Professional Documents
Culture Documents
net/publication/309212124
CITATION READS
1 1,117
1 author:
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Olav Bjarte Fosso on 18 October 2016.
Prepared by
Summary:
The document describes some applications of the Continuation Power Flow (CPF), the
equation systems and implementations aspects as well as computational experience. The
foundation of this work was done during a research stay at the department Analytical Methods
& Specialized Studies, Power System Planning Division, Ontario Hydro, 1992/1993
sponsored by SINTEF Energy Research (EFI) and The Norwegian National Committee of
CIGRE.
This document is used for educational purposes at the Department of Electric Power
Engineering at NTNU.
1
TABLE OF CONTENT:
1 INTRODUCTION ....................................................................................... 3
2 BASIC PRINCIPLE ...................................................................................... 4
2.1 MATHEMATICAL FORMULATION ........................................................................................................................ 5
3 INITIATION............................................................................................... 6
3.1 INCREASE OF REACTIVE LOAD ............................................................................................................................ 6
3.2 INCREASE OF ACTIVE LOAD ............................................................................................................................... 6
3.3 INCREASE OF ACTIVE AND REACTIVE LOAD ........................................................................................................... 7
3.4 EQUATION SYSTEMS ........................................................................................................................................ 7
Some of the methods used to calculate margins to voltage instability are based on optimization [4],
load flows [2, 5] or sensitivities [6]. The optimization methods identify the point of bifurcation
directly by maximizing the reactive load that can be supplied. The load flow methods apply
approaches as to increase the load in steps or the schedule the voltage on a condenser and monitor
the corresponding reactive power. The first of these approaches may have a problem with
convergence close to the maximum load point while the latter does not have this problem. The initial
implementation based on sensitivities estimated the next point based on sensitivities of the state
variables (voltage and angles) and from these the dependent variables (active and reactive injections)
were calculated. On the new operating point, load flows iterations could be conducted if higher
accuracy could be achieved. Without the load flow iterations, a too optimistic solution could be
achieved if the steps were too large. In the paper [7] the sensitivity method was generalized with a
general increase in combinations of active and reactive load on a number of nodes without any lack
of efficiency. This was done though an initialization of the right-hand side of the equations which
gave the sensitivities of the voltage and angles as a linear combination of the different rows of the
inverse Jacobian matrix. The computations only involved a forward and backward substitution of the
Jacobian matrix. The inverse Jacobian matrix was never explicitly calculated and the sensitivity
matrices never evaluated as all formations were available from the linearized load flow equations.
However, though the method was robust and quite efficiency, it had the deficiency that the margin
could be over-estimated if load flows were not done to get back to the real curve describing the
voltage as a function of increasing loads.
This document will describe the principle of the continuation power flow, computational steps, some
implementation details and some application areas. The mathematical formulation and solution
strategy have many similarities with the approach suggested in [10]. However, the loading strategies
and applications are different as well as implementation.
The computation sequence is built around a step-wise increase in load from a basic operating point
until the maximum system load is found.
The computational steps are the same as the ones used in the sensitivity method, implemented
by [7], when this method is combined with load flow iterations to account for discrete events and
non-linearities.
However, a modification is done in the formulation of the load flow equations to avoid the ill-
conditioning close to collapse. This modification permits the load flow to be solved for voltages also
below the maximum load.
The blue line indicates the exact voltage on the given bus for an increase in the load. The load
increase can typically represent an increase on a few buses or in an area and may include a
combination of active and reactive loads. The first point is found based on a linear prediction
due to the chosen load pattern and step size. In the next step, load flow iterations are
performed while assuming the load to be fixed. This corresponds to the vertical move down to
the exact curve.
This problem is solved by keeping the voltage fixed and next find the load corresponding to the
specified voltage. Since the voltage is kept fixed at the monitored bus, this corresponds to a
horizontal movement in the figure. At other buses the movement would be a combination of
changes in voltage and load.
To switch properly between fixed load and fixed voltage is the main challenge of this method. The
implemented approach and experiences will be described in a separate section.
It can be difficult or at least time consuming to find the exact maximum load. The approach is to
decrease the step size in order to come closer to the maximum load. Dependent of the required
accuracy, the number of steps can be large.
The change in formulation compared to the standard load flow is that the increase in load is added
as a variable in the equations.
By separating the load flow equations into a real and an imaginary part, the equations are given by:
Pinj (V ) g p ( ,V ) S 0 (1)
Qinj (V ) g q ( ,V ) S 0 (2)
Pinj(V) Net active injection based on base case load and generation
Qinj(V) Net reactive injection based on base case load and generation
gp(δ,V) Balance equations for active power flow
gq(δ,V) Balance equations for reactive power flow
α Vector giving the distribution of increase in reactive load
δ Vector giving the distribution of increase in active load and generation
The α and β give the distributions of the increase in load. The distributions are initialized as
described in the following sections.
