EE-521 Term Paper 1

Review of Load Flow Analysis Schemes for

Distribution Networks
Syed Muhammad Hur Rizvi [syed.rizvi@wsu.edu]
Washington State University

newton based scheme.
Abstract—Distribution networks are mostly radial and have In this paper following schemes have been reviewed:
High R/X ratios. Load flow Analysis techniques normally used A.) Implicit Z-Bus based, B.) Forward/Backward sweep
for transmission network load flow analysis are not suited for based, C) Direct method, D) Modified Newton and E)
distribution network load flow analysis. Jacobian matrix for Modified Fast decoupled.
traditional load flow techniques end up getting ill-conditioned.
Load flow techniques for distribution networks devised for Paper is organized as follows. In Section II, Mathematical
dealing with such issues can be grouped in to Newton like and basics and algorithms of above mentioned schemes have been
Non-newton techniques. Some of the techniques in both reviewed. In Section III merits and demerits along with
categories have been analyzed in this paper. suitability of various schemes for various applications have
been discussed. Section IV presents the Conclusion.
Index Terms—Distribution Load flow Analysis, Implicit Z-
Bus, Forward-Backward, Newton, Fast decoupled
II.MATHEMATICAL DESCRIPTION OF DISTRIBUTION
LOAD FLOW SCHEMES
I. INTRODUCTION Based on the type of iterative method adopted Distribution
load flow analysis schemes can be grouped into two
Distribution networks are mostly radial or at most weakly
categories:
meshed in most cases. Distribution laterals can be 3-phase, 2-
I. Newton Type Iterative techniques
phase or 1-phase depending on the needs of consumers.
II. Non-Newton Type Network Topology based
Distribution networks comprise of un-transposed lines and
schemes.
despite efforts to allocate equal loads on all phases,
Category-I includes schemes that are either extension of
distribution networks are unbalanced. Moreover R/X ratio of
conventional newton based load flow schemes in transmission
typical distribution networks is much higher than transmission
networks or their modifications for distribution networks.
networks.
Techniques in Category-II exploit the radial nature of
Because, of above stated reasons transmission load flow
distribution network to enhance the computational efficiency
schemes aren’t suitable for distribution networks and
of distribution load flow problem.
demonstrate poor convergence properties. Conventional
In this section mathematical background of some of the
Newton Raphson and fast decoupled schemes are hampered
distribution load flow schemes is presented.
by ill-condition of Jacobian matrices.
These issues with traditional schemes resulted in work on A. IMPLICIT Z-BUS METHOD
many non-Newton schemes for distribution networks. The This scheme [1] is based on principle of superposition and
most common and most reported scheme in distribution load has similarities in algorithm to Gauss-seidel method in
flow literature is Forward Backward sweep based load flow transmission load flow analysis. This scheme is based on the
analysis of radial networks. Scheme as reported in [1] is based fact that voltage at each bus has two contributions:
on application of ladder network analysis like technique to 1. Voltage deviation due to current injections
non-linear distribution networks. In [2] Mr. X et. al have 2. Contribution due to swing bus voltage source
presented an improved version of forward/backward sweep It is assumed while applying this approach that there is only
algorithm. one voltage source in the system.
Another non-Newton scheme is implicit Z-bus based load Algorithm:-
flow scheme which is based on principle of superposition and I. Factorization of Y-Bus (Ybus=L*U)
sparse factorization of Y-bus matrix. T-cheng et. al [3] have II. Computation of injected currents
used this scheme in development of their program (GDAS) for k
I inj=conj¿
distribution system analysis.
III. Computation of voltage deviation due to injected
Among Newton type schemes specially developed for currents
distribution networks are the Modified Newton Method [4] k k
and Modified Fast decoupled scheme [5]. Both these schemes I inj=L . U . D V
are modified to deal with ill-condition of Jacobian Matrix and IV. Superposition principle:
k +1 (k )
are more suitable to distribution networks than the traditional V =V . NL+ D V
V. Convergence check.

EE-521 Term Paper 2

 If Tol<V(k+1)-V(k). Solution found VII. Check convergence. If achieved terminate else go to

 If Tol>V(k+1)-V(k). Go to Step II. step III.

