You are on page 1of 6

PowerTech 2011, 19 – 23 June 2011, Paper accepted

Trondheim, for presentation at the 2011 IEEE Trondheim PowerTech


Norway 1
Paper ID: 546

Critical Comparison of Robust Load Flow


Methods for Ill-conditioned Systems
Jorge F. Gutiérrez, Student Member, IEEE, Manfred F. Bedriñana, Member, IEEE, and
Carlos A. Castro, Senior Member, IEEE

Abstract-- In this work, the performances of robust load proximity of the load level to the system’s
flow methods are compared, in special when applied to ill- maximum loadability or maximum loading point
conditioned systems. In such cases, the iterative process may (MLP), also known as the saddle-node bifurcation
diverge or oscillate when standard Newton Raphson (SNR) point; ii) choosing an initial state that conducts the
based load flow calculation methods are used. Some known standard calculation method to a trajectory that
approaches to solve these systems are the Load Flow with Step
Size Optimization (LFSSO) and Continuation Power Flow
diverges and the condition number increases. This
(CPF). Also, the Continuous Newton Power Flow (CNPF) behavior occurs although the system loading is
method was proposed recently as a novel approach to solve ill- within the feasible region and a operating point does
conditioned systems. Even though the performances of these exist.
methods have been reported in several sources, a critical 3) Infeasible cases. The operating point does not exist
comparison of the convergence characteristics and results for because the load exceeds the system MLP. Robust
ill-conditioned systems are presented in this work. A modified methods are supposed to detect and provide
version of the CNPF is also tested in order to improve its information about the insolvability situation.
robustness. Simulation results for some IEEE test cases and In order to improve the SNR based methods performance
realistic systems are shown to validate the analysis.
the Load Flow with Step Size Optimization (LFSSO) was
Index Terms—Load flow analysis, ill-conditioned systems, proposed in [5]. LFSSO uses a factor (optimal multiplier)
step size optimization, continuous Newton method. that minimizes a quadratic cost function based on the power
mismatches. This method worked very well, however the
I. INTRODUCTION voltages appeared in rectangular coordinates, which is not a

W ELL known facts such as the increase in peak load common feature of load flow programs. An approach based
demand, transmission congestion, power transfers on polar coordinates representation has been proposed in [6]
between utilities and the deregulation process with some and its advantages have been already demonstrated [7]. In
redundancy reduction in several systems, have led the [8], the authors recommended the implementation of the
power industry to an increased concern about power optimal multiplier modification to the SNR based methods
systems security and adequacy. In order to evaluate the with polar coordinates (rather than rectangular coordinates)
power system condition in system planning and operation, to get the fastest, most robust performance, regardless of
robust offline and online tools such as: contingency system solvability or size.
analysis, voltage stability and dynamic analysis are needed Recently in [3] a novel approach to the load flow
and their efficiency and reliability are still an issue [1]. The solution based on continuous Newton method has been
load flow is a basic tool for power system analysis, and proposed. The Continuous Newton Power Flow (CNPF)
there are two important features this tool should exhibit: exhibits a formal analogy between the solution of a system
efficiency and robustness. A robust tool will supply the of ordinary differential equations and the solution of
system operator with a load flow tool that: converges to an nonlinear systems based on the Newton-Raphson method.
answer only in cases when the operating point lies within This approach presents a more robust behavior than SNR
the feasible region, and in some cases will provide methods, but its robustness seems to be associated to
appropriate information when the load level corresponds to heuristic rules that could lead to divergence of the iterative
an infeasible situation. process or to an UEP for some ill-conditioned cases, as it
From the solvability point of view, load flow cases can will be shown later.
be classified into one of the following categories [2]-[4] : The goal of this work is to compare some robust solution
1) Well-conditioned cases. The system’s operating techniques to the power flow problem, and to show that a
point can be obtained using classical approaches, combination of the LFSSO with the CNPF (referred to as
such as the standard Newton-Raphson (SNR) the Modified Continuous Newton Power Flow - MCNPF)
method. This point could be: i) a stable equilibrium presents a better performance to solve ill-conditioned and
point (SEP) or operating point ii) an unstable infeasible cases than the original CNPF method.
equilibrium point (UEP). The stability of the final
state depends on the initial point and the numerical II. THEORETICAL CONCEPTS
method employed.
A. Standard Newton Raphson Load flow (SNR)
2) Ill-conditioned cases. Linear systems theory defines
an ill-conditioned system as the one for which its The load flow equations are formulated as:
condition number is sufficiently large. This large g ( x) = 0 (1)
condition number could be a result of: i) the where x∈ℜ (2nPQ+nPV)
is the vector of state variables, x = [șt