If only the reactive load is increased, the requirement to the initiation is:
k 1
k 1.0 (3)
This way of loading the system may be a measure of system robustness but it may be vulnerable to
the user’s experience and knowledge of the system. An improper distribution can either give an
unrealistically small or large margin. The existing area distribution may be used as a key for loading
and by this avoiding too extreme distributions either way.
i gi li (4)
where:
βi Decrease in injection at bus ’i’
βgi Increase in generation at bus ’i’
βli Increase in load at bus ’i’
i 1
gi 1.0 a (5)
i 1
li 1.0 (6)
i 1
i a (7)
It may be possible to change the loading distribution during the simulation. If you assume that the
simulation was not primarily for calculating margins to voltage collapse but to simulate a dynamic
loading sequence over a period, the priority of loading may be defined by other schedules.
In this case, it was found convenient to use the active load as a parameter. This means that the
requirements to initiation of equation (13) as given in the previous section are the same. By
assuming a fixed ratio between active and reactive power on each load bus, all the a-coefficients
are determined by:
αi = tan ϕi (8)
where:
ϕi Phase angle of the increase in load at bus 'i'
There is no requirement to the sum of the αi when the complex load is increased in the system.
An option is of course to define the profiles of active and reactive load independently. In the case
where you want to simulate the dynamic load sequence between for example system areas, this
would be the most flexible method.
The load parameter is an additional variable in the calculations. The standard Jacobian matrix will
be extended by one column. This is indicated in equation (9):
P P
V V b
Q Q
S
V
(9)
This equation system is singular since the number of columns is larger than the number of rows. The
equation system is therefore extended by one equation to force the solution to be non-trivial.
The right hand side will take different forms in the predictor and the corrector phases of the
solution.
The equation system (1) is used but extended with one equation. The right hand side is set to zero.
The additional equation forces one of the variables to be different from zero. Which one of the
- variables depends on the selection of continuation parameter. In the initial steps, the load will
normally be the continuation parameter.
The equation system used for the sensitivity calculation when the load is the continuation
parameter is given in equation (10).
P P
V
0
Q Q
V 0
V
S 1
0 0 1
(10)
After a forward and a backward substitution, the solution vector will contain the sensitivities in
voltage angles and magnitudes for a 1.0 pu increase in load.
Note: These sensitivities are exactly the same as those calculated in the sensitivity method [7]. The
difference is only that the sensitivity method uses α and β on the right hand side to enforce a
proper linear combination of the columns of the inverse Jacobian matrix (J-1).
The sensitivity vector is now monitored to identify if a 1.0 pu increase in load will cause any
voltage to change more than a maximum permitted change.
Δx = Δxo + σΔ (12)
More sophisticated schemes for choosing step-length could be used. The advantage of this
method is that you can monitor different quantities in the system, choose step-length according to
for example transformer-step changing, phase-shifting, maximum transmission line flow, defined
transfer corridor limits or other quantities, and then adapt the loading strategies according the
system conditions. Of course all of this has to be done based on steady-state equations and will
not give the complete picture og the system dynamics performance but still valuable from the
quasi-static description.
The previous description is based on that the load is the continuation parameter. As the maximum
load is approached, the continuation parameter will normally change and a voltage will typically be
the parameter with the largest rate of change. The change of continuation parameter is done by
setting the voltage with the largest rate of change to 1.0 in the equation (9).
P P
V
0
Q Q
V 0
V
S 1
0 1i 0
(13)
σ =ΔVmax
(14)
After an approximate solution is found in the predictor phase, corrections are found by doing load
flow iterations.
Again the equation system (9) is used. The right hand side will in this case be the mismatch of the
load flow equations.
The equation system when the load is the continuation parameter, is now given by:
P P
V
P
Q Q
V Q
(15)
V
S 0
0 0 1
P P
V
P
Q Q
V Q
(16)
V
S 0
0 1i 0
The additional equation will in the corrector phase ensure that the selected continuation parameter
is kept equal to its estimated value. If the continuation parameter is the increase in load, the
correction will be vertical according to Figure I. If the voltage is the continuation parameter, the
move in Figure 1 will be horizontal, i.e. to find the load flow corresponding to the specified voltage
on a bus.
The selection of continuation parameter is based on the size of the elements in the sensitivity
vector. If the rate of change of voltage becomes the limiting factor for the step-length, the voltage
at the node with the largest change is chosen as continuation parameter.
Using the load as a continuation parameter in the region of maximum load can cause problems. If
the estimate exceeds the maximum capability, the algorithm will diverge.