Comments:- Comments:-
This method assumes that system has only one voltage  Forward backward sweep algorithm is normally used as
specified Bus. If that assumption holds this method a benchmark to compare convergence characteristics of
demonstrates convergence properties as good as Newton’s load flow schemes.
method. Convergence characteristics worsen if the system has  The drawback associated with this method is that it can
more voltage controlled buses. With increasing penetration of not handle meshed or weakly meshed networks. It is
Distributed generation Implicit Z-Bus method may not only suited for distribution networks that are entirely
demonstrate good convergence properties. radial.
B. FORWARD BACKWARD SWEEP BASED TECHNIQUES  Moreover it’s mathematical representation is not
friendly for applications like state-estimation and
Forward-Backward sweep [2] based scheme is extension of
optimal power flow.
ladder network theory to solve non-linear distribution
networks. In this technique following two steps are iteratively Traditional 2 port Representation:-
performed after ordering the network properly: For both transmission and Distribution systems sending end
 Backward sweep: Based on KCL to find all the and receiving end quantities are normally represented in terms
branch currents based on current injections. of two port matrices. Consider node “m” to be a sending node:
Calculations in this step traverse from end node I m=c . V n+ d . I n (BACKWARD SWEEP).
towards the source node (Hence the name
Backward sweep).
V n= A .V m −B . I n (FORWARD SWEEP)
 Forward Sweep: Based on KVL to find out Here numerical value of “A”, “B”, “c” and “d” would depend
upon the link between sending end and receving end.
voltages at all the buses in the network.
Constants will have different values for transmission lines,
Calculations in this step traverse from source node
Voltage regulators and transformers different configurations
towards the end node (Hence the name Forward
sweep).
Another similar technique [3] proposed by Cheng et al C.DIRECT METHOD BASED ON BIBC & BCBV MATRICES
enhances the computational efficiency of forward sweep by This method [4,5] basically tries to generalize forward
using linear proportion principle. In the forward sweep, backward sweep in-terms of 2 system matrices:-
following 6 ratios are calculated at the end of backward sweep  BIBC (Bus injection to Branch current Matrix)
and multiplied with calculated quantities in backward sweep:  BCBV (Branch current to Bus Voltage)
Real (V s ( a) ) imag(V s (a )) The solution algorithm for load flow is as follows:
 γ 1= , γ 2=
Real(V 1( a) ) imag (V 1 (a ))

 γ 3=
Real (V s ( b) )
Real (V 1 (b ))
, γ 4=
imag(V s( b ))
imag (V 1 (b ))
k
I i =conj
( Pi + j. Qi
Vi
k
)
Real (V s ( c )) imag(V s (c ) ) [ B(k ) ]=[ BIBC ] [ I k ]
 γ 5= , γ 6=
Real (V 1 (c ) ) imag (V 1 (c ))
Algorithm:- [ ∆ V k ]=[ BCBV ] [ Bk ]
The Algorithm for scheme proposed by Cheng et. al [2] is as
follows: k +1 0 k
V =V +¿ ∆ V
I. Sort buses according to distance from substation BIBC & BCBV are constant matrices based on the network
II. Initialize topology.
III. BACKWARD SWEEP: Start backward sweep from Computation of BIBC & BCBV
end buses. Apply KCL/KVL, to compute currents Consider a single feeder radial distribution network shown in
and voltages Continue until substation/source voltage Figure 1. BIBC and BCBV can be computed as stated below.
is computed.
IV. Check if voltages are with in tolerance. If yes
terminate, else continue
V. Real and Imaginary parts of estimated substation
voltages are compared to actual voltages for each
phase. This results in generation of 6 ratios. Figure 1
VI. FORWARD SWEEP: Multiply each ratio with
corresponding quantity at all buses to get new value.
Continue until the end bus.
EE-521 Term Paper 3

COMPUTATION OF BIBC: The Algorithm is based on definition of these 3 quantities:

For the network given in Figure 1, if we apply KCL we can j∆V
write branch current equations as: E=∆ θ+ , S=∆ P+ j ∆Q , W =DB + j DG
V
 I 12=I 2 + I 3 The modified Newton method’s Jacobian equation (1) can be
I 23=I 3 represented as below in terms of above defined quantities:

[ ] [ ][ ]
T
I 12 1 1 I2 An−1 W ( An−1 ) E=S
= [ B ] =[ BIBC ] [I ]
I 23 0 1 I3 A( n−1) S L =S ( Analogous ¿ BACKWARD SWEEP )
T
 BIBC is an upper triangular matrix, it has ones as fills W ( A n−1 ) E=S L ( Analagous ¿ FORWARD SWEEP )
on points depending upon system topology.
 This step is generalization of backward sweep in Updated values are given as:
terms of BIBC matrix. k+1 k
θ=θ + Real ( E )
COMPUTATION OF BCBV: V k +1=V k +imag ( E )∗V k
For network in Figure 1, starting from node adjacent to source
node we can write Bus voltages as: Convergence check: Like the conventional NR convergence is
 V 2=V 1 −I 12 . Z 12 checked upon Del-P and Del-Q.
E. MODIFIED FAST DECOUPLED METHOD
 V 3=V 2−I 23 . Z 23=V 1−I 12 . Z 12−I 23 . Z 23 In order to demonstrate the mathematical basis consider a
single feeder distribution system in Figure 2.


[ ][ ][
V1 V 2
−
V1 V 3
Z
= 12
0 I 12
Z 12 Z 23 I 23 ][ ]
 ∆ V =BCBV .[B]
 BCBV is a lower triangular matrix with Line Figure 2
impedance of corresponding section as fills.
Assuming nth bus is the last point in the network In+1=0. If
 This step above is basically generalization of forward we know voltage Vn then we can compute all the voltages in
sweep. the network. Using this argument authors [5] have considered
Vn as the only independent variable in the network.
D.MODIFIED NEWTON BASED METHOD
Because of high R/X ratio of distribution lines, the
conventional load newton Raphson scheme does not
Consider w k =
[ ]
Vk
I k +1
Using same representation:
then w k−1=gk (w k )

demonstrate good convergence properties due to ill-

conditioning of Jacobian. Scheme proposed by Zhang et. al []
addresses this problem by decomposing the Jacobian into
T
UD U form to avoid the ill-condition. The method is based
w n−1=gn ( w n )=gn
([ ])
Vn
0
on following assumptions:-
w n−2=gn−1
Assumptions:-
1. All the shunt branches in the network are ignored.
2. It is assumed that voltage difference between
( wn−1 )=g n−1 . g n ([ ]) [ ]
Vn
0
V
w 0= 0 =g1 ( w1 ) =g1 . g2 … … g n
I1

([ ])
adjacent nodes is negligible Vn
Based on these assumptions and approximations based on
them the traditional Newton Raphson form can be written as: 0

][ ][ ][
∆θ
[ ][ ]
T
A n−1 0 DB −DG A n−1 0 ∆ P (1)Based on the functions given on previous slide we can
∆V =
0 A n−1 DG D B 0 T
An−1 V ∆ Q compute the voltage at source Vo given the end voltage Vn
∂f
f ( V n )=V 0 ( V n ) −V 0 J ( V n) =
D D
Here B and G are block diagonal matrices whose entries ∂V n
are given as: ❑
∂f
DB (i, j)=V i V j Bij cos ( θij ) DG (i , j)=V i V j B ij sin(θ ij ) ∆ V k+1
n = ∗(V 0 ( V kn )−V 0 )
∂Vn
Algorithm:- k +1 k
V n =V n +∆ V n
k+1
First step of the algorithm is ordering of the distribution

network in form of layers as we traverse from source node
MULTI-LAYER CASE
towards end nodes. In this step An-1 matrix is also obtained
Same concept discussed for a single feeder can be extended to
which has 1’s at diagonal entries and -1’s at other connection
multi-layer case. If there are “m” end voltages, then there will
points.
EE-521 Term Paper 4