978-1-4244-8417-1/11/$26.00 ©2011
PowerTech 2011, 19 – 23 June 2011, Trondheim, Norway 2
Paper ID: 546

Vt]t, also ș∈ℜnPQ+nPV and V∈ℜnPQ are vectors of bus C. Continuous Newton Power Flow (CNPF)
voltage angles and magnitudes, respectively; nPQ and nPV It was shown in [9] that the simplest solution of a first-
are the number of PQ and PV buses, respectively; g(x) is order differential equation given by
defined as g = [ǻPt ǻQt]t, where ǻP∈ℜnPQ+nPV and x = f (x) (8)
ǻQ∈ℜnPQ are the real and reactive power mismatches, could be obtained by forward Euler’s method as:
respectively. Eq. (1) can also be written as:
Δx( r ) = Δt ⋅ f ( x ( r ) ) (9)
ª ΔP(x) º ª Psch − Pcal (x) º ª 0 º
g ( x) = « »=« »=« » (2) Considering Δt = 1 and if f (x) is defined as:
¬ ΔQ(x) ¼ ¬Q sch − Q cal (x) ¼ ¬ 0 ¼
f ( x ) = − [∇ x g ]
-1
where subscripts sch and cal stand for scheduled and g (x) (10)
x
calculated terms, respectively. Also, the reactive power then, the solution of a nonlinear system by SNR method is
generation limits are taken into account, so PV buses are analogous to the solution of numerical integration by
switched to PQ whenever some reactive power limit is Euler’s method. Also, considering that Δt = ȝ then, the
reached and can be switched back to PV whenever analogy will be between LFSSO and the numerical
appropriate. integration method with a variable time step. Therefore,
Expanding (1) into Taylor series up to the first order solving the initial value problem with some variable time
terms and making the assumption that the initial guess is step could be viewed as solving a nonlinear system using a
close enough to the system solution, the high order terms robust method.
can be neglected, yielding the standard Newton-Raphson The fourth-order Runge-Kutta (RK4) will show a better
method for n-dimensional systems as: performance than the forward Euler method. A generic RK4
Δ x( r ) = − ª¬∇x g( r ) º¼ g ( x( r ) )
−1
(3) step at the rth iteration is obtained by the following
(r ) expression.
where ∇xg is the system’s Jacobian matrix at iteration r.
k1 = f (x( ) )
r

B. Load Flow Method with Step Size Optimization (LFSSO)


k2 = f (x( ) + 0.5Δtk1 )
r

LFSSO was developed for solving ill-conditioned power


k3 = f ( x( ) + 0.5Δtk2 )
r
systems [5], [6]. For those, the conventional load flow (11)
methods exhibit a poorer performance, or simply diverge, (r)
k4 = f (x + 0.5Δtk3 )
although the system indeed operates in a stable equilibrium Δt (k1 + 2k2 + 2k3 + k4 )
x( ) = x( ) +
r +1 r
point. At the rth iteration of the LFSSO, the state variable
vector x(r+1) is calculated as: 6
In order to improve RK4 efficiency when solving load
x ( r +1) = x ( r ) + μ ( r ) Δx ( r )
(4)
flow problems, a variable time step (Δt) was proposed in
Δ x ( r ) = − [∇ x g ]
-1
g(x( r ) ) [3], using a heuristic rule based on the truncation error value
x = x( r )
(ȟ) at each iteration, as:
where ȝ(r) is a step size optimization factor that multiplies
the state variable correction vector ǻx(r). Also, ȝ is { (
ξ = max abs k2 − x ( r +1) )}
computed to minimize the following quadratic function
if ξ > 0.01 then ǻt ← {0.985ǻt , 0.75} (12)
based on the power mismatches. max