If the voltage is used as continuation parameter, the algorithm can fail if too large steps in the
permitted voltage change are used. In this case, it is possible to try to keep the voltage fixed on a
level where no load flow solution exists. On the lower part of the curve, the voltages sometimes
start to increase when the load is decreasing found in the prediction phase.
The variable set to 1.0 in the predictor phase only provides a scaling of the sensitivity vector. For
the base case, the sensitivities of the voltages will normally be small compared to the sensitivity in
the global load increase. On the maximum load level, the sensitivity in the load will be zero. Using
the load as a scaling parameter will cause at least one of the voltage sensitivities to be -∞. It is
therefor necessary to change this scaling parameter before the maximum load is reached.
A possibility is to always select the continuation parameter with the largest rate of change. The
experience with this approach is that the switching to the voltage as a continuation parameter
occurs too late. Particularly can this be a problem when the load is increased on many buses. In
this case, it is likely that a non-converged load flow occurs and a decrease in step size is necessary
to find a solution.
The experience is therefore that the voltage has to take over as a continuation
parameter significantly before the sensitivity becomes the largest one.
These two criteria are combined in the selection procedure where the most constraining criterion
is selected.
A model of the Norwegian system. This model has 511 buses and 806 transformer
and transmission lines.
A model of Ontario Hydro's system with 560 buses and 1120 transformer and
transmission lines.
Figure 2 indicates the voltage on two buses in the Norwegian system for an increase in active
power transfer. The load is increased on Bus 405 and on Bus 455. These two buses constitute an
interface to the Swedish system. The increased active load is distributed among the generating
units in the Norwegian system according to the unit reserve. Alternatives could be to use a
distribution based on:
A maximum of 100 MW and maximum change in voltage of 0.005 are used. Many load flows are
therefore required to establish the complete curves.
The load is used as the continuation parameter up to 800 MW with full step size. The voltage now
becomes a limiting factor and the step size is decreased.
The sensitivity of the load will become negative after the critical point.
The problem of using too large steps of the voltage is illustrated by the curve for Bus 405. Just after
the maximum load, the voltage change gets a decreasing rate and becomes positive. Using this
voltage as continuation parameter can cause bad convergence properties or even fail to converge.
After the maximum load in the system, a simulation achieved by find the corresponding load on the
lower part of the equations will unload the system and therefor the voltages on most of the nodes
would increase again. This is physical solutions but more an illustration of the algorithms capability to
find the shape of the curves in the vicinity of the maximum load point.
Figure 3 gives an example when the reactive load is increased on a single node.
The step size can as mentioned be increased by using a scheme where the previous solution is
saved. This approach makes it possible to take a step backward and use reduced step size if the
algorithm fails to converge. This is important close to the maximum loading point as the equation
system will become ill-conditioned.
The implementation described here is relatively simple and quite many extensions are possible to
increase robustness and flexibility. However, as a tool the Continuation Power Flow (CPF) has
many useful applications areas.
9 REFERENCES
[1] Voltage Stability of Power Systems: Concepts, Analytical Tools, and Industrial Experience,
IEEE Catalog Number 90TH0358-2.
[2] Working Group 38.01 task force 03: “Planning against voltage collapse”, Electra no 11 March
1987, p 55-75
[3] P-A Løf, T.Smed, G.Anderson, D.J. Hill; “Fast Calculation of Voltage Stability Index”, 91 WM
2013-0 PWRS
[4] T. van Cutsem; “A method to Compute Reactive Margins with respect to Voltage Collapse”,
IEEE Transactions of Power Systems, Vol 6, No. 1, February 1991.
[5] O.O. Obadina, G.J. Berg: “Determination of Voltage Stability Limits in Multi-machine
Systems”, IEEE/PES 1988 Winther Meeting, New York, Paper no. 88WM 194-3.
[6] N. Flatabø, R. Ognedal, T. Carlsen; «Voltage Stability Conditions in a Power System Calculated
By Sensitivity Methods”, IEEE Transactions in Power Systems, Vol. 5, No. 4, November 1990.
[7] N. Flatabø, O.B. Fosso, R. Ognedal, T. Carlsen, K.R. Heggland: “A method for Calculation of
Voltage Instability Applied on the Norwegian System for Maintaining Required Security
Level”, IEEE Transactions on Power Systems, Vol. 8, Issue 3, pp 920-928, 1993.
[8] Jef Van Hecke et al: “Indices Predicting Voltage Collapse including Dynamic Phenomena”,
CIGRE Task Force 38.02.11, June 1994.
[9] Carson W. Taylor et al: “Modelling of Voltage Collapse including Dynamic Phenomena”,
CIGRE TF 38.02.10, 1993.
[10] V. Ajjarapu, C. Christy, "The Continuation Power Flow: A Tool for Steady State Voltage
Stability Analysis", Transactions on Power Systems, Vol. 7, No. 1, February 1992.