Dis-
6 CONCLUSIONS tribution, vol. 149, pp. 7 - 14, 2002.
There is an increased interest in the sensitivity analy- [8] Q. Zhou and J. W. Bialek, "Generation Curtailment to
sis of distribution networks due to increased penetration Manage Voltage Constraints in Distribution Networks,"
by Distributed Generation. However application of IET Proc. Generation, Transmission and Distribution, Vol.
Jacobian-matrix based sensitivity factors encounters 1, May 2007, pp. 492-498.
[9] T. Gonen, Modern Power System Analysis: John Wiley &
serious problems in distribution networks as generally it
Sons, Inc., 1988.
is difficult to perform load flow calculations in distribu- [10] T.-H. Chen, M.-S. Chen, K.-J. Hwang, P. Kotas,
tion networks due to a lack of on-line information. With and E. A. Chebli, "Distribution System Power Flow
the lack of an up-dated Jacobian matrix, sensitivity fac- Analysis - A Rigid Approach," IEEE Trans. on Power De-
tors derived using e.g. a constant Jacobian matrix may livery, vol. 6, pp. 1146-1152, 1991.
be highly inaccurate. [11] SSE, "Long Term Development Statement for
This paper has considered sensitivities of voltages Scottish Hydro electric Power Distribution Ltd's Electric-
and losses to power injections. We have proposed to use ity Distribution System," Nov. 2002.
[12] J. J. Grainger and J. William D. Stevenson, Power
the Gauss-Seidel method and Z-matrix to derive the
System Analysis: McGraw-Hill, Inc., 1994.
sensitivity factors. The advantage is that the sensitivity [13] D. Shirmohammadi, H. W. Hong, A. Semlyen, G.
factors are derived in a closed form as explicit functions X. Luo: "A Compensation-Based Power Flow Method for
of real and reactive power injections and they can be Weakly Meshed Distribution and Transmission Networks,
calculated in a relatively straightforward form, without IEEE Trans. Power Systems, Vol. 3, May 1988, pp. 753-
having to solve the full ac load-flow problem. Gauss- 762.
Seidel method is known to be less sensitive to the choice [14] A. J. Wood, B. F. Wollenberg: "Power Generation
of the initial operating point than the Newton-Raphson Operation and Control" Wiley, 1996.
method and tends to be more robust in distribution net- [15] Christie, R. D., Wollenberg, B. F. and Wangen-
steen, I.: "Transmission Management in Deregulated Envi-
works. All these features make it more convenient to be
ronment," Proceedings of the IEEE, vol. 88, pp. 170-195,
used by DNOs. 2000.
The proposed methodology has been illustrated using [16] P. R. Bijwe, G. K. Viswanadha Raju: "Fuzzy
a model of 71-bus real distribution network. Comparison Distribution Power Flow for weakly meshed system",
of sensitivity factors derived using the proposed IEEE Trans on Power Systems, Nov, 2006, vol 21, No.4,
methodology with those calculated using the Jacobian pp1645-1652
matrix showed that the two methods produced almost

16th PSCC, Glasgow, Scotland, July 14-18, 2008 Page 6

You might also like