if ξ < 0.01 then ǻt ← {1.015ǻt , 0.75}


¦ g ts2 , i
1 2 1
min F (μ ) = 2
g st 2
= 2 (5) min
i∈Ωg
After the first step this heuristic rule corrects the time
where gts is g expanded in Taylor series, considering up to step size (Δt) for the next iteration based on the truncation
the second-order terms, as: error in the middle of the step. The first step will always
g ts ( μ ) = g ( x ( r ) ) + μ ∇ x g( x ( r ) )t Δx ( r ) + μ 2T ( x ( r ) ) (6) have a unitary size which could be a disadvantage. It seems
Also, T(x) corresponds to second order terms of g, given that the method’s performance could be improved with a
by: technique that evaluates the step size from the beginning,
1§ ∂ ·
2
and that uses a stop criterion when the step size becomes too
¨ ¦ Δxi
T ( x) = ¸ g ( x) (7) small in order to reduce the number of iterations.
2 © i∈Ωg ∂xi ¹
Substituting (5) in (4) and applying the local minimum D. Two-bus example
condition ˜F/˜ȝ = 0, a cubic equation is obtained and solved In order to analyze the continuous Newton approach to
for ȝ. power system solution, a simple two bus example system
For well-conditioned systems, ȝ assumes values close to will be used (Fig. 1). The system has a single lossless
one and does not affect the iterative process in a significant transmission line with no shunt charging connecting the two
way. In the case of ill-conditioned systems, ȝ assumes buses.
values such that the iterative process is smoothed out and
the solution is obtained, while the SNR method would have E∠ 0º V∠ ș jQg Pg = 0
failed. For infeasible cases, ȝ tends to zero and LFSSO jx
C -’ ” Qg ” ’
provides a point on the boundary between solvable and
unsolvable regions. 1 2
Pc + jQc

Fig. 1. Two bus example power system


PowerTech 2011, 19 – 23 June 2011, Trondheim, Norway 3
Paper ID: 546
Bus 1 is treated as a slack bus with constant voltage E switching from RK4 to Euler’s when the system correction
(E= 1.0 p.u., ș1= 0.0º). Bus 2 is a PV bus with variable P-Q (ǻx) becomes smaller than some threshold was proposed in
demand but with an ideal synchronous condenser that holds [3].
its voltage magnitude V (V = 1.0 p.u.) for any load value. For a system load Pc = 1.0 p.u. the analytical and
For simplicity the line reactance will be x= 1.0 p.u.. numerical solutions are shown in Fig. 4 Now the system
Two load levels at bus 2 will be considered in order to behavior plot have more pronounced changes at the
analyze continuous Newton approach: Pc = 0.8 p.u. (well- beginning that before (for Pc = 0.8 p.u.) and Euler method
conditioned case), Pc= 1.0 p.u. (it can be easily verified that will also take more iterations to find a solution. Methods
this load level corresponds to the MLP). based on variable time step size will exhibit a better
The power flow equation for this system is: performance than the ones with fixed step. One possible
ΔP(2) = Pspc 2 − VEB12 sin θ (13) approach to obtaining variable step methods is by using (12)
where Psch2 is the scheduled power at bus 2 (Psch2 = - Pc2) as proposed in [3]. Another one referred here as the
Modified Continuous Newton Power Flow - MCNPF will
and B12 is the admittance matrix element (equals to 1/x). For
this system (10) becomes: be tested in Section IV.
Graph Of Theta By Analitical And Numerical Methods
§ 1 · 0
θ = f (θ ) = ¨ ¸ ( Psch − VEB12 sin θ ) (14) Analytical sol.