be m independent variables in the network. In-case of end meshed distribution networks. Another drawback is their un-
voltages other than main feeder, Bus branching voltages serve suitability for applications like state estimation and optimal
as input/source voltages. In such a case voltage mismatch is power flow which are important building blocks for power
taken as difference of Branching bus estimated voltage from system analysis and operation.
two ends. Distribution load flow analysis techniques in other category
DECOUPLED NATURE are Newton type techniques. Authors have proposed
Authors have shown that off diagonal entries in the Jacobian modifications to deal with ill-conditioning of Jacobian matrix.
are nearly zero, thus they can be assumed to be equal to zero. These techniques are suited for state estimation and optimal
This approximation means that mismatch for any bus on power flow like applications. Some examples of such
lateral “i” is only dependent on end voltage of lateral “i” and methodologies are modified Newton Raphson and Modified
end voltage of lateral “i’s” parent. This approximation speeds
fast decoupled techniques discussed in this paper.
up the algorithm greatly.
ADVANTAGES
REFERENCES
Authors have shown that off diagonal entries in the Jacobian
are nearly zero, thus they can be assumed to be equal to zero. [1] T. H. Chen, M. S. Chen, K. J. Hwang, P. Kotas and E. A. Chebli,
"Distribution system power flow analysis-a rigid approach," in IEEE
This approximation means that mismatch for any bus on Transactions on Power Delivery, vol. 6, no. 3, pp. 1146-1152, Jul 1991.
lateral “i” is only dependent on end voltage of lateral “i” and [2] D. Thukaram, H.M. Wijekoon Banda, Jovitha Jerome, A robust three
end voltage of lateral “i’s” parent. This approximation speeds phase power flow algorithm for radial distribution systems, In Electric
up the algorithm greatly. Power Systems Research, Volume 50, Issue 3, 1999, Pages 227-236,
ISSN 0378-7796.
[3] G. W. Chang, S. Y. Chu and H. L. Wang, "An Improved
Backward/Forward Sweep Load Flow Algorithm for Radial Distribution
F. OTHER METHODS Systems," in IEEE Transactions on Power Systems, vol. 22, no. 2, pp.
Compensation based scheme (Shiro-Mohammadi et. al): 882-884, May 2007.
Based on conversion of weakly meshed networks to radial [4] Jen-Hao Teng, "A direct approach for distribution system load flow
solutions," in IEEE Transactions on Power Delivery, vol. 18, no. 3, pp.
networks before application of forward/backward iterative 882-887, July 2003.
scheme. [5] J. H. Teng, "Modelling distributed generations in three-phase
distribution load flow," in IET Generation, Transmission & Distribution,
Modified Gauss Seidel Method : An extension of implicit Z- vol. 2, no. 3, pp. 330-340, May 2008.
[6] T. H. Chen, M. S. Chen, K. J. Hwang, P. Kotas and E. A. Chebli,
Bus scheme in which while applying implicit Z-bus most "Distribution system power flow analysis-a rigid approach," in IEEE
recently updated values by gauss seidel based step are used. Transactions on Power Delivery, vol. 6, no. 3, pp. 1146-1152, Jul 1991.
Moreover instead of doing LU factorization of whole Y-bus, [7] D. Thukaram, H.M. Wijekoon Banda, Jovitha Jerome, A robust three
YAA,YBB and YCC are factorized separately affording better phase power flow algorithm for radial distribution systems, In Electric
Power Systems Research, Volume 50, Issue 3, 1999, Pages 227-236,
computational efficiency. ISSN 0378-7796.
[8] G. W. Chang, S. Y. Chu and H. L. Wang, "An Improved
Backward/Forward Sweep Load Flow Algorithm for Radial Distribution
Systems," in IEEE Transactions on Power Systems, vol. 22, no. 2, pp.
III. DISCUSSION
882-884, May 2007.
[9] Jen-Hao Teng, "A direct approach for distribution system load flow
Load flow is one of the most important tools used to analyze solutions," in IEEE Transactions on Power Delivery, vol. 18, no. 3, pp.
882-887, July 2003.
the steady state behavior of Power systems. The load flow [10] J. H. Teng, "Modelling distributed generations in three-phase
analysis schemes for Transmission networks are well distribution load flow," in IET Generation, Transmission & Distribution,
developed and techniques like Newton Raphson and Fast vol. 2, no. 3, pp. 330-340, May 2008.
decoupled demonstrate excellent convergence properties.
Same techniques when applied to solve distribution load flow
problem face convergence issues because of higher R/X ratio
of distribution networks. Higher R/X ratio results in ill-
conditioning of Jacobian matrices.
In order, to avoid the convergence issues techniques specially
devised for distribution networks are preferred. Distribution
load flow techniques can be grouped into 2 categories which
are Newton type and Non-Newton Type techniques.
The advantage associated with most Non-newton techniques is
their suitability to analyze radial distribution networks. These
methods demonstrate excellent computational performance if
distribution network are radial. Forward-Backward sweep,
Implicit Z-bus method and Direct method based on
BIBC/BCBV matrices are some of the methods in this
category. This disadvantage associated with techniques in this
category is that they are not suited for analyzing weakly