© VEB12 cos θ ¹
-0.2
RK4
-0.4 Euler
The Jacobian matrix for this system will be singular
when ș= -ʌ/2. -0.6

T heta
Considering that ș(t=0) = 0.0 (flat start) and using -0.8

classical integration techniques the system analytical -1

solution becomes: -1.2

§ P ·
θ ( t ) = sin −1 ¨ c ( e −t − 1) ¸
-1.4
(15)
© VEB12 ¹ -1.6
0 5 10 15
Consider that the Euler method is used to solve this Time

example. From (8) and (9) the solution trajectory is defined Fig. 3. Maximum loadability case - Two bus system Pc = 1.0 p.u.
by the following expression:
In order to improve the performance of CNPF it is
( k +1) (k ) § − P − VEB12 sinθ · desirable to reduce the number of iterations. The plots
θ (t ) =θ ( t ) + h⋅¨ c ¸
© VEB12 cosθ ¹
(16) shown in Fig. 4 are for RK4 and Euler methods trajectories
with h = 2 and h = 3. When using Euler method with h = 2
Fig. 2 shows the plot of the analytical solution and the the system operating point can be obtained after almost half
numerical solution obtained by Euler and RK4 methods of the iterations of when h = 1, but with h = 3 the method
with a fixed time step h = 1.0 and a load at bus 2 of Pc= 0.8 becomes unstable. This is a consequence of using the
p.u. explicit Euler method, which is a first order method. RK4
Graph Of Theta By Analitical And Numerical Methods approach is well-known for exhibiting a more stable
0
Analytical sol. behavior than Euler method, and even with h = 3 the
-0.2
Runge-Kutta sol method reaches the operating point. This result could
Euler sol. motivate the simulation of realistic systems increasing time
-0.4 step size.
T heta

Effect of Bigger Time Step Lengths


0
-0.6
-0.5
-0.8
-1

-1
T h eta

0 5 10 15 -1.5
Time
Fig. 2. Well-conditioned case - Two bus system with Pc = 0.8 p.u. -2 Analytical sol.data1
RK4 h= 2
Euler h= 2
As discussed before, solving an initial value problem by -2.5
RK4 h=3
Euler method with unitary time step is equivalent to solving Euler h= 3
-3
a nonlinear equations system by SNR. 0 5 10 15
Time
Fig. 4. Comparison of different integration step sizes for the two bus
system with Pc = 1.0 p.u
By analyzing the plots several conclusions can be
obtained. It is well known from classical ODE’s theory that
III. TEST RESULTS
RK4 methods give closer and more stable trajectories than
Euler methods. As we are not interested in system time In order to compare the robustness of the
evolution but in its final steady state solution, it seems that abovementioned methods (SNR, LFSSO and CNPF using
the Euler approach will take less iterations to reach the load RK4) some tests were conducted with an ill-conditioned
flow solution’s neighborhood. In order to take advantage of system (11-bus Scudder system) [6].
this feature thus reducing the number of CNPF iterations, Fig. 5 shows the V-Q curve obtained from the SNR
PowerTech 2011, 19 – 23 June 2011, Trondheim, Norway 4
Paper ID: 546
method by varying the reactive power injection at bus 8 IV. MODIFIED CONTINUOUS NEWTON POWER FLOW
(Q8) from 20 to 140 MVAr. The SNR iterative process (MCNPF)
started from flat start for each value of Q8. Due to the convergence problems that CNPF method
1.5
V-Q Curve by SNR Method
exhibited in the simulation tests, a modification for the time
step size calculation is tested. As the method’s robustness
relies on the variable time step, the heuristic rules of (12) to
1 determine Δt were replaced by the optimal multiplier in
polar coordinates proposed in [6]. The algorithm for
Voltage at 8 Bus

MCNPF is succinctly described as follows.


0.5 1) Store x (r )
2) Compute k1 as in (11).
0
3) Compute μ using polar coordinates as explained in
Sec. II-B (using x(r) and the classical power flow
20 40 60 80 100 120 140
Reactive Power Injected at 8 Bus

Fig. 5. V-Q curve for bus 8 of Scudder 11-bus system using SNR equations).

The asterisks represent the solution points provided by 4) Compute k2, k3, k4, and x (r + 1) as in (11), replacing
the SNR method (those with V8= 0 represent divergent Δt by μ.
loading situations). The continuous line represents the A routine based on this algorithm was implemented and
system complete V-Q curve, containing the actual voltages for the 11-bus system the results were the same as those
for each value of Q8. From Fig. 6 it is clear that the system obtained with LFSSO in Fig. 6(a).
is ill-conditioned for some load conditions. The SNR In order to compare the differences between the time
method provides either SEP points (high voltages), UEP steps size calculations when using CNPF, MCNPF and
points (low voltages), or simply diverges (shown in Fig. 6 LFSSO, the step size evolution for three different loadings
with V8= 0, for the sake of simplicity), depending on the were plotted.
scheduled value of Q8. The first loading is shown in Fig. 7 and corresponds to a
For the same system and loading conditions, LFSSO and 11-bus Scudder system with 120 MVAr generation at bus 8
CNPF methods provided the results shown in the following (Q8 = 120 MVAr). For this case CNPF and MCNPF
figures. Fig. 6 (a) was obtained by LFSSO and flat start, methods converged to the same operation point as can be
while Fig. 6 (b) was obtained by CNPF method and flat seen from Fig. 6 a) and b) after 10 iterations (recall that
start. By comparing these figures, one can conclude that MCNPF and LFSSO always converge to the same operation
LFSSO is more robust than CNPF. While LFSSO provided point). The LFSSO method converged after 4 iterations.
Scudder case with Q8 = 120 MVAr

the correct solution for the whole range of Q8 starting from a


1.1

reasonable point (flat start), the solution obtained by CNPF 1

in some cases correspond to unstable operation conditions, 0.9

while in other cases it is not able to find the solution.


Time Step Size

0.8

Finally, Fig. 6 (c) and (d) were obtained by CNPF, but 0.7

modifying the initial condition. In Fig. 6 (c) the initial 0.6


voltage magnitudes were set to 1.05 p.u., while in Fig. 6 (d) CNPF
MCNPF
0.5
they were set to 0.85 p.u. For those cases CNPF could find a LFSSO

solution for the whole reactive power injection range. 0.4


1 2 3 4 5 6 7 8 9 10 11
Iteration
However, the solution provided may correspond to unstable Fig. 7. Step Size Evolution for 11-bus Scudder system Q8= 120 MVAr
cases, which is not a desirable feature.
The second loading is shown in Fig. 8 and corresponds to
V-Q Curve by LFSSO Method V-Q Curve - CNPF - Heur. Min. Step= 0.75
the 11-bus Scudder system with 60 MVAr generation at bus
1.2 1.2
8 (Q8 = 60 MVAr). For this case CNPF solution differs
from the one obtained by MCNPF and LFSSO methods.
1 1
Voltage at 8 Bus

Voltage at 8 Bus

0.8 0.8

0.6 0.6
While CNPF converged after 11 iterations to a low voltage
0.4 0.4 point (UEP), MCNPF and LFSSO converged to the high
0.2 0.2
voltage point (SEP) after 9 and 4 iterations respectively.
0 0
Scudder case with Q8 = 60 MVAr
20 40 60 80 100 120 140 20 40 60 80 100 120 140 1.1
Reactive Power Injected at 8 Bus Reactive Power Injected at 8 Bus

a) LFSSO method (flat start) b) CNPF (flat start) 1


V-Q Curve - CNPF - V initial= 1.05 V-Q Curve - CNPF - V initial= 0.85

0.9
1.2 1.2
Time Step Size

1 1 0.8
Voltage at 8 Bus

Voltage at 8 Bus

0.8 0.8
0.7
0.6 0.6

0.6
0.4 0.4 CNPF
MCNPF
0.2 0.2 0.5 LFSSO
0 0
20 40 60 80 100 120 140 20 40 60 80 100 120 140
Reactive Power Injected at 8 Bus Reactive Power Injected at 8 Bus 0.4
1 2 3 4 5 6 7 8 9 10 11 12

c) CNPF with V = 1.05 p.u. d) CNPF with V = 0.85 p.u. Iteration

Fig. 6. V-Q characteristics. Fig. 8. Step Size Evolution for 11-bus Scudder system Q8= 60 MVAr
PowerTech 2011, 19 – 23 June 2011, Trondheim, Norway 5
Paper ID: 546
The third loading is shown in Fig. 9 and corresponds to a robustness as LFSSO, however with a larger number of
11-bus Scudder system with 80 MVAr generation at bus 8. iterations. Besides, every MCNPF iteration requires four
For this situation CNPF diverged after 200 iterations while Jacobian matrix factorizations.
the MCNPF converges after 9 iterations and LFSSO TABLE I. CONVERGENCE AND NUMBER OF ITERATIONS
converged after 4 iterations. Case SNR LFSSO CNPF MCNPF
11-bus [5] (Q8= 80 MVAr) * 4 * 10
Ill-Cond
Scudder case with Q8 = 80 MVAr UCTE Summer * 7 24 20
1.1
IEEE-30 buses 3 3 9 8
1 IEEE-118 buses 4 4 11 9
Peruvian System 5 5 14 12
0.9 Well-Cond
Brazilian System 6 8 15 12
Time Step Size

0.8 UCTE Winter Off-peak 7 5 19 16


0.7
UCTE Winter Peak 6 6 19 16
11-bus [10] * 8 * 10
0.6
CNPF IEEE-30 buses (r = 1.56) * 4 * 5
MCNPF IEEE-118 buses (r = 2.25) * 5 * 25
0.5
LFSSO
Infeasible Peruvian System (r = 1.30) * 11 * 5
0.4
5 10 15 20 25 Brazilian System (r= 1.05) * 10 * 9
Iteration
UCTE-Winter Off-peak (r = 1.55) * 9 * 15
Fig. 9. Step Size Evolution for 11-bus Scudder system Q8= 80 MVAr UCTE Winter Peak (r = 1.15) * 5 * 22

For all of these cases a flat start point seems as an initial Table II shows the Euclidean norms of the power
guess that lies “far” from the operating points because every mismatches for the final operating points provided by
one of them is above the 1.2 p.u. level. In order to correct LFSSO and MCNPF in the case of infeasible systems. In
this poor initial guess, the optimal multiplier (μ) inside the both of these methods as the corrected vector value gets
LFSSO and MCNPF methods calculate an initial small time closer to the feasibility boundary, the Jacobian matrix
step size for all the cases, while the heuristic rule embedded minimum eigenvalue is getting closer to zero and the step
in CNPF begins with a unitary initial step. size calculated becomes very small. Both methods provide
points on the feasibility boundary rather than simply diverge
V. ADDITIONAL SIMULATION RESULTS [13].
Simulation results for the following test systems will be TABLE II. EUCLIDIAN NORM OF THE FINAL OPERATING POINT
shown: (a) ill-conditioned 11-bus [6] and (b) well- Case LFSSO MCNPF
conditioned 30-bus and 118-bus IEEE test systems [11]. 11-bus [10] 8 ||ΔS ||0.4076 10 ||ΔS ||= 33.5512
IEEE-30 buses (ρ = 1.56) 4 ||ΔS||= 5.4136 5 ||ΔS||= 4.9386
Also, simulations for three realistic systems, namely one IEEE-118 buses (ρ = 2.25) ||ΔS||= 124.4259 ||ΔS||= 124.6711
5 25
Brazilian, one Peruvian and one approximate load-flow Infeasible Peruvian System (ρ = 1.30) 11 ||ΔS||= 227.5238 5 ||ΔS||= 669.3644
model of interconnected European (Union for Coordination Brazilian System (ρ= 1.05) 10 ||ΔS||= 106.0807 9 ||ΔS||= 111.0202
UCTE-Winter Off-peak (ρ = 1.55) 9 ||ΔS||= 12.3891 ||ΔS||= 11.8508
of Transmission of Electricity - UCTE) have also been 15
UCTE Winter Peak (ρ = 1.15) 5 ||ΔS||= 248.0603 22 ||ΔS||= 159.2410
carried out and the results are shown in this section.
The Brazilian system data corresponds to an operating
Fig. 10 illustrates the Euclidian norms obtained with both
point of the year 2011 [15]. The system consists of 4526
LFSSO and MCNPF for some of the infeasible cases listed
buses, 6312 branches and 500 generators. The total base
in Table II.
case load is 68.5 GW and 20.5 GVAr.
The Peruvian system data corresponds to an operating 800
point of the year 2010 [16]. The system consists of 512 LFSSO
700
buses, 611 branches and 60 generators, and the total base MCNPF
600
case load is 3.8 GW and 1.1 GVAr.
The UCTE system contains 1254 buses, 1944 branches 500

and 378 generators. Three cases are available, namely the 400
Summer peak (ill-conditioned), Winter peak, and Winter 300
off-peak 0.
200
Also, infeasible situations based on the well-conditioned
100
systems were created by scaling system demand beyond the
system’s MLP and using another 11-bus systems described 0
IEEE-118 buses Peruvian System Brazilian System UCTE Winter
in as ill-conditioned [10] but reported in [4] as infeasible. Peak
A solution tolerance of 0.1 MW/MVAr (0.001 p.u.), flat Fig. 10. Final power mismatches Euclidian norms.
start initial point, and generator reactive power limit
enforcement were used in all simulations. As the trajectory of solution of nonlinear systems
The simulation results presented in Table I illustrate the solution and initial value problems depends on the initial
performance of the tested methods. point supplied (flat start or other) and the method itself, it
cannot be assured which one of those approaches will
Ill-conditioned and infeasible cases resulted in SNR
supply a closer state to the scheduled state. There are some
method divergence. CNPF method also diverged in some of
approaches to find system MLP considering reactive power
the ill-conditioned cases and cannot handle infeasible cases
generation limits and geometric characteristics of feasibility
unless the heuristic rules are modified. LFSSO method
boundary [14].
converged for all cases. The MCNPF exhibits the same
PowerTech 2011, 19 – 23 June 2011, Trondheim, Norway 6
Paper ID: 546
VI. CONCLUSION BIOGRAPHIES
The analysis and results obtained for the robust load flow Jorge F. Gutiérrez (M'01) was born in Cali, Colombia. He received the
B.S. and the M.S. degrees in Electrical Engineering from Universidad
methods showed that the LFSSO presents a better general Industrial de Santander in 1994 and 2000, respectively and has been with
performance as compared to the basic CNPF in all cases. Universidad Nacional de Colombia - Manizales since 2001 where he is an
The CNPF considered here uses RK4 integration technique Assistant Professor. Currently he is working towards his Ph.D. degree in
Electrical Engineering at UNICAMP. His research areas are: power
that is based in an equivalent slope evaluated from four load
systems operation, power flow techniques and transmission lines.
flow solutions with first order approximation. On the other
hand, LFSSO is based in second order Taylor expansion, Manfred F. Bedriñana (S'00, M'09) was born in Lima, Peru. He received
resulting in a more efficient search which results in a better the B.S. degree in Electrical Engineering (first-class honors) from National
performance over the other ones when used in realistic University of Engineering, Peru in 2000. He received the M.S. degree and
Ph.D. degree in Electrical Engineering from Federal University of
applications, such as in the real time operation of power Maranhão and UNICAMP, Brazil, in 2003 and 2009, respectively. His
systems. Also CNPF was not able to handle ill-conditioned research areas are security of electrical energy systems and electricity
and infeasible situation in an appropriate way. A modified markets.
version of CNPF, referred to as MCNPF was tested.
Carlos A. Castro (S'90, M'94, SM'00) received the B.S. and M.S. degrees
MCNPF used the optimal multiplier of LFSSO for defining from UNICAMP, in 1982 and 1985 respectively, and the Ph.D. degree
the time step, and this resulted in better performance than from Arizona State University, Tempe, AZ, in 1993. He has been with
CNPF as far as the handling of ill-conditioned and UNICAMP since 1983, where he is currently an Associate Professor.
infeasible case. However, the efficiency is still an issue
since in general MCNPF converges after a larger number of
iterations as compared to LFSSO (see Table 1).

REFERENCES
[1] B. Stott,“On-line Network Analysis Applications – A Mini-Tour Of
Core Technologies” – CEPEL Roadmap workshop, Florianopolis SC
Brasil, May 20, 2006. Available:
http://roadmap.cepel.br/present/Stott_Online Applications.pdf.
[2] P. R. Bijwe and S. M. Kelapure, ”Nondivergent Fast Power Flow
Methods”, IEEE Trans. Power Syst., vol. 18, no. 1, pp. 633 – 638,
May, 2003.
[3] F. Milano, “Continuous Newton’s Method for Power Flow Analysis”,
IEEE Trans. Power Systems Vol. 24. Feb 2009.
[4] Y. Wang, L. C. P. da Silva, W. Xu and Y. Zhang, “Analysis of ill-
conditioned power-flow problems using voltage stability
methodology, IEE Proceedings, vol 148 Issue 5, pp 384 - 390, Sep
2001.
[5] S. Iwamoto and Y. Tamura, “A load flow calculation method for ill-
conditioned power systems”, IEEE Trans. Power App & Syst., vol.
PAS-100, pp. 1736 – 1743, Apr. 1981.
[6] C. A. Castro and L. M. C. Braz, “A new approach to the polar
Newton power flow using step optimization”, in Proc. 29th North
Amer. Power Symp., Laramie, WY, USA, 1997.
[7] L. M. Braz, C. A. Castro and C. A. F. Murari, “A critical evaluation
of the size optimization based load flow methods”, IEEE Trans.
Power Syst., vol. 15, no. 1, pp. 202 – 207, Feb. 2000.
[8] J. E. Tate, T. J. Overbye, “A comparison of the optimal multiplier in
polar and rectangular coordinates”, IEEE Trans. Power Syst., vol. 20,
no. 4, pp. 1667 – 1674, Nov. 2005.
[9] S. Hetzler, “A Continuous Version of Newton’s Method”, College
Mathematics Journal, vol 28, no. 5, pp 348-351, Nov. 1997.
[10] S. C. Tripathy, G. Durga, O. P. Malik and G.S. Hope, “Load Flow
Solutions For Ill-Conditioned Power Systems By a Newton-Like
Method”, IEEE Trans. Power App & Syst., vol. 101, no. 10, pp.
3648–3657, Oct. 1982.
[11] Power Systems Test Case Archive, University of Washington College
of Engineering. [Online]. Available:
http://www.ee.washington.edu/research/pstca/.
[12] Power Systems Test Case Archive, University of Washington College
of Engineering. [Online]. Available:
http://www.ee.washington.edu/research/pstca/.
[13] T. J. Overbye, “A power flow measure for unsolvable cases”, IEEE
Trans. Power Syst., vol. 9, no. 3, pp. 1359–1365, Aug. 1994.
[14] M. F. Bedriñana, Carlos A. Castro. “Maximum Loading Point
Computation based on Load Flow with Step Size Optimization”. In
Proceeding: 2008 IEEE Power & Energy Society General Meeting,
Pittsburgh, PA, USA, 20–24 Jul. 2008.
[15] Brazilian National System Operator (ONS). [Online]. Available:
http://www.ons.org.br/.
[16] Peruvian National System Operator (COES). [Online]. Available:
http://www.coes.org.pe/.
Approximate Model Of European Interconnected System. [Online].
Available: http://www.see.ed.ac.uk/~jbialek/Europe_load_flow/.

You might